Users Guide

Provalis Research

Copyright 1989-2014 Provalis Research. All right reserved.

 DISCLAIMER

This software and the disk on which it is contained are licensed to you, for your own use. This is copyrighted software
owned by Provalis Research. By purchasing this software, you are not obtaining title to the software or any copyright
rights. You may not sublicense, rent, lease, convey, modify, translate, convert to another programming language,
decompile, or disassemble the software for any purpose. You may make as many copies of this software as you need for
backup purposes. You may use this software up to two computers, provided there is no chance it will be used
simultaneously on more than one computer. If you need to use the software on more than one computer simultaneously,
please contact us for information about site licenses.

WARRANTY

The WORDSTAT product is licensed "as is" without any warranty of merchantability or fitness for a particular purpose,
performance, or otherwise. All warranties are expressly disclaimed. By using the WORDSTAT product, you agree that
neither Provalis Research nor anyone else who has been involved in the creation, production, or delivery of this
software shall be liable to you or any third party for any use of (or inability to use) or performance of this product or for
any indirect, consequential, or incidental damages whatsoever, whether based on contract, tort, or otherwise even if we
are notified of such possibility in advance. (Some states do not allow the exclusion or limitation of incidental or
consequential damages, so the foregoing limitation may not apply to you). In no event shall Provalis Research's liability
for any damages ever exceed the price paid for the license to use the software, regardless of the form of claim. This
agreement shall be governed by the laws of the province of Quebec (Canada) and shall inure to the benefit of Provalis
Research and any successors, administrators, heirs, and assigns. Any action or proceeding brought by either party
against the other arising out of or related to this agreement shall be brought only in a PROVINCIAL or FEDERAL
COURT of competent jurisdiction located in Montral, Qubec. The parties hereby consent to in personam jurisdiction
of said courts.

COPYRIGHT

Copyright 1998-2015 Provalis Research. All rights reserved. No part of this publication may be reproduced or
distributed without the prior written permission of Provalis Research, 1255 University Avenue, Suite #1255, Montreal,
QC, CANADA, H3B 3W9.
 TABLE OF CONTENT
Introduction to WordStat .......................................................................................................................... 5
Programs Capabilities ............................................................................................................................. 7
The Content Analysis & Categorization Process ..................................................................................... 11
A Quick Tour: Performing Your First Content Analysis .......................................................................... 14

Preparing and Importing Data................................................................................................................. 19

Preliminary Text Preparation ........................................................................................................... 19
Importing Spreadsheet Files ............................................................................................................. 20
Importing Database Files ................................................................................................................. 21
Importing Text and Word Processor Files ........................................................................................ 21
The Working Environment ..................................................................................................................... 24
First Screen - Dictionaries ................................................................................................................ 26
Second Screen - Options................................................................................................................... 33
Third Screen - Frequency ................................................................................................................. 37
Using the Dictionary panel ....................................................................................................... 41
Working with the auto Suggest Panel........................................................................................ 42
Fourth Screen - Extraction ............................................................................................................... 44
Extracting Topics ..................................................................................................................... 45
Extracting Phrases .................................................................................................................... 49
Extracting Named Entities ........................................................................................................ 56
Extracting Misspellings & Unknown Words ............................................................................. 58
Fifth Screen - Co-occurrences.......................................................................................................... 65
Dendrograms............................................................................................................................. 65
2-D & 3-D Mapping .................................................................................................................. 71
Link Analysis ............................................................................................................................ 74
Proximity Plots.......................................................................................................................... 80
Statistics page ........................................................................................................................... 83
Sixth Screen - Crosstab ................................................................................................................... 85
Seventh Screen - Keyword-in-Context ............................................................................................. 93
Eighth Screen - Classification ........................................................................................................... 97
Select Features Page ................................................................................................................ 100
 Learn & Test Page ................................................................................................................... 103
History & Experiment Page...................................................................................................... 106
Classification Experiment Dialog Box ...................................................................................... 108
Apply Page .............................................................................................................................. 110
Exporting a Classification Model to Disk ................................................................................. 113

Common Tasks
Creating and Maintaining Dictionaries ........................................................................................... 114
Working with Rules ....................................................................................................................... 124
Using Lexical Tools for Dictionary-Building................................................................................... 127
Monitoring and Customizing Substitutions ...................................................................................... 135
Configuring External Preprocessing Routines.................................................................................. 139
Viewing and Editing Text ............................................................................................................... 142
Displaying keyword distribution using barcharts or pie charts.......................................................... 145
Creating and Using Norm Files ....................................................................................................... 148
Performing Text Retrieval Using Keywords .................................................................................... 150
Creating Bubble Charts ................................................................................................................... 155
Using Heatmap Plots ...................................................................................................................... 158
Performing Correspondence Analysis.............................................................................................. 162
Editing the Case Descriptor............................................................................................................. 166
Filtering Cases ................................................................................................................................ 167
Expression Operators and Rules ............................................................................................... 170
Supported xBase Functions ...................................................................................................... 171
Performing Analysis on Manually Entered Codes ............................................................................ 177
Computing Inter-rater Agreement Statistics ..................................................................................... 178
Exporting Frequency Data ............................................................................................................. 181
Exporting Categorization Models .................................................................................................... 183
Using the WordStat Document Classifier ........................................................................................ 184
WordStat Software Developers Kit (SDK) ..................................................................................... 188
Performing Multivariate Analysis on Words or Categories .............................................................. 189
Managing Outputs with the Report Manager ................................................................................... 191
References............................................................................................................................................. 198
Technical Support ................................................................................................................................ 199
 Introduction to WordStat
WordStat is a text analysis module specifically designed to study textual information such as responses to
open-ended questions, interviews, titles, journal articles, public speeches, electronic communications, etc.
WordStat may be used for automatic categorization of text using a dictionary approach or various text
mining methods. WordStat can apply existing categorization dictionaries to a new text corpus. It also may
be used in the development and validation of new categorization dictionaries. When used in conjunction
with manual coding, this module can provide assistance for a more systematic application of coding rules,
help uncover differences in word usage between subgroups of individuals and assist in the revision of
existing coding using KWIC (Keyword-In-Context) tables.
WordStat includes numerous exploratory data analysis and graphical tools that may be used to explore the
relationship between the content of documents and information stored in categorical or numeric variables
such as the gender or the age of the respondent, year of publication, etc. Relationships among words or
categories as well as document similarity may be identified using hierarchical clustering and
multidimensional scaling analysis. Correspondence analysis and heatmap plots may be used to explore
relationship between keywords and different groups of individuals.
WordStat is a module that must be run from either of the following base products:
SimStat -This statistical software provides a wide range of statistical procedures for the analysis of
quantitative data. It offers advanced data file management tools such as the ability to merge data files,
aggregate cases, perform complex computation of new variables and transformation of existing ones.
When used with SimStat, WordStat can analyze textual information stored in any alphanumeric, plain
text and rich text memo variable (or field). It includes various tools to explore the relationship
between any numeric variable of a data file and the content of alphanumeric ones. Its close integration
with SimStat facilitates further quantitative analysis on numerical results obtained from the content
analysis (ex.: factor analysis or correspondence analysis on keyword frequencies, multiple regression,
etc.).
QDA Miner - The text management and qualitative analysis program allows one to create and edit
data files, import documents, and perform manual coding of those documents. Several analysis tools
are also available to look at the frequency of manually assigned codes and the relationship between
those codes and other categorical or numeric variables. When used with QDA Miner, WordStat can
perform content analysis on whole documents or selected segments of those documents tagged with
specific user defined codes.
WordStat module may be accessed in both of these programs from the CONTENT ANALYSIS command
in the ANALYSIS menu.
A few additional utility programs are also included with WordStat that may be run as standalone
applications or be accessed directly through WordStat:
Report Manager - This application has been designed to store, edit and organize documents, notes,
quotes, tables of results, graphics and images created by QDA Miner and WordStat or imported from
other applications.
Document Conversion Wizard - This utility program provides an easy way to import numerous
documents and create a project file. It can also be used to split large files into smaller units and to
extract various numeric and alphanumeric data from structured documents.

WordStat Users Manual 5

Dictionary Builder - This tools allows the development of comprehensive categorization dictionary for
automatic content analysis. The program may be run as standalone application but also from dictionary
page of WordStat by pressing the SUGGEST button. To obtain more information on this software see page
127.
Document Classifier - This utility program is a stand-alone application that may be used to perform
content analysis and automatic text classification on a text pasted from the clipboard or stored in a file. For
more information on this utility program, see WordStat Document Classifier on page 184.
Chart Editor - The chart editor is a standalone application that may be used to further customize charts
created using WordStat.

6 WordStat Users Manual

 Program's Capabilities
TEXT PROCESSING CAPABILITIES
Performs analyses on Rich Text documents stored QDA Miner projects or SimStat data files.
Perform analyses on alphanumeric variables containing short textual information such as responses to
open ended questions, titles, descriptions, etc.
Stemming in 18 human languages (Basque, Catalan, Czech, Danish, Dutch, English, Finnish, French,
German, Hungarian, Italian, Norwegian, Polish, Portuguese, Romanian, Russian, Spanish, Swedish,
and Turkish).
Automatic lemmatization (available in English, French, Spanish, Swedish, German, Norwegian, and
Italian, contact us if you need support of other languages).
Substitution process for customized lemmatization of words or automatic spell correction of common
misspellings.
Optional exclusion of pronouns, conjunctions, expressions, etc, by the use of existing or user defined
exclusion lists.
Categorization of words or expression using existing or user defined dictionaries.
Word categorization based on Boolean (AND, OR, NOT ) and proximity rules (NEAR, AFTER, BEFORE)
Word or expression substitution and scoring using wildcards and weighting.
Frequency analysis of words, derived content categories or concepts.
Interactive development and validation of multi-level dictionaries or categorization schema.
Ability to restrict an analysis to specific portions of a text or to exclude comments and annotations.
Option to perform a content analysis on a random sample of cases.
Integrated spell-checking with support for 20 languages.
Integrated thesaurus (English, French, Spanish, German, Portuguese, Italian) to assist the creation of
categorization schema.
Case filtering on any numeric or alphanumeric variable and on keyword occurrence (with AND, OR,
and NOT Boolean operators).
Prints presentation quality tables.
Export any output table to SPSS, Stata, Excel, HTML, ASCII, Tab separated or comma separated
value files.
All graphics may be saved to disk in BMP, JPEG or PNG file format.

FEATURE EXTRACTION TOOLS

Automatic topic extraction using factor analysis with comparative analysis.
Easy identification of technical terms, proper nouns and common misspellings.
Phrase finder allows identification of the most recurring phrases.
Pattern-based named-entity extraction tool.

WordStat Users Manual 7

UNIVARIATE KEYWORD FREQUENCY ANALYSIS
Univariate keyword frequency analysis (keyword count and case occurrence).
Keyword co-occurrence matrix (within documents, paragraphs, sentences)
Integrated clustering and dendrogram display of keyword similarities

KEYWORD RETRIEVAL FUNCTION

A powerful keyword retrieval function allows identification of text units (documents, paragraphs or
sentences) containing one keyword or a combination of keywords with optional filtering of cases.
Ability to attach QDA Miner codes to retrieved segments.
Retrieved segments may be exported to disk in tabular format (Excel or delimited text files) or as text
reports (Rich Text Format).

MULTIPLE RESPONSES AND COMPARISONS

Can perform a single frequency analysis on information stored in several alphanumeric variables
(memo or string variables).
Comparison of keyword occurrence between different variables.
Compute inter-rater agreement measures on codes manually entered in different variables (pct. of
agreement, Cohen's Kappa, Scott's Pi, Krippendorf's R and r-bar, free marginal, and intraclass
correlation).

KEYWORD CO-OCCURRENCE AND ANALYSIS

Integrated clustering and dendrogram display of keyword co-occurrences.
Proximity plot to easily identify all keywords that co-occurs with one or several target keywords.
2-D and 3-D multidimensional scaling on co-occurrences of words or content categories.
Link analysis using force-based graphs, multidimensional scaling or circular graph displays.
Flexible keyword co-occurrences criteria (within a case, a sentence, a paragraph, a window of n
words, a user defined segment) as well as clustering methods (first- and second-order proximity,
choice of similarity measures).
Easy text retrieval directly from dendrogram or proximity plots.

ANALYSIS OF CASE OR DOCUMENT SIMILARITY

Hierarchical clustering, multidimensional scaling and proximity plot may be used to explore the
similarity between documents or cases.

8 WordStat Users Manual

RELATIONSHIP TO NUMERICAL, CATEGORICAL DATA, AND DATES
Comparison between any text variable and any nominal or ordinal variable (such as sex of the
respondent, specific subgroups, years of publication, etc.).
Automatic recoding of date variables into days, weekdays, weeks, months, quarters, years or decades.
Choice between 12 different association measures to assess the relationship between keyword
occurrence and nominal or ordinal variables (Chi-square, Likelihood ratio, Student's F, Tau-a, Tau-b,
Tau-c, symmetric Somers' D, asymmetric Somers' Dxy and Dyx, Gamma, Pearson's R, and
Spearman's Rho)
Correspondence analysis allows examination of relationships between words or categories and other
nominal or ordinal variables.
Ability to sort keyword matrix in alphabetical order, by keyword frequency or case occurrence, on
the obtained statistics or on its probability.

KEYWORD-IN-CONTEXT
Ability to display a Keyword-In-Context (KWIC) table of any included, leftover or user defined
word, word pattern or phrase.
KWIC tables may be sorted in ascending order of case number, words with context, or on values of
independent variables.
Ability to jump from a specific occurrence in the KWIC table to the original text variable in order to
view or edit the selected word.
KWIC tables may be saved in data files for further processing.
Customizable KWIC and report function to display all hits as lists of paragraphs, sentences or user
defined segments.

AUTOMATED TEXT CLASSIFICATION

Machine learning algorithms (Naive Bayes and K-Nearest Neighbors) for document classification.
Flexible feature selection for automatic selection of best subsets of attributes.
Numerous validation methods (leave-but-one, n-fold crossvalidation, split sample).
Experimentation module allows easy comparison of predictive models and fine-tuning of
classification models.
Classification models may be saved to disk and applied later using either a standalone document
classification utility program, a command line program or a programming library. Note: The
command line and the programming library are part of WordStat Software Developer's kit (SDK)
which is sold separately.

WordStat Users Manual 9

NORM CREATION AND COMPARISON
Ability to create norm files based on frequency analysis of words or content categories.
Comparison of obtained frequencies to previously saved norm files.

FULL INTEGRATION WITH A SIMTAT & QDA MINER

Document and alphanumeric variables are stored in the same file as all other numeric variables.
The same data file format is used by SimStat, QDA Miner and WordStat.
Variable selection and analysis are performed within SimStat using a simple 3-step operation:
1. Open the existing data file.
2. Select one or several alphanumeric variables as dependent variables and, optionally, other
nominal or ordinal variables to be treated as independent.
3. Execute the CONTENT ANALYSIS command from the STATISTICS drop-down menu.
New variables representing frequency or occurrence of words, keywords or concepts can be added to
the existing data file or exported to a new data file in order to be submitted to more advanced
analysis (such as cluster analysis, correspondence analysis, multiple regression, etc.).
Data can be imported from and exported to different file format including dBase, Paradox, Excel, MS
Access, SPSS, Stata, comma or tab separated text files, XML, HTML, etc.
Ability to perform numeric and alphanumeric transformation or to apply filters on cases of the data
file to restrict the analysis to specific subgroups.

10 WordStat Users Manual

 The Content Analysis & Categorization Process
The most basic form of content analysis WordStat can perform is a simple frequency analysis of all words
contained in one or several text variables of a data file. However, WordStat offers several features that
permit the user to accomplish more advanced forms of content analysis that may involve automated
categorization, different weighting of words and phrases, inclusion or exclusion of words based on
frequency criteria, etc. To fully understand the possibilities offered by the program, one first needs to
understand the various underlying processes involved in a typical WordStat frequency analysis and how
these processes may be combined to achieve various kinds of content analysis tasks.

WordStat's categorization involves up to seven consecutive processes:

1- TEXT PREPROCESSING (including stemming, n-grams, etc.)

The preprocessing option allows users to access external text preprocessing routines that are not
part of the WordStat program. This option is useful to perform custom transformation on the text
to be analyzed. WordStat includes a few sample text processing routines, such as a Porter stemmer
which remove common English suffixes and prefixes as well as a character n-grams routine which
decomposes every word into sequences of 3, 4 or 5 characters. Please note that the Porter stemmer
routine available in the preprocessing process is for demonstration purpose only and will greatly
reduce the processing speed of Wordstat. We recommend using instead the integrated stemming
routine (see #2 below).
2- STEMMING
The stemming process is a natural language processing routine that reduces inflected words to a
common stem or root form. The English stemmer, for example, returns "write" for "write",
"writes", "writing", and "writings". Stemming routines are available for these languages: Basque,
Catalan, Czech, Danish, Dutch, English (Porter & snowball methods), Finnish, French, German,
German2 (different Umlaut handling), Hungarian, Italian, Norwegian, Polish, Portuguese,
Romanian, Russian, Spanish, Swedish, and Turkish.

3- SUBSTITUTION PROCESS (including lemmatization and automatic spell correction)

The substitution process takes individual words and replace them with another word form or with
a sequence of words. Such a process is typically used for lemmatization, a procedure by which all
plurals are transformed into singular forms and past-tense verbs are replaced with present-tense
versions. It may also be used for derivational stemming in which different nouns, verbs, adjectives
and adverbs derived from the same root word are transformed into this single word. Custom
substitution process may also be created to perform automatic spelling corrections of common
misspellings.

4- EXCLUSION PROCESS

An exclusion process may be applied to remove words that you do not want to be included in the
content analysis. This process requires the specification of an exclusion list. Such a process is used
mainly to remove words with little semantic value such as pronouns, conjunctions, etc., but may
also be used to remove some words or phrases used too frequently or with little discriminative
value.

WordStat Users Manual 11

 5- CATEGORIZATION PROCESS

The categorization process allows one to change specific words, word patterns or phrases to
other words, keywords or content categories and/or to extract a list or specific words or codes.
This process requires the specification of an inclusion dictionary. This dictionary may be used
to remove variant forms of a word in order to treat all of them as a single word. It may also be
used as a thesaurus to perform automatic coding of words into categories or concepts. For
example, words such as "good", "excellent" or "satisfied" may all be coded as instances of a
single category named "positive evaluation", while words like "bad", "unsatisfied" or
expressions like "not satisfied" may be categorized as "negative evaluation".

6- ADDITION OF FREQUENT WORDS

The fifth process is the application of a frequency criterion that is used to add to the included
words or categories words that are used more than a specific number time or that are found in more
than a specific number of cases. When an inclusion dictionary is used, this option will append to
this list of included words or categories, all words that meet the minimum frequency criterion. If
no inclusion or categorization dictionary is used, all words that meet this minimum requirement
and that have not been excluded (see process #4) will be added to the final word/category list. Note
that this process can only be used to add new words to the actual list of words and categories found
in this inclusion dictionary. It cannot be used to remove any of those items (see process #7).

7- REMOVAL OF WORDS OR CATEGORIES

When this process is applied, all words or categories that do not meet a minimum frequency or
case occurrence criterion will be removed from the final word/category list. It can also remove
items occurring in too many cases. This process may be combined with the categorization process
(#5) to remove infrequent categories. It may also be used in conjunction with the addition criterion
(see process #6) to provide a composite criterion of inclusion that involves both a minimum
keyword frequency and a minimum case occurrence.

12 WordStat Users Manual

Since the application of each process is optional, numerous combinations are possible, each combination
allowing the researcher to perform different types of content analysis. For example, here are the minimal
requirements for different forms of content analysis:

TYPE OF ANALYSIS LEMMATIZATION ADD COMMENT

OR STEMMING EXCLUSIONDI INCLUSIONDI WORDS REMOVE
CTIONARY CTIONARY WORDS
Simple word frequency *
analysis (most frequent words)

Simple frequency analysis of *

semantically significant words

Word count with *

lemmatization

Word count of specific words

Automatic categorization of
texts

Frequency analysis on the

most frequent categories.

Frequency analysis of Codes may optionally be
manually entered codes or inserted between brackets.
keywords
Rating of texts on specific Different weights may be
attributes assigned to different words

*recommended to restrict the analysis to the most frequent words or categories.

WordStat Users Manual 13

 A Quick Tour
A CONTENT ANALYSIS OF PERSONAL ADS
For this example, we will produce a content analysis on personal ads published in a Montreal cultural
newspaper on January 22 and January 29, 1998, and we will examine whether there is a relationship
between words used and the gender and age of the person who wrote the ad. The required data has been
stored in a data file named SEEKING.DBF.

From QDA Miner

Step #1 - Open the data file
Start the QDA Miner program. One will be presented with a dialog box like this:

Click the OPEN AN EXISTING PROJECT button and select the SEEKING.WPJ file located in the
default folder.
If you closed or disabled this introductory dialog box, then from the main screen select the OPEN
PROJECT command from the file menu and select the SEEKING.WPJ file located in the default
folder.

Step #2 - Select the variables

Execute the CONTENT ANALYSIS command from the ANALYZE menu. A dialog box similar to
this one will appear:

14 WordStat Users Manual

 Set the text to analyze to ALL TEXT.
In the IN RELATION WITH group box, select the OTHER VARIABLES radio button.
Click the drop down list and select GENDER and AGEGROUP.

Move to Step #4

From SimStat
Step #1 - Open the data file
From within SimStat, select the FILE | DATA | OPEN command sequence and select the
SEEKING.DBF file

Step #2 - Select the variables

Execute the STATISTICS | CHOOSE X-Y command
Set the Variable List box to ALL to view all variable types.
Move the GENDER and AGEGROUP variables to the INDEPENDENT list box
Move the AD_TEXT variable to the DEPENDENT list box
Press the OK button

Step #3 - Run the content analysis module

Execute the STATISTICS | CONTENT ANALYSIS command.

Step #4 - Choose the proper dictionaries

WordStat consists of an application window with eight pages. The first page allows one to select,
view, and edit the dictionaries used in this specific content analysis. Set the dictionaries to the
following values:
Exclusion: ENGLISH
Categorization: SEEKING
and make both of them are enabled (see check boxes to the left side of the dictionaries edit boxes).

Step #5 - Setting the proper options

The second page allows you to specify various options such as whether numeric values should be
included, whether frequent words should be added, etc. Disable all options by removing any check
mark in the various check boxes.

WordStat Users Manual 15

Step #6 - Perform an univariate frequency analysis on categories
Click the third tab (Frequencies). The program will perform a categorization of words found in the ads
and compute a frequency analysis on those categories.
To sort the frequency matrix in alphabetical order, set the SORT BY option to Words. You can also
display those words in descending order of Keyword Frequency or Case Occurrence.
By default, the words displayed in the matrix are those specified in the Inclusion list. To display words
that have been left out, click the Leftover words tab.
To move a word to the inclusion or the exclusion list, you can click it in the frequency table and drag it
to the desired location, or press the right button of the mouse.

Step #7 - Examining the relationship between included categories and the gender of the
author.
Press on the Sixth tab (Crosstab).
Click the WITH drop-down list box and select GENDER. to display a contingency table of categories
frequency by gender.
The TABULATE option allows one to choose whether the table should be based on the total
frequency of included words or on the total number of cases containing those words.
The SORT BY option allows one to sort the table on the word or category name (alphabetical order)
or by descending order of keyword frequency. You may also click any column header to sort the grid
in ascending or descending order of the values found in this column.
The DISPLAY option allows one to specify the information displayed:
Count
Row percent
Column percent
Total percent
Category percent (for case occurrences)
Percent of total words (for keyword frequency)

Step #8 - Estimating the strength of the relationship

Use the STATISTIC drop-down list box to select an association measure, such as a Chi-square or a
Pearson's R statistic.
To sort the table on the chosen statistic or on its probability, use the SORT BY drop-down list box.

Step #9 - Visualizing the relationship between categories and the age of the author.
Use the mouse to highlight cells of the categories you would like to compare.

Click the button or press the right button of the mouse and select the Chart Selected Rows menu
item.

16 WordStat Users Manual

Step #10 - Performing correspondence analysis on age groups
Click the WITH drop-down list box and select AGEGROUP to display keyword counts by age group.

Click the button to access the correspondence analysis dialog box.

Press on the 2-D Map or 3-D Map tabs to examine a 2 axis or a 3 axis solution or on the STATISTICS
tab to browse through the correspondence analysis statistics.

Click the button to close this dialog box and return to WordStat main window.

Step #11 - Displaying a keyword by keyword matrix or a keyword by case matrix

Click the WITH drop-down list box and select <other words> to display a keyword by keyword matrix
or on <case no> to view a keyword by case matrix.
Step #12 - Viewing a Keyword-In-Context (KWIC) list of specific words or categories
Press on the Keyword-In-Context tab to access the KWIC table.
Set the LIST option to Included and select the word or category for which you would like to obtain a
KWIC table.

Click the button to display the KWIC table for this word or category.
To sort the table on the case number, on the keyword along with the prior or subsequent words, or on
the sex of the respondent, use the SORT BY drop-down list box.
To display KWIC tables of any user defined word or word pattern, set the LIST option to "User
defined", enter your word pattern (with or without wildcards) in the WORD edit box and click the
button.

Step #13 - Editing a text from the KWIC list

To modify the word or keyword or the text surrounding it, select it from the KWIC list, right-click and
select the EDIT command. (You may also double-click the specific line you wish to edit).
To save the modified text, click the OK button. Clicking the CANCEL button restores the original
text.

Step #14 - Creating a concordance report

Make sure the Keyword-In-Context page is active and that the KWIC table displays the proper
information.
Set the amount of context that should be displayed around each word by setting the CONTEXT
DELIMITER option.

Click the button. Note: If this button is inactive, click the button to refresh the

WordStat Users Manual 17

 content of the KWIC table and then click the button.

Step #15 - Examining relationship between words or content categories

Move to the Co-Occurrence page.
Click the DENDROGRAM tab to perform a hierarchical cluster analysis on included categories. You
may change the number of partitions displayed using the No clusters option.
Click the MAPPING tab to perform a multidimensional scaling, and display a plot in two or three
dimensions.

Step #16 - Saving the keyword frequencies on disk

Move to the Frequencies page.

Press the button and select the DATA MATRIX command.

Set the options to export to the existing data file or a new one.

Step #17 - Quitting the module and returning to QDA Miner or SimStat

Click the button in the upper left-hand corner of WordStat and select the EXIT command or
click the X mark in the upper right-hand corner.

18 WordStat Users Manual

 Preparing and Importing Data
This section provides general information on how to prepare textual data or specific instruction on how to
import data into QDA Miner and SimStat.

Preliminary Text Preparation

While interview transcripts, responses to open-ended questions, or any other kind of textual information
may be typed directly within SimStat or QDA Miner, there are many situations where electronic versions
already exist either in the form of text files or as data files accessible only through specific applications
such as word processor, spreadsheet or database programs. All this information must be transferred into a
QDA Miner project or SimStat data file for further processing. However, prior to using WordStat for
content analysis, some modification or adjustments may need to be made.

Uppercase and lowercase letters

WordStat is case-insensitive and therefore accepts files in either upper- or lowercase.

Check spelling of documents

The automatic content analysis feature of WordStat involved numerous operations of word recognition and
generally requires each word to be spelled correctly. Any misspelled word may be left uncoded and leads to
imprecise or invalid conclusions. Two strategies may be used to deal with misspellings:
1. One may run documents through a spell-checker to make sure all words are spelled correctly.
WordStat provides spell checking for more than 20 human languages. The spell-checking may be
performed in QDA Miner or through the Text Editor feature of WordStat. An even more efficient
approach is to use the Misspelling & Unknown Words feature of WordStat to quickly retrieve all
potentially misspelled words and to replace them all at the same time.

2. An alternative approach would be to build a content analysis process that would take into account
the misspelling of words. To achieve this, one may use the Substitution feature to automatically
replace those misspelled words with their correct forms or add the most commonly misspelled
keywords into the content analysis dictionary.

Remove hyphenation
While WordStat can be configured to accept compound words with dashes, it cannot differentiate dashes
and hyphens. As a consequence, a hyphenated word will often be treated as two separate words. It is thus
recommended to revise the text to ensure no hyphenation is present.

Add or remove square brackets ( [ ] ) and braces ( { } )

Square brackets and braces have special meanings for WordStat. For example, braces are often used to
remove a section of the text that you don't want to process while square brackets may be used to restrict the
analysis to specific portions of text. If these symbols are used in a text for other purposes, they should be
replaced with other symbols.

WordStat Users Manual 19

If there are specific parts of your text that you do not what to process, such as some explanation notes,
interviewer questions and probes, comments, etc.), enclose them in braces (ex. {comment} ). Also, if you
want to perform a content analysis on only a small portion of the entire text, such as on manually entered
codes, enclose this portion of text in square brackets. QDA Miner's coding feature may also be used to
restrict the analysis to some sections or exclude specific text segments from the content analysis process.
Once those text segments have been manually tagged in QDA Miner, one could then specify, when calling
WordStat, to ignore sections tagged with specific codes or to only analyze segments associated with one or
several codes.

Importing Spreadsheet Files

Most spreadsheet programs allow for entry of both numeric and alphanumeric data into cells of a data grid.
QDA Miner as well as SimStat can import spreadsheet files produced by EXCEL (*.xls; *.xlsx).

To import an Excel spreadsheet from QDA Miner

Choose the NEW command from the PROJECT menu.
Click the Import from an Data File button.
Select the file format using the List File of Type drop down list.
Select the file you want to import and click the OK button.

To import an Excel spreadsheet from SimStat:

Choose the DATA | IMPORT command from the FILE menu.
Select the file format using the List File of Type drop down list.
Select the file you want to import and click the OK button.

The program displays a dialog box where one can specify the spreadsheet page and the range of cells where
the data are located. You must specify a valid range name or provide upper left and lower right cells,
separated by two periods (such as A1..H20). If you set the Range Name list box to ALL, the program
attempts to read the whole page.

Formatting spreadsheet data

The selected range must be formatted such that the columns of the spreadsheet represent variables (or
fields) while the rows represent cases. Also, the first row should preferably contain the variable names
while the remaining rows hold the data, one case per row. QDA Miner and SimStat will automatically
determine the most appropriate format based on the data it finds in the worksheet columns. Cells in the first
row of the selected range are treated as variable names. If no variable name is encountered, QDA Miner and
SimStat will automatically provide one for each column in the defined range.
When reading the data for analysis, all blank cells and all cells that do not correspond to the variable type
(e.g., alphanumeric entries under a numeric variable, or a numeric value under a string variable) are treated
as missing values.

20 WordStat Users Manual

Importing Database Files

MS Access, dBase and Paradox files

QDA Miner and SimStat can directly import MS Access, dBase and Paradox data files. For the last two file
formats, the length of alphanumeric variables should not exceed 256 characters and memo variables are not
supported. If you do have such kind of data, you may use the exporting capabilities of your database
program to create a date file more compatible with SimStat (such as a Visual FoxPro data file or a tab
delimited text file).
Other database files
QDA Miner and SimStat offer ways to import from various database formats by connecting directly to the
database system using an ODBC connection. For database systems for which no ODBC driver exists, it is
still often possible to import data by exporting the data to a common file format that can be read by QDA
Miner or SimStat. The recommended file formats are, in descending order of preference, Excel, Tab-
delimited text files, or CSV files.

Importing memo variables

Memo variables that have not been successfully imported may be transferred to the data file either by using
cut and paste operations or by retrieving text files from disk.

Importing Plain Text or Word Processor Files into SimStat

QDA Miner provides an easy way to import documents stored in various formats, including MS Word,
WordPerfect, Rich Text, HTML, PDF and plain-text files. When using SimStat as the base module, such a
task is not as obvious. One way to transfer data from a word processor document into SimStat is to open
simultaneously both applications and use cut and paste operations to transfer data through the clipboard.
However, this may not be the most efficient way, especially when one needs to import a large amount of
information. The following section presents four additional methods to transfer text information into memo
variables:
Using the Document Conversion Wizard program
Retrieving a text file into a memo variable
Importing comma or tab delimited text file
Importing page delimited memo files

While the first method can read textual data stored in word processor documents, the last three methods
require the data to be stored on disk in plain ASCII files without any formatting or typesetting code. Most
word processors offer an option to save a document as a plain text file. If you don't know how to create such
a text file, please refer to your word processor manual.

Using the Document Conversion Wizard program

WordStat includes a conversion utility program that can assist you in the importation of text files stored in
either word processor documents such as MS Word, MS Write, WordPerfect, RTF, or Acrobat PDF files,
but also of text stored in ASCII (plain text), HTML or even Excel spreadsheet files. To run this program:

WordStat Users Manual 21

 Point to the Programs folder in the Windows' Start menu, then select Provalis Research and then click
Document Conversion Wizard.
This utility program will guide you through the process of importing one or numerous text files.

Retrieving a text file into a memo variable

This method should be used to retrieve a single unit of text into a memo variable for a specific case. If
textual data for several cases need to be retrieved, they should be stored in different text files. To retrieve
the text file from SimStat:
Open the data file where the information should be stored.
Position the cursor on cell in which you would like to store the text. A memo editor should appear at
the bottom of the data sheet.
Click inside the memo editor or press F2.
Click the Import Text Into Memo button, select the text file you wish to retrieve and click OK.

Importing comma or tab delimited text files

If you wish to retrieve a text file containing several numeric and alphanumeric variables, you may have to
transform this file into a comma or tab delimited text file. There are, however, several limitations to this
transfer method. If commas are used as delimiters, then all existing commas within text variables should
ideally be removed. If a tab delimited format is chosen, all tab characters already present in a text variable
should be removed. Another important limitation is that all the information of a single case must be stored
in a single line. For this reason, hard returns in long texts should be removed so that the entire text is stored
on a single line. (There is no limitation on the total number of columns per line, so it is possible to store
very long texts on a single line).
QDA Miner as well as SimStat can read up to 2000 numeric and alphanumeric variables from a plain ASCII
file (text file). The file must have the following format:
Every line must end with a carriage-return.
The first line must include the variable names, separated by tabs or commas.
Variable names may have a length of not more than 10 characters. Longer strings are truncated to 10
characters.
The remaining lines must include the numeric or alphanumeric values, separated by tabs or commas.
Each line must contain data for one case and variables must be in the same order for all cases. All
invalid data and all blanks encountered between commas or tabs are treated as missing values. A
single dot can also be use to represent a missing numeric value.
Comments can be inserted anywhere in the file by putting a * at the beginning of the line.
Blank lines can also be inserted anywhere in the file.
Comma delimited text files require a .CSV extension while tab delimited files require a .TAB
extension.

22 WordStat Users Manual

Importing page delimited memo files
SimStat provides a simple method to import numerous texts by the use of page delimited memo files. This
file format consists of a plain text file which contains the textual data of numerous individuals for a single
memo variable. The text for each case must be separated by page break characters (ASCII 12). The file
name extension of this text file should be .MMO. To import such a file:
Choose the DATA | IMPORT command from the FILE menu.
Set the file format to Page Delimited Memo using the List File of Type drop down list.
Select the file you want to import and click the OK button.

The resulting file consists of a SimStat data file with two variables: RECNO, a numeric variable containing
a sequential number going from 1 up to the total number of cases encountered in the input file, and TEXT, a
memo variable containing the textual data for this case.
Note: Importation of numerous text variables may be achieved by performing successive importations of
page delimited memo files and then using the APPEND VARIABLES command to merge the resulting files
into a single one. In order to achieve this, great care should be taken to give unique names to the various
TEXT variables and to assure that the case sequence of the various text files is identical.

WordStat Users Manual 23

 The WordStat User Interface
WordStats user interface is built around a eight pages workspace that provides an integrated environment
for developing, testing, validating and applying content analysis dictionary and to perform various text
mining tasks.

Dictionaries This page allows one to adjust various text analysis processes, create and modify
dictionaries, exclusion and substitution lists, as well as add, remove and edit existing entries in those
dictionaries.
Options This page contains various options controlling how the text data will be processed. It also
includes options affecting linguistic tools as well as program appearance.
Frequencies This page displays a table of the frequency of keywords or content categories. One may
also get a list of leftover words, allowing one to modify the current categorization dictionary, the
exclusion list or the substitution list.
Extraction This page allows one to perform topic modeling, find the most common phrases, extract
named entities, as well as misspellings and uncommon words, and assign them to the current
categorization dictionary, the exclusion list or the substitution list. One may also use the misspelling
page to batch replaced misspellings in the original documents.
Co-occurrences This page allows one to explore connections between words, keywords, phrases or
content categories using hierarchical clustering, multidimensional scaling, link analysis and proximity
plots
Phrase Finder This page allows one to extract the most common phrases and idioms and to easily
add them to the current dictionaries. Co-occurrence and comparison techniques such as clustering,
multidimensional scaling and correspondence analysis are also available from this page.
Crosstab This page allows one to compare keyword frequencies across values of numerical,
categorical or date variables. Along with a table of frequencies, several statistics and graphical
techniques may be applied including correspondence analysis, heatmaps, bubble charts, bar charts and
line charts. The automatic document classification feature can also be accessed from this page.
Keyword-in-Context This page allows one to display a concordance table word patterns or phrases,
or of all items related to a content category. Such a table is very useful to validate a dictionary by
allowing one to examine in context how words are being used.
Classification This page gives access to the automated text classification module that allows one to
apply a machine-learning approach to the existing textual database. Options allow one to develop a
classification model that can later be used to accurately classify uncategorized documents into
predefined classes.

Two additional pull down menus can be accessed to perform various tasks:

Clicking the button in the upper left-hand corner of the main window displays a menu that allows one
to leave WordStat and return to the calling application as well as to perform various tasks such as editing

24 WordStat Users Manual

case descriptors (page 166), filtering cases (page 167), accessing the Report Manager (page 191), or
accessing the text editor (page 142).

The button in the upper right-hand corner provides access to this help file, which can also be accessed
at any time by pressing the F1 key. In addition, this menu allows you to check whether you are using the
latest version of WordStat and also gives access to specific important information and some useful links to
the Provalis Research website.

WordStat Users Manual 25

 Dictionaries Page
Without further information, WordStat can perform a frequency analysis on each of the words encountered
in the chosen document or alphanumeric variables. However, it is also possible to apply various
transformations on the words before performing the frequency analysis. The first two pages of the main
window (see below) allow one to specify how the textual information should be processed. For example,
one can tell the program to lemmatize words, to ignore words found in an exclusion list (also known as a
stop list) or categorize them using a categorization dictionary. The Dictionaries page allows using or
creating dictionaries, exclusion and substitution lists, add, remove or edit existing entries in those
dictionaries. (For more information of other analysis options available see Options Page on page 33)

The following section provides a description of the four processing steps involved in the transformation of
textual data into keywords or content categories. Additional information about dictionary creation and
maintenance can be found on page 114.

STEP #1 - PREPROCESSING
The preprocessing option allows for the custom transformation of the text to be analyzed prior to, or in
place of the execution of the other three standard processes provided by WordStat: lemmatization,
exclusion and categorization. This transformation is accomplished by the execution of specially
designed external routines accessible in the form of an external EXE file or a function in a DLL
library. This feature is provided to offer greater flexibility by allowing any user with programming
skills or resources to customize the processing of textual information. For more information on this
feature see Configuring External Preprocessing Routines on page 139.

26 WordStat Users Manual

STEP #2 - STEMMING
Stemming is a natural language processing routine that reduces inflected and derived forms of words to
a common root form or word stem. The English stemmer, for example, returns "write" for "write",
"writer", "writing" and "writings". Stemming can be especially useful in some exploratory text-mining
tasks or when developing automatic document classification models by grouping related words together,
and, thus reducing the total number of word forms. However, it may also decrease the precision in the
measurement of some topics associated with specific inflected forms. Plural and singular forms of some
nouns are often used to refer to different concepts or ideas. The same is true for various tenses of verbs.
For example, in a sentiment analysis project, we found that the verb "improve" was often associated
with negative comments, while its past-tense form "improved" was generally associated with positive
comments. Since stemming is based on a limited number of morphological rules, stemming algorithms
are also prone to errors. For example, the English Porter stemmer will group words like "universal",
"universe" and "university" into the single word root: "univers". Stemming may also fail to group related
words that do not follow typical grammatical rules.

STEP #3 - SUBSTITUTION
The substitution process may be used to automatically replace specific words with other word forms. It
may be used to substitute common misspellings or perform lemmatization. One could also use this
process to perform a simple type of categorization where specific words are replaced with keywords.
WordStat provides four predefined substitution processes to perform lemmatization on documents in
English, French, Italian and Spanish. Lemmatization is a process by which various forms of words are
reduced to a more limited number of canonical forms. A typical example of lemmatization would be the
conversion of plurals to singulars and past tense verbs to present tense verbs. The lemmatization
algorithm implemented in WordStat is a dictionary-moderated method, partly inspired by Krovetz's
KSTEM suffix substitution algorithm. Since the lemmatization algorithm does not rely on a prior part-
of-speech tagging of words, it is much faster than traditional lemmatization routines. It may, however,
result in a few invalid word substitutions, but usually, those errors will have no major consequences on
the result of an analysis. However, it is important to remember that lemmatization, like stemming, may
decrease the measurement precision of some concepts or topics (please refer to Step #2 for more
information on this issue). WordStat offers a way to monitor all substitutions performed by this routine
and to override any by creating a list of custom substitutions. For more information on such a feature,
see Monitoring and Customizing Substitutions on page 135.

STEP #4 - EXCLUSION OF WORDS

The exclusion dictionary (also known as a stop list) is used to remove all words that are not to be
included in the content analysis. It is used mainly to remove words with little semantic value such as
pronouns, conjunctions, etc. Wildcard symbols such as * , ?, # and [ ] are supported. The * character
(also call the "asterisk" or "star") matches zero or more characters. For example, the following
expression:
REPORT*
will exclude all words beginning with REPORT (such as REPORT, REPORTS, and REPORTER),
while this expression:
COLO*R
will match both COLOR and COLOUR.
The question mark character (?) matches exactly one character. For example:

WordStat Users Manual 27

 EXP?RT
will match both EXPORT and EXPERT.
The number sign (#) wildcard stands for a single numerical digit, so for example:
B##
will match words like B12 (the vitamin) as well as B52 (the aircraft), while:
#####
will match any five digit number, typical of US zip codes. Please note that the use of the # wildcard will
work as long as one sets the Accept Numeric Characters check box on the OPTIONS page.
The square brackets "[" and "]" are used to match a single character out of a list of characters. For
example,[AEIOU] will match any one of those vowels, while [A-E] will match any letter between A and
E. The following pattern:
[A-Z][0-9][A-Z]_[0-9][A-Z][0-9]
will match typical Canadian postal codes, like H3B 1W9. Note that the underscore character in the
above example represents a space.
An expression that includes several words may also be excluded by joining the various words with
underline characters. For example:
NOT_*
Will exclude all words preceded by the word "not".
The currently opened exclusion dictionary may be deactivated by removing the check mark in the check
box at the left of the exclusion dictionary name.

STEP #5 - CATEGORIZATION OF WORDS AND PHRASES

The inclusion dictionary allows one to change specific words, word patterns (such as REPORT* or
EXP?RT), or expressions, to another word, category or concept. This feature may be used to remove
variant forms of a word in order to treat them as a single instance or to group related words under
meaningful categories. Inclusion dictionaries may also be used to perform a frequency analysis on
manually entered codes. By manually entering specific keywords (such as "EVAL_POS" , or
"EVAL_NEG") in a text variable and by entering those same keywords in the inclusion list, it becomes
possible to extract those codes and perform frequency and contingency analysis on them.
A categorization dictionary may also contain rules delineating the conditions under which specific
words or phrases should be categorized. Those rules may consist of complex expressions involving
Boolean (AND, OR, NOT) and proximity operators (NEAR, BEFORE, AFTER). Those kinds of rules
allow one to eliminate basic ambiguity in words by taking into account the presence of other words that
may alter the meaning. A good example would be the presence of a negative word form (such as "rarely"
or "never") close to an adjective. Another example would be the differentiation of the various meanings
of the word BANK by identifying other words like "river", "money" and "deposit" surrounding "bank".
For more information on rules, see section Working with Rules, page 124.
The inclusion dictionary is structured as a hierarchical tree where words, word patterns, phrases, and
rules are grouped in a folder that represents a category name. Categories and individual words may also

28 WordStat Users Manual

be included in a higher order category, allowing one to create multi-level dictionaries like the following
one:
COUNTRY
NORTH-AMERICA
CANADA (1)
UNITED-STATES (1)
USA(1)
MEXICO (1)
SOUTH-AMERICA
BRAZIL (1)
CHILI (1)

In the above example, words like CANADA, USA, or MEXICO may be coded as either NORTH-
AMERICA or COUNTRY, depending on whether the categorization is performed up to the first or second
level of the dictionary (see Level of Analysis, page 30).
Wildcards such as *, ?, # and the square brackets [ and ] are supported. For example, the following
item under the support category:
SUPPORT
SUPPORT*
will change SUPPORT, SUPPORTS, SUPPORTING, SUPPORTIVE, SUPPORTER, etc. into a single
word SUPPORT, while the following word pattern:
SUPPORT
*SUPPORT*

will also substitute all words with the substring "SUPPORT" in it, such as UNSUPPORTEDLY,
UNSUPPORTED, etc.
An expression that includes several words may also be substituted by joining the various words with
underline characters. For example, you may change the expression "going out" with the category
"NIGHTLIFE" by specifying the following item:
NIGHTLIFE
GOING_OUT
You may also use wildcards in expressions such as:
NIGHTLIFE
GO*_OUT
to substitute several forms of an expression at once.
Integer weights can also be assigned to specific items so that a specific word or word pattern may
count for more than one instance of the category. For example, in order to compute an

WordStat Users Manual 29

 aggressiveness score on specific texts, you may choose to assign a weight of 5 points to word
patterns such as KILL* or MURDER* but only a single point to word patterns like INSULT*.

CATEGORIZATION SETTINGS
LEVEL OF ANALYSIS - This option allows one to specify up to which level the coding should be
performed. For example, in the following dictionary:
COUNTRY
NORTH-AMERICA
CANADA (1)
UNITED-STATES (1)
USA(1)
MEXICO (1)
SOUTH-AMERICA
BRAZIL (1)
CHILI (1)
if a level of 1 is specified, all words that are stored at a higher level than the root level will be
coded as the parent category at this first level. For example, words like CANADA and MEXICO
will be coded as COUNTRY along with other country names like BRAZIL. Setting the level of
analysis to a numeric value of 2 will results in the coding of those two words as NORTH-
AMERICA, while BRAZIL will be coded as SOUTH-AMERICA. Items stored at the same or
at a lower level than this option will remain unchanged.
Setting the LEVEL option to AS SHOWN instructs WordStat to match the level of
categorization performed to the level of details currently displayed in the tree view of the
categorization dictionary. This option allows one to set different levels of categorization by
expanding broad categories that should be broken down and by collapsing categories for
which finer details are not needed. For example, if we modify the above tree by collapsing the
NORTH-AMERICA category, WordStat will display it the following way:
COUNTRY
NORTH-AMERICA
SOUTH-AMERICA
BRAZIL (1)
CHILI (1)

The program will report frequencies of individual countries like BRAZIL or CHILI but will
categorize every instance of CANADA, UNITED-STATES, USA and MEXICO as NORTH-
AMERICA.
Please note that it is possible to prevent a category from being broken down into subcategories
or items, even if the level of analysis is set to a higher setting, or if it is set to AS SHOWN and
the items contained in this category are visible. Such a feature is useful when the content of a
category consists of different ways of referring to the exact same thing (for example

30 WordStat Users Manual

 UNITED_STATES, UNITED_STATES_OF_AMERICA, US and USA) or consists of
various misspellings.

To make a category unbreakable, select it in the dictionary tree, click the button,
and put a check mark in the Unbreakable box. The folder icon normally used to represent
categories will be transformed into a folder icon with a key inside. You may also select the
category, right click, and then select UNBREAKABLE | YES from the pop-up menu. To
unlock the folder, follow the previously described steps for editing the category and remove
the check mark in the Unbreakable box or select UNBREAKABLE | NO from the pop-up
menu.
CATEGORIES ONLY - When the LEVEL OF ANALYSIS option is set to a value higher than one,
this option instructs WordStat to limit the level increase to the coding of the last category at or
below the specified level. This option is especially useful when working with unbalanced
hierarchical categorization systems where individual words are stored at different levels. For
example, in the following dictionary:
SENSATION
ODOR
AROMA (1)
BREATH (1)
FRAGRANCE (1)
NOSE (1)
ANXIETY
AFRAID (1)
TREMOR (1)
setting the level of analysis to 2 without enabling this option would code words like AROMA
or BREATH as ODOR, but would include in the final results individual words like TREMOR
or AFRAID. Enabling the CATEGORIES ONLY option ensures that individual words won't
be included but will be coded as their parent category.
USE FULL PATH AS CATEGORY NAME - When the LEVEL OF ANALYSIS option is set to a
value higher than one, this option instructs WordStat to substitute the full path of an item as
the category name. The slash ( / ) character is used to separate the various levels. For example,
in the above example, setting this option to true and the level analysis to 2 will code the word
AROMA as SENSATION/ODOR. Increasing the level of analysis up to 3 will return
SENSATION/ODOR/AROMA.
ALLOW OVERLAP - By default, categories are mutually exclusive such that a word can only be
entered in a single category. Enabling this option allows one to create overlapping categories
where words can be classified simultaneously into two or more categories. However, please
take note that current multivariate techniques available in WordStat such as clustering,
correspondence analysis and multidimensional scaling as well as other multivariate statistical
procedures make the assumption that categories are statistically independent. Using
overlapping categories creates data that clearly violate this assumption and may yield dubious
results.
SHOW WARNINGS - Some items in an exclusion list or categorization dictionary may remain

WordStat Users Manual 31

 undetected in documents because of their incompatibility with some analysis options. This
occurs, for example, when an item is found both in the categorization dictionary and the
exclusion list, or when this item includes non-alphabetic characters that have not been
specified as valid. The following table displays the various types of problems that may be
identified by WordStat:

TYPE DESCRIPTION

Item includes invalid characters WordStat identifies individual words using

alphabetic characters and other special characters
specified by the user in the Valid Characters option.
So, to make sure any item containing non-alphabetic
characters is properly recognized, this special
character must be added to the list of valid characters.

Item includes numeric characters An item in the categorization dictionary or the

exclusion list that includes numeric characters cannot
be recognized since the Accept Numeric Characters
option is currently disabled.

Item also in the exclusion list An item found in a categorization dictionary cannot
be recognized if it matches an item found in the
exclusion list.

Phrase starts with an excluded word In order to be recognized, a phrase cannot start with a
word found in the exclusion list. Therefore, this
excluded word should preferably be removed from
the exclusion list in order for the phrase to be
recognized.
Enabling the Show Warnings option instructs WordStat to identify potential compatibility
problems affecting items in a dictionary, and it displays a list of those problems in a special
dialog box. This dialog box is displayed prior to the application of dictionaries for a content
analysis.
For more information on how to open, activate or deactivate a dictionary or how to add, edit or remove an
entry in a dictionary, see Creating and Maintaining Dictionaries, page 114)

32 WordStat Users Manual

 Options Page
This page offers different options that control how the textual information should be processed. The
options are grouped under three different pages:
1) Analysis
2) Languages
3) Miscellaneous

DICTIONARY OPTIONS
ADD WORDS - When the inclusion dictionary is disabled, all words that are not found in the
exclusion list will be included in the final keyword frequency analysis. This option allows one
to restrict the number of words included to the most frequent ones by setting a minimum
Frequency or Case Occurrence criterion for inclusion. This option may also be used while
the inclusion list is active to add to this list, other words that are used at a high frequency.
However, this option can only be used to add new words to the list of words and categories
found in this inclusion dictionary and cannot be used to remove any of those items. To
remove items in this inclusion dictionary based on a frequency or case occurrence criterion
see the REMOVE WORDS option below.

REMOVE WORDS - This option allows one to restrict the number of included words or categories
to the most frequent ones by setting a minimum Frequency or Case Occurrence criterion for

WordStat Users Manual 33

 inclusion. This criterion is applied both to items in the inclusion dictionary and words that
meet the criterion specified with the ADD WORDS option.
Examples:
If no inclusion dictionary is used and you want to include any word that appears at least
10 times, but in no less than 5 different cases, you need to activate the ADD WORDS
option and set its criterion to a minimum FREQUENCY of 10. You then have to set the
REMOVE WORDS criterion to a minimum Case Occurrence of 5. Only words that
meet both criteria will be included.
When an inclusion list is used to lemmatize words, but you only want to obtain
frequency information on those words that appear a specific number of times, you have
to activate the inclusion dictionary and set the minimum frequency criterion of both the
ADD WORDS and REMOVE WORDS options to the required frequency.
When an inclusion list is used to categorize words, but you only want to analyze the
most frequent categories, you have to activate the inclusion dictionary and set the
REMOVE WORDS option to the required frequency. In this situation, the ADD
WORDS option should be deactivated.
LEAVE CATEGORIES EQUAL TO ZERO - By default, WordStat removes from the frequency
table any keyword or category in the categorization dictionary that had not been encountered
in the analyzed text. Enabling this option instructs the program to leave those items with a
zero frequency in the table. Such an option is especially useful when comparing obtained
frequencies to normative data or to other samples. This option should also be enabled when
creating norm files (see Creating and Using Norm Files on page 148).
REMOVE ITEMS OCCURRING IN MORE THAN n PERCENT OF CASES - This option allows
one to remove keywords or categories appearing in more than a specified percentage of cases.
This criterion is applied both to items in the categorization dictionary and to words that meet
the criterion specified in the ADD WORDS option. Such an option is especially useful to
remove words that are too common to have any informative or discriminative value.
KEEP A MAXIMUM OF n ITEMS - This option allows one to restrict the number of included
words or categories to a maximum number of items, based either on their total frequency,
number of case occurrences, or on the computed TFxIDF index. This selection occurs only
after all the previous frequency options have been assessed and only if the total number of
remaining items is higher than the specified maximum. If the cutting point falls on a
frequency or a case occurrence shared by many items, those with the highest TFxIDF values
will be selected.

TEXT TO INCLUDE
DON'T PROCESS TEXT WITHIN BRACES - This option can be used to instruct the program to
skip all text found between braces (i.e. { and } ). This option is especially useful when you
want to insert comments or annotations in the text variable without affecting the content
analysis. It can also be used to ignore in an interview transcript all questions, prompts, and
other verbal interventions made by the interviewer.

34 WordStat Users Manual

 DONT PROCESS TEXT WITHIN BRACKETS - This option can be used to instruct the program to
skip all text found between brackets (i.e. [ and ] ). Since WordStat can also be configured to
analyze only text found between such brackets (see option below), these two options may be
used to toggle between an analysis of keywords entered manually between those brackets and
of the surrounding text.
PROCESS ONLY TEXT WITHIN BRACKETS - This option can be used to instruct the program to
process only the text found between brackets (i.e. [ and ] ). This option may be used to
perform an analysis on keywords entered manually in the text by one or several coders.

CHARACTERS
ACCEPT NUMERIC CHARACTERS - By default, every word consisting of numeric values or of a
mix of letters and numbers is excluded from the analysis. This option can be used to include
those words.
ADD CHARACTERS APPEARING - This set of options allows one to specify which characters,
besides letters of the alphabet, should be considered as an integral part of a word. For
example, the word "ex-wife" can be treated as a single word or as two separate words ("ex"
and "wife") if the hyphen is included in the list of valid characters. Two edit boxes may be
used to specify additional characters. The ANYWHERE option is used to specify special
characters that will be considered as part of a word, no matter where they appear, while the
EMBEDDED IN WORDS option should be used to specify characters that should be enclosed
within other valid characters and not at the beginning or the end of a word. For example,
adding the period and comma to the list of characters embedded in words, will allow one to
retrieve numeric values such as 97.5 or 1,000,000 or domain names like www.google.com as
a single token without the risk of retrieving words immediately followed by commas or
periods.
CASE SENSITIVE - By default, WordStat internally converts all text to uppercase letters so that
processing of words is cases insensitive. This may be inappropriate if one wants to identify
proper nouns or analyze text written in some European languages like German where
differences in letter cases may denote different meaning. Enabling this option prevent the
internal conversion to uppercase letters and will treat two instances of the same word different
in their case (lower or upper case) as two distinct words.

CASE PROCESSING
RANDOM SAMPLE - When this option is activated, the program will randomly select a fraction of
all cases and performs the content analysis on this subsample. The proportion of cases can be
specified using the spin button located at the right of the checkbox. This option reduces the
processing time for large files and is especially useful during the initial phase of an analysis
where dictionaries are constructed and categorization schema are developed and revised. It
also allows one to preview the kind of results that would be obtained on very large data files.
INCLUDE RECORDS WITH MISSING VALUES - When examining the relationship between
textual data and categorical or numerical variables, WordStat will skip any cases with a
missing value on any one of these variables. Enabling this option instructs WordStat to
include all cases, whether or not values are missing. All missing values are assigned to an
additional class labeled as "MISSING." Any analysis involving comparisons between classes

WordStat Users Manual 35

 of categorical variables (cross-tabulation, correspondence analysis, etc.) will include this
additional class.
WEIGHTING VARIABLE - This option allows the selection of a variable that will be used to apply
weight to the cases. When the program reads a case, the value of the weighting variable for
this case is truncated to an integer. This integer value specifies how many times the case will
be duplicated. If the value is less than one, the case is excluded from the analysis. This option
is especially useful when the textual data to be analyzed have already been reduced to a
frequency list, such as when analyzing a list of the most frequent queries on a search engine.

LANGUAGES PAGE
ADD OR REMOVE LANGUAGE RESOURCE FILES - WordStat may use various resources (such as
spelling dictionaries, thesauri, and lemmatization routines) to process some natural languages
more efficiently. By default, WordStat installs resources for processing English documents.
Clicking this button allows one to download and install resource files for other languages or
remove language resources previously installed.
ACTIVE SPELLING DICTIONARIES - WordStat makes use of language dictionaries to spell-check
existing textual data and to suggest inflected forms of words found in the user dictionary. This
group of options lets one specify which dictionary to use with the current data file.
ACTIVE THESAURUS - WordStat's SUGGEST feature can use a thesaurus to suggest synonyms of
existing words in the text collection or in the user categorization dictionary. This option allows
one to select the language of the thesaurus (for more information on this feature, see Using
Lexical Tools for Dictionary Building, page 127).

OTHERS PAGE
COLOR SCHEME - A color scheme is a set of colors used for background, page tabs, borders and
buttons. WordStat comes with several color schemes. To choose a desired color scheme, simply
select one from the drop-down list. To disable this feature and use the default Windows theme, set
this option to System Default.
FLAT TABLES (WITHOUT GRID LINES) - When this option is unchecked, tables in WordStat are
displayed with grid lines while column and row headers are displayed in 3-D. Enabling it,
removes grid lines and flattens row and column headers.
PERCENT DECIMAL PLACES - Use this option to modify the number of decimal places used to
display percentages in frequency tables and in crosstabulation tables.
SHOW HARD RETURNS AS - In KWIC lists and reports, hard returns normally used to mark the
beginning of a new paragraph are represented by a symbol. This option allows toggling on and
off the display of this symbol.
TREATMENT OF ITEMS NOT FOUND IN NORM FILES - When comparing keyword frequencies with
normative data, a specific word in the frequency list may be absent from the normative data file.
This may occur for neologisms, technical terms, proper nouns, low frequency words as well as
misspelled words. This option allows one to choose whether, in such a situation, the cells for
expected frequency and comparison statistics should be left empty or if those statistics should be
extrapolated by setting the expected frequency to the lowest possible frequency (based on the size
of the normative body of text that was used to compute the normative data).

36 WordStat Users Manual

 Frequencies Page
The Frequencies page is used to display a frequency table of words or category names. This can be used to
perform a univariate frequency analysis on words or categories and also to modify any of the active
dictionaries or word lists.

By default, the table shows the included words in descending order of frequency. The table includes the
following statistics:
FREQUENCY Number of occurrences of the word or category names.
% SHOWN Percentage based on the total number of words displayed in the table
% PROCESSED Percentage based on the total number of words encountered during the
analysis.
% TOTAL Percent based on the total number of words less those excluded by list.
NO CASES Number of cases where this keyword appears.
% CASES Percentage of cases where this keyword appears.
TF*IDF Term frequency weighted by inverse document frequency. Such a weighting
is based on the assumption that the more often a term occurs in a document,
the more it is representative of its content yet, the more documents in which
the term occurs, the less discriminating it is.
Tabs at the top of the table allow you to access (1) a frequency table of all Included content categories or
keywords, (2) a frequency table of Leftover Words consisting of individual words that have not been

WordStat Users Manual 37

categorized or included in the analysis.
To the left of the main grid, a panel lists the content categories of the current dictionary following additional
entries representing the substitution and exclusion processes. This panel may be used to quickly assign
items in the table to any one of these locations using the drag-and-drop operation (For more information on
how to use this panel, see Using the Dictionary Panel on page 41).

SORT BY - This option allows a display of words in the frequency table in alphabetical order, on
keyword endings, or by descending order of keyword frequency (NO WORDS column) or
case occurrence (NO CASES column). Sorting the table by keyword endings facilitates the
identification of plural form of words that should be substituted by their singular form or the
substitution of verbs by their infinitive form. One may also sort on any column in ascending
order on any column values by clicking this column header. Clicking a second time on the
same column header sorts the rows again in descending order.

The button can be used to move one or several words to the exclusion or substitution list or
to add or remove a word from the inclusion list. The permitted moves depend on the words
currently displayed. If you want to remove a word from the inclusion list, the DISPLAY option
should be set to INCLUDED or ALL. To add a word to the inclusion list, the DISPLAY should be
set to LEFTOVER or ALL. This button is also used to display a Keyword-In-Context table of the
selected keyword.
It is also possible to quickly access the pop-up menu invoked by this button by pressing the right
button of the mouse anywhere on the grid (see below).

The button allows one to produce barcharts or pie charts to visually display the distribution of
specific keywords or categories. To produce such charts:

38 WordStat Users Manual

 Set the Sort By option to the order in which you wish the values be shown graphically.
Select the rows you would like to plot (multiple but separate rows can be selected by
clicking while holding down the CTRL key)

Click the button.

For further information see Displaying Distribution Using Barcharts or Pie Charts on page 145.

The button allows one to create normative frequency data from the current file, to store them
on disk and to compare currently displayed frequencies with previously saved norms. See Creating
and Using Norm Files on page 148 for more information on this topic.

The button may be used to append frequency information to the current data file, save to disk
a matrix of word or keyword frequency by cases or export the current categorization model. For
more information on one of these topics see Exporting Frequency Data (page 181) or Exporting
Categorization Models (page 183).

The button allows one to access a keyword retrieval feature to retrieve all documents,
paragraphs or sentences containing a specific keyword or a combination of keywords. See
Keyword Retrieval (page 150) for more information on this topic.

The button allows one to automatically attach QDA Miner tags to all paragraphs or sentences
associated with currently displayed content categories or keywords. When clicked, a dialog box
asks whether the coding should be applied to whole paragraphs or to individual sentences. If some
WordStat keywords have no corresponding QDA Miner codes, new codes will be created under a
special codebook category before the autocoding process begins.

The button is used to draw color guidelines on alternate rows in order to facilitate the reading
of large tables. When clicking this button color guidelines are shown. Clicking this button again
removes the color guidelines.

The button allows one to view an automatic suggestion panel displaying leftover words
potentially related to the currently selected item. For more information on the auto-suggest panel,
see Working with the Auto-Suggest Panel on page 42.

The button displays various statistics on the text categorization process, such as the total
number of words processed, the number and proportion of words that have been excluded that
have been categorized. The dialog box also displays document statistics - such as the average word
length of sentences, paragraphs and documents as well as several coverage statistics, including the
percentage of cases, paragraphs and sentences containing at least one keyword and the proportion
of words that have been categorized or included. The coverage statistics are especially useful when
one applies a content analysis dictionary developed to describe a specific data set on new data sets.
A significant decrease in coverage may indicate the need to update a dictionary in order to better
reflect changes over time or specific differences in this new data set.

WordStat Users Manual 39

 The button is used to reapply the content analysis process on the current data set. This button
is disabled by default and becomes enabled when changes are made to any one of the currently
active text analysis processes (such as the categorization dictionary, the exclusion list or the
substitution process). Clicking this button will instruct WordStat to reprocess the text collection
and update the current table.

To export the frequency table to disk:

Click the button. A Save File dialog box will appear.

In the Save as Type list box select the file format in which to save the table. The following formats
are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited file (*.CSV),
SPSS, Stata, HTML file (*.HTM;*.HTML), and Excel spreadsheet file (*.XLS).
Type a valid file name with the proper file extension.
Click the Save button.

To copy the entire table to the clipboard:

Right-click anywhere in the frequency table.
Select the COPY TO CLIPBOARD | TABLE command from the pop-up menu

To copy selected rows to the clipboard:

Select the rows you would like to copy (multiple but separate rows can be selected by clicking
while holding down the CTRL key).
Right-click anywhere in the frequency table.
Select the COPY TO CLIPBOARD | SELECTED ROWS command from the pop-up menu.

To search for a specific item:

Right-click anywhere in the frequency table.
Select the FIND command from the pop-up menu. A search dialog box will appear.
Type the search string in the Find What edit box. To restrict the search to items starting with the
search string, enable the Match Beginning of Item option. To restrict the search to whole words
matching the search string, enable the Match Whole Word Only.
Click the Find button to search the first item matching the typed string. Clicking this button again
finds the next occurrence of the search string, starting at the currently selected item.

Using the Dictionary Panel

The dictionary panel provides an easy way to assign words or phrases to the current categorization
dictionary and to the exclusion or substitution lists. It may also be used to remove or edit existing items.
This panel is located to the left of the Frequencies, Crosstab and Phrase Finder pages and looks like the
one shown below.

40 WordStat Users Manual

The main section of this panel is a tree representing the structure of the current content analysis along
with the substitution and exclusion processes (if active). When an item on this list is selected, its content is
listed in a resizable window below the tree structure. At the top of the panel, a button allows one to undo the
last change made to any one of those lists.

To move items to a list or to an existing content category:

In the table to the right of this panel, click the word or phrase you wish to assign to a list or content
category. To select a group of adjacent entries, move the mouse cursor over the first item in the list,
click and hold the mouse button, drag the mouse to the last entry to highlight the block of rows you
want to assign, and then release the mouse button. To select disjointed items, hold down the CTRL
key while clicking each one of the items.
Once the words or phrases have been highlighted, drag and drop them on the category of your choice.
To add an item to the root folder of a categorization dictionary, simply drop it into the < ROOT >
folder.

To move items to a new category:

Select the words or phrases you would like to assign to this new category.
Drag and drop those items on the item. An dialog box will appear, allowing you to
type the name of this new category.

WordStat Users Manual 41

To rename or delete an existing content category:
In the tree representation of the dictionary, right-click the category you would like to rename or
delete.
Select the appropriate command.

To rename, delete or move an item in a list or in a content category:

In the window below the tree display, right-click the item you would like to modify.
Select the appropriate command.

To cancel a modification:
If you want to undo the last assignment, deletion or modification made using this panel, click the
button. Note: If you leave the mouse cursor over this button, a hint window will appear
showing you which modification will be canceled by clicking this button.

Working with the Auto-Suggest Panel

One of the biggest challenges of quantitative content analysis lies in the fact that a single idea may be
expressed in many different ways, like using synonyms, paraphrases, or idioms. When one needs to identify
all instances of such an idea, a critical task becomes identifying those numerous forms. WordStat provides
several tools to support this task such as the Suggest feature (see page 127) that can be used to retrieve a list
of all known synonyms, related words and inflected forms of the items already in a dictionary from another
source such as a thesaurus or a lexical database. The Auto-Suggest feature is an optional panel on the
Frequencies page that also lists suggestions. It differs however, from the Suggest feature in several ways.
First, it only displays suggested words that were found in the current text collection and that are not already
in the categorization dictionary (leftover words). It may also be used not only on content categories but also
on any word extracted by WordStat and displayed in the Included words and Leftover Words lists. It may
thus be used from the very beginning of the dictionary construction process to quickly identify potential
groupings of words and assign the relevant ones to existing or new content categories.
To display this panel, push down the button. A panel to the right of the frequency table will appear. To
display suggestions, simply select the appropriate row in the frequency grid. When the selected item is a
word, the panel will display synonyms, antonyms, related items and words with a similar beginning
(potentially related items, inflected and misspelled forms). When the selected row consists of a content
category, then the panel displays the same information but for all words currently in this content category.
Selecting more than one row will also result in a compound view of suggestions of all selected items.
At the top of the panel, radio buttons allow you to choose the extent of the suggestions. By default, the
panel returns the most likely synonyms, while related words consist of hypernyms, hyponyms, holonyms
and meronyms. Choosing the More option retrieves other potential synonyms, as well as coordinate terms
and possible attributes of the selected item.

Assigning suggestions to a dictionary

There are two methods to assign suggested words to the categorization dictionary or to the exclusion list.

42 WordStat Users Manual

The first method will perform these operations on items in this panel only, while the second method will
also include selected items in the main frequency list.
To perform one of the above mentioned operations on the suggested items only, select one or several
suggestions, right-click anywhere in this panel and choose the appropriate command.
To include with those suggestions, selected words in the frequency table, click the button on the
tool bar and choose the appropriate command. You may also drag and drop words from the frequency
table to the appropriate location in the Dictionary panel to the left. When dropping items from the
frequency table, all selected suggestions will also be moved to the selected location.

WordStat Users Manual 43

 The Extraction Page
The Extraction page groups tools that may be used to extract useful features from the text collection.

The Topics modeling tool will automatically extract the most important topics from a text collection
using factor analysis. Results may be saved as a content analysis dictionary or may be further
examined using co-occurrence analysis or crosstabulation.

The Phrases extraction feature will identify idioms and common phrases and will allow one to add
them to a content analysis dictionary as well a perform co-occurrence analysis and comparison
analysis of those phrases.

The Named Entities extraction feature can identify proper nouns, names of people, locations or
organizations as well as acronyms. One may then select relevant ones and move them to the
categorization dictionary.

The Misspellings & Unknowns extraction feature provides a tool that identifies misspellings and
some technical terms by comparing the list of word forms encountered in the entire text collection
against a list of common words. Extracted words may be added to the current categorization dictionary
or to a substitution process. They may also be replaced in the original documents with a properly
spelled form.

44 WordStat Users Manual

Extracting Topics
The Topic Extraction feature of WordStat attempts to uncover the hidden thematic structure of a text
collection by applying a combination of natural language processing and statistical analysis. The main
statistical procedure used for topic extraction in Wordstat is a factor analysis. Technically speaking, such an
extraction is achieved by computing a word x document frequency matrix, or alternatively by segmenting
documents into smaller chunks and computing a word x segment frequency matrix. Once this matrix is
obtained, a factor analysis with Varimax rotation is computed in order to extract a small number of factors.
All words with a factor loading higher than a specific criterion are then retrieved as part of the extracted
topic. While in hierarchical cluster analysis, a word may only appear in one cluster, topic modeling using
factor analysis may result in a word being associated with more than one factor, a characteristic that more
realistically represents the polysemous nature of some words as well as the multiplicity of context of word
usages.

The current implementation of the topic modeling procedure has a limit of 2500 words or content
categories. We are working on ways to increase the capability to at least twice this amount). To insure the
stability of the factoring solution, low frequency items should preferably be excluded. It is thus strongly
recommended to remove any word occurring less than 10 times on smaller data sets, ideally less than 30 to
50 times on larger ones. Stemming, lemmatization or the creation of a categorization dictionary may also be
used to group words or phrases, including less frequent ones, prior to the topic extraction.

WordStat provides the following analysis options to control the topic modeling process:

Segmentation - This option allows one to specify whether the data to be used for topic modeling will
be based on the co-occurrence of words in the same document, or whether they will be based on
co-occurrence within paragraphs or sentences. The choice of segmentation should ideally reflect
how topics are being distributed in a typical document and across documents as well as the
objective of the analysis. When the text collection consists of long document containing multiple
topics (such as long political speeches) and one needs to identify all topics in order to compare
their relative frequencies, then performing a segmentation by paragraph or by sentence may be
more sensitive than computing co-occurrences by documents. Alternatively, if one attempts to
differentiate documents by identifying domains or disciplines, or to identify the dominant issue of
documents, then performing the analysis at the document level may be more appropriate. When
analyzing responses to open-ended questions, which may include several topics listed in a single
paragraph, segmenting by sentence may also result in a more precise extraction of the various
topics they contain.

No. of Topics - Setting this option allows one to specify how many topics to extract.

Loading - This option allows one to set the minimum factor loading an item (typically a word) should
reach in order to be retained in the factor solution. By default, this value is set to 0.4. Increasing
the cutoff value will reduce the number of words, keeping only the more representative ones, while
reducing it may include words that are somewhat less characteristic of the extracted topic.

Once the options have been set, click the button to perform the analysis. Please note that
extracting topics on more than a few hundred words can take several minutes. Once extracted, the Topic

WordStat Users Manual 45

Modeling page should looks like this

The table located on the left contains the following information:

NO The factor number. Please note that some factor numbers may be omitted if none of
their items attained the factor loading cutoff criteria. When factors are being merged
by the user, this column displays the numbers of all factors that have been merged
together.
NAME WordStat use an algorithm to automatically provide a label for the extracted topic.
This label may be edited by cliking the button.

KEYWORDS This column list all keywords meeting the factor loading cutoff criteria in descending
order of factor loading.
% VAR This column shows the percentage of variance explained. Please note that the smaller
the segment one chose, the lower the percentage.
FREQ This column displays the total frequency of all items listed in the keywords column.
CASES This column displays the number of cases containing at least one of the items listed in
the keywords column.
% CASES This column displays the percentage of cases with at least one of the items listed in
the keywords column.

46 WordStat Users Manual

Topic modeling buttons:

This button allows you to delete the topic on the selected row. .

Click this button to merge a topic in another one. You first need to select the row containing the
first topic you would like to merge and then click this button. A dialog box will appear with a
list of all other topics. Select the second topic and click OK .

To rename a topic, select it first and then click this button. Type the new name and click OK.

To retrieve segments associated with a topic, select it and click this button. All text segments
containing at least two keywords of the selected topic will be retrieved and presented in a table
format. You may however change both the type of segments retrieved (paragraphs, sentences or
full documents) or the minimum number of topic words needed for retrieval.

This button allows one to perform cooccurrence analysis of all the extracted topics including
clustering and multidimensional scaling, and create proximity plots as well as link charts. For
more information on the various features available, see the Co-Occurrence Page topic.

This button allows one to perform full crosstabulation analysis of all the displayed topics with
structured data, apply statistical analysis and create various charts such as correspondence plots,
heatmaps, bubble charts and bar charts. For more information on the various features available
for crosstabulation analysis, see the Crosstab Page topic.

This button stores the extracted topics currently displayed into a new categorization dictionary
where folders at the first level correspond to different topics, and where each of those folders
contains the associated words. A dialog box allows one to save

Press this button to append a copy of the topic table in the Report Manager. A descriptive title
will be provided automatically. To edit this title or to enter a new one, hold down the SHIFT
keyboard key while clicking this button (for more information on the Report Manager, see the
Report Management Feature topic).

This button allows to store the topic table to disk in various formats, including Excel, tab and
comma delimited files, plain text, HTML, XML, SPSS or Stata files.

Clicking this button allows you to print a copy of the displayed chart

Using the right panel

On the right of this table, a panel allows one to look at the distribution of the selected topic among values of
up to two structured variables. One may chose to display this distribution using either a vertical bar chart, a
horizontal bar chart or a line chart by clicking on the corresponding button. Four statistics may also be
represented on those charts:

WordStat Users Manual 47

 Case Occurrence - number of cases in this subgroup containing at least one of these words.
Category percent - percentage of cases in this subgroup containing at least one of these words.
Word Frequency - total number of these words in this subgroup.
Rate per 10,000 words - rate of words in this subgroup per 10,000 words.
Right-clicking anywhere in the chart areas displays popup menu that allows one to edit the chart, save it to
disk or in the report manager, or copy it to the clipboard. Clicking a specific bar or a data point of a line
chart also allows one to retrieve text segments associated with the selected class and containing words of
the selected topic.

48 WordStat Users Manual

Extracting Phrases
To accurately represent the meaning of a document, it is sometimes not good enough to rely on words alone
but one should look at idioms and phrases. While obtaining comprehensive list of words is easy, finding
common phrases in a specific text corpus is often much more difficult. The phrase finder feature of
WordStat provides such a tool. It will scan an entire text corpus and identify the most frequent phrases and
idioms and allow one to easily add them to the currently active categorization dictionary. In order to reduce
redundancy in such a table, short phrases that are part of larger ones are automatically removed from the
list, provided that their frequency is lower than or equal to the frequency of a longer version.

Before scanning for phrases, one has to set various options that will be used to determine the extent of the
scanning process. The first two options that need to be set are the minimum and maximum number of words
a phrase can have (Min words and Max words). These two values determine both the processing time, the
memory requirement as well as the number of resulting phrases. The larger the range between those
minimum and maximum values, the longer it will take to collect all possible sequences of words. The Min.
Frequency or Min. Cases options allow one to eliminate from the list phrases that appear only a few times
by setting a minimum frequency criterion. When set to Min. Frequency, the criterion specifies the
minimum number of times a phrase must appear regardless of whether it comes from a single document or
from multiple documents. Setting it to Min. Cases allows one to require those occurrences to appear in a
minimum specified number of cases.
When these options have been set properly, click the button to perform the search.
By default, found phrases are presented in descending order of frequency. The Sort By list box can be used
to reorder the obtained list in descending order of frequency, in alphabetical order or in descending order of

WordStat Users Manual 49

phrase length (number of words in the phrase). One can also sort on any column of the table by clicking its
header once to sort the rows in ascending order and a second time to sort these rows on the same column
but in descending order.
To add a phrase or idiom to the currently selected categorization dictionary or to the exclusion dictionary
simply drag it to the proper location in the Dictionary Panel located to the left of the screen (see Working
with the Dictionary Panel). You may also press the button on the tool bar or right-click and select the
desired location.
Please note that because of the processing sequence used by WordStat, a phrase added to a categorization
dictionary may still not be recognized if it starts with a word currently in the exclusion list. When WordStat
encounters a word to be excluded, it automatically ignores further processing and moves to the next word.
For this reason, a phrase should never begin with a word in the exclusion list. However, WordStat will
recognize any other phrases containing but not starting with a word in this list. To prevent the identification
of phrases that would be normally ignored by WordStat, the phrase finder deliberately ignores any phrase
that starts with one of the words found in the active exclusion list. To override this and identify all possible
phrases, go back to the Dictionary page and disable temporarily this exclusion list. If important phrases or
idioms are found to start with words in the exclusion list, it may then be justified to remove those starting
words from the exclusion list or simply disable this list when performing content analysis.
Another important fact that one should keep in mind is that phrases in a dictionary are always treated as a
single token (or unit) and always have precedence over single words or word patterns. This means that if a
word is included in one content category (A) and some phrases containing this word have been added to
another category (B), the word will only be categorized into A if it is not part of a phrase found in B.
This feature is essential to perform disambiguation of words, since it allows one to remove some false
positives associated with a word by identifying phrases associated with those false positives. For example,
if a content category measuring references to money contains the word "BILL", then adding phrases like
"BILL OF RIGHTS" or "BILL CLINTON" to another category will prevent those instances of "BILL"
being categorized as "money". Note: When two phrases start with the same words, longer phrases have
precedence over shorter ones, since they are likely more specific. However, when two phrases partially
overlap, the first one encountered in the text will be categorized, preventing the second one from being
recognized.

Finding overlaps
While WordStat tries to reduce redundancy in the list of phrases by automatically removing short phrases
that are part of larger ones, the resulting list may still contain items that are not independent of each other
such as phrases that sometimes overlap. In order to allow users to take into account potential overlaps when
selecting phrases, WordStat provides a display option that allows one to see when a selected phrase includes
a shorter one, is part of a larger one, or sometimes overlaps other phrases. Such information is especially
useful when one needs to identify idioms that are more specific, often found in longer phrases, or more
generic ones, usually composed of shorter phrases.
To enable the display of information regarding overlaps, simply click the button. A window appears to
the right of the table. Selecting a phrase in the table automatically shows all other items that overlap this
selected item. Each phrase is accompanied by a ratio indicating the total number of times this other phrase
occurs and how many times it overlaps with the selected item. For example, if one selects the phrase I'M
LOOKING FOR in the table showing it occurs 26 times in a document collection, one may notice that it

50 WordStat Users Manual

overlaps with another phrase, LOOKING FOR SOMEONE, with a ratio of 11 out of 12. This suggests that
LOOKING FOR SOMEONE occurs 12 times, but on 11 occasions, both phrases overlap (I'M LOOKING
FOR SOMEONE). This ratio also indicates that on one other occasion, this second phrase occurs without
overlapping the first one. It is also useful to compare the total number of overlaps with the total frequency
of the target phrase. In the above example, we can conclude that the phrase I'M LOOKING FOR - occurring
26 times - is followed by SOMEONE on 11 occasions. Thus, on 15 other occasions, it is followed by
something else.
To hide information about overlaps, click the button again to raise it up.

Assigning overlapping phrases to a dictionary or obtaining a KWIC table

There are two methods of assigning phrases listed in the overlap panel to the categorization dictionary or
the exclusion list or of producing a keyword-in-context table. The first method performs these operations on
items in this panel only, while the second method also includes selected items in the main phrase list. To
perform one of the above-mentioned operations on the overlapping items only, select one or several
overlapping phrases, right-click anywhere in this panel and choose the appropriate command. To include
phrases in the main table, select one or several phrases in the main table, then the overlapping phrases listed
in the overlap panel and then click the button. You may also drag and drop phrases from the main table
to the appropriate location in the Dictionary panel to the left. All selected overlapping phrases will be
added.

Comparing frequencies or case occurrences of phrases

The distribution of a phrase among classes of a categorical variable may be quite useful when choosing
whether or not to include it in a categorization dictionary. For example, one may want to identify phrases
that are typical of some topics in order to better describe them or to differentiate them from other topics.
While the Crosstab page in WordStat is normally used for such a purpose, it can only be used for items
already included in a categorization dictionary or selected by the content analysis process. In other words,
one way to compare the frequency of phrases identified by the phrase finder among classes of a categorical
variable is to move all those phrases to a categorization dictionary, and then use this dictionary to obtain the
cross frequency of those phrases. However, the phrase finder page offers a convenient way to obtain such
information without the need to move those phrases to the categorization dictionary.

To compare frequencies or case occurrences of phrases:

Once phrases have been extracted, click the button. A dialog box similar to the following one will
appear:

WordStat Users Manual 51

The Compare list box shows all categorical variables that are available for comparison. Select the variable
on which the comparison will be performed.
Use the Tabulate list box to specify whether the comparison will be based on the frequency or case
occurrence of those phrases and to specify whether data will be presented using absolute or relative
frequencies. Four options are currently available:

FREQUENCY Total number of times this phrase occurs

CASE OCCURRENCE Total number of cases in which this phrase occurs
FREQ PER 100 cases Number of times this phrases occurs per 100 cases
% OF CASES Percentage of cases in which this phrase occurs

Choose the Statistic that should be used to assess the relationship between the frequency or case occurrence
of the phrases and classes of the categorical variable. The Chi-square is the overall chi-square value
computed on all classes of the categorical variable, while the Max Chi option is the chi-square value
computed on the class with the highest case occurrence or frequency against all the other classes. Select
None if you don't want to display any comparison statistic.
Check the Show highest class option to display a column indicating the label of the class with the highest
relative frequency or case occurrence. In the event that two or more classes obtain the same high
percentage, the cell will list all the labels associated with each of those classes.
Click the OK button to perform the computation.
Once the computation is completed, several additional columns are added to the right side of the table. To
sort rows based on values in any of the newly created columns, click the appropriate column heading.
Clicking several times on the same column heading toggles between ascending and descending order.

Filtering the table

Extracting phrases from a large collection of documents can result in a very large table containing
thousands of phrases. Clicking the button brings a dialog box offering filtering options that allow one
to view only phrases containing either a key word or phrases that are characteristic of a specific class.
Filtering conditions are specified in a dialog box similar to this one:

Enabling the Phrase containing option and entering a string in the edit box allows one to display only
phrases containing the specified string. If a comparison has been performed between classes of a

52 WordStat Users Manual

categorical variable, one may also view phrases that are characteristic of a class by enabling the Scoring
high for option and selecting the value associated with this class. In the above example, both filtering
options were used, restricting the phrases displayed in the table to those containing the string HUMOUR
and found to be characteristic of the 30-39 age group.
To apply the filtering condition, click the Apply button. When a filtering condition is active, the
button is down. To remove filtering conditions and display all extracted phrases, click this button.

Phrase co-occurrences and correspondence analysis

WordStat offers the possibility to perform co-occurrence analysis (clustering, multidimensional scaling,
proximity plot) and correspondence analysis on defined content categories as well as to words meeting
specific frequency criteria. To apply those operations on phrases, one could add them to a user-defined
dictionary, enable this dictionary and then move to the Frequencies or Crosstab page in order to access the
appropriate command. The Phrase Finder page allows one to perform those operations on extracted phrases
without the need to assign them to a dictionary. Special dialog boxes also allow one to add frequent words,
define additional selection criteria or analysis options and save the resulting list of items into a new content
analysis dictionary.

To perform a co-occurrence analysis on phrases:

Click the button. A dialog box similar to this one will appear:

The Processing options are almost identical to those found at the top of the Options page (see page 33),
allowing one to add to the extracted phrases, to single words occurring more than a specific number of
times or in a specific number of cases, or to remove items too frequent or not frequent enough. One may
also restrict the analysis to a specific number of items.
The Saving into a new dictionary options may be used to store phrases and words into a new dictionary
file. Enabling this option and selecting All Phrases will allow the opportunity to store all phrases extracted
using the Phrase Finder page, whether or not they meet the specified frequency criteria. Selecting the other
option will store only phrases meeting those criteria as well as all words that were also included in the
analysis.

WordStat Users Manual 53

For further information on the various tools available for annualizing co-occurrences, see Hierarchical
Clustering and Multidimensional Scaling (page 65-73)

To perform a crosstabulation analysis:

Click the button. A dialog box similar to this one will appear:

The Processing and Saving options are identical to the ones available when performing co-occurrence
analysis on phrases (see previous page). Two additional options are displayed to select the base statistic for
the correspondence analysis (Frequency or Case Occurrence) as well as the categorical variable
containing the classes to be compared.
For more information on crosstabulation, see Crosstab page (page 85)

Other table operations

To copy the entire table to the clipboard:

Right-click anywhere in the frequency table.
Select the COPY TO CLIPBOARD | TABLE command from the pop-up menu.

To copy selected rows to the clipboard:

Select the rows you would like to copy (multiple but separate rows can be selected by clicking while
holding down the CTRL key).
Right-click anywhere in the frequency table.
Select the COPY TO CLIPBOARD | SELECTED ROWS command from the pop-up menu.

54 WordStat Users Manual

To search for a specific item:
Right-click anywhere in the table.
Select the FIND command from the pop-up menu. A search dialog box will appear.
Type the desired search string in the Find What edit box. To restrict the search to items starting with
the search string, enable the Match Beginning of Item option and to restrict it to whole words
matching the search string, enable the Match Whole Word Only option.

Click the Find button to search the first item matching the typed string. Clicking this button again finds the
next occurrence of the search string, starting at the currently selected item.

WordStat Users Manual 55

Extracting Named Entities
The Named Entities extraction tool uses pattern-based algorithms to identify persons, locations,
organizations and acronyms. This feature relies on the existence of mixed cases words and all uppercase
words in text and will thus ignore sentences or paragraphs in full uppercase. This approach to named entity
recognition has the benefit of wording in many languages but may not be appropriate for some of those,
such as German where uppercase letters are used in common nouns. It may also miss some named entities if
some of their significant words are not capitalized. For example, "Ministry of education" will not be
recognized, while "Ministry of Education" will be). The phrases extraction tool may however be able to
retrieve those named entities if they occur more than once.

Clicking the button brings a dialog box with the following three options:
Remove items in the dictionary - This option removes from the final list named entities that are
already in the active categorization dictionary. By default, this option is enabled.
Remove known common words - Enabling this option remove extracted single words entities that are
also common words. This option is especially useful to remove capitalized words in titles or words
in social media data that have been put in full capital typically to stress their importance. It may
however miss the identification of named entities that are also common nouns such as "Apple",
"Windows" or some acronyms.
Minimum frequency - This option allows one to eliminate from the list named entities that appear only
a few times.
When these options have been set properly, click the button to perform the search. Retrieved
items are listed in descending order of frequencies. The Total column indicates the total frequency of this
named entity, while the Unique column reports how many of those do not overlap with other ones.

56 WordStat Users Manual

Adding named entities to a dictionary or obtaining a KWIC table
To add a named entity to the currently selected categorization dictionary or to the exclusion dictionary,
simply drag it to the proper location in the dictionary panel located to the left of the screen (see Using the
Dictionary Panel, page 41). To add them to a new category, drop those on the New Category item. You
may also press the button on the toolbar or right-click and select the desired location.
To produce a keyword-in-context table of a specific named entity, first select it and then click the
button or right-click and select KWIC list.

WordStat Users Manual 57

Extracting Misspellings and Unknown Words
The Misspellings & Unknown feature of WordStat offers a tool that identifies common misspellings by
comparing the list of word forms encountered in the entire text collection against a list of common words.
This feature may also retrieve single words that represent or are part of technical terms, company and
product names, as well as abbreviations. It thus partly overlap the Named Entity extraction tool that will
often provide more complete list of those entities by adding multi-word terms representing those entities as
well as common words with irregular capitalization. When a categorization dictionary is active, this feature
will also attempt to match unknown words to existing items in this dictionary and will present the extracted
words in two separate lists: the Categories list will contain all unknown words that are potentially related to
existing item in your dictionary, presented in a tree view organized by categories while all other unknown
words will be listed on the Others page.
Once extracted, Identified misspellings may be added to the categorization dictionary, to the exclusion list,
or a substitution process. One may also replace those words in the original documents with a proper
spelling.
Several options are available to control how unknown words are extracted and how they are matched to
potential replacements. Clicking the button brings a dialog box similar to this one:

Dictionary - By default, the extraction is performed in reference to common English words (British and
American English). To identify unknown words in documents written in another language or to
exclude technical terms from a specific domain click the language name(s) listed beside the
Dictionary option or modify the settings on the Language page located on the main Options page
of WordStat.:
Remove words already in the categorization dictionary - Some misspellings or uncommon words
may have already been added to an exclusion list or a categorization dictionary. This option
automatically remove those items from the lists of retrieved words.
Find match for words in the categorization dictionary - WordStat will attempt to match unknown
words with existing ones in the language spelling dictionary and identify the most likely

58 WordStat Users Manual

 replacements. Enabling this option will also attempt to match uncommon words in the
categorization dictionary, such as technical terms, proper names, or even misspelled forms.
Confidence of suggestions - This option allows one to adjust the confidence level used by WordStat
when identifying potential replacements. Setting it to Highest, will result in less suggestions by
applying a stricter matching criterion, while setting it to Moderate will provide more suggested
replacements at the risk of increasing the number of irrelevant suggestions.
Min frequency for uncategorized words - This option allows one to eliminate from the Others list,
named entities that appear only a few times allowing you to focus on the most frequent items.
Please note that this setting has no effect on the Categories list, since matches will be reported
even for words appearing only once.
Identify misspelling of words in categorized phrases - For a phrase in a categorization dictionary to
be recognized, all its words must be properly spelled. It is thus important to pay closer attentions to
those misspellings that may result in a failure to recognize such a phrase. Setting this option will
add an additional column in the Others list to indicate whether it matches a word that is part of
one of the phrases in the categorization dictionary.
Batch Replacements Settings - WordStat remembers text replacements made in prior sessions and
can reapply them in batch. Clicking this button brings a dialog that allows one to edit or export this
list of text replacements as well as import such a list created by another user or on another
computer. (see Managing the Batch Replacement List on page 62 for more information).

To Search unknown words, click the button. A dialog similar to this one will appear.

WordStat Users Manual 59

WordStat will retrieve all unknown words and displayed them in two lists located in the middle panel of the
dialog: The Categories list and the Others list. One can move from one list to the other by selecting the
corresponding tab.

The Categories list

If a content analysis dictionary is available and active, WordStat will attempt to match unknown words with
existing items in this dictionary and present them in a tree list view, with the unknown forms stored
underneath the original dictionary form it was matched with. Those suggestions are identified with a ?
symbol and are followed by a numerical value representing their frequency in the current text collection.
Existing items and suggested forms are also grouped under their containing content category. Check boxes
on the left are used to select multiple items in order to perform some operations on them. To select a single
item, put a check mark beside it by clicking its check box. Remove this check mark to unselect it. Selecting
a content category, subcategory or on an existing dictionary item automatically select all items underneath
it. If there is no active categorization dictionary, the Categories page will be disabled.

The Others list

The second list contains all words that could not be matched to an existing item of the content analysis
dictionary, either because no dictionary was currently active or because they were not similar enough to be
matched confidently. On the right of the word, the software shows the most likely word replacement that
has been found, This column will remain empty if no replacement was similar enough. A third column
contains a numerical value representing the frequency of the unknown form in the current text collection.
Finally a fourth column will contains the word "Yes", if the word may be matched to a word that is part of a
phrase in the content dictionary. Such an indication may be useful to identify situations where a phrase may
go undetected because one of its word has been misspelled.

Available operations
Up to four types transformation or processing are allowed on these words: 1) You can replace all instances
of a selected word in the original document by another word or phrase; 2) You can assign the words to an
existing content categories; 3) you can instruct WordStat to automatically replace this word with another
one by adding it to a live substitution process; or 4) you can add this word to a custom list of valid words
causing the program to ignore those words the next time there is a search for vocabulary words. You may
also obtain a keyword-in-context list associated with a specific word in order to decide how that word
should be treated.
Except for the Keyword-in-context list, none of the other four operations are performed immediately.
Instead they are added to an action list allowing you to review, modify or cancel previously defined actions
prior to the application of all the specified changes.

To replace words in the original documents:

Select the word to be replaced.

Click the button. A dialog box similar to this one will appear.

60 WordStat Users Manual

 Type the new replacement word or phrase or choose from the Suggestions list box on the left side of
the dialog box.
Click the OK button to confirm this replacement and add this operation to the list of actions to
perform.

To add a word to a category of your content analysis dictionary:

Select the word you would like to categorize.

Click the button. A dialog box will appear with a list of all categories and
subcategories of the current content analysis dictionary.
Select the category under which this item should be stored and then click OK.
You may add words from this list to the current categorization dictionary or to the exclusion list, or
assign it to the substitution process in order to have it replaced automatically by another word. To
perform any one of those assignments, simply select and drag the item into the proper location in the
dictionary panel to the left of the table (see Using the Dictionary Panel on page 41).

To add a word to an automatic substitution process:

Select the word you would like to add.

Click the button. A dialog box similar to this one will appear.

WordStat Users Manual 61

 Type the new replacement word or phrase or choose from the Suggestions list box on the left side of
the dialog box.
Click the OK button to confirm this substitution and add this operation to the list of actions to
perform.

To add a word to the custom list of words to ignore:

Select the word you would like to add to the custom dictionary.

Click the button.

To remove operations previously defined:

Select the operations that you would like to remove.

Click the button. All words associated with the removed actions are moved
back to the list of unknown words and positioned at the bottom of the list.

To perform all the defined word replacements and categorization:

Click the button.

It is also possible to perform most of those actions or to produce a keyword-in-context table by selecting an
item and either clicking the button or right-clicking.

Managing the Batch Replacement List

WordStat remembers text replacements made in prior sessions and can reapply them in batch. One can edit
entries in this list of replacements, remove items, merge lists by importing lists created by other users or
allow one to export this list to a file. To perform those operations, click the Setup button on the
Misspellings & Unknown page and then click the Batch Replacements Settings button. A dialog box
similar to this one will appear:

62 WordStat Users Manual

To manually add a replacement
Click the Add button. A dialog box like this one will appear:

Type the word to be replaced in the Substitute edit box.

Type the word that will replace it in the With edit box.
Click the Ok button to add the replacement to the existing list.

To edit a replacement
Select the row containing the item you want to edit.
Click the Edit button. A dialog box similar to the one showed above will appear.
Edit any one of the entries in the Substitute or the With edit box.
Click the Ok button to confirm the change.

WordStat Users Manual 63

To delete replacements
Select the rows containing the items you want to delete. You can hold down the CTRL key to
select disjoint rows, or press the SHIFT key to select a range of successive rows.
Click the Delete button.

To import a replacement list

Click the Import button. An Open File dialog box will appear.
Select the file containing the replacements you want to import, and click Open. All replacements in
the imported file will be appended to the current replacement list. Duplicate items won't be
imported, while conflicting items will be showed, allowing you to either keep the existing
replacement or overwrite it with the one in the imported file.

To export a replacement list

Click the Export button. A Save File dialog box will appear.
Type a valid file name and select the location where the file should be stored.
Click the Save button.

The Min Frequency option allows one to eliminate from the list all words appearing only a few times by
setting a minimum frequency criterion.

Once this option has been set, click the button to start searching for vocabulary words. The list of
words retrieved are then listed in a frequency table on the left of the screen and presented in descending
order of frequency. To sort this list in alphabetical order, click the top of the first column.

64 WordStat Users Manual

 Co-occurrences Page
The Co-occurrences page allows one to perform hierarchical cluster analysis and multidimensional scaling
on all keywords and display a dendrogram, a proximity plot, or concept map of those items based on their
co-occurrences. It also allows one to compute the similarity of cases based on keyword use.
The first page of the dialog box is used to specify whether the clustering should be performed on keywords
or on cases and to set various analyses and display options for both types of analysis.

Clustering Cases/Documents
When the clustering is set to be performed on cases or documents, the distance matrix used for
clustering and multidimensional scaling consists of cosine coefficients computed on the relative
frequency of the various keywords. The more similar two documents are in term of the distribution of
keywords, the higher the coefficient. The case label that is used to identify the various cases can be set
by choosing the Edit Case Descriptors command from the WordStat main menu (see page 166).
Clustering keywords
When clustering keywords or content categories, several options are available to define co-occurrence
and choose which similarity index will be computed from the observed co-occurrences.

CO-OCCURRENCE - This option allows you to specify how a co-occurrence will be defined. By
default, a co-occurrence is said to happen every time two words or two categories appear in the same
case (by case option). You may also restrict the definition of co-occurrence to entries that appears in
the same paragraph or the same sentence, or to words or categories that are located in the same user
defined section (delimited by a / character). Finally, you may restrict even further the definition of
co-occurrences by limiting the co-occurrences to a small window of words of specified length. Such
a small window is especially useful when doing an analysis directly on words (rather than categories)
since it allows to identify idioms or phrases that may need to be added to the categorization
dictionary. Co-occurrence on larger text segments such as cases or paragraphs may be more
appropriate to identify the co-occurrence of themes in individual subjects.
INDEX - The Index option lets you select the similarity measure used in clustering and in
multidimensional scaling. Four measures are available. The first three measures are based on the
mere occurrences of specific words or categories in a case and do not take into account their
frequency. In all those indexes, joint absences are excluded from consideration.
Jaccard's coefficient - This coefficient is computed from a fourfold table as a/(a+b+c)
where a represents cases where both items occur, and b and c represent cases where one item is
found but not the other. In this coefficient equal weight is given to matches and non matches.
Sorensen's coefficient - This coefficient (also known as the Dice coefficient) is similar to
Jaccard's but matches are weighted double. Its computing formula is 2a/(2a+b+c) where a
represents cases where both items occur, and b and c represent cases where one item is present
but the other one is absent.
Phi coefficient - This coefficient is a measure of association for two binary variables. It is
similar to the Pearson correlation coefficient in its interpretation.

WordStat Users Manual 65

 .
Cosine theta - This coefficient measure the cosine of the angle between two vectors of values.
It ranges from -1 to +1. This coefficient takes into account not only the presence of a word or
category in a case, but also how often it appears in this case.
Inclusion Index - This coefficient measures the conditional probability that a document that
contains an item X will also contain an item Y. It will take the maximum value of 1 when one of
these items always appear when the second one appear, even if the reverse is not necessarily
true. The inclusion index is optimal for analyzing fields that are organized hierarchically.
Association Strength - This coefficient measures the co-occurrence of items taking into
account the possibility that two items will sometimes co-occur by chance.

Probabilistic - Traditional co-occurrence measures do not take into account the possibility that two
words will sometimes co-occur by chance. As a consequence, clustering solutions obtained using
those metrics are biased toward the formation of clusters of high-frequency items. While the problem
may remain undetected or negligible when clustering low frequency words or when analyzing
co-occurrence within a limited context (such as within a sentence, within a window or within a few
words), the problem becomes much more apparent when clustering broad content categories or
frequently used words. Enabling this option applies a correction to either the Jaccard or the Sorensen
coefficient.

CLUSTERING TYPE - Two broad types of keyword clustering are available. The first method is based
on keyword co-occurrences (First Order Clustering) and will group together words appearing near
each other or in the same document (depending on the selected co-occurrence window). The second
clustering method is based on co-occurrence profiles (Second Order Clustering) and will consider
that two keywords are close to each other, not necessarily because they co-occur but because they
both occur in similar environments. One of the benefits of this clustering method is its ability to
group words that are synonyms or alternate forms of the same word. For example, while TUMOR
and TUMOUR will seldom or never occur together in the same document, second order clustering
may find them to be pretty close because they both co-occur with words like BRAIN or CANCER.
Second order clustering will also group words that are related semantically such as MILK, JUICE,
and WINE because of their propensity to be associated with similar verbs like DRINK or POUR or
nouns like GLASS (for more information, see Grefenstette, 1994).

REMOVE SINGLE WORD CLUSTERS - One way to extract potentially interesting knowledge from
dendrograms is to focus on the aggregation of items at an early stage of the clustering process.
However, when clustering hundreds or thousands of items, the identification of those items requires
the user to scroll through a very long dendrogram which includes many clusters of isolated items.
Enabling this option simplifies the use of the dendrogram by hiding all single item clusters and
allowing one to concentrate only on the strongest associations. Setting this option also removes
isolated items from multi-dimensional scaling plots, greatly enhancing their value when analyzing a
large number of items. Please note, however, that when this option is enabled, changing the number
of clusters while viewing a 2-D or 3-D MDS plot will cause the program to recompute the distance
and location of remaining items.

66 WordStat Users Manual

REAL TIME ANIMATION - When this option is enabled, the multidimensional plots are updated after
every iteration allowing the user to monitor the progress made during the analysis at the cost of
higher computing time.

TOLERANCE - This option specifies the tolerance factor that is used to determine when the algorithm
has converged to a solution. Reducing the tolerance value may produce a slightly more accurate
result but will increase the number of iterations and the running time.

MAXIMUM ITERATIONS - This option allows one to specify the maximum number of iterations that
are to be performed during the fitting procedure. If the solution does not converge to the limit
specified by the TOLERANCE option before the maximum number of iterations is reached, the
process is stopped and the results are displayed.

INITIAL CONFIGURATION - This option allows one to specify whether the multidimensional scaling
will be applied on a random configuration of points or on the result of a classical scaling.
Selecting the Classical Scaling option instructs WordStat to perform a classical scaling first on the
similarity matrix, and then use the derived configuration as initial values for the ordinal
multidimensional scaling analysis.
Selecting the Randomized Location option instructs WordStat to perform the multidimensional
scaling analysis on a random configuration of points. By default, WordStat initializes the random
routine before each analysis with a new random value. The seed value used for the creation of this
initial configuration is stored along with the final stress value in the history list box, located at the
bottom of the dialog box. The Seed option may be used to specify a starting number that will be
used to initialize the randomization process and produce a fixed random sequence. To recall a
specific seed value used previously, double-click the proper line in the history list box.

WordStat Users Manual 67

Dendrogram
WordStat uses an average-linkage hierarchical clustering method to create clusters from a similarity
matrix. The result is presented in the form of a dendrogram (see below), also known as a tree graph. In
such a graph, the vertical axis is made up of the items and the horizontal axis represents the clusters
formed at each step of the clustering procedure. Words or categories that tend to appear together are
combined at an early stage while those that are independent from one another or those that don't appear
together tend to be combined at the end of the agglomeration process.

NO CLUSTERS - This option allows you to set how many clusters the clustering solution should have.
Different colors are used both in the dendrogram and in the 2-D and 3-D maps to indicate
membership of specific items to different clusters. However, if the option to remove single item
clusters is enabled, an increase in the number of clusters may in fact result in a decrease in the
number of clusters displayed and in the overall height of the dendrogram since all single item clusters
will be hidden.
DISPLAY - This option lets one choose whether the vertical lines of the dendrogram represent the
agglomeration schedule or the similarity indices.

When clustering keywords or content categories, clicking this button displays bars beside
each dendrogram item to represent their relative frequencies.
Use this button to increase the dendrogram font size and focus on a smaller portion of the
tree.
Use this button to reduce the dendrogram font size and view a larger portion of the tree.
This button allows one to perform full crosstabulation analysis with structured data, apply
statistical analysis and create various charts such as correspondence plots, heatmaps, bubble
charts and bar charts. A dialog box allows one to restrict the analysis to specific clusters
containing a minimum number of items and cluster names are automatically generated using
characteristic words and phrases of each cluster. For more information on the various
features available for crosstabulation analysis, see the Crosstab Page topic on page 85.
This button stores the cluster solution currently displayed into a new categorization
dictionary where folders at the first level correspond to different clusters, and where each of

68 WordStat Users Manual

 those folders contains the associated words or expressions. A dialog box allows one to save
only clusters containing a minimum number of items. Cluster names are automatically
generated using characteristic words and phrases of each cluster. One may then edit those
cluster names and their content from the Dictionary Page.

Press this button to append a copy of the graphic in the Report Manager. A descriptive title
will be provided automatically. To edit this title or to enter a new one, hold down the SHIFT
keyboard key while clicking this button (for more information on the Report Manager, see
page 191).
This button allows storing the displayed dendrogram into a graphic file. WordStat supports
three different file formats: .BMP (Windows bitmap files), .PNG (Portable Network Graphic
compress files) and .JPG (JPEG compressed files).
To retrieve text segments or documents associated with a specific cluster, click anywhere on
a cluster to select it (the selected cluster is displayed using thicker black lines), and then click
this button to retrieve the associated documents. When performing first order clustering on
keywords, this operation retrieves all text segments containing at least two keywords of the
selected cluster .When performing second order clustering of keywords, all text segments
containing a single one of those keywords will be retrieved.

The slide ruler provides another way of quickly changing the number of clusters included in
the clustering solution. Moving the slider to the left increases the minimum distance required
to form a cluster and thus produces a dendrogram with more clusters. Moving the slider to
the right aggregates smaller clusters into bigger ones. However, if the option to remove
single-item clusters is enabled, an increase in the number of clusters may, in fact, result in a
decrease in the number of clusters displayed and in the overall height of the dendrogram.
Note: Clustering using other similarity or distance measures or agglomeration methods may be achieved
using the MVSP cluster analysis procedure (see Performing Multivariate Analysis, page 189).

Using the right panel

On the right of this table, a panel allows one to look at the distribution of the selected topic among values of
up to two structured variables as well a network graph representing the relationship between all items in the
selected cluster.
For the first two panel, one may chose to display this distribution of this cluster using either a vertical bar
chart, a horizontal bar chart or a line chart by clicking on the corresponding button. Four statistics may also
be represented on those charts:
Case Occurrence - number of cases in this subgroup containing at least one of these words.
Category percent - percentage of cases in this subgroup containing at least one of these
words.
Word Frequency - total number of these words in this subgroup.
Rate per 10,000 words - rate of words in this subgroup per 10,000 words.

WordStat Users Manual 69

Right-clicking anywhere in the chart areas displays popup menu that allows one to edit the chart, save it to
disk or in the report manager, or copy it to the clipboard. Clicking a specific bar or a data point of a line
chart allows one to retrieve text segments associated with the selected class and containing words of the
selected cluster.
In the lower right panel, relationship between words within the selected cluster are represented using a
network type graph. The layout of the nodes can be based on a multidimensional scaling analysis, a force
based graph, or may be spread in circle. A track bar allows one to hide the weaker connections, while other
buttons may be used to zoom in and out.

Clicking the button allows one to jump to the full size Link Analysis page (page 74).

70 WordStat Users Manual

Mapping
The concept maps are graphic representations of the proximity values computed on all included keywords
using multidimensional scaling (MDS). In those maps, a point represents an item (keyword or content
category) and the distances between pairs of items indicate how likely those items are to appear together. In
other words, items that appear close together on the plot usually tend to occur together, while words or
categories that are independent from one other or that don't appear together are located on the chart far from
each other. Colors are used to represent membership of specific items to different partitions created using
hierarchical clustering. An option also allows one to vary the size of each data point in order to take into
account the observed frequency of each item. The resulting maps are useful to detect meaningful underlying
dimensions that may explain observed similarities between items.
Please note that since multidimensional scaling attempts to represent the various points into a two- or
three-dimensional space, some distortion may result, especially when this analysis is performed on a large
number of items. As a consequence, some items that tend to appear together or that are parts of the same
cluster may still be plotted far from each other. Also, performing a multidimensional scaling on a large
number of items usually produces a cluttered map that is hard to interpret. For these reasons, interpretation
of concept maps may be feasible only when applied to a relatively limited number of items.

2-D and 3-D Map buttons and controls

NO CLUSTERS - This option allows the setting of the number of clusters that the clustering solution
should have. Different colors are used both in the dendrogram and in the 2-D and 3-D maps to
indicate membership of specific items to different clusters. The slide ruler located on the top toolbar
of the dialog box may also be used to quickly change the number of clusters. Please note that when
the Remove single word clusters option is enabled, changing the number of clusters either way
often causes the program to recompute MDS maps to take into account the different number of items
displayed.
The actual orientation of axes in the final solution is arbitrary. The map may be rotated
in any way we want, as long as the distances between items remain the same. The
rotating knob can be used to adjust the final orientation of axes in the plane or space in
order to obtain an orientation than can be most easily interpreted.

Clicking this button enables to zoom in a plot. To zoom an area of the plot, hold the left
mouse button and drag the mouse down/right. You'll see a rectangle around the selected
area. Release the left mouse button to zoom.

Clicking this button restores the original viewing area of the plot.

Clicking this button performs another multidimensional scaling on a new random

configuration of points. This button is visible only when the initial configuration is set
to Random Location.

This button is used to create a copy of the chart to the clipboard. When this button is
clicked, a pop-up menu appears, allowing you to select whether the chart should be
copied as a bitmap or as a metafile.

WordStat Users Manual 71

 This button allows editing of various features of multidimensional scaling plots such as
the appearance of value labels and data points, the chart and axis titles, the location of
the legend, etc. (see Multidimensional Scaling Plot Options)

Pressing down this button creates a constrained multidimensional scaling. This mapping
algorithm allows one to preserve the clustering structure in multidimensional scaling
plots, making the interpretation of 2-D and 3-D MDS maps a lot easier and more
consistent with the clustering solutions. Enabling this option allows one to use the MDS
module to create maps of concepts similar to those suggested by Trochim, in its
Concept Mapping procedure.

Pressing down this button displays lines to represent relationships between data points
of the multidimensional scaling plot. When the button is down, a cursor will appear in a
tool panel below the plot, allowing you to select the minimum association strengths to
be displayed.

Clicking this button creates a bubble plot where the areas of data points are proportional
to the relative frequency of those items. This type of display is especially useful when
one needs to take into account a third variable, in this specific case the frequency of
items, when interpreting the distance between data points.

Press this button to append a copy of the graphic in the Report Manager. A descriptive
title will be provided automatically. To edit this title or to enter a new one, hold down
the SHIFT keyboard key while clicking this button (for more information on the Report
Manager, see page 191).

This button allows storing the displayed multidimensional scaling plot into a graphic
file. WordStat supports four different file formats: .BMP (Windows bitmap files), .PNG
(Portable Network Graphic compress files) and .JPG (JPEG compressed files) as well
as .WSX a proprietary file format (WordStat Chart file). Charts stored in the latter
format may be opened, further edited and customized using the Chart Editor external
utility program.

Clicking this button allows you to print a copy of the displayed chart.

3-D Map buttons

This button can be used to show or hide left, bottom and back walls.

Clicking this button exchanges the data of the X, Y and Z axis.

Clicking this button draws anchor lines from the floor to the data point to better locate
data points in all 3 dimensions.

Clicking this button allows you to change the viewing angle of the chart. To rotate the
chart, make sure this button is selected, click any area of the chart, hold the mouse
button and drag the mouse to apply the desired rotation.

72 WordStat Users Manual

Locating a data point on the depth dimension of a 3-D plot can be very difficult especially
when the plot remains static. One often has to rotate this plot constantly on the various axes
to get an accurate idea of where the data point is located on this third axis. Clicking this
button forces WordStat to rotate the plot automatically. To disable the automatic rotation,
click a second time.

WordStat Users Manual 73

Link Analysis
The link analysis feature allows one to visualize the connections between keywords or dictionary items
using a network graph. It offers a high level of interactivity, allowing you to explore relationships as well as
detect underlying patterns and structures of co-occurrences using three layout types: a multidimensional
scaling, a force-based graph, and a circular layout.
This page is initially associated with the clustering features and the dendrogram page, such that selecting a
specific cluster in the dendrogram will result in a network view of its elements where each element is
represented as a node, while their relationship will be represented as a line connecting those node (also
called an "edge"), the thickness of this line representing the strength of this relationship. Nodes and edges
can be edited, deleted or moved, and new nodes and edges may also be added, either manually or using
statistical criteria.

Layout selection
The following three buttons allow you to select one of the three layout types.

Clicking this button assigns the location of nodes in a multidimensional space such that nodes
that co-occur more often are plotted close together, while those co-occurring less often are
plotted far from each other.

74 WordStat Users Manual

 Clicking this button draw a force-directed graph in which links are more of less of equal
length and there are as few crossing links as possible.

Clicking this button draws nodes in a circle. Nodes that co-occur often are plotted close
together.

Links and Node Selection

Clicking this button allows you to select nodes and links, either manually, or based on
statistical criteria. Please see the section Selecting Nodes and Links below for more
information on this features

This track bar can be used to hide links. By default, the cursor is positioned to the right.
Moving it to the left gradually remove the weakest links, allowing you to identify more
clearly the strongest ones.

By default, a graph includes links reaching a specific statistical threshold. Clicking this
button add additional links by gradually reducing this statistical threshold.

Clicking this button remove all links that have been hidden using the track bar control (see
above), delete all nodes that are no longer connected, and refresh the display to take into
account the lower number of elements in the graph.

Navigation mode
The following three buttons allow you to select the default mouse cursor behavior.

Click this button down to select specific nodes and links. To select multiple items, hold the
CTRL key down while clicking on additional items or select the rectangular region of the
graph containing the items you want to select.
To select a node along with all nodes connected to it (neighbors), hold the ALT key while
click the node. One may also select a node, right-click and choose GET NEIGHBORS.
Once items are selected, they can then be moved, deleted or edited. One may also search for
associated text segments either by right-clicking to call a contextual menu or by clicking the
button on the toolbar.
Click this button down lets you control which part of the image is visible in the image
window.

Click this button down allows you to zoom in a specific region. Once this button is down,
click the upper left corner of the rectangular region you want to zoom in, drag the mouse
down the lower right corner of this region and then release the mouse button.

Zooming in and out

Clicking this button zooms in or out so you can see the entire map.

WordStat Users Manual 75

 Click this button to increase the zoom level of the map.

Click this button to decrease the zoom level of the map.

This list box let you set the zoom level to predefined levels. One can also type in the desired
zoom level.

Editing buttons
This buttons allows you to display or hide the numerical value associated with links.

This button allows you to change the size of nodes, links and their associated text. When
clicked, a dialog box with four track bars appear. Moving their cursor to the right increases
the size of its associated element while moving it to the left decreases it size.
Clicking this button delete the selected nodes and links.

Clicking this button brings a dialog box that allows you to change the style, the color and the
size of the selected nodes and links.
To retrieve segments associated with a topic, select it and click this button. When more than
one node or when a link is selected, all text segments containing at least two keywords will
be retrieved and presented in a table format. You may however change the type of segments
retrieved (paragraphs, sentences or full documents) or the minimum number of topic items
needed for retrieval.
This button is used to create a copy of the graph to the clipboard. When this button is
clicked, a pop-up menu appears, allowing one to select whether the graph should be copied
as a bitmap or as a metafile.
Press this button to append a copy of the graph in the Report Manager. A descriptive title
will be provided automatically. To edit this title or to enter a new one, hold down the SHIFT
keyboard key while clicking this button (for more information on the Report Manager, see
the Report Management Feature topic, page 191).
This button allows storing the displayed multidimensional scaling plot into a graphic file.
WordStat supports four different file formats: .BMP (Windows bitmap files), .PNG (Portable
Network Graphic compress files) and .JPG (JPeg compressed files).
Clicking this button allows you to print a copy of the displayed graph.

Selecting nodes and links

There are four methods available to select nodes to be plotted:
1. Selecting a cluster in the Dendrogram page
The link analysis graph is initially connected to the dendrogram page, so that selecting one of its
clusters results in its items being represented as nodes on the link graph. The similarity criterion is
also automatically set to the value connecting the last item of this cluster to the other elements. To
map connections between items of another cluster, simply move back to the Dendrogram page and

76 WordStat Users Manual

 click anywhere on the new cluster to select it and plots it. A smaller version of the link map should
appear in the lower right panel of the dendrogram page.You may then move back to the Link
Analysis page, or click the button in this lower right panel.

2. Double-clicking a node
When the graph is in selection mode, double-clicking a node will set the node being clicked on as the
target item and display all nodes associated with it according to the last criterion settings (primary
links). All currently displayed nodes not associated with this new target item will be removed.
If the shift key is being hold down while double-clicking a node, the graph will also include nodes
associated to a second degree to the target node (secondary links) as long as they also reach the same
criterion.

3. Selecting nodes and right-clicking

When one or several nodes are selected, right-clicking brings a popup menu that allows one to either
ADD to the current graph, primary and secondary links or REPLACE the currently displayed nodes
with those associated directly (one level) or indirectly (second level) with the selected nodes.

4. Using the nodes and link filtering dialog box

Clicking the button on the toolbar brings a dialog box that allows one to manually select items
to be graphed. This dialog box looks like the one below:

WordStat Users Manual 77

The following groups of options are available for selecting nodes and links to be plotted:
Source List This list box contains all available items that are available to be plotted. To
include them in a graph, simply move them to the Destination List using the
arrow button. A search edit box at the top of this list allows one to quickly find
specific items.
Destination List This list box contains items that are currently plotted. When the Show Target
Items Only option is checked,the dialog box will offer the possibility to select
additional nodes and links based on similarity or proximity criteria. When not
checked, the graph will display the connection of only the nodes currently
listed. Arrow buttons may be used to remove existing or add additional items.
Selection level When the destination list is set to include only target items, one may choose
whether additional nodes should be plotted and how those will be selected.
When set to Target Only, no additional items will be plotted,. Selecting
Primary Links will allow the selection of additional nodes directly linked to
target items based on a similarity criterion. When the Secondary Links is
chosen, additional items may be included if they are indirectly connected to the
target nodes through the primary links, as long as they reach the same similarity
criteria. By default, all links reaching such a criteria will be plotted, including
links among the primary and secondary linked nodes. Selecting Direct Links
Only will omit most of those links to keep only the ones that connect them to
the target items.
Node Selection This set of options allows you to adjust the similarity criterion to be used for
selecting additional items. The similarity criteria displayed is the one currently
set on the Options page of the Co-Occurrence section (see page 65). One can
type a specific value, typically between 0 and 1 in the edit box, increase or
decrease the value using either the up and down arrow buttons, or by sliding the
cursor on the track bar on the right of the edit box. The Number of Nodes
option is automatically adjusted to indicate how many items meet the current
similarity criterion. Editing its value will reciprocally adjust the similarity
criterion.
Link Selection This set of options allows you to adjust the similarity criterion to be used for
selecting which links will be plotted. The similarity criterion for displaying
links should always be equal or lower than the one used for selecting nodes.
Reducing this criterion will plot additional links. One can type a specific value
in the edit box, increase or decrease this value using either the up and down
arrow buttons, or by sliding the cursor on the track bar on the right of the edit
box. The Number of Links option is automatically adjusted to indicate how
many links will be plotted. Editing its value will reciprocally adjust the
similarity criterion.

Once the options have been set, click the to update the graph and display the selected
nodes and links.

78 WordStat Users Manual

WordStat Users Manual 79
Proximity plot
Cluster analysis and multidimensional scaling are both data reduction techniques and may not accurately
represent the true proximity of keywords or cases to each other. In a dendrogram, while keywords that
co-occurs or cases that are similar tends to appear near each other, one cannot really look at the sequence of
keywords as a linear representation of those distances. One has to remember that a dendrogram only
specifies the temporal order of the branching sequence. Consequently, any cluster can be rotated around
each internal branch on the tree without in any way affecting the meaning of the dendrogram. The best
analogy here is to think of a Calder mobile. Different photos of such a mobile will yield different distances
between hanging objects. While multidimensional scaling is a more accurate representation of the distance
between objects, the fact that it attempts to represent the various points into a two- or three-dimensional
space may result in distortion. As a consequence, some items that tend to appear together or be very similar
may still be plotted far from each other.
The proximity plot is the most accurate way to graphically represent the distance between objects by
displaying the measured from one or several target objects to all other objects. It is not a data reduction
technique but a visualization tool to help one extracts information from the huge amount of data stored in
the distance matrix at the origin of the dendrogram and the multidimensional scaling plots. In this plot, all
measured distances are represented by the distance from the left edge of the plot. The closer an object is to
the selected one, the closer it will be to the left.

To select a keyword or a case that will be used as the point of reference, one can choose from the
KEYWORD or CASE drop down check list located at the top of the page. One can also freely browse
through different keywords or cases by double-clicking its bar in the Proximity Plot. The co-occurrence or
similarity to more than one target item may be displayed in a single chart allowing easy comparisons. When
several target items are selected, the proximity plot may consist of bars clustered side by side (clustered
bars), or stacked, representing either the total amount (stacked bars), or the relative distribution of scores

80 WordStat Users Manual

(100 percent stacked bars). When two target items are selected, it is also possible to display the bars on both
sides of a central axis like the sample chart above (mirrored bars).
By default, the chart displays the proximity of the target items to a maximum of 30 related items. Clicking
the button located in the upper left-hand corner of the chart, displays a dialog box that allows one to
either manually choose items to be plotted, or to automatically select a specific number of items.

When looking at keyword co-occurrences, selecting a bar enables the button. Clicking this button
retrieves every document or text segment containing both keywords, allowing one to further explore the
factors that may explain this co-occurrence. When examining the similarity of documents rather than
keywords, clicking this button retrieves both documents and displays them side by side in a dialog box.
Right-clicking any existing bar displays a menu that allows one to remove the selected item, move it to the
list of target items either by adding it to the existing bars or replacing one of them. One may also retrieve
documents or text segment using this popup menu.
The Table page allows one to examine in more detail the numerical values behind the computation of those
plots. When the distance measure is based on co-occurrences, the table provides detailed information, such
as the number of times a given keyword co-occurs with another one (CO-OCCURS) and the number of
times it appears in the absence of this selected keyword (DO NOT). Such a table also includes the number
of time the selected keyword appears in the absence of the given keyword (IS ABSENT). In the example
below (computed using the paragraph as the frequency criteria), we can see on the highlighted line that the
word MILITARY co-occurs 107 times with IRAQ, but this word is encountered in 285 paragraphs without
the word IRAQ, while IRAQ is found in 1,182 paragraphs in the absence of MILITARY. The Jaccard
coefficient of 0.109 indicates that of all paragraphs containing either one of these words, 10.9 percent
contains both words. Note, however, that not all proximity measures can be interpreted as easily. To
facilitate the interpretation of this table, the status bar provides a textual interpretation of some of the
statistics.

WordStat Users Manual 81

The following list provides a brief description of the buttons found on these two pages

Proximity plot controls

By default items in the proximity plot are sorted in descending order of proximity scores.
Clicking this button sorts items in alphabetical order.
This button is used to create a copy of the chart to the clipboard. When this button is clicked,
a pop-up menu appears, allowing one to select whether the chart should be copied as a
bitmap or as a metafile.
This button allows editing of various features of the proximity plot, such as the appearance of
value labels and bars, the chart and axis titles, and the location of the legend.

Proximity plot and proximity table controls

Press this button to append a copy of the chart or the proximity table in the Report Manager.
A descriptive title will be provided automatically. To edit this title or to enter a new one, hold
down the SHIFT keyboard key while clicking this button (for more information on the
Report Manager, see page 191).
When the proximity plot is displayed, this button allows storing the chart on disk in one of
the supported graphic file formats. When the proximity table is shown, the table can be saved
to disk in an Excel, text delimited, XML, HTML, SPSS, or Stata file.
Clicking this button allows you to print a copy of the displayed chart or table.

82 WordStat Users Manual

Statistics page
The Statistics page displays the co-occurrence and similarity matrices used for building the dendrogram,
MDS plots, as well as the proximity plot and table.

The first two pages displays by default only the lower triangular part of the matrix of co-occurrence and
similarity. Selecting the Full Matrix option will display data on both sides of the diagonal.

To export a matrix to social network analysis software tools:

Select the matrix you would like to import by activating either the co-occurrences or the similarity
page.
Click the button.
In the Save As Type list box, select the file format under which to save the table. The following
formats are supported: UCINET file (*.DL), Pajek Network file (*.NET), NetDraw file (*.VNA) and
NetMiner file (*.SNF).
Type a valid file name with the proper file extension.
Click the SAVE button.

WordStat Users Manual 83

To append a copy of the table in the Report Manager:

Click the button. A descriptive title will be provided automatically for the table. To edit this
title or to enter a new one, hold down the SHIFT keyboard key while clicking this button.
For more information on the Report Manager, see the Report Management Feature (page 191).

To export the table to disk:

Click the button. A Save File dialog box will appear.
In the Save As Type list box, select the file format under which to save the table. The following
formats are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited file
(*.CSV), MS Word (*.DOC), HTML file (*.HTM; *.HTML), XML files (*.XML) and Excel
spreadsheet file (*.XLS).
Type a valid file name with the proper file extension.
Click the SAVE button.

To print the table:

Click the button.

84 WordStat Users Manual

 Crosstab Page
The Crosstab page is used to display a contingency table of words or categories. This contingency table is
computed only on items that have been included. If an inclusion dictionary has been specified, this grid will
display only the words or keywords in this list. If no inclusion list has been specified, the grid will display
all words that have not been explicitly excluded. Along with absolute and relative frequency of keyword
occurrence or keyword frequency, several statistics may be displayed to assess the relationship between
independent variables and word usage or to assess the reliability of coding made by several human coders
or a single coder at different times.

Options
TABULATE - The TABULATE option allows choosing whether the values in the table should be based
on the total frequency of keywords or the number of cases containing those keywords.
WITH - The WITH drop down list allows choices on how the keyword count should be broken down.
The following options are available:
<other keywords> - display a square table showing the number of co-occurrence of words in the
same case.
<case number> - display the keyword occurrence or frequency for each individual cases.
ANY INDEPENDENT VARIABLE - If numeric, categorical or date variables were selected as
independent variables, their names will appear in this list box. Selecting any of those variable
names will display a contingency table allowing for the assessment of the relationship between this
variable and the keywords or content categories. When a date variable is selected, a dialog box like
the one below will appear allowing one to automatically recode those dates into various time
periods or date units like week days or months.

WordStat Users Manual 85

 <variables> - When the content analysis is performed on several alphanumeric variables, this
option will allow for comparison of the occurrence of words according to variable name.
Select variables... By default, the variables listed in the With list box are the independent
variables that have been selected when calling WordStat. Choosing this option displays a dialog
box that allows one to select other numeric, categorical or date variables.
Combine variables... This option allows one to compare the frequency of keywords or content
categories among the combined values of two variables. When this item is selected, a dialog box
appears allowing one to choose the two variables whose values will be combined.

SORT BY - The SORT BY option presents the opportunity to sort the table by keyword or category
names (alphabetical order) or by descending order of frequency or case occurrence. When a statistic
is displayed (see option STATISTICS), the table can also be sorted based on the value of this statistic
or on its statistical probability. It is also possible to sort on the values of any specific column by
clicking this column heading. Clicking several times of the same column heading toggles between
ascending and descending sort orders.

DISPLAY - The DISPLAY list box allows one to specify the information displayed in the table. The
following options are available:
Count
Row percent
Column percent
Total percent
When the TABULATE option is set to Case Occurrence, two additional statistics are also available:
Percent of cases (percentage of all cases or individuals)
Category percent (percentage of cases or individuals in this subgroup)

STATISTIC - When keyword frequency or occurrence is broken down by an independent variable (see
WITH option), a drop down list box will appear. This list box allows one to choose among 12
association measures to assess the relationship between this independent variable and the utilization
of each word or category.

86 WordStat Users Manual

 Nominal level statistics
Chi-square
Likelihood ratio
Student's F
Ordinal or internal level statistics
Tau-a
Tau-b
Tau-c
Somers' D (symmetric)
Somers' Dxy (asymmetric)
Somers' Dyx (asymmetric)
Gamma
Spearman's Rho
Pearson's R

PROBABILITY - The probability option allows one to select whether the probability value should be
computed using a 1-tailed or 2-tailed test. Probabilities of Chi-square, Likelihood ratio, and Student's
F are always computed using a 2-tailed test.

AGREEMENT - When comparing word or category usage between different alphanumeric variables, a
drop down list box will appear. This list box allows one to choose among 8 different inter-rater
agreement measures to assess the reliability of coding.

Nominal level agreement statistics

Percentage of agreement
Cohen's Kappa
Scott's pi
Free marginal (nominal)
Ordinal or internal level agreement statistics
Krippendorf's R
Krippendorf's r-bar
Free marginal (nominal)
Intraclass correlation
For more information on how to assess reliability with those statistics see Computing Inter-Rater
Agreement Statistics (page 178).
To display column labels vertically, click the button located to the right of the Tabulate With list box.

The button is used to reapply the content analysis process to the current data set. This button is
disabled by default and becomes enabled when changes are made to any one of the currently active text
analysis processes, such as the categorization dictionary, the exclusion list or the substitution process.
Clicking this button will instruct WordStat to reprocess the text collection and update the current table.

WordStat Users Manual 87

Creating barcharts or line charts
The Crosstab page also allows one to produce barcharts or line charts to visually compare the
distribution of specific words or categories among values of an independent variable such as subgroups
of individuals (male vs. female) or time periods. To produce such charts:

Set the TABULATE and DISPLAY options so that the information you want to visualize is
displayed in the table.
Using the mouse, select the rows you would like to display. Multiple disjoint rows can be
selected through clicking while holding down the CTRL key.
Click the button or press the right button of the mouse and select the Chart Selected Rows
command.
A dialog box like this one will appear:

This window allows one to graphically examine the relationship between codes and values of an
independent variable. The bar chart should preferably be used to display the distribution of various
categories within subgroups as defined by a nominal independent variable, while the line chart should
preferably be used to examine the relationship between those categories and an ordinal or quantitative
variable.
A quick way to retrieve documents, paragraphs or sentences associated with a specific bar or pie slice
is by right-clicking and selecting the keyword retrieval command.

88 WordStat Users Manual

 The following table provides a short description of available buttons and controls:

Controls Description

Press this button to vertically display the labels on the bottom axis

Press this button to append a copy of the graphic in the Report Manager. A
descriptive title will be provided automatically. To edit this title or to enter a new
one, hold down the SHIFT keyboard key while clicking this button. For more
information on the Report Manager, see the Report Management Feature (page
191).

Press this button to retrieve a chart previously saved on disk.

Press this button to save a chart on disk. Charts are saved in a proprietary format and
may be edited and customized using the Chart Editor.

Pressing this button allows you to print a copy of the displayed chart.

Pressing this button causes the values represented on the bottom axis to be exchanged
with those of represented by different lines or bars (legend).

Click this button to turn on/off the 3-D perspective for the current chart.

This button allows you to edit various features of the chart such as the left and bottom
axis , the chart and axis titles, the location of the legend, etc.

This button is used to create a copy of the chart to the clipboard. When this button is
clicked, a pop-up menu appears allowing you to select whether the chart should be
copied as a bitmap or as a metafile.

Pressing this button closes the chart dialog box and returns to WordStat's main
screen.

Customizing barcharts and line charts

Clicking the button on the chart dialog box gives access to a dialog box to customize the
appearance of barcharts and line charts. The options available in this dialog box represent only a small
portion of all settings available.

To further customize the chart, modify data points, value labels, or series order, click the button
located to the right-hand side of the dialog box.

WordStat Users Manual 89

LEFT AXIS
Minimum / Maximum - WordStat automatically adjusts the vertical axis scale to fit the range of values
plotted against it. To manually set these values, type the desired minimum and maximum.
Increment - Increasing or decreasing this value affects the distance between numbers as well as tick
marks. Horizontal grid lines are also affected by modification of this value.
Horizontal Grid - This option turns horizontal grid lines on and off. Grid lines extend from each tick
mark on an axis to the opposite side of the graph. To increase or decrease the number of grid lines or
the distance between those lines, change the Increment value of the axis. A list box also allows a
choice among five different line styles to draw those grid lines.

LEGEND
Location - This option positions the legend. Legends may be placed at Top, Left, Right and Bottom
side of the chart.
From top - When the legend is displayed on the left or the right side of the chart, this option specifies
the legend's top position in percent of total chart height.
From left - When the legend is displayed on the top or the bottom chart, this option specifies the
legend's top position in percent of total chart width.

TITLES
Proper titles and axis labels are of utmost importance when describing the information displayed in a
chart. By default, WordStat uses variable names and labels as well as other predefined settings to
provide such descriptions.
The title page allows one to modify the top title, as well as the labels on the left, bottom and right axis.
To edit the title, select the proper radio button. Enter several lines of text for each title by pressing the
<Enter> key at the end of a line before entering the next line.
The Font button to the right-hand side of the edit box allows changing the font size or style of the
related title.

3-D VIEW
Orthogonal - Turning this option off disables the free elevation and rotation of the 3-D chart.

Zoom - This option zooms the whole chart. Expressed as a percentage, increasing the value positively
will bring the chart towards the viewer, increasing the overall chart size as the Zoom value increases.
3-D Percent - The 3-D Percent property indicates the size ratio between chart dimensions and chart
depth by specifying a percent number from 1 to 100.
Perspective - Use this property with Orthogonal unchecked to modify the 3-D perspective of the
Chart. Larger values add more depth perspective.
Bar shadow - Enabling this option adds dark shades to the sides of 3-D bars. Turning it off will color
the sides of the bar the same as the front.
Bar width - This option determines the percent of total bar width used. Setting this value to 100 makes
joined bars.

90 WordStat Users Manual

 Bar depth - Use this property to limit the depth that each bar series uses. By default, bars will take up
the part proportional to the number of bar series in the chart so that the back of a bar will join the
front of the bar immediately behind it. To insert a gap between series of bars, decrease this value.

Creating Bubble Charts

Bubble charts are graphic representations of contingency tables where relative frequencies are
represented by circles of different diameters.
Set the TABULATE, COUNT and DISPLAY options so that the information to view is
displayed in the table.
Click the button.
For more information, see Creating Bubble Charts (page 155).

Creating heatmaps with clustering of rows and columns

A heatmap plot is a graphic representation of crosstab tables where cell frequencies are represented by
different color brightness or tones. When combined with clustering of rows and/or columns, this
exploratory tool allows one to identify functional relationships between specific keywords and
subgroups defined by values of the independent variable. To create a heatmap:
Set the WITH option to an independent variable.
Set the TABULATE option to either CASE OCCURRENCE or KEYWORD FREQUENCY.
Click the button to access the heatmap dialog box.
For more information on heatmaps, see Using Heatmap Plot (page 158).

Performing correspondence analysis

Correspondence analysis is an exploratory technique that provides a graphic overview of relationships
in large crosstabulation tables of frequency. To perform a correspondence analysis:
Set the WITH option to an independent variable.
Click the button to access the correspondence analysis dialog box.
For more information on correspondence analysis, see Performing Correspondence Analysis (page
162).

To export the table to disk:

Click the button. A Save File dialog box will appear.

In the Save as Type list box, select the file format in which to save the table. The following
formats are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited file
(*.CSV), HTML file (*.HTM;*.HTML), and Excel spreadsheet file (*.XLS).
Type a valid file name with the proper file extension.
Click the Save button.

WordStat Users Manual 91

To append the table to the Report Manager:

Click the button. A descriptive title will be provided automatically for the table.
To edit this title or to enter a new one, hold down the SHIFT keyboard key while clicking this
button.
(for more information, see the Report Management Feature topic on page 191).

To copy the entire table to the clipboard:

Right-click anywhere in the frequency table.
Select the COPY TO CLIPBOARD | TABLE command from the pop-up menu.

To copy selected rows to the clipboard:

Select the rows you would like to copy (multiple but separate rows can be selected by clicking
while holding down the CTRL key).
Right-click anywhere in the frequency table.
Select the COPY TO CLIPBOARD | SELECTED ROWS command from the pop-up menu.

To search for a specific item:

Right-click anywhere in the table.
Select the FIND command from the pop-up menu. A search dialog box will appear.
Type the desired search string in the Find What edit box. To restrict the search to items starting
with the search string, enable the Match Beginning of Item option and to restrict it to whole
words matching the search string, enable the Match Whole Word Only option.
Click the Find button to search the first item matching the typed string. Clicking this button
again finds the next occurrence of the search string, starting at the currently selected item.

92 WordStat Users Manual

 Keyword-In-Context Page
The Keyword-In-Context (KWIC) technique allows one to display in a table the occurrences of either a
specific word, or of all words related to a category, with the textual environment in which they occur. The
text is aligned so that all keywords appear aligned in the middle of the table. This technique is useful to
assess the consistency (or lack of consistency) of meanings associated with a word, word pattern or
category. In the example below, we can see that the word pattern KILL*, which may have been assigned to
a category like "aggressiveness", refers to words with different meanings, some of them quite distant from
the concept of "aggressiveness":

I have decided to KILL a few hours before...

He said that he would KILL me if I call the police.
Too much garlic KILL the taste of the meat.
The Black Death was a disease that KILLED millions.
My shoes are KILLING me
The French skier Jean Claude KILLY won 3 gold medals.

When displaying rules, only the keywords or key phrases associated with the first item of those rules are
displayed. For example, in a rule like:

#SATISFACTION before #TEACHER and not near #NEGATION

the KWIC list will contain only items in the SATISFACTION category meeting the conditions specified by
this rule.

Once an inconsistency has been detected, it becomes possible to reduce it by making changes to the textual
data or to the dictionaries. For example, the researcher may change all occurrences of the word KILL in the
original text for either KILL1 or KILL2 in order to differentiate the different meanings and then add only
one of these modified words (say KILL1) to the substitution or inclusion dictionary. The word KILLY may
also be added to the dictionary of excluded words. The categorization of phases may also be used to
distinguish various meanings of a word. For example, the use of KIND to refer to the adjective
("considerate and helpful nature") may be reliably differentiate from the use of KIND as a noun ("category
of things") or as an adverb by categorizing the phrase "KIND OF" as instances of this word used as a noun
or as an adverb and by categorizing the remaining instances of KIND as the adjective. Disambiguation may
also be performed by identifying words in close proximity that are associated with specific meanings and by
creating categorization rules (see Working with Rules on page 124).
The KWIC technique is also useful to highlight syntactical or semantic differences in word usage between
individuals or subgroup of individuals. For example, candidates from two different political parties may use
the word "rights" in their discourses at the same relative frequency, but we may find that these two groups
use this word with quite different meanings. We may also find that the meaning of a word like "moral"
evolves with the age of a child.

WordStat Users Manual 93

The Keyword-In-Context page has been designed to facilitate the various tasks involved in content analysis.
The page looks like this:

The upper-right part of the screen provides a list of all instances of keywords associated with a dictionary
category or of a specific word or phrase along with its surrounding text. The panel on the left shows a tree
view of items and their context in descending order of frequency and may be used to browse through and
filter through the KWIC list on the right. The text panel below the KWIC table displays the full document
from which the selected keyword comes from and highlights it. The text panel can be used to examine the
full context of a keyword, but may also be used to add words and phrases to the current categorization
dictionary or to the exclusion list. To assign a word or a phrase to a list or content category, position the text
cursor on the word you want to assign, or select one or several words with the mouse and right-click to
display a contextual menu. Select the To Categorization Dictionary or the To Exclusion List menu item.
LIST - This option allows for specifying whether the words for display in the KWIC table either should
be selected from the list of included words or from the list of all remaining words that have not been
explicitly excluded. The option User Specified allows one to enter a word or word pattern at the
keyboard and search for all instances of this expression.
WORD - This option allows one to choose among all keywords belonging to the list of Included or
Leftover words (see above). When the LIST option is set to User Specified, this option becomes an
edit box where one can type a word or word pattern. (Wildcards such as * and ? are supported).
SORT BY - This option allows for sorting the keyword-in-context table in either ascending order on
any of the following options:
Case number - The KWIC table is sorted in ascending order of case position.

94 WordStat Users Manual

 Keyword & Before - The KWIC table is sorted on the keyword as well as the words appearing
immediately before it.
Keyword & After - The KWIC table is sorted on the keyword and the words appearing
immediately after it.
Keyword & Variable - When several text variables have been selected, the KWIC table
includes a column indicating in which variable the keyword was found. When this option is
selected, the KWIC table is sorted so that all words associated with a category or matching a
word pattern are displayed in alphabetical order. Lines with identical words are sorted on the
variable name from which they come. This display is useful to examine whether specific words
are used with the same meaning in different variables.
Variable & Keyword - When several text variables have been selected, the KWIC table
includes a column indicating in which variable the keyword was found. This option displays a
KWIC table where all lines are sorted on the variable name from which they originate. Lines
extracted from a single variable are sorted by keywords. This display is useful to establish
whether different variables contain different information. For a more detailed analysis of
difference in usage of specific words, use the Keyword & Variable sort order.
Keyword & VARNAME - This option displays a KWIC table where lines are sorted by words.
Lines with identical words are sorted on the value of the selected independent variable. This
display is useful to highlight differences between subgroups in the meanings associated with a
specific word.
VARNAME & Keyword - This option displays a KWIC table where lines are sorted by the
values of the selected independent variable. Lines with identical values on this variable are
sorted by keywords. This display is useful to establish whether subgroups differ on the use of
words associated with a category. For a more detailed comparison of usage of specific word, use
the Keyword & VARNAME sort order.
CONTEXT DELIMITER - This option allows one to select the amount of context displayed in the KWIC
table as well as in the concordance report. In the KWIC table, context strings, either before or after
the keyword, are limited to 255 characters.
None - This option instructs WordStat to display as much context as possible, up to a limit of
255 characters.
Paragraph - When this option is selected, the program will limit the context displayed to the
paragraph in which a specific keyword appears.
Sentence - When this option is selected, the program will limit the context displayed to the
sentence in which a specific keyword appears. A sentence must end with a period followed by a
space or a hard return, or by an exclamation or a question mark.
User defined - When this option is selected, the program will retrieve text found before and
after the keyword until a slash character is encountered.

Once the settings have been set, click the button to start searching all instances of the selected
keyword.

WordStat Users Manual 95

Using the tree list
The upper left panel displays a tree list view where each keyword is presented in descending order of
frequency along with either its context or some associated variable. The content by which keywords are
sorted depends on the sort setting, so that when the SORT BY option is set to "Keyword & Before" or
"Keyword and After" this tree list will be sorted and broken down by up to 5 words of its surrounding text
(like the screen shot above). When the SORT BY option is set to the keyword and a variable, the tree list
will be sorted on keywords first, and each keyword will be broken down by the values of the selected
variable.
Selecting any item in this tree list will filter the keyword-in-context list on the right to display only the
selected item. Moving to SHOW ALL ITEMS will remove any filtering condition and show all hits.
By default, the tree display displays all items. One can filter this list and display items meeting a minimum
frequency criterion. To adjust this criterion, right-click anywhere in this panel and select SET MINIMUM.
A submenu allows one to increase the minimum frequency criteria, reset it to 1, or adjust the default
minimum frequency criterion to be used for all other KWIC analysis to the current value.
Any KWIC table may be saved to disk in Excel, plain ASCII, text delimited, or HTML format by clicking
the button. To export the content of the table to a new SimStat/QDA Miner data file, press the Export
button located at the top of the Frequencies page.

To print the KWIC table, click the button.

Clicking the button produces a concordance report on the keywords currently displayed in the KWIC
table. The sort order and context delimiter of the current KWIC table are used to determine the display
order and the amount of context displayed in this concordance report. This report is displayed in a text
editor dialog box (see below) and may be modified, stored on disk in RTF, HTML or plain text format,
printed, or cut and pasted to another application. Graphics may also be pasted anywhere in this report.

96 WordStat Users Manual

 Classification Page
Automated text categorization is a supervised machine-learning task by which new documents are classified
into one or several predefined category labels based on an inductive learning process performed on a set of
previously classified documents. This machine-learning approach of classification has been known to
achieve comparable if not superior accuracy than classification performed by human coders, yet at a very
low cost in manpower. It has been used to automatically classify documents into proper categories or to find
relevant keywords describing the content and nature of a document. It has also been used to automatically
file or re-route documents or messages to their appropriate destinations, to classify newspaper articles into
proper sections or conference papers into relevant sessions, to filter emails or documents (like spam
filtering), or to route a specific request in an organization to the appropriate department. Automated text
categorization may also be used to identify the author of a document of unknown or disputed authorship.
For a good overview of automated text classification, see Sebastiani (1999).
The automated text categorization module in WordStat allows one to apply either Naive Bayes or K-Nearest
Neighbors learning algorithms on an existing textual database in order to develop a categorization model (or
classifier). The program also provides features to test the accuracy of the classification and to optimize the
various parameters. Once optimized, the obtained classification model may be used immediately to classify
uncategorized documents or may be saved on disk to be applied later outside WordStat using the WordStat
Document Classifier utility program. The classifier may also be incorporated into a desktop or web
application or within a document management system using the WordStat Software Developer's Kit.
The development and application of a text classifier often involve the following steps:
1. Removal of function words, words that appear in only a few documents and words that appear too
often.
2. Dimension reduction, through lemmatization, stemming, categorization, word clustering or other
dimension-reduction techniques.
3. Feature selection, which consists of a selection of terms based on their capability to discriminate
between categories of documents.
4. Training the classifier on the train set.
5. Testing the accuracy of the classification on a test set.
6. Applying the classifier to new documents.
While the basic content analysis features of WordStat may be used to deal with the first two steps, the
Automated Text Categorization dialog box allows one to accomplish tasks related to the last four steps. This
dialog box consists of four pages:
The Select Features page allows one to apply various feature selection methods to select a subset of
terms to be used by the classifier.
The Learn & Test page is the location where machine-learning algorithms are set and tested. This
page also allows the storing of classification models to disk.
The History & Experiment page keeps track of every learning test performed during a session
allowing one to choose the best setting and algorithm for a specific classification task. It also gives
access to a batch experiment dialog box that may be used to define numerous tests and perform them
all at once.
The Apply page is used to apply a classifier to an external document, a list of documents or to the
current data file.

WordStat Users Manual 97

Accessing the Automated Text Classification dialog box
To develop a classifier for a specific categorical variable, you need to select a categorical variable
containing the values you want to predict. While it can be done in WordStat, the most common way to
select it is from QDA Miner or SimStat while calling WordStat. In QDA Miner, one has to choose this
categorical variable in the In Relation with section. In Simstat, this variable should be assigned to the
independent list box and assign the text variables on which the prediction should be based to the dependent
list box. Once in WordStat, set the various text processing options (such as the lemmatization, the exclusion
and categorization lists, and all the other analysis options needed) to obtain the desired list of keywords or
content categories. Then move to the Classification page.

Settings
The Settings page allows one to select which variable contains the values to predict and choose a validation
method.

Selecting the variable to predict

To select the variable containing the values to predict, set the first list box to the name of this variable. If its
name is not listed, choose the <Select Variables> item to display a list of all available variables, select the
variable to predict and click OK. Then set the drop down list to this newly added variable.

Selecting the validation Method

The evaluation of a classifier consists of measuring its effectiveness at classifying documents that have
already been classified. Those documents should, however, not be part of the training set used to develop
the classification model, since it would likely overestimate the real performance of the classifier. Yet,
training a classifier on only a portion of the available training set may result in a less than optimal classifier.
Cross-validation methods have been proposed as a compromise solution that allows one to develop a
classification model on all the available documents in the training set yet provide a somewhat more realistic
estimate of the classifier performance. WordStat offers three broad types of validation methods:
Leave-one-out - This cross-validation method consists of 1) removing a document from the training
set, 2) developing a classification model on the remaining documents, 3) applying this model to predict
the membership of this single document and 4) comparing the decision made by the classifier to the
actual class to which this document belongs. This procedure is then repeated for each document in the
training set and the different decisions are combined to estimate the performance of the classifier. While
this method logically involves the computation of a large number of models and may seem to be time
consuming, in practice the classification model is computed only once but adjusted analytically to
remove the contribution of the test document prior to its classification. This cross-validation method
will often overestimate the performance of a classifier if the training set includes duplicate documents
or if included documents are not totally independent from one another.

n-folds - This method consists of splitting the training set into smaller partitions and testing each
partition on the classification performance obtained by a model developed on the remaining ones. For
example, when using a five-fold cross-validation method, the training set is divided randomly into five
subsets, each containing approximately 20% of the documents. For each subset, the program tests the
accuracy obtained by a classification model developed on the remaining 80% of the original training
set. The performances obtained on all five classifiers are then used to estimate the performance of the

98 WordStat Users Manual

 classifier computed on the full training set. WordStat provides a choice between five-fold, 10-fold and
20-fold cross-validation.

External file - A more conventional method for assessing the performance of a classifier is to test the
accuracy of the classifier on an entirely different set of documents that have also been classified but are
totally independent of the training set on which the categorization model is based. To perform such a
test, WordStat requires the test set to be stored in a different data file. When this option is selected, an
Open File dialog box is displayed allowing one to identify the file containing the external set. WordStat
then displays a dialog box like the one below allowing one to choose the text variable containing the
documents to be used for classification and the numerical variable containing the class to which this
document belongs. Once set, click OK to return to the classification page.

Once the variable and the validation method have been set, click the button to continue.
WordStat will compute all statistics needed, and will then automatically move to the Select Features page.

WordStat Users Manual 99

Select Features page
The Select Features page allows one to view the strength of the relationship between words, keywords
or content categories and classes of the selected categorical variable. It also allows you to select a
subset of items based on those statistics.

The strength of the relationship between an item and the classes of the categorical variable can be
computed either on the occurrence (present or absent), on the frequency of items in each class, or on
the percentage of words. To change the base statistic used for assessing differences among classes, set
the Compute statistics on list to the proper option.
The discriminative strength of each item is assessed using three statistics and is presented in a table
containing the following information:

Name The word, keyword or content category.

Global Chi The overall chi-square value computed on all classes of the categorical variable.

P The probability of the above chi-square value.

Max Chi The chi-square value computed on the class with the highest case occurrence or
frequency against all the other classes.

100 WordStat Users Manual

 P The probability of the Max Chi value.

Biserial The biserial correlation computed between the class of the categorical variables
with the highest case occurrences and the remaining classes. This coefficient
assumes that the presence or absence of a class is determined by a trait normally
distributed. Contrary to the standard correlation coefficient, this measure of
association may yield a value lower than -1.0 or higher than +1.0.

Predict Indicates the class in which the item most frequently occurs. When the highest
case occurrence appears for more than one class, the column includes the labels
of all those classes.
Clicking any column header sorts the table in ascending order of the data in that column. Clicking the
same column header a second time sorts its content in descending order. The check boxes in the first
column show, by default, that all items are to be included in the classification model. To manually
remove an item, simply click in the box to remove the check mark.
The FILTER TABLE option allows one to display only the terms characteristic of a specific class. It
may also be set to <selected> to display only items that have been selected for inclusion in the
categorization model. To display all items set this option to <all classes>.
The lower portion of the page displays a bar chart with the percentage of cases in each class of the
categorical variable containing the selected item. Classes are presented in descending order of case
occurrence. This graph is synchronized with the above table so that changing the selected item in this
table results in the display of the corresponding distribution chart.

To display the chart on the right-hand side of the table, click the button. To bring the bar chart
back to the bottom of the table, click the button.

Performing Automatic Feature Selection

It has been shown that reducing the number of terms in classification models can reduce not only the
cost of recognition by reducing the number of features that need to be collected, but also sometimes can
improve the classification accuracy and can reduce the risk of overfitting. Several methods have been
proposed to select a subset that yields the best performance for a specific classification system.
WordStat allows one to apply a filter on available items in order to select a specified number of those
based on the computed association with the classes to predict.

To access the feature selection dialog box, click the button. The following dialog box will
appear:

WordStat Users Manual 101

 To include all items, set the selection criterion to All and click OK. To remove all the check marks
beside the items, set the selection criterion to None and click OK. When the Select criterion is chosen,
specify how many items should be selected and which statistic will be used to select them. Three
statistics are available for performing such a selection: The global chi-square value, the max chi-square
and the bi-serial correlation (see above for a description of these statistics). By default, the selection is
performed by extracting items with the highest values on the selected statistic. Enabling the
Optimization option instructs the program to apply a special algorithm that will take into account this
statistic as well as current contrasts between classes. This option has been found in most situations to
improve the performance of the classifier for an equal number of selected features.
To close this dialog box and apply the selection criteria, click OK. You may also leave this dialog box
without affecting the current selection by clicking the Cancel button.
Once features have been selected, move to the Learn & Test page to select a learning algorithm and
test it.

102 WordStat Users Manual

Learn & Test Page
The Learn & Test page is where machine-learning algorithms are chosen and tested and from which the
classifier may be exported to disk.

The panel at the top of the page allows one to select the machine-learning algorithm to use, set
various analysis options and choose how the classification model will be tested.

Learning options

METHODS - Two machine-learning algorithms are available for text classification:

The Naive Bayes algorithm classifies text by estimating the probability of a class, given the
presence or absence of specific words or keywords in the document to be classified. It first
computes the probability of each term to occur in documents of specific classes in the training set.
It then combines the probabilities associated with words found in the document to classify to
estimate the probability that this document belongs to different classes. Finally, it assigns the
document to the class with the highest probability. A multinomial Naive Bayes model has been
chosen to handle both binomial and multinomial classification tasks as well as binary and
numerical weights of items.
The k-Nearest neighbor classification method compares a document to be classified to all
documents in the training set, retrieves the k most similar documents, and then assigns the new
document to the most common classes in this retrieved set. This method is usually known to
provide accurate classification when the training set is large enough, yet can be very time-
consuming because of the need to compare and rank the entire training set for similarity with the
test document. It also usually requires a larger storage space since it must keep frequency

WordStat Users Manual 103

 information for all documents in the training set rather than just a few classification rules or
mathematical formulas like many other machine-learning methods. However, WordStat uses a
very efficient K-NN algorithm that drastically improves the computing speed and reduces the disk
space and memory requirement. When this method is chosen, an NO edit box appears below the
Method list box, allowing one to set the number of similar documents on which this classification
will be based. Values higher than 20 or 30 are typically used in text classification tasks.

USE - This option is used to select the item statistic to be used in training and classification. Choosing
Case Occurrence results in the use of binary weights, indicating whether or not a word or keyword
occurs in the document. Selecting Keyword Frequency allows one to use additional information
related to how often this item occurs in each document. Percentage of Words and Percentage of
Keywords provide two methods to normalize the obtained frequency to take into account the
document length. Such normalization is performed by dividing the frequency either by the total
number of words found in the document or the total number of keywords that have been extracted by
WordStat.

FEATURE WEIGHTING - Feature weighting has been presented as an alternative to feature selection or
as a way to further improve classification accuracy from selected item sets. This method consists of
giving more weight to items that are rather good at differentiating documents from distinct classes
and negligible weight to those that are distributed evenly among classes. The most frequently used
weight in information retrieval is the TF*IDF measure where the frequency of an item is adjusted to
take into account the number of documents containing this item. However, such a weighting can be
considered to be only a crude approximation of the capacity of the item to differentiate documents
from distinct classes. More accurate performance of the classifier can be expected from using a
weight based on a more direct indicator of this discriminative capability such as the Global Chi-
square or the Max Chi described previously.

Results
A common way of assessing the accuracy of a classifier is by comparing the accuracy of predicted
class membership against actual membership. Such information is provided by the Confusion Matrix
where each predicted class is plotted against the actual class. Accurate predictions are plotted in the
diagonal going from the top left to the bottom right of the table. Values in this diagonal are printed in
bold characters for easy identification. Values in cells below or above this diagonal represent
classification errors. Besides the actual number of documents in each cell, the table shows the row,
column and total percentages. Row percentages represent the number of documents in a class that have
been classified in a specific way, while column percentages express the percentage of a specific
prediction actually belonging to a known class. This table may be used to identify which classes are the
easiest or hardest to predict, as well as which classification errors are the most common. To facilitate
comparisons across the classes of the categorical variable, two related statistics are printed on the right
of the table: Precision is the probability that documents identified as belonging to a class are correctly
classified and Recall is the probability of documents in a class to be correctly identified.
Several statistics are provided to assess the global performance of the classifier. The Nominal
Accuracy measure is the proportion of documents correctly classified It is considered a micro-average
statistic since it gives equal weight to documents regardless of how they are distributed among classes
of the categorical variable. The Average Precision and Average Recall measures are macro-average
statistics obtained by computing the mean precision and recall obtained for every class. The Ordinal

104 WordStat Users Manual

Accuracy measure weights disagreements so that errors in prediction will be considered higher when
the predicted value is far from the original value, while predictions that are closer to the original value
will be counted as partial disagreements.
The Confusion List page presents information already found in the confusion matrix but in the form of
a single list that allows one to identify more easily the most common errors. The table may be sorted on
the actual class of the documents, the predicted classification, the number of times such a classification
error occurred, or the proportion of documents that have been misclassified this specific way. By
default, the table is sorted in descending order of frequency. To sort the table on values in another
column, simply click this column header. Clicking the same column header a second time sorts its
content in descending order.
The Review Errors window displays a list of all documents that have been misclassified, allowing one
to examine for each document the classification error made by the classifier as well as the computed
values associated with every class of the categorical variable. A text window in the bottom of the list
also allows one to review the text on which the classification has been made and potentially identify
some of the reasons why the document had been misclassified.

WordStat Users Manual 105

History & Experiment Page
There is, unfortunately, no rule or rationale to help one chose a classifier and its options and
parameters, nor to indicate how many or which items should be selected for inclusion in the
development of a classification model. For this reason, the selection of classifiers and their
optimization must rely on experimentation. In other words, in order to obtain the best classifier, one
needs to perform numerous tests involving systematic variations of the various settings and compare
the results. The History page offers a way to keep track of previously performed trials, the options
used, as well as the obtained performances. From this page, one can also access an Experiment dialog
box that provides a convenient way to define and run a large number of classification experiments on
the same data set.

Data from prior trials are presented either in the form of a table (Table page) or as a line chart (Graph
page). Clicking any column header of the table sorts it in ascending order of the data in this column.
Clicking the same column header a second time sorts its content in descending order. By default, the
displayed statistics are computed for all classes of the categorical variable. To restrict the display of
either the table or the line chart to statistics related to a single class, set the Class list box to the desired
class. Setting this option to <all classes> brings back the micro- and macro-average statistics computed
on all classes.
Data from specific trials may be deleted from the table by selecting their rows and clicking the
button.
The Graph page allows one to compare the performance of various settings and the relationship
between those settings using a line chart like the one shown below.

106 WordStat Users Manual

By default, the chart displays the relationship between accuracy and the number of features in the
classification model. One may, however, examine the relationship between other parameters (such as
precision versus recall) by changing the information plotted either on the horizontal or the vertical axis
of the chart.
The following table provides a short description of buttons available:

Control Description
Press this button to save either the table or the chart on disk. The table may be saved
to disk in Excel, plain ASCII, text delimited, or HTML Charts may be saved in BMP,
JPG or PNG graphic file format or may be stored on disk in a proprietary format
(.WSX file extension) that may later be edited and customized using the Chart Editor.
Pressing this button prints a copy of the displayed table or chart.

This button allows the editing of various features of the chart such as the left and
bottom axis, the chart and axis titles, the location of the legend, etc.
This button is used to create a copy of the chart to the clipboard. When this button is
clicked, a pop-up menu appears allowing one to select whether the chart should be
copied as a bitmap or as a metafile.
The History page also gives access to the Experiment feature where one can quickly perform a series
of classification experiments. To access this feature click the button.

WordStat Users Manual 107

Classification Experiment Dialog Box
The Classification Experiment feature allows one to quickly perform a series of classification
experiments in order to choose among classification methods, analysis settings and selected sets of
items (or features).
To access this dialog box click the button. A dialog box similar to the one below will
appear:

The basic principle of this dialog box is to create a list of classification experiments involving different
settings and then to instruct WordStat to perform all those experiments one after the other. Experiments
first need to be defined in the upper part of the dialog box and then to be moved to the table of
experiments located at the bottom of the dialog box. After several experiments have been defined and
added to the list, one can execute all of them at once.
The first steps involve setting the experiment options. The Feature Selection option located at the top of
the dialog box provides automatic feature selection options similar to those available from the first
page of the document classification dialog box with only one exception: while the original dialog box
allows one to select only one feature set size at a time, the current dialog box allows one to set
numerous feature set sizes at once. For example, by entering the following string in the Select edit box:
50 100 150 200 300
five classification experiments will be performed using the same analysis settings but on features set
sizes of 50, 100, 150, 200 and 300, picked out using the chosen selection method. For more information
on the Statistics, Based on and Optimization option, see Performing Automatic Feature Selection
(page114).
The Learning and Testing groups of options also provide similar settings as those available on the
Learn & Test page. Please refer to this section for information on the available algorithms and options.

108 WordStat Users Manual

To add a classification experiment to the list:
Set the various classification experiment settings.

Click the button to move the defined experiment to the list.

Clicking the button displays a dialog box that allows one to quickly define numerous variations of
the current classifier and to add all those at once to the list of experiments to be performed.

To remove an experiment from the list:

In the list of previously added experiments, select the one to delete.

Click the button.

To run experiments in the list:

Click the button. The program performs all the experiments in the list that had not been
executed before. Once an experiment is completed, the Executed column for this item is set to Yes
preventing the program from executing the same experiments twice. Results of every experiment are
automatically appended to the History page.

Click the button to close the dialog box and return to the History page of the classification dialog box.

WordStat Users Manual 109

Apply Page
The Apply page allows one to use the most recently tested classifier to categorize either a single
document, a list of files or documents stored in the current data file or another SimStat/QDA Miner
data file. To perform similar tasks using previously saved classification models, use the WordStat
Document Classifier utility program.

The document classification feature supports numerous file formats such as plain ASCII text files as
well as HTML, Rich Text, MS Word, WordPerfect, Acrobat PDF files. Detailed results of
classifications are displayed in a table at the bottom of the dialog box and may be either saved to disk
or printed. When applied to the current database or another database, the automatic classification
feature may be useful to categorize unclassified documents or to review existing classifications based
on the results of the new classifier.

To classify a single document:

Set the Text To Classify list box to Single Document.
Click the Open File button to locate and import the file containing the text to be classified. You
may also type directly in the text editing window or paste a text previously copied to the
clipboard (by moving to the text editing window and pressing Ctrl-V).
Click the Classify button to apply the current classifier to the displayed text.

To classify a list of documents:

Set the Text to Classify list box to List of Documents.

110 WordStat Users Manual

 Click the Edit List button to display a dialog box like the one below that allows you to browse
through your computer and select certain documents. You may add documents located in
different folders by successively adding documents located in a specific folder and then moving
to a new location where the other documents are located. Click OK to confirm the changes to the
file list.

Click the Classify button to apply the current classifier to all documents in the list.

To classify documents in the current data file:

Set the Text to Classify list box to Current Data File.
Click the Classify button to apply the current classifier to all documents contained in the selected
text variables. Please note that those are the same text variables used for developing the current
classifier. However, some documents may have been ignored during the classification phase,
either because they had been filtered out or because they had not been classified before and
contained missing values in the categorical variable. Those documents, as well as all other
previously classified documents, will be categorized by the current classifier and the result of
this classification will be displayed in a result table.
To store the predicted class or the computed score obtained for every class, click the button.
The following dialog box will appear:

WordStat Users Manual 111

 To save the predicted class, put a check mark beside Save Predicted Class and enter a variable
name.
To save the scores associated with each class and upon which the classification has been made,
put a check mark beside Save Scores and enter a variable prefix (up to 7 characters). Variable
names are created by adding successive numeric values to this prefix. For example, if the edit
box at the right of the Variable Prefix option is set to "CLASS_", the variable names will be
CLASS_1, CLASS_2, CLASS_3, etc.
If any one of the specified variables does not exist, WordStat will create new ones and store the
numerical values associated with either the predicted class or the class scores. A confirmation
dialog box will ask for confirmation of the creation of those new variables as well as the
overwriting of any existing ones.

To classify documents in another data file:

Set the Text to Classify list box to External Data File.
Click Open File to locate the SimStat/QDA Miner data file containing the documents to be
classified. A dialog box similar to the following one will appear:

Select one or several text or document variables that will be used for classification purposes and
click OK. The content of the data file is displayed in a table, while the text to be classified is
displayed on its right. You can resize this text window by dragging its left border.
Click the Classify button to apply the current classifier to all documents contained in the selected
text variables.
To store the predicted class or the computed score obtained for every class, click the button.
A dialog box similar to the following will appear:

112 WordStat Users Manual

 To save the predicted class, put a check mark beside Save Predicted Class and enter a variable
name.
To save the scores associated with each class and upon which the classification has been made,
put a check mark beside Save scores and enter a variable prefix (up to 7 characters). Variable
names are created by adding successive numeric values to this prefix. For example, if the edit
box at the right of the Variable Prefix option is set to "CLASS", the variable names will be
CLASS1, CLASS2, CLASS3, etc.
If any one of the specified variables does not exist, WordStat will create new ones and store the
numerical values associated with either the predicted class or the class scores. A confirmation
dialog box will ask to confirm the creation of those new variables, as well as to overwrite any
existing variables.

To export the table to disk:

Click the button. A Save File dialog box will appear.

In the Save as type list box select the file format in which to save the table. The following
formats are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited file
(*.CSV), HTML file (*.HTM;*.HTML), and Excel spreadsheet file (*.XLS).
Type a valid file name with the proper file extension.
Click the Save button.

Exporting a classifier to disk

Once developed and optimized, a document classification model may be saved to disk and later be used
outside the WordStat main program to categorize new documents. The saved categorization model
includes all word exclusion, extraction and categorization settings needed to accurately retrieve items
used in the classification model as well as either the classification rules (Naive Bayes) or the keyword
indexing of documents in the training set (K-Nearest Neighbors) needed to perform document
classification.

To save the classifier on disk:

Click the button locate on the top of the Learn & Test page. A standard Save File dialog
box will appear.
Enter the file name under which you would like to save the classifier and then click the Save
button. WordStat will automatically provide a .wclas file extension.
Document classifiers may be retrieved and applied to new documents using the WordStat Document
Classifier utility program (see page 184). A special Software Developer's Kit (SDK) is also available
upon request from Provalis Research allowing any programmer to integrate WordStat categorization
and classification technologies into one's own database or document management system (see page
188).

WordStat Users Manual 113

 Creating and Maintaining Dictionaries
Please note that in the instructions below the exclusion list and the categorization dictionary are both
referred to as "dictionary".
To open an existing dictionary
Select the dictionary from the Categorization drop down list. If the dictionary is not listed, click the
button to display a dialog box that will let you browse through folders and select the
dictionary.

To create or copy a dictionary

Click the button located to the right of the exclusion list or of the categorization dictionary.
A dialog box enables specifying the name and location of the new dictionary file. If a dictionary file
is already active, it will ask whether existing entries should be copied to the new dictionary file. If
you answer Yes, all entries in the previously opened dictionary will be retrieved and stored in the
new one. Answering No will result in an empty dictionary.

To add new words to the dictionary

From the Dictionaries page

In the Dictionary Viewer group box, select the dictionary to which you would like to add new words.

Press on the button and select Words/Phrases...

From the Frequencies page

Select the rows containing the words you would like to add.

Press on the button or press the right button of the mouse.

Select the dictionary to with you would like to add the selected words.

From the Crosstab page

Select the rows containing the words you would like to add.
Press the right button of the mouse to display a pop-up menu.

Select the dictionary to with you would like to add the selected words.

Note: You may also drag and drop a word into the dictionary panel to the right of the frequency table (see
Using the Dictionary Panel, page 41).
From the text editor

114 WordStat Users Manual

 Select the word you would like to add.

Click the button or press the right button of the mouse to display a pop-up menu.
Select the dictionary to with you would like to add the selected words.
If you choose to add a word to the exclusion list, the word will automatically be stored in this file
without any dialog box. If the Inclusion dictionary is selected, the program will display a dialog box
similar to the following:

To add new items to an existing category:

Type de words or phrases you would like to add in the edit box, one item per line. Spaces are
automatically replaced by an underscore character.
Select the proper category from the right panel.
Click the Add button.

To add new items to a new content category:

Type the words or phrases you would like to add in the edit box, one item per line. Spaces are
automatically replaced by an underscore character.
Set the check box on the right panel to New Category.
Type the name of the new category as select the existing category under which this new category
should be stored. To create a main category, select the < ROOT > item.
Click the Add button.

Wildcards such as * and ? are supported.

Weights can also be assigned to specific categorization, so that a specific word, word pattern, or
expression may count for more than one instance of the concept. The default value for this option is 1.

WordStat Users Manual 115

 To use a lower or higher value, edit the Weight option either by entering a new numeric value in the
edit box or by clicking the spin buttons to increase or decrease this value. Valid weight can be any
floating point value higher than zero.

If you want to add a word to a non existing category, you first need to create such a category (see
below) and then follow the above steps to add the word to this new category.

To add categories to the inclusion dictionary:

From the Dictionaries page

In the Dictionary Viewer group box, select the Inclusion radio button.
Press on the button and select Category. The Add Categories dialog box will appear.
Select the Main Category or the Subcategory radio button depending on whether you want this new
category to appear at the main level or whether you want it to be created under an existing category.
If you choose to create a sub-category, you then need to select from the Location outline the category
under which you would like to store it.
Type the category names you would like to add in the edit box, one item per line, and click the Add
button.

To remove an entry or a category from a dictionary:

Select the Dictionaries page by clicking the first tab at the top of the dialog box.
In the Dictionary Viewer group box, select the dictionary from which you would like to remove
words or categories.

Select the words or categories you would like to delete and click the button. If a
non-empty category is selected, you will be asked to confirm its deletion. If you answer Yes, all
words and subcategories belonging to this category will also be erased.

To edit an entry in a dictionary:

Select the Dictionaries page by clicking the first tab at the top of the dialog box.
In the Dictionary Viewer group box, select the dictionary containing the words or categories you
would like to edit.

Select the item you want to modify and click the button.

116 WordStat Users Manual

To search for an entry in a dictionary:

Right-click anywhere in the categorization dictionary.

Select the FIND command from the pop-up menu. A search dialog box will appear.
Type the desired search string in the Find What edit box. To restrict the search to items starting with
the search string, enable the Match Beginning of Item option and to restrict it to whole words
matching the search string, enable the Match Whole Word Only.
Click the Find button to search the first item matching the entry. Clicking this button again finds the
next occurrence of the search string, starting at the currently selected item.

Moving words or categories using drag and drop

The easiest way to change the structure of an inclusion dictionary is by using drag and drop operations.
Using the mouse, you can move a word to a different category, move an existing category or
sub-category to another location on the main level or under an existing category. There are two ways to
perform such kind of drag and drop operation:

1) For single move operation

Press and hold down the ALT key
Click the item in the dictionary you would like to move and keep the mouse button down
Move the mouse to the desired location and release the mouse button.

2) For multiple move operations

Another way to achieve multiple drag and drop operations is by enabling the Drag & Drop check box
located on the left of the dictionary.
To perform such operations, you first need to enable the drag & drop editing feature:
Select the Dictionaries page by clicking the first tab at the top of the main window.
In the Dictionary Viewer group box, select the categorization dictionary.
Check the Drag & Drop Editing check box.
Click the item you want to move, and hold the mouse button down.
Then, drag the item over its new location and release the mouse button.

By default, the dragged item is stored under the category at the cursor position. To move a word or a
category to the main level or to the same level as the category under the cursor, simply hold the ALT
key while dropping the dragged item.

WordStat Users Manual 117

 Moving a word to a distant category
Select the item you would like to move.
Click the right button of the mouse to display the contextual menu.
Select the Move To command.
Choose the category where the selected item should be moved and click OK.

Importing Dictionaries
To import a dictionary stored in an Excel, CSV or Tab delimited file, the data file has to be formatted such
that the name of categories are listed under up to four columns, and the items such as words and phrases are
listed in a separate column. The data file may also contain an additional column containing weights to be
used for each item, this weight being represented by a positive integer or floating point numerical value.
The first row must contain a header that will help one identifies the content of each column. A typical
dictionary table consisting items stored in a hierarchical dictionary with two levels (main categories and
subcategories) along with individual weights may look like this:

MAIN CATEGORY SUBCATEGORY ITEMS WEIGHT

EMOTION ANXIETY NERVOUS 0.6
EMOTION ANXIETY EMBARRASS* 2.1
EMOTION ANXIETY DISCOMFORT* 2.7
EMOTION SADNESS ABANDON* 1.0
EMOTION SADNESS ALONE 1.3
BEHAVIOR AGGRESSION ABUS* 1.2
BEHAVIOR AGGRESSION HUMILIATE 0.8
BEHAVIOR AGGRESSION BEAT* 2.5
BEHAVIOR ASSERTION CLAIM* 1.0
BEHAVIOR ASSERTION REQUEST* 3.2

A sparse version of it that omits repetitions of category names may look like this:

MAIN CATEGORY SUBCATEGORY ITEMS WEIGHT

EMOTION ANXIETY NERVOUS 0.6
EMBARRASS* 2.1
DISCOMFORT* 2.7
SADNESS ABANDON* 1.0
ALONE 1.3
BEHAVIOR AGGRESSION ABUS* 1.2
HUMILIATE 0.8
BEAT* 2.5
ASSERTION CLAIM* 1.0
REQUEST* 3.2

To import such kinds of files, move to the Dictionaries page of WordStat and click the IMPORT button
located on the far right of the Categorization check box. An Open File dialog box will appear allowing you
to select the file containing the dictionary you want to import. Once selected, click Open. A dialog box
similar to this one appear:

118 WordStat Users Manual

The drop down lists on the left of this dialog box allows one to identify the columns containing the category
and subcategory names. Up to four levels of categories can be specified this way. It is however possible to
import dictionaries with higher levels of categories by separating the deeper levels (fifth or more) with a
backslash character and storing those at the fourth level as part of this subcategory name. For example, a
fourth level subcategory that would be stored as: SENTIMENT\NEGATIVE\UNFAIR would create two
additional levels underneath SENTIMENT, with NEGATIVE as a subcategory of SENTIMENT and
UNFAIR at the sixth level underneath NEGATIVE.

If the name of the categories are not fully specified like in our first example, but resemble more to the
second table where only the first occurrence of the category name has been typed in, then one has to select
the Sparse Entry of Category check box to make sure that the items will be imported and stored in their
proper category.

Once all the options have been set, click the OK button. A Save File dialog box will ask you to enter a
dictionary file name. WordStat will point to the default folder where other dictionaries are stored. You may
browse to a different folder to store the new dictionary at another location. Once save, the newly created
dictionary will become the default one.

Export Dictionaries
To export a dictionary, move to the Dictionaries page of WordStat and click the EXPORT button located on
the far right of the Categorization check box. A Save File dialog box will appear. Select the proper file
format under which you would like to save the file, type the name of the file and click OK to create it.

Merging Dictionaries
The WordStat Merge feature allows one to append categories and items contained in one categorization
dictionary into another dictionary. To merge dictionaries:
From the dictionary page of WordStat, open the dictionary into which you would like to import
new categories.

Click the button. An Open dialog box is displayed.

WordStat Users Manual 119

 Locate the dictionary containing the items that you would like to import and click OK. A dialog
box similar to the one below is displayed:

Select the categories you would like to import by clicking in the box beside the desired
category(ies) and clicking OK. To select all items, right click anywhere on the list of categories
and select Check All. Choosing Uncheck All removes all check marks previously entered.
If an imported category already exists in the currently active dictionary, WordStat ignores duplicate
items and only imports new items not already found in the original category. New categories are
appended to the existing structure along with all their items.

Printing the dictionary

To create a printed version of the categorization dictionary, go to the Dictionaries page and click the
button. A print dialog box like the one below will appear, allowing you to set various
options of what and how the dictionary should be printed.

120 WordStat Users Manual

PRINTER - Select the desired printer. To adjust the printer settings, such as the page size and
orientation or printer resolution, select the PROPERTIES button.
CONTENT - This option allows you to specify what should be printed. Selecting All prints the entire
dictionary along with all its items. Selecting As Shown will print only currently visible items. When
the Select Items option on the dictionaries page is enabled, a third option Selected Items becomes
available allowing one to restrict the printing to previously selected categories and items.
SUPPRESS WEIGHTS - When this option is disabled, weights of each item are printed within
parentheses. Enabling this option prevents those weights from being printed.
TITLE - Use this option to specify a particular line of text that will appear at the top of each page.

FONT SIZE - This option may be used to adjust the font size used to print dictionary items.

NUMBER OF COLUMNS - Dictionaries may be printed with up to seven columns per page, allowing
one to print large dictionaries on fewer pages. Please note that when increasing the number of
columns per page, it may be necessary to decrease the font size to prevent the overlapping of items in
adjacent columns.
START NEW PAGE ON ROOT CATEGORY - Root categories are the dictionary categories still visible
when the dictionary is fully collapsed. Selecting this option instructs the program to start the printing
of all items starting from this root category at the top of a new page.
FOOTER - Enable this option to print a footer at the bottom of each page. A footer can consist of up to
three items: The Filename (printed on the left margin of the footer), the Page Number (located at the
bottom center of the page), and the Date (printed on the right margin of the footer).

WordStat Users Manual 121

Setting dictionary options
The dictionary options dialog box allows one to assign a description to a categorization dictionary and
some options to be set automatically set upon its opening. This dialog box may also be used to prevent
modification to the dictionaries or to some analysis options that are needed for the categorization to be
performed accurately.

To access the dictionary options of the currently active dictionary, click the button located
on the Dictionary page.

Description page

The first page of this dialog box allows one to enter text to describe the dictionary. Use this
description to inform other users about the intended use of a created dictionary, its assumptions, its
strengths and limitations, etc. Such an option may also be used to document for personal use how the
dictionary was created, what remain to be done, etc. A check box located on the upper left-hand
corner of this page can be used to automatically display this description in a dialog box when a user
opens it. The Version edit box allows you to insert a short alphanumeric string that will be used to
identify the version number of the dictionary.

Lock Options

Once a categorization dictionary has been developed and validated, one may want to restrict further
changes to the dictionary or prevent other users from setting content analysis options that would be
incompatible with this dictionary. For example, a categorization dictionary may require the presence
of a specific exclusion list to handle exceptions or may assume that the program accepts specific
characters or requires prior lemmatization of the documents. The Lock Options page allows one to
restrict operations and changes in analysis options for the currently active dictionary. However,
please note that those restrictions are neither permanent nor password protected. They may be
overridden at any time by accessing the Dictionary Options dialog box again and by removing any of
the restrictions that has been set previously.

122 WordStat Users Manual

The first set of options (Changes) contains two check boxes that allow one to prevent the addition of
new entries or the deletion and editing of existing ones. When those options are enabled, all buttons
or command on the dictionary page or elsewhere that give access to editing operations are disabled.

To regain access to those editing operations, click the button and disable the items that
correspond to the dictionary to edit.
All the remaining options on this dialog box are used to prevent further changes to various options
found on the Dictionary or the Options page. Before locking any item, you have to make sure the
corresponding option is properly set. For example, if the categorization dictionary is not compatible
with prior stemming and requires some special characters to be treated as valid, you first need to
disable the Stemming option on the dictionary page, move to the Options page and enter all those
special characters in the Valid Characters edit box prior to locking those options in the Lock
Options page.
Please note that the dictionary description as well as any restriction applied to a dictionary are
automatically saved in a file with the same name as the categorization dictionary file but with an
.NFO file extension. To make sure the description and the various options follow the dictionary,
always make sure to include this .NFO file along with the .CAT categorization dictionary file.

WordStat Users Manual 123

 Working with Rules
The WordStat Rules editor may be used to define complex coding rules allowing one to specify under
which conditions a particular item or category of items should be coded. Such a feature may be useful to
differentiate between numerous meanings of a single word (disambiguation). For example, one may limit
the coding of the word "bank" to situations where the word refers to the financial institution. This can be
done by restricting the coding of "bank" to documents containing vocabulary related to monetary or
financial transactions ("cash," "money," "mortgage," "investment," etc.) or by excluding alternate meanings
such as when "bank" appears in close proximity to words like "river" or "canoe." Rules may also be used to
measure various forms of a phrase. For example, the idiom "TURN OFF may be expressed in many
different ways ("turn it off," "turned off," "turned this off," "turned his radio off"). While figuring out all the
possible forms of such an idiom may be very difficult, if not impossible, a single coding rule to look for the
word pattern "TURN*" followed by "OFF" within the same sentence could very well cover most of those
situations. Rules can also take into account the presence of words that may alter the power of an adjective,
such as negations or qualifiers like "rarely," "numerous," "few," etc. Rules may even be used to identify
sequences of events or complex actions.
In WordStat, a rule can refer to individual words, word patterns, or phrases, or it may also refer to several
items belonging to a content category of the current dictionary. A reference to a category is always
preceded by the number or pound ('#') character. For example, in the rule:

SATISFIED NEAR #PROFESSOR

the first item, SATISFIED, refers to a single word while #PROFESSOR will match any item found in the
PROFESSOR content category.

Just like words or phrases, rules may be stored anywhere in a categorization dictionary. A rule consists of a
target item and from one up to four conditions, each condition consisting of another item linked to the first
item using a Boolean (AND, NOT) or a proximity operator (NEAR, BEFORE and AFTER, or their
negative forms, NOT NEAR, NOT BEFORE, NOT AFTER). The context in which those conditions will be
tested also needs to be specified, allowing one to either consider the content of the entire document or
restrict the test to a single paragraph or a single sentence. When a proximity operation is used, one also has
to specify the maximum distance in number of words that must separate the two items in order for this
proximity condition to be tested as true or false.

To create a rule, click the button and select the RULES menu item. A dialog box similar to the
one below will appear:

124 WordStat Users Manual

The minimum requirements for a rule to be valid are:
A unique name
At least one statement consisting of a target item, a Boolean or proximity operator and a second item.
The conditions under which this rule should be tested
The weight given to items meeting those conditions
The content category where the rule will be stored be stored
The ampersand ("@") is used as a prefix to denote the presence of a rule and will be added automatically to
the rule name. In the above dialog box, the rule item @SATISFIED_PROFESSOR will be stored under the
content category POSITIVE_FEELING and will be considered as true if the word SATISFIED occurs in
the same paragraph as one of the items in the content category #PROFESSOR and if there is no item in the
#NEGATIONS category in the same sentence and within five words before the target word (i.e.
SATISFIED).
To enter a specific word, word pattern or phase, simply type the desired item. Spaces between words are
automatically converted to underscore characters. To enter a content category, type the number or pound
('#') character immediately followed by the name of the category. An existing category may also be selected
from a drop-down list by clicking the down arrow located to the right of the edit box and clicking the
appropriate category name, listed in alphabetical order.

WordStat Users Manual 125

The following operators may be used in a rule:

RULE CONDITION IS TRUE IF...

item1 AND item2 ...both items occur in the same document, paragraph or sentence.

item1 NOT item2 ...the first item occurs in the document, paragraph or sentence but not the
second one.

item1 NEAR item2 ...both items occur in the same document, paragraph or sentence, and are
no more than n words apart.

item1 BEFORE item2 ...both items occur in the same document, paragraph or sentence, and the
second item appears after the first one within the next n words.

item1 AFTER item2 ...both items occur in the same document, paragraph or sentence and the
first item appears after the second one within the next n words.

item1 NOT NEAR item2 ...the first item occurs in a document, paragraph or sentence, and is not
found within n words of the second item.

item1 NOT BEFORE ...the first item occurs in a document, paragraph or sentence, and is not
item2 followed within n words by the second item.

item1 NOT AFTER item2 ...the first item occurs in a document, paragraph or sentence, and does not
occur within n words after the second item.

By default, operators are set to <none>. To add an additional criterion, set its operator to a valid Boolean or
proximity operator. To remove criteria, set the operator immediately below the last desired criteria to
<none>. When more than one condition is set, you will be asked to specify whether you want to match all
criteria or match any one of those.

Once the rule has been properly defined, click the button located in the lower right-hand corner
of the dialog box to append the rule definition to the selected content category and to clear the form. Once
you have finished entering rules, click the close button to quit this dialog box and return to the WordStat
main screen.
Please note that, in order to prevent any recursive or cross-reference problems in rules, content categories
can only refer to words, word patterns or phrases stored in categories and will thus ignore the presence of
other rules. For example, if a category named #SATISFACTION contains 10 word patterns and three rules,
any reference to this category in a rule will take into account those 10 words and will ignore instances
where any one of the three rules have been found to be true.

126 WordStat Users Manual

 Using Lexical Tools for Dictionary-Building
Creating a comprehensive categorization dictionary is quite often a difficult, time-consuming and subjective
task. WordStat can assist you in finding words that may be related to existing words in your categories by
the use of several lexical tools:

A spelling dictionary is used to propose inflected forms of existing words already in your dictionary.
Several dictionaries are currently available for different human languages such as English, French,
Italian, Dutch, etc.

Two English thesauri are also used to propose synonyms of words already in your dictionary.

A WordNet based lexical database is used to find synonyms, antonyms as well as hypernyms,
hyponyms, coordinate terms, holonyms, meronyms, etc. This database contains over 150,000 root
words (including many proper nouns) and offers over 120,000 synonym sets. The availability of word
sense definitions allows for manual as well as automatic filtering of proper word senses.

These three tools are available through the auto suggest panel on the frequency list (see page 42) as well a
through two dictionary-building commands.

The Basic command uses the selected spelling dictionaries and the two thesauri to identify related
synonyms and inflected forms.

The Advanced command gives you access to a more powerful dictionary-building tool that uses a
WordNet based lexical database to find, not only synonyms, but all related words such as hypernyms,
hyponyms, holonyms, meronyms, coordinate terms as well as the selected spell-checking dictionaries
to find inflected forms of those words.

You will find below a description of these two dictionary-building tools:

Basic Dictionary-building Tools

To access the basic dictionary-building tool:

Select the Dictionaries page by clicking the first tab at the top of the main WordStat screen.

Press on the button.

Select the Basic command.

WordStat will immediately start looking for synonyms and inflected forms of all words in your
inclusion dictionary and will report them in a dialog box like this one:

WordStat Users Manual 127

 This dialog box displays on the first page a list of synonyms that were found to be related to existing
words in the various categories. Synonyms for a specific category are sorted so that those that were
related to several existing words in this category are located at the top of the list while synonyms
related to only a single word are located at the bottom. The numeric value under the Score column
indicates the number of existing dictionary entries to which it was related, while the value under the
Frequency column indicates how often this word has been found in the current text collection.
The second page lists all words whose spelling begins with the same letters as existing words and that
were not already included in the actual dictionary. For example, if the word "understanding" is found in
the dictionary, the program will suggest words like "understandings", "understandingly",
"understands", "understanded", "understandable", and "understandably". The frequency score indicates
how often this word has been found in the current text collection.
To display only words existing in the current text collection, select the Show Existing Words Only
option, in the upper right order of the dialog box. Please note that if this dialog is accessed prior to any
WordStat text analysis, this option will be grayed out. Running a simple frequency analysis on the
current text collection will collect the frequency information needed to allow this option to be used.
To add suggested words to the dictionary, place a check mark beside the words you would like to add
and click the Add button.
Click the Close button to return to WordStat.

128 WordStat Users Manual

Advanced Dictionary-Building Tools
The advanced dictionary-building tool can be accessed either as a stand-alone application of from
within WordStat. To run the stand-alone version:

Point to the Programs folder in the Windows' Start menu, then select Provalis Research and then click
Dictionary Builder.
To access the advanced dictionary-building tool from within WordStat:

Select the Dictionaries page by clicking the first tab at the top of the main WordStat screen.

Press on the button.

Select the Advanced command.

A dialog box like this one will appear:

The first page is used to set various dictionary and search options. The second and third pages are
used to find words and idioms semantically related to existing entries in the dictionary, while the last
page is used to find derived form of those entries.

DICTIONARY PAGE
The first page of the dictionary builder program allows you to select or change the WordStat
dictionary, specify the words and categories you want to work with, along with the type of relationship
to look for. It also allows you to specify how the program will search for inflected forms of existing
words in your dictionary.

WordStat Users Manual 129

 To select a dictionary

Click the button. A standard Open dialog box will appear.

Select the WordStat dictionary file you want to work with.

Selecting words and/or categories

By default, the dictionary-building program will search for related words and idioms for all existing
words and categories in your WordStat dictionary. To restrict the search to specific categories or
words within a category, simply deselect the words and categories you want to exclude by removing
the check marks beside them. Clicking a category check box to change its state also changes the
check box state of all words and subcategories within this category.

Specifying the type of relationship to look for

The Search for group box allows you to specify what type of relationship the program will look for.
For example, you may choose to search only for synonyms and similar terms or decide to also search
for hypernyms, hyponyms, coordinate terms, etc.

Setting how Inflected forms will be retrieved

The Match Partial Word option affects how inflected forms are found. When this option is
deactivated, the program only retrieves words that start with the whole word. For example, if the
dictionary includes the word INTELLIGENT, the program will suggest words like
INTELLIGENTLY and INTELLIGENTSIA. If the Match Partial Word option is activated, the
program will also suggest words like INTELLIGENCE, INTELLIGENCES, INTELLIGIBLE, and
INTELLIGIBLY.

Showing existing words only

By default, suggested words and phrases are presented whether or not they were found in the current
text collection. Selecting the Existing Leftover Words Only option restricts the list of suggestions to
those present in the text collection and not yet captured by the dictionary. This option will be greyed
out if no text processing has been done yet in WordStat. Running a simple frequency analysis in
WordStat prior to running this program will collect the frequency information needed to allow this
option to be available.

DEFINITIONS PAGE
Using a comprehensive lexical database such as WordNet to find related words and phrases has one
major drawback. Searching for numerous types of relationship for even small WordStat dictionaries
can yield a huge number of suggested words. For example, when searching for suggested words for a
dictionary containing 129 words grouped under 13 categories, more than 12,000 new words and
phrases were obtained, many of them unrelated to the existing categories. Browsing through such a
huge number of suggestions to find the most relevant ones can be an overwhelming task. The
Definitions page was created to somewhat reduce this burden by providing an intermediary step where
the user can select, for each of the words, the word senses that are the most relevant to the containing
category. The program offers both manual and automatic selections of word senses and also allows one
to combine both methods.

Automatic selection of word senses

130 WordStat Users Manual

WordStat dictionary builder uses a basic disambiguation algorithm to try to identify, among all word
senses, those that are the most likely to be related to the containing category. This algorithm involves
the computation for each word sense of a relevance score. The higher is this score, the more likely
the word sense will be related to the category, while a score equal to zero suggests that this word
sense is unrelated to the category. Once those relevance scores have been computed, the program can
use one of three different rules to select proper word senses.

Best - This rule instructs the program to select for each word, the sense that has obtained the
highest relevance score. When selecting the highest score, a 20% tolerance is used so that, on some
occasions, more than one word sense will be selected. This selection rule is the most conservative
one and ensures that relevant word senses are the most likely to be selected. However, we have
also found that this selection method may lack some sensitivity and may fail to select other
relevant word senses (false negatives).
Relevance > 0 - This rule instructs the program to select all word senses that have been found to be
related, even slightly, to the category. This selection rule is very liberal in that it is the most likely
to select most relevant word senses at the cost of a lack of specificity (too much false positives).
Relevance > 0.1 - This rule is slightly more conservative that the previous one, in that it also
rejects all word senses that have obtained a score of 0.1. Besides a score of zero, 0.1 is the lowest
score that may be obtained. Experiences have shown that, very often, word senses with such a low
score are unrelated to the category. Removing those word senses thus results in an increase in
specificity along with only a marginal decrease in sensitivity.
The application of any of these three rules is performed by selecting the proper rule from the Select
drop down list. This list box may also be used to select or unselect all definitions.

WordStat Users Manual 131

 Manual selection of word senses

Manual selection of word senses can be carried out either alone or after an automatic selection has
been made by the program. Manual selection is performed simply by browsing through the list of all
definitions and selecting those that are related to the current category while making sure unrelated
definitions are unselected. The decision to include or exclude a specific word sense may rely on the
displayed definition, on the relevance score, and also on the examination of all words that have been
found to be related to this specific word sense. Those suggested words are automatically displayed in
the right panel of the Definition page when the word definition is highlighted.

Selected word senses may be saved on disk by clicking the button, and later retrieve by clicking
the button.
Once the word senses have been chosen, activating the Words page will start the search, extract all
words and phrases related to the selected word senses, and will display them by categories and by the
type of relationship (synonyms, antonyms, etc.)

WORDS PAGE

The Words page displays a list of suggested words and idioms that were found to be related to existing
words in the various categories and allows you to select suggestions and add them to the existing
dictionary. The "All words" page includes a list of all words and idioms that were suggested,
irrespective of their relationship with the existing entries. The remaining pages allow one to examine
those same words by the nature of their relationship with existing entries.

132 WordStat Users Manual

 Relevance ranking and sorting
For each suggestion, a Word relevance score is computed that takes into account the number of times
an item has been suggested as well as the relevance score obtained by the word senses from which it
was derived. Those suggestions are presented in descending order of relevance so that the
suggestions that are the most likely related to the containing category are located at the top of the list
while suggestions that are less likely to be relevant are found at the bottom of this list.

Specificity index
Very often, a word is suggested in more than one category. This is especially true when the
dictionary includes categories that are semantically close each other. One good example of such a
categorization system is the Lasswell dictionary that tries to differentiate ten different forms of power
relations (power gain, power loss, cooperation, authoritative, conflict, doctrine, etc.). When making a
decision on whether a word should be added to a given category, it is important to consider whether
this word is specific to this category or whether it has also been suggested in other categories. The
Compute Specificity button allows one to obtain a specificity index as well as a list of all the other
categories in which this item also appears. This specificity index is computed by making the sum of
all relevance scores obtained by this word in the various categories and computing the proportion of
this total score that is related to the current category. A specificity of 1.0 indicates that this item has
only been suggested for this category. When the item has been found to be related to more than one
category, a list of all other categories in which it also appears is displayed in the Other Categories
column along with the relevance score obtained in each of those categories. You can use this
information to decide to which category this word should be added.

To add words or idioms to categories

Place check marks beside the item you would like to add.

Click the Add button.

INFLECTED FORMS PAGE

The Inflected Form page lists all words whose spelling begins with the same letters as existing words
and that were not already included in the actual dictionary. For example, if the word "understanding" is
found in the dictionary, the program will suggest words like "understandings", "understandingly". If the
Match Partial Word option is enabled (see Dictionary page), this same word will also yield words like
"understands", "understandable", and "understandably". The From column displays the original word
from which the inflected form has been derived.

WordStat Users Manual 133

 To add suggested words to the dictionary, place a check mark beside the words you would like to add
and click the Add button.
The Auto Select button allows one to automatically select from all the suggested forms those with
specific suffixes such as all suggested forms ending with 's' or 'ed'. When searching inflected forms of
English words, it is also possible to use WordNet to automatically select words that share the same
meaning as the original word from which it was derived. As an example, the program will
automatically select words like BEHAVIORS and BEHAVIOURS as valid forms derived from
BEHAVIOR since all three forms will yield the same WordNet definitions. One can also set this
feature to accept any new word form for which there is at least one WordNet definition containing the
original word. For example, when enabling this option the word COMPETING would be automatically
selected as a valid inflected or derived form of COMPETITION since one of WordNet definitions
associated with COMPETING (i.e. "Being in competition") contains the original word from which it
was derived.

Click the button to return to quit the dictionary builder program and return to WordStat.

134 WordStat Users Manual

 Monitoring and Customizing Substitutions
The substitution process may be used for automatic spelling correction or for lemmatization. Since
WordStat does not rely on a prior part-of-speech tagging of words to perform lemmatization, but rather on
some suffix substitution rules and lists, some improper word substitutions may occur. In specific situations,
substitution may be linguistically conceivable yet semantically invalid. For example, the noun "ground,"
referring to the solid part of the earth's surface, may be erroneously taken as the passive form of the verb
"grind" and be replaced with this infinitive form. WordStat offers a way to monitor all substitutions
performed by this substitution routine and to override those deemed necessary by creating a list of custom
substitutions, or exceptions. Such a tool may also be used to review and edit previously entered entries in
the substitution process.

To review or edit manually defined substitutions:

Enable the substitution process on the Dictionary page by putting a check mark in the box found to
the left of the Substitution list box.
Apply it to your text collection by moving to the Frequencies page.
Move back to the Dictionary page.

Click the to the right of the substitution list to review all substitutions performed. A dialog
box similar to this one will appear:

The Substitution Process page of this dialog box provides information about the internal process (if any)
involved in the lemmatization or substitution routine as well as a list of all manual substitutions.

To remove a substitution rule:

Select the rule and click the button.

WordStat Users Manual 135

To add a new substitution rule:

Click the button. The following dialog box appears:

Type in the Original edit box the word you would like to replace, then type the replacement word in
the Replace with edit box and click OK to create the new substitution rule. This new rule is
automatically added to the substitution process.

To cancel a modification:
All changes are automatically saved to disk. To cancel any change made to the list of manual
substitutions during the current WordStat session, click the button. A list of all changes
performed in this substitution process will be displayed.
Select the modification you would like to cancel and click the Undo button.

To review all performed substitutions:

Move to the Substituted Words page of the dialog box.

The table presents all performed substitutions in alphabetical order as well as other information (such as
the substituted word, how frequent the substitution has been made, the length of the original word as well as
the inverted version of the original word). Clicking any column header sorts the table in ascending order of

136 WordStat Users Manual

the data in that column. Clicking the same column header a second time will sort its content in descending
order.
Because of the algorithm used, lemmatization errors are more likely to occur on shorter words. Sorting on
the Length column allows one to focus more specifically on those short words. Also, suffix substitution
may be more problematic for some suffixes. For example, lemmatizing words ending with ING may
introduce confusion between noun, adjective and verb forms. Sorting on the Inverted column allows one to
quickly review all substitutions made to words sharing the same ending.

To correct an invalid substitution:

Select the row with the substitution you would like to override.
Press the right button of the mouse and select either Keep Invariant to instruct WordStat to keep the
word in its original form or Substitute With to specify what word will be substituted with the
selected word. When this last option is selected, the following dialog box appears:

The initial word is automatically entered in the Original edit box. Then type the replacement word
and click OK to create the rule. This new substitution rule is automatically added to a list of
exceptions and will automatically be accessed when using the currently selected lemmatization
routine.

To export the table to disk:

Click the button. A Save File dialog box will appear.

In the Save As Type list box, select the file format under which to save the table. The following
formats are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited file
(*.CSV), MS Word (*.DOC), HTML file (*.HTM; *.HTML), XML files (*.XML) and Excel
spreadsheet file (*.XLS), SPSS data files (*.SAV), or STATA data files (*.DTA).
Type a valid file name with the proper file extension.
Click the SAVE button.

WordStat Users Manual 137

To append a copy of the table in the Report Manager:

Click the button. A descriptive title will be provided automatically for the table. To edit this
title or to enter a new one, hold down the SHIFT keyboard key while clicking this button (for more
information on the Report Manager, see page 191).

To print the table:

Click the button.

To leave this dialog box

Click the button. If modifications have been made to the list but have not been saved, you will
be prompted whether those modifications need to be saved. Choosing NO will result in the loss of all
changes made to the list since you entered this dialog box or since the last time those modifications
had been saved. .

138 WordStat Users Manual

 Configuring External Preprocessing Routines
The preprocessing option allows users to access external text preprocessing routines that are not part of the
WordStat program. This option is useful to perform custom transformations on the text to be analyzed. For
example, a routine may be created to remove all foreign accents, to segment a document in a particular way,
to perform part-of-speech tagging, word disambiguation, stemming or transforming words into n-grams
(sequences of letters). Those transformations are not applied to the original documents stored in the
database but are instead performed live immediately after the textual information has been read into
memory and prior to any text processing available in WordStat (lemmatization, exclusion of words,
categorization, etc.).
Such external routines may be written in any programming language that can create or be called from a
stand-alone EXE file or a DLL. The programmer is responsible for matching the formal parameters and
result type of the conventions used by WordStat for data interchanges. Technical information on those
programming conventions is available on request from Provalis Research. The following section describes
how to configure WordStat to call an already existing text a preprocessing routine.

Calling an Executable program

The first method by which an external routine may be integrated within WordStat is through the calling of
an executable program (typically a console application with an EXE file extension). With such a method,
information is transferred between the two programs by way of temporary files generated on the fly and
stored in a default temporary folder. WordStat first creates a text file (WORDSTAT.IN) in the temporary
folder containing the text to be processed. It then calls the external routine, with optional parameters (which
may or may not include the input and output file name and their locations). Once the external program ends,
WordStat retrieves another text file (WORDSTAT.OUT) created by the external program and containing
the text to be processed in place of the original one. The two temporary files are then deleted.

To configure WordStat to call an EXE file:

Click the button located to the right of the Preprocessing list box.

Type the name you would like to give to this preprocessing routine and click OK. This name will be
added to the list box of available routines.

Enter the name of the program file including the full path or click the button to display a dialog
box that will allows browsing through folders and then select the appropriate program file.
In the Working Dir edit box, specify the working directory for the program if necessary. Specifying
$TEMP as the working directory instructs the program to set the working directory to the temporary
folder.

Enter the parameters to transfer to the program at start-up. Typically, you will transfer the input and
output file names, as well as any command line options needed for the external routine, to perform

WordStat Users Manual 139

 the required transformation. You can specify multiple parameters and can use any one of these three
string constants:

CONSTANT STANDS FOR

$TEMP The system temporary folder.

$IN The temporary text file created by WordStat and to be processed by the
external routine (the actual file name used is WORDSTAT.IN)

$OUT The name of the text file created by the external routine and retrieved by
WordStat (the actual file name used is WORDSTAT.OUT).

Calling a function in a DLL

The second method used to call a preprocessing routine is to execute an external function stored in a
dynamic library (DLL). This will manipulate the text stored in the computer memory at a specific memory
address. Since no file input and output operations are necessary to transfer information, this method is often
much faster than running an EXE file. However, because this external routine has access to the same
memory space as WordStat, great care should be taken when running or creating such a routine. To
minimize the risks involved in calling external functions, a programming convention has been imposed
where the name of the function to be called must begin with the two uppercase letters 'WS', thus reducing
the risk of calling a function that has not been written specifically for WordStat.
To configure WordStat to call a DLL function:

Click the button located to the right of the Preprocessing list box.

Type the desired name for this preprocessing routine and click OK. This name will be added to the
list box of available routines.

Enter the name of the DLL file including the full path or click the button to display a dialog box
that allows browsing through folders and then select a DLL.
Once a DLL file has been entered, the dialog box will provide a list of functions that are likely to be
compatible with WordStat (with names starting with "WS"). Select the function containing the
transformation routine needed.

WordStat must set apart in advance a "buffer size" that will contain the transformed text. By default,
the memory space is equal to the length of the original document. For many text transformation
routines, such as stemming or lemmatization, which often result in shorter text, this space should be
large enough. However, for other types of text preprocessing (e.g., part-of-speech tagging or
transformation of words into n-grams), the size of the transformed text may be twice or three times
larger than the original. The Buffer Size option allows you to specify how much larger the memory
space should be in order to hold the transformed text. A numerical value between 1 and 10 can be
used to represent the value by which the original text should be multiplied. For example, if this

140 WordStat Users Manual

 option is set to 3 and the size of the original text into memory is 10 kilobytes, then 30 kilobytes of
memory will be reserved for holding the transformed text. The calling routine should run a test to see
whether the reserved space is sufficient and should return an error message if not, allowing WordStat
to respond with an error message to the user.

Once all the options have been set, click the OK button.

Other tasks
To edit the settings of an external routine:

Select the preprocessing routine you would like to edit from the dropdown list.

Click the icon to edit the external routine settings.

To remove an external routine:

Select the preprocessing routine you would like to remove from the dropdown list.

Click the button.

WordStat Users Manual 141

 Viewing and Editing Text
WordStat's integrated text editor allows browsing and editing alphanumeric variables and documents
submitted to content analysis, as well as spell checking of text found in a specific case or in the entire data
file. When viewing plain text documents, the editing window consists of a single editing view where text
can be edited and keywords are highlighted. When viewing rich text documents, the editing and keyword
highlighting features are accessed through two separate views. The keyword highlighting feature allows one
to identify all words or phrases that have been coded as well as those belonging to specific coding
categories. The text editor can also be used for dictionary maintenance tasks by allowing the addition of
words or expressions to active dictionaries. One can also jump directly from a selected word to a
keyword-in-context table of all instances of this word.

It is also possible from this dialog box to examine and edit all numeric and alpha numeric values stored in
other variables of the data file. To view and edit those values for the current case, the Show all variables
check box should be selected. When enabled, the screen will split vertically. On the left side, a panel with a
list of all variables with their values for this case will be shown. To edit any of those values, press the F2
key or double-click the value to edit.
The following table provides a short description of available buttons and controls:

CONTROL DESCRIPTION
This button allows the importation of text from various file formats including plain
text file, RTF, MS Word, WordPerfect, MS Write or HTML. If the variable containing
the document supports only plain text documents then all formatting options and
unsupported features, such as bullets, graphics or headers are removed.
Export the current document to disk. Plain text document may only be saved as plain
ANSI document while RTF documents may be saved in plain text or RTF format.

142 WordStat Users Manual

 Print the current document.

Cut the selected text to the clipboard.

Copy the selected text to the clipboard.

Paste text from the clipboard at the current cursor position.

Reverse the last action made to the text.

Search for a specified word or phrase.

Search for and replace a specified word or phrase.

Spell-check all cases.

Spell check only the current text.

Pressing this button allows you to add the selected word to an active dictionary. It may
also be used to produce a KWIC table of the currently selected word or expression.
Variable: When more than one text variable is analyzed, this drop-down list box allows
selecting the variable to display in the edit box.
Highlight: WordStats text editor displays all words that have been coded using bold characters
while words belonging to the active category are showed in blue. To change the active
category and highlight all words that belong to a selected category, simply choose the
proper category from this list box. To highlight all categories using different colors,
set this option to <all categories>.
This button allows you to access the color coding dialog box that lets you to assign to
each specific category in the dictionary specific font and background colors (see
below).

Move to the first case of the data file.

Move to the last case of the data file.

Move to the previous case.

Move to the next case.

WordStat Users Manual 143

Assigning color codes to categories

The color code dialog box allows you to assign specific font and background colors to each category of
the current inclusion dictionary.

To access the color code dialog box

Set the Highlight drop down list to <all categories>.

Click the button.

To change the colors of a category

Select the category in the categories list box

Use the Font and Background color selectors to choose from a list of predefined colors or click the
Other button to define a custom color.
Information on color codes are automatically saved with the project options

144 WordStat Users Manual

 Displaying distribution using Barcharts or Pie Charts
WordStat allows one to produce barcharts or pie charts to visually display the distribution of specific
keywords or categories. To produce such charts:
Move to the Frequencies page.
Set the Sort By option to the desired graphic order of the values.
Select the rows you would like to plot (multiple but separate rows can be selected by clicking
while holding down the CTRL key)

Click the button.

Three types of charts may be used to depict the distribution of keywords or content categories:

The vertical bar chart is the default chart used to display absolute or relative frequencies
of keywords or content categories.
The horizontal bar chart displays the same information as the vertical one but is especially
useful when the number of keywords is high and their labels cannot be displayed entirely
on the bottom axis.
The pie chart is useful to display the relative frequency of each keyword and compare
individual values to other values and to the whole. Numerical values displayed in pie
charts are always expressed in percentages of either the total frequency or case
occurrences.

The Plot option allows one to select the values that will be used as the scale for the length of bars in
barcharts or as the percentage base for pie charts. For barcharts the options are:
FREQUENCY Number of occurrences of the keyword
% SHOWN Percentage based on the total number of keywords displayed in the table
% PROCESSED Percentage based on the total number of words encountered during the analysis
% TOTAL Percentage based on the total number of words that have not been excluded
NO OF CASES Number of cases where this keyword appears
% CASES Percentage of cases where this keyword appears

For pie charts, two options are available to specify how percentages will be computed:
FREQUENCY Percentage based on the total frequency of keywords
NO OF CASES Percentage based on the total number of case occurrences

The View Others option displays an additional bar or slice representing all items in the frequency table that
have not been selected.

The following table provides a short description of available buttons and controls:

WordStat Users Manual 145

 Controls Description
Press this button to vertically display the labels on the bottom axis.

Press this button to append a copy of the graphic in the Report Manager. A
descriptive title will be provided automatically. To edit this title or to enter a new
one, hold down the SHIFT keyboard key while clicking this button (for more
information on the Report Manager, see page 191).
Press this button to retrieve a chart previously saved on disk.

Press this button to save a chart on disk. Charts are saved in a proprietary format and
may be edited and customized using the Chart Editor.
Pressing this button allows you to print a copy of the displayed chart.

Click this button to turn on/off the 3-D perspective for the current chart.

This button allows you to edit various features of the chart such as the left and bottom
axis , the chart and axis titles, the location of the legend, etc.
This button is used to create a copy of the chart to the clipboard. When this button is
clicked, a pop-up menu appears allowing you to select whether the chart should be
copied as a bitmap or as a metafile.
Pressing this button closes the chart dialog box and returns to WordStat's main
screen.

Customizing barcharts and pie charts

Clicking the button on the chart dialog box gives access to a dialog box to customize the
appearance of barcharts and line charts. The options available in this dialog box represent only a small
portion of all settings available.
To further customize the chart, modify data points, value labels, or series order, click the button
located to the right-hand side of the dialog box.

LEFT OR BOTTOM AXIS

Minimum / Maximum - WordStat automatically adjusts the vertical axis scale to fit the range of values
plotted against it. To manually set these values, type the desired minimum and maximum.
Increment - Increasing or decreasing this value affects the distance between numbers as well as tick
marks. Horizontal grid lines are also affected by modification of this value.
Horizontal Grid - This option turns horizontal grid lines on and off. Grid lines extend from each tick
mark on an axis to the opposite side of the graph. To increase or decrease the number of grid lines
or the distance between those lines, change the Increment value of the axis. A list box also allows
a choice among five different line styles to draw those grid lines.

146 WordStat Users Manual

LEGEND
Location - This option positions the legend. Legends may be placed at Top, Left, Right and Bottom
side of the chart.
From top - When the legend is displayed on the left or the right side of the chart, this option specifies the
legend's top position in percent of total chart height.
From left - When the legend is displayed on the top or the bottom chart, this option specifies the
legend's top position in percent of total chart width.

TITLES
Proper titles and axis labels are of utmost importance when describing the information displayed in a
chart. By default, WordStat uses variable names and labels as well as other predefined settings to
provide such descriptions.
The title page allows one to modify the top title, as well as the labels on the left, bottom and right axis.
To edit the title, select the proper radio button. Enter several lines of text for each title by pressing the
<Enter> key at the end of a line before entering the next line.
The Font button to the right-hand side of the edit box allows changing the font size or style of the
related title.

3-D VIEW
Orthogonal - Turning this option off disables the free elevation and rotation of the 3-D chart.
Zoom - This option zooms the whole chart. Expressed as a percentage, increasing the value positively
will bring the chart towards the viewer, increasing the overall chart size as the Zoom value
increases.
3-D Percent - The 3-D Percent property indicates the size ratio between chart dimensions and chart
depth by specifying a percent number from 1 to 100.
Perspective - Use this property with Orthogonal unchecked to modify the 3-D perspective of the
Chart. Larger values add more depth perspective.
Bar shadow - Enabling this option adds dark shades to the sides of 3-D bars. Turning it off will color
the sides of the bar the same as the front.
Bar width - This option determines the percent of total bar width used. Setting this value to 100
makes joined bars.
Bar depth - Use this property to limit the depth that each bar series uses. By default, bars will take up
the part proportional to the number of bar series in the chart so that the back of a bar will join the
front of the bar immediately behind it. To insert a gap between series of bars, decrease this value.
Pie depth - Use this property to change the thickness of the pie chart.

WordStat Users Manual 147

 Creating and Using Norm Files
A useful element in interpreting the results of a content analysis is the possibility of comparing the obtained
results to some normative data and identifying how similar or dissimilar the observed frequencies are
compared to those norms. For example, one may wish to compare the vocabulary of an adult victim of a
brain injury to vocabularies of normal adults or the mission statement of a business to a collection of
mission statements of Fortune 500 companies. One may also establish the reading level of a school manual
by comparing its vocabulary to collections of books read by children of various ages. Normative data are
typically computed on a large sample of documents and represent either general norms with data from a
wide variety of sources or are computed on a more specific text corpus related to the channel, the domain
area or the specific situation being studied. For example, one could compare the speeches of candidates of a
presidential election to a large collection of English text from different sources (newspapers, novels,
technical documents, etc.) or to a more specific corpus of spoken English or to a collection of political
speeches. Comparison of word frequencies to norms established on a general corpus may be especially
useful to identify the specific terminology of a set of documents. On the other hand, using a more specific
collection may allow one to identify subtler differences or nuances.
WordStat allows one to create normative data on a collection of documents based on either the content of a
categorization dictionary or on the frequency of individual words. Those norms may be stored on disk and
later be compared to the results of a content analysis performed on other documents. When comparing
results to norms established using a categorization dictionary, it is highly recommended to use the same
dictionaries and the same analysis options as the ones used to create those norms. Using different settings
may result in invalid comparisons. To prevent such a situation, WordStat will detect any difference in
settings and issue a warning message, pinpointing all differences in those settings. On the other hand,
comparing the results to a norm file based on a comprehensive list of word frequencies may provide a more
flexible solution than using a norm established using a categorization dictionary since a single word
frequency norm file may be used to compare results obtained using various categorization systems. In such
a situation, WordStat automatically computes from the words in the norm file the expected frequencies for
each content category. However, it is important to remember that if the categorization dictionary contains
phrases or rules, the expected frequency will likely be underestimated and may thus be invalid.
When a comparison to a norm file is performed, WordStat appends four columns to the right of the
frequency table and computes each item's expected frequency, the deviation from the observed frequency,
the Z value (standardized deviation) and its two-tailed probability.

To create a norm file based on content categories:

From SimStat or QDA Miner, open the corpus on which the norms will be established.
Call WordStat.
Select the categorization dictionary on which the norms should be computed and set the various
analysis options (exclusion list, lemmatization, etc.).
Move to the FREQUENCIES page to force the computation of frequencies on the content categories.

Click the button and select the SAVE AS A NORM FILE command. A file-saving dialog box
will be displayed.

148 WordStat Users Manual

 Enter the name of the file under which you would like to store the norms (by default, the .wnorm file
extension is added to the file name), then click SAVE to create the file.

To create a norm file based on word frequencies:

From SimStat or QDA Miner, open the text collection on which the norms will be established.
Call WordStat.
If a categorization dictionary is active, disable it and set the various analysis options (exclusion list,
lemmatization, etc.). It is recommended to set the minimum frequency or record occurrence to "1"
and to disable the option to remove words under a specific frequency or occurrence in order to obtain
a detailed frequency list of all words in the normative sample.
Move to the FREQUENCIES page to force the computation of word frequencies.
Click the button and select the SAVE AS A WORD FREQUENCIES command. A file-saving
dialog box will be displayed.
Enter the name of the file under which you would like to store the norms (by default, the .wfreq file
extension is added to the file name), then click SAVE to create the file.

To compare the obtained frequencies with existing norms:

Set the required dictionary and options and move to the FREQUENCIES page to instruct WordStat
to compute the frequencies of words or of content categories.
Click the button and select COMPARE TO NORM FILE or COMPARE TO WORD
FREQUENCIES, depending on whether you would like to compare the obtained frequencies to
norms established on content categories or on words.
Select the norm file to which you would like the comparison to be made and click OK.

To Remove comparison statistics:

Click the button and select the REMOVE NORM STATISTICS command.

WordStat Users Manual 149

 Performing Text Retrieval using Keywords
The KEYWORD RETRIEVAL feature can retrieve any document, paragraph or sentence containing a
specific keyword, a combination of keywords or no keyword at all. Text units corresponding to the search
criteria are returned in a table on the Results page. This table may then be printed or saved to disk. It may
also be used to create tabular or text reports as well as to attach QDA Miner codes to the retrieved text
segments.
To start the Keyword Retrieval feature, go to the Frequencies page and click the button. This dialog
box can also be accessed from other parts of the program to retrieve text units associated with a cluster or
with a specific association. It may also be accessed by selecting an item in the frequency page,
right-clicking and selecting KEYWORD RETRIEVAL from the pop-up menu. When calling this function,
a dialog box similar to this one appears, allowing you to specify the desired search criteria:

RETRIEVE - This option determines the text unit on which the search will be performed as well as what
will be retrieved. You can select three different text units:
The Documents search unit allows WordStat to apply the search expression on each document
associated with a specific case and, if a specific document meets the search condition, its location
will be displayed.
Setting this option to Paragraphs allows WordStat to display any paragraph meeting the search
condition.
When selecting Sentences as the search unit, WordStat returns sentences meeting the search
condition.

150 WordStat Users Manual

KEYWORD FILTERING - This group of options allows one to select the keywords on which the
retrieval will be based. Setting the first list box to No Keyword will retrieve all text units for which
no keyword has been found. This option is especially useful to identify topics or themes that have not
been covered by the current categorization dictionary or new keywords that should be added to
enhance the coverage of existing categories in the dictionary. Setting the filtering to Include at least
allows one to retrieve text units containing a minimum number of keywords from a selected list. To
select the keywords, click the arrow button to show all available keywords and then click the desired
items. The minimum number of keywords a specific unit must contain in order to be retrieved is
specified using a small edit box with spin buttons to the right. Setting this numerical value to the total
number of keywords selected will force the program to retrieve only those units containing all those
keywords. Setting this number to a lower value will retrieve all text units containing at least this
number of keywords from the selected list. In the example shown above, any unit containing
keywords from two or more of the four categories HUMOR, NIGHTLIFE, OUTDOOR or SPORT
will be retrieved.

To enter a second filtering condition, click the button. You can choose to link the two
filtering conditions using either one of the three Boolean expressions: AND, OR or NOT. Choosing
AND will retrieve all text units fulfilling both criteria; selecting OR will result in a retrieval of text
units meeting either the first or the second condition, or both, while choosing the NOT Boolean
operator will retrieve text units meeting the first condition but not the second one.

To limit the filtering conditions to a single statement, click the button.

VARIABLE FILTERING - The second group of options allows one to use other variables restrict the
retrieved text units to specific cases selected according to some logical condition. This filtering
condition may consist of a simple expression, or may include up to two expressions joined by a
logical operator (i.e., AND, OR). In the above screen shot, only text units from cases where the
variable GENDER is equal to MEN will be retrieved.
The following table shows the various operators available for each data type:

DATA TYPE AVAILABLE OPERATORS

NOMINAL / ORDINAL Equals

Does not equal
Is empty
Is not empty

NUMERIC and DATE Equals

Does not equal
Is greater than
Is lesser than
Is greater than or equal to
Is lesser than or equal to
Is empty
Is not empty

WordStat Users Manual 151

 BOOLEAN Is true
Is false

STRING Contains
Does not contain
Is empty
Is not empty

ADD VARIABLES - This dropdown checklist box may optionally be used to add the values stored in
one or more variables to the table of retrieved segments for the specific case from which a text
segment originated.

Once all the search options have been set properly, simply click the button to retrieve the
selected text units.

Working with the Retrieved Text Units

The retrieved text units are displayed in a table found on the RESULTS page.

This table contains the case number and the variable from which the segment originates, as well as the value
of all additional variables selected by the user (see the ADD VARIABLES option above). When searching
for paragraphs or sentences, the table also displays the text associated with the retrieved unit and its location
(its paragraph and sentence number). By using arrow keys or by clicking a row the associated text is
displayed in a separate window at the bottom of the screen with all keywords in bold. Selecting specific
words or phrases in this text window by right-clicking displays a pop-up menu to assign them to a content
category, the exclusion list or to obtain a keyword-in-context list.

152 WordStat Users Manual

To sort the table of retrieved units in ascending order by any column, simply click the column header.
Clicking the same column header a second time sorts the rows in descending order. Tables may also be
printed, stored as a text report, or exported to disk in various file formats such as Excel, ASCII, or HTML.

To remove a search hit from the hit list:

Select its row and then click the button.

To assign a QDA Miner code to a specific search hit:

In the table of search hits, select the row corresponding to the text segment you want to code.
Use the CODE drop-down list located above this table to select the code to assign.

Click the button to assign the selected code to the highlighted text segment.

To assign a QDA Miner code to all search hits:

Use the CODE drop-down list located above this table to select the code to assign.

Click the button to assign the selected code to all text segments matching the search expression.

NOTE: To automatically attach QDA Miner tags to all paragraphs or sentences associated with all
currently displayed content categories or keywords, you may use the autocoding feature available by
clicking the button on the Frequencies page (see page 37).

To create a new QDA Miner code:

Click the button. A dialog box will appear allowing you to create a new QDA Miner code (for
more information see Adding a QDA Miner codes).

To create a report of retrieved segments:

Click the button. The sort order of the current table is used to determine the display order in the
report. This report is displayed in a text-editing dialog box and may be modified, stored on disk (in
RTF, HTML or plain text format), printed, or cut-and-pasted into another application. Graphics and
tables may also be inserted anywhere in this report.

To export the table to disk:

Click the button. A Save File dialog box will appear.

WordStat Users Manual 153

 In the Save As Type list box, select the file format under which to save the table. The following
formats are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited file
(*.CSV), HTML file (*.HTM; *.HTML) and Excel spreadsheet file (*.XLS).
Type a valid file name with the proper file extension.
Click the SAVE button.

To print the table:

Click the button.

To close the Keyword Retrieval dialog box:

Click the button.

154 WordStat Users Manual

 Creating Bubble Charts
Bubble charts are graphic representations of contingency tables where relative frequencies are represented
by circles of different diameters. This type of graphic chart allows one to quickly identify high- and low-
frequency cells and is thus especially useful for presentation purposes. Many features of the chart can be
customized to highlight specific findings. Rows and columns can be moved freely or deleted, and one can
adjust the color of each cell as well as the fonts used in the chart.
The bubble chart represents graphically the underlying table and displays the corresponding measure used
(code occurrences or frequencies, word counts or percentages) and will also reflect the currently selected
display settings (such as count, percentage of rows or columns, etc.).

To create a bubble chart:

Move to the CROSSTAB page.
Set the Tabulate option to either Case Occurrence or Total Frequency.
Set the With option to the desired independent variable.
Set the Display option to specify how this information will be displayed.

Click the button. A dialog box similar to this one will appear:

WordStat Users Manual 155

To adjust the size of the bubbles
In the Bubbles group box at the top of the window, adjust the Size option to Small, Medium or
Large.

To adjust the color of the bubbles

Select the cell or group of cells you would like to alter.
Right click and select the SET COLOR command. A dialog box will appear letting you choose a
specific color value.

To move a row or a column

Click anywhere on the cell header containing the row or column label and hold the mouse down.
Drag the mouse cursor over the desired new location and release the mouse button.

To delete selected rows or columns

Select a cell range covering the rows or columns you would like to delete.
Right-click to display a popup menu.
Select the REMOVE command and then choose SELECTED ROWS or SELECTED COLUMNS
depending on your desired action.

To adjust the font size and style of the title and labels
Click the Change button beside the Title or Labels options. A Font setting dialog box will appear,
letting you change the font, the font size, style, and color.

To edit the title

Click anywhere in the title region and type in the new title. The height of the title region is
automatically adjusted to the number of lines in the title.

The following table provides a short description of additional buttons:

Press this button to append a copy of the chart in the Report Manager. A descriptive title
will be provided automatically. To edit this title or to enter a new one, hold down the SHIFT
keyboard key while clicking this button (for more information on the Report Manager, see
page 191).
Click this button to save a chart on disk. Charts may be saved in BMP, JPG or PNG graphic
file format or may be saved in a proprietary format (.WSX file extension) that can later be
edited and customized using the Chart Editor.

156 WordStat Users Manual

Clicking this button allows you to print a copy of the displayed chart.

This button creates a copy of the chart to the clipboard. When this button is clicked, a
shortcut menu appears allowing you to select whether the chart should be copied as a bitmap
or as a metafile.
Pressing this button closes the chart dialog box and returns to WordStat's main screen

WordStat Users Manual 157

 Using Heatmap Plots
Heatmap plots are graphic representations of crosstab tables where relative frequencies are represented by
different color brightness or tones and on which a clustering is applied to reorder rows and/or columns. This
type of plot is commonly used in biomedical research to identify gene expressions. When used for text
mining, such an exploratory data analysis tool facilitates the identification of functional relationships
between related keywords and group of values of an independent variable by allowing the perception of
cells clumps of relatively high or low frequencies or of outlier values.
The heatmap plot implemented in WordStat allows one to graphically examine the relationship between
keywords (rows) and values of an independent variable (columns). While the clustering available through
the WordStat Frequency (see Hierarchical Clustering and Multidimensional Scaling, pages 65-73) is
performed on individual cases and documents, the clustering analyses used in the heatmap plot are
performed directly on the crosstabulation tables. As a consequence, the similarity index computed for two
keywords and used for clustering does not represent their co-occurrences within cases but measures the
similarity of their distribution among the various groups of the independent variable. Likewise, two
subgroups defined by values on the independent variable will be considered near to each other if the
distributions of keywords in those two groups are similar.
The heatmap plot can be performed either on keyword frequency or occurrences within cases (present or
absent). For both types of analysis, the observed frequency is transformed into a percentage by dividing the
frequency by either the total number of words in a specific subgroup (keyword frequency) or by the total
number of cases in this subgroup (case occurrences).

158 WordStat Users Manual

To create a heatmap:
In the WordStat main window, move to the crosstab page.
Set the TABULATE option to either Keyword Frequency or Case Occurrence.
Set the WITH option to the desired independent variable.

Click the button.

HEATMAP PAGE
The main section to the right of this page has shown in the screen shot below, displays the heatmap grid
representing the relative frequencies of each cell (row and column intersection) using different brightness or
color tones. Optional dendrograms are displayed at the top and to the left margin of this grid. The size of
these dendrograms may be adjusted by moving the mouse cursor over the bottom edge (upper dendrogram)
or the right edge (dendrogram on the left) of the dendrogram and dragging its limit to the desired size.
The font size used to display the row and column values may also be adjusted by clicking the or
buttons located on the top toolbar. The size of cells and the distance between dendrogram leaves are
automatically adjusted to the new font size.
To identify which specific cases or text documents are associated with a cell or group of cells, simply select
the rectangular area for which you would like to obtain such a list, click the button and select
documents, paragraphs or sentences. WordStat locates the documents or text segments associated with those
cells and displays them in the Keyword Retrieval dialog box (see page 150).

ROWS OPTIONS GROUP

Clustering of rows - Enabling this option reorders the rows according to the result of a cluster analysis
and display a dendrogram at the left of the heatmap plot. Keywords that are distributed across the
various subgroups in a similar way will tend to be grouped under the same cluster.
Sort by - When the clustering of keywords is disabled, rows may be sorted in alphabetical order, in
descending order of frequency or case occurrence or displayed in their original categorization
dictionary order.
No Clusters - This option allows setting how many clusters the clustering solution should have. When
two clusters or more are selected, horizontal red lines are drawn in the heatmap to delineate those
clusters.

COLUMNS OPTIONS GROUP

Clustering of columns - Enabling this option reorders the columns according to the result of a cluster
analysis and display a dendrogram at the top of the heatmap. Columns with similar distribution of
keywords will tend to be grouped under the same cluster. When the clustering option is disabled, the
columns are presented in their ascending order of values. Disabling the clustering of columns is
especially useful to preserve the ordinal nature of the values. For example, when looking for a
relationship between some words or keywords and publication years, then disabling the clustering of
columns will automatically sort those columns in chronological orders allowing the identification of
temporal trends.

WordStat Users Manual 159

 No Clusters - This option allows setting how many clusters the clustering solution should contain.
When two clusters or more are selected, vertical red lines are drawn in the heatmap to delineate those
clusters of keywords.

HEATMAP OPTIONS GROUP

Colors - The colors list box allows the selection of various color schemes to represent differences in
percentages. Monochrome schemes will express differences using levels of brightness of a single
color where black always represents the lower limit of the selected range. Multicolor spectrums may
be used to represent greater nuances in the selected range of percentages.
Minimum / Maximum - The minimum and maximum slide bars allows the setting of the range of values
that will be used to display variations of color tones or brightness. By default, the minimum value is
set to zero while the maximum value is set to the highest observed percentage. To increase the
contrast at a specific location of this range, minimum and maximum limits may be adjusted. All cells
with values lower than the minimum will be represented with the color located on the left end of the
selected color spectrum while cells with percentages higher than the maximum limit will be
displayed with the color locate to the right end of this spectrum. Reducing the range of values
increases the contrast in a specific region of percentage values.
KEYWORD AND GROUP DENDROGRAM PAGES
The second and third pages of the heatmap dialog box allow a more detailed examination of the
clustering of words or categories (Keyword Dendrogram) and of all values on the independent variable
(Group Dendrogram). WordStat use an average-linkage hierarchical clustering method to create
clusters from a similarity matrix. The result is presented in the form of a dendrogram (see below), also
known as a tree graph. In such a graph, the vertical axis is made up of the items and the horizontal axis
represents the clusters formed at each step of the clustering procedure. Items that tend to be distributed
similarly within the other variable appear together are combined at an early stage while those that have
dissimilar distributions tend to be combined at the end of the agglomeration process.

160 WordStat Users Manual

 No Clusters - This option sets how many clusters the clustering solution should have. Different colors
are used in the dendrogram to indicate membership of specific items in different clusters.
Display - This option sets whether the vertical lines of the dendrogram represents the agglomeration
schedule or the similarity indices.
At times, it is desirable to see some areas of a dendrogram. The and buttons may be used to
zoom in or out of the dendrogram.

CLUSTERING STATISTICS PAGE

The fourth page of the heatmap dialog box allows the close examination of the matrices of similarities
among keywords or among groups as well as the statistics associated with the agglomeration processes of
both the keywords and the groups.

WordStat Users Manual 161

 Performing Correspondence Analysis
Correspondence analysis is a descriptive and exploratory technique designed to analyze relationships
among entries in large frequency crosstabulation tables. Its objective is to represent the relationship among
all entries in the table using a low-dimensional Euclidean space such that the locations of the row and
column points are consistent with their associations in the table. The correspondence analysis procedure
implemented in WordStat allows one to graphically examine the relationship between words or content
categories and subgroups of an independent variable. The results are presented using a 2 or 3-dimensional
map. Correspondence analysis statistics are also provided to assess the quality of the solution. WordStat
currently restricts the extraction to the first three axes. This insures that the results remain easily
interpretable. To further differentiate among some values of the independent variable, we recommend
applying a filter to restrict the analysis to a subset of values.
The first two pages of the dialog box, provides graphical displays of correspondence maps. When the
number of words or categories or the number of subgroups is less than four, only the 2-D Map page can be
accessed. When the comparison is restricted to two groups of individuals, only one axis can be extracted. In
this situation, the axis is plotted diagonally in the two-dimensional space. We have done this mainly for
readability reasons: plotting all the keywords on a single horizontal axis would have produced a cluttered
list of keywords that would have made the graph useless.
The 2-D correspondence plot offers the ability to remove a keyword or a class of the independent variable
and to automatically recompute the analysis on the remaining items. It also allows one to obtain a KWIC
table or perform a keyword retrieval for a specific keyword. To perform any of these operations, simply
click the desired item to display a popup menu and choose the proper command. If more than one item is
close by the mouse click, the menu will offer the possibility to choose on which keyword or class the
operation should be performed.
You will find immediately below a description of the various controls in the dialog box, followed by some
guidelines for interpreting correspondence plots.

2-D Map control

PLOT When more than two axes have been extracted, this control allows you to select all the
possible axe combinations that can be graphed on the two axes of the plot.

2-D and 3-D Map Controls

Words This checkbox allows you to display or hide the row points (i.e., words or category names)

Groups This checkbox allows you to display or hide the column points (i.e., subgroup labels).

Clicking this button enables to zoom in a plot. To zoom an area of the plot, hold the left
mouse button and drag the mouse down/right. You'll see a rectangle around the selected area.
Release the left mouse button to zoom
Clicking this button restores the original viewing area of the plot.

162 WordStat Users Manual

 Items distributed in a very similar way among subgroups of the categorical variables may be
plotted on top of each other, making them hard to differentiate. Clicking down this button
adds some random noises to the location of individual words or keywords, allowing one to
clearly identify those that overlap. To remove the random noises, click the button again.
Moving this slider allows the adjustment of the scaling used to plot groups or classes of the
categorical variable on the correspondence plot. While the positions of keywords and groups
relative to the origin of the axes are significant, it is important to remember that the scaling
used for keywords and groups are independent from each other. This slider may thus be used
as a reminder of this fact. It may also be used to increase the readability of some plots, by
moving group names so that they do not overlap keyword names.
Clicking this button brings up a dialog box to customize the appearance of the
correspondence plots (click here to obtain more information on the various settings that may
be changed).
Press this button to append a copy of the correspondence plot in the Report Manager. A
descriptive title will be provided automatically. To edit this title or to enter a new one, hold
down the SHIFT keyboard key while clicking this button (for more information on the
Report Manager, see page 191).
This button is used to create a copy of the chart to the clipboard. When this button is clicked,
a pop-up menu appears, allowing you to select whether the chart should be copied as a
bitmap or as a metafile.
Clicking this button allows you to print a copy of the displayed chart.

Clicking this button closes the Dendrogram & Concept Map dialog box and returns to
WordStats main window.

3-D Plot controls

This button can be used to show or hide left, bottom and back walls.

Clicking this button draws anchor lines from the floor to the data point to better locate data
points in all 3 dimensions.

Clicking this button allows you to change the viewing angle of the chart. To rotate the chart,
make sure this button is selected, click any area of the chart, hold the mouse button and drag
the mouse to apply the desired rotation.
Locating a data point on the depth dimension of a 3-D plot can be very difficult especially
when the plot remains static. One often has to rotate this plot constantly on the various axes
to get an accurate idea of where the data point is located on this third axis. Clicking this
button forces WordStat to rotate the plot automatically. To disable the automatic rotation,
click a second time.

WordStat Users Manual 163

Interpreting correspondence analysis results
Interpretation of correspondence analysis maps can be somewhat tricky and should be made with great
care, especially when examining the relationship between row points and column points. Here are some
basic rules that should help you interpret such maps:

Relationship among words or categories (row points):

The more similar is the distribution of a word or a content category among subgroups to the total
distribution of all keywords within subgroups, the closer it will be to the origin. Words or
categories that are plotted far from this point of origin have singular distributions.
If two words or computed categories have similar distributions (or profiles) among subgroups of
the independent variable (columns), their points in the correspondence analysis plot will be close
together. For example, if the words consist of artist names and the studied subgroups represent
different age groups, then if the form of the distribution of two different artists among those age
groups is similar, then they will tend to appear near each other. Words with different profiles will
be plotted far from each other. Please note however that two points may appear close to each other
on a two or three axes solution, but may in fact be far apart when taking into account an additional
dimension.

Relationship among subgroups (column points)

The more singular a profile of words/categories for a subgroup is, compared to the distribution of
those words/categories for the entire sample, the farther this subgroup will be from the point of
origin.
If two subgroups of individuals have similar profiles of word usage or content categories, they will
be plotted near each other.
Subgroups with different profiles will be plotted far from each other.

Relationship between words/categories and subgroups (row and column points):

Great caution should be taken when interpreting the distances between two sets of points (row and
column points). The fact that the name of a subgroup is near a specific word or content category
should not necessarily be interpreted as an indication that they are closely related.
While the distance between words or content categories and subgroups has no interpretable
meaning, the angle between such a word point and a subgroup point from the origin is meaningful:
An acute angle indicates that the two characteristics are correlated.
An obtuse angle, near 180 degrees, indicates that the two characteristics are negatively correlated.

164 WordStat Users Manual

 In the example below, REM, U2 and RadioHead could be viewed as related to 18 years old
listeners. However, while both REM and U2 points are closer to the 18 years old subgroup,
Radiohead should be identified as more characteristic of those listeners since it is farther from the
origin.

Words or categories closely associated with two subgroups will be plotted in an angle from the
origin that will lie between those two groups. In the above example, the rock groups Korn and
Metallica seem to be characteristics of both 15 and 16 years old listeners.
For a more comprehensive description of this method, its computation and applications, see Greenacre
(1984). For an application of correspondence analysis to the analysis of textual data see Lebart, Salem,
and Berry (1998).

WordStat Users Manual 165

 Editing the Case Descriptor
The EDIT CASE DESCRIPTOR command allows you to define a case descriptor that will be used to
identify each case based on the values it contains in one or several variables. Such a string is used to
identify cases when retrieving segments as well as when analyzing case similarities using clustering,
multidimensional scaling or proximity plots.
This command can be accessed by clicking the button in the upper left-hand corner of the main
window. A dialog box similar to this one will appear:

This dialog box allows you to specify a label that will be used to describe each case. The label may be
changed by editing the text in the DESCRIPTION STRING edit box. To insert the value stored in a specific
variable into the description, simply enter the variable name in uppercase letters and enclose this name
between braces. Alternatively, you can insert a variable name at the current caret location by clicking the
corresponding item in the VARIABLES list located just above the edit box.
If you enter the following string:
{GENDER} subject - {AGE} years old
The {GENDER} and {AGE} strings will be replaced with their corresponding value for this specific case.
If the current case contains information about a seventeen-years-old male, the above string will be displayed
as:
Male subject - 17 years old
It is also possible to insert the following string:
{CASENUM}
This string will display a unique case number, representing the physical order of this case in the project file.

166 WordStat Users Manual

 Filtering Cases
The FILTER CASES command temporarily selects cases according to some logical conditions. You can use
this command to restrict analyses to a subsample of cases or to temporarily exclude some group of subjects.
Two types of filtering are available in WordStat: a basic filtering dialog box that provides some guidance
for easily building filters, and a powerful xBase filtering tool that allows one to create more complex
filtering expressions that contains more advanced mathematical functions, as well as Boolean and relational
operators.
If no filtering expression has been set or if the current filtering condition is relatively simple, a dialog box
similar to this one will be displayed:

This dialog box consists of two major sections. The upper section allows one to specify up to four filtering
conditions joined by logical operators (such as AND, OR). Each condition consists of a variable, an
operator and, if needed, some numerical, categorical or string values. The following table presents the
various operators available for each data type.

DATA TYPE AVAILABLE OPERATORS

NOMINAL / ORDINAL Equals

Does not equal
Is empty
Is not empty
NUMERIC and DATE Equals
Does not equal
Is greater than
Is lesser than
Is greater than or equal to
Is lesser than or equal to
Is empty
Is not empty
BOOLEAN Is true
Is false

WordStat Users Manual 167

 STRING Contains
Does not contain
Is empty
Is not empty
DOCUMENT Is empty
Is not empty
Is coded
Is uncoded
IMAGE Is empty
Is not empty

The lower section allows one to filter cases based on the presence of words or phrases associated with
specific content categories in the document being processed by WordStat. This section is disabled by
default and becomes active as soon as a content analysis has been performed. The AND, OR and NOT
Boolean operators are used to determine how those two categories should be combined.
For example:
APPEARANCE AND ART Selects only cases that contain words in both categories.
APPEARANCE OR ART Selects cases that contain words in either one of these two categories.
APPEARANCE NOT ART Selects all cases containing a word in category APPEARANCE but that
do not contain any word found in the ART category.

Once a filtering expression has been entered you can apply the filter and leave this dialog box by clicking
the Apply button. If the filter expression is invalid, an error message will appear and exiting from the dialog
box will not occur.
To temporarily deactivate the current filter expression, click the Ignore button. The filter expression will be
kept in memory and may be reactivated by selecting the FILTER CASES command again and clicking
Apply.
To exit from the dialog box and restore the previously active filtering expression, click the Close button.

A more advanced xBase filtering dialog box can be accessed by clicking the button. The
filtering expression can be typed directly into the Filter edit box using the proper syntax, or one can use any
element displayed in the upper part of the dialog box to build a valid expression. To obtain more
information on expression operators and evaluation rules and supported xBase functions, see page 171.

168 WordStat Users Manual

You can directly type the filtering expression in the Case Filter edit box using the proper syntax, or use any
elements displayed on the upper part of the dialog box to build a valid expression. To restore previously
used filtering conditions, click the down arrow button located to the right of the Filter edit box.
The upper part of the Filter dialog box contains various elements to help you build a valid xBase filtering
expression:

VARIABLE NAME LIST BOX - Double-clicking a variable name from the list box located to the left of
the dialog box inserts that name in the edit box at the current caret position.
FUNCTION LIST BOX - A list of valid xBase expressions is displayed to the right of the dialog box.
Double-clicking an xBase function in the list box, inserts that function at the current caret position.
When a function requires one or several arguments, the argument section remains highlighted. To
replace the highlighted text with a value, an expression or a variable name, simply type the proper
text on the keyboard or select a variable name or function.
NUMERIC, BOOLEAN AND RELATIONAL OPERATORS BUTTONS - Clicking any relational or
Boolean operation or on any numeric button inserts the corresponding symbol in the edit box at the
current caret position.

The following section provides a description of xBase syntax rules used in the FILTER command and a
detailed description of each xBase function.

WordStat Users Manual 169

Expression Operators and Rules
Operators used in xBase expressions are standard in every xBase dialect.

String Operators

+ Joins two strings. Trailing spaces in the strings are placed at the end of each string.
- Joins two strings and removes trailing spaces from the string preceding the operator and
places them at the end of the string following the minus sign operator.

Numeric Operators
+ Addition
- Subtraction
* Multiplication
/ Division
^ Exponentiation (or **)
Relational Operators
= Equal to
== Exactly equal to
<> Not equal to
# Not equal to
!= Not equal to
< Less than
> Greater than
<= Less than or equal to
>= Greater than or equal to
$ Is contained in

Logical Operators (Notice the periods surrounding the operator)

AND both expressions are true

OR either expression is true
NOT either expression is false

Evaluation Order

When more than one type of operator appears in an xBase expression, the order of evaluation is as
follows: Expressions containing more than one operator are evaluated from left to right. Parentheses
are used to change the evaluation order. If parentheses are nested, the innermost set is evaluated first.

170 WordStat Users Manual

Numeric operators are evaluated according to generally accepted arithmetic principles:
operators contained in parentheses
exponentiation
multiplication and division
addition and subtraction
Order of evaluation may be altered with parentheses:
3+4*5+6 = 29
(3+4)*5+6 = 41
(3+4)*(5+6) = 77

Logical operators are evaluated as NOT first, AND second, and OR last. Logical evaluation order may
also be altered with parentheses. In multiple conditional expressions that contain the NOT operator,
always use parentheses to enclose the NOT operator with the expression to which it applies.

Supported xBase functions

The following xBase functions are supported in the SORT CASES and FILTER CASES command

NOTE: Memo variable names are not allowed in xBase expressions.

ALIAS()

Returns the Alias name of the current work area as a string.

ALLTRIM (String)

Trims both leading and trailing spaces from a string. The string may be derived from any valid xBase
expression.

ALLTRIM(" Provalis ") returns 'Provalis'.

AT (SearchString, TargetString)

Determine whether a search string is contained within a target. If found, the function returns the
position of the search string within the target string (relative to 1). If not found, the function returns 0
(zero).

AT("gh", "defghij") returns 4.

CHR (Val)

Converts a decimal value to its ASCII equivalent.

CHR(83) returns 'S'

WordStat Users Manual 171

CTOD (String)

Converts a character string into an xBase date. The string must be formatted according to the Windows
date format settings.

CTOD("12/31/94")

DATE ()

Returns the system date (today). Use DTOC(DATE()) to retrieve today's date formatted according to
the Windows settings.

DAY (DateVariable)
Returns the day portion of an xBase date as an integer.

DELETED ()
Returns True if the case is deleted and False if not deleted.

DESCEND (String)
An xBase function that inverts a key value using 2's complement arithmetic. The result of the operation
is the arithmetic inverse of the key value. When inverted keys are sorted in ascending sequence, the
result is in descending order. A filter expression could be
DESCEND(DTOS(billdate)) + CUSTNO

DTOC (DateVariable)
Converts an xBase date into a character string formatted according to the Windows settings. For
example, if the date format was American and the date variable contained March 21, 1995,
DTOC(datevariable) would return '03/21/1995'.

DTOS (DateVariable)
Converts an xBase date into a string formatted according to standard xBase storage conventions
(CCYYMMDD). For example, December 21, 1993 would be returned as '19931221'. Indexes that
contain date elements should use the DTOS() function, which naturally collates into oldest date first.

EMPTY (Variable)
Reports the empty status of any xBase variable. Character and date variables are empty if they consist
entirely of spaces. Numeric variables are empty if they evaluate to zero. Logical variables are empty if
they evaluate to False.
Memo variables that contain no reference to a memo block in the associated memo file are empty.

IF (Logical, True Result, False Result)

This is the immediate if function. If the Logical expression is true, return the True result, otherwise
return the False result. The types of the True Result and the False Result must be the same (i.e., both
numeric, or both strings, etc.) The logical expression must of course evaluate as True or False.
IF(DATE() - CTOD("12/31/93") > 0,"This Year", "Last Year")

172 WordStat Users Manual

IIF (Logical, True Result, False Result)
Supported exactly like IF() as noted above.

INDEXKEY ()
Returns the current index key as a string. (Same as ORDKEY()).

LEFT (String, Length)

Returns the leftmost characters of the expression for the defined length.
LEFT("xyzabc", 3) returns 'xyz'.

LEN (Expression)
Returns the length of the expression result as an integer.

LOWER (String)
Converts the string expression into lower case.

MONTH (DateVariable)
Returns the month portion of an xBase date as an integer.

ORDER ()
Returns the current index order as an integer.

ORDKEY ()
Returns the current index key as a string. (Same as INDEXKEY())

PADC (String, Length, Character)

Centers the passed string between a number of the passed character to make the string the specified
length.
'[' + PADC("Smith", 9 ,"-") + ']' returns '[--Smith--]'.

PADL (String, Length, Character)

Pads the passed string to the specified length with the specified characters. If the string is longer than
the value specified by Length, the string is truncated to this length.

'[' + PADL("Smith", 8, "*" ) + ']' returns '[***Smith]'.

'[' + PADL("John Smith", 8, " " ) + ']' returns '[John Sc]'.

WordStat Users Manual 173

PADR (String, Length, Character)
Pads the passed string to the specified length using the specified character. If the string is longer than
the value specified by Length, the string is truncated to this length.

'[' + PADR("Smith", 8, " " ) + ']' returns '[Smith ]'.

'[' + PADR("John Smith", 8, " " ) + ']' returns '[John Sc]'.

RAT (SearchString, TargetString)

Determine whether a search string is contained within a target, starting from the right side of the target
string. If found, the function returns the position of the search string within the target string (relative to
1). If not found, the function returns 0 (zero).

RAT( "ab", "abzaba" ) returns 4.

RECCOUNT ()

Returns the number of cases in the table as a long integer.

RECNO ()

Returns the current physical record number as a long integer.

RIGHT (String, Length)

Returns the rightmost characters of the expression for the defined length.

RIGHT("xyzabc", 3) returns 'abc'.

SELECT ()

Returns the workarea number for the current work area as a long integer.

SPACE (Length)

Returns a string consisting entirely of spaces for the defined length.

STOD (String)

The inverse of DTOS(). STOD() converts a string formatted according to standard xBase storage
conventions (CCYYMMDD) to an xBase Date formatted according to the Windows settings.
STR (Number, Length, Decimals)

Converts a number into a right-justified string with decimal digits following the decimal point. The
total length of the string is defined by the length parameter. STR(RECNO(), 5, 0) is a common
indexing element that ensures creation of unique keys if appended to another variable element.

174 WordStat Users Manual

 An index key using this expression could be built with NAME + STR(RECNO(),5,0)

If the decimals parameter is omitted, the function defaults to zero decimal places. If the length
parameter is omitted as well, the length of the result is the length of the variable.

STRZERO (Number, Length, Decimals)

Converts a number into a, zero-padded right justified string with decimals digits following the decimal
point. The total length of the string is defined by the length parameter.
STRZERO( 1234, 10, 2 ) returns '0001234.00'
If the decimals parameter is omitted, the function defaults to zero decimals. If the length parameter is
omitted as well, the length of the result is the length of the variable.

SUBSTR (String, Start, Length)

Returns a portion of the string expression starting at the defined start location for the defined length.
SUBSTR('xyzabcd', 3, 4) returns 'zabc'.

TIME ()
Returns the system time as a string in the form HH:MM:SS.

TRANSFORM (Expression, Picture)

Transform converts strings and numeric values into formatted character strings. The function
transforms the result of the first expression in accordance with the second picture string.
The picture string is made up of two parts. The first part is the Function string and it is optional for both
strings and numeric values (as long as the second Template string is present).
A character string transformation picture may consist of only a Function string or only a Template or
both.
A numeric picture must contain a Template string; the Function string is optional.
A logical value must contain only a Template string with Template characters L or Y.
The Function string consists of a leading @ character followed by one or more formatting characters. If
the Function string is present, the @ character must be the first character in the picture string with its
formatting characters immediately following and it may not contain spaces.
If a Template string exists as well, it follows the Function string. A single space separates the Function
string and the Template string.
Function string characters allowed for numeric values are:
B left justify;
C display CR after positive numbers;
X display DR after negative numbers;
Z blank a zero value;
( enclose negative numbers in parentheses.

Function string characters allowed for strings are:

R inserts unassigned template characters;

WordStat Users Manual 175

 ! converts all alpha characters to upper case.

The @R Function requires a Template; the ! Function does not.

The Template string describes the format on a character by character basis. The Template string is
made up of special characters which have specific results and optional unassigned characters which
either replace characters or are inserted in the formatted string depending upon the absence or presence
of the @R Function string.
Template assigned characters are as follows:
A,N,X,(,# are place holders and are interchangeable;
L displays logical values as T or F;
Y displays logical values as Y or N;
! converts the corresponding character to upper case;
, (comma) or a space (in Europe) in a numeric template separate the elements of a number;
. (period) or , (comma - in Europe) in a numeric template specify the decimal position;
* fills leading spaces with asterisks in a numeric template;
$ as the leading character in a numeric template results in a floating dollar sign being placed in
front of the formatted number.

Example: Where "phone" is a character variable holding a phone number with no formatting characters.
'transform(phone, "@R (###) ###-####")' returns '(909) 699-6776'.

If the formatting characters were actually present in the variable, the "@R" function would be omitted

For numeric variables,

'transform(123456.78, "$9,999,999.99")' returns ' $123,456.78'.

TRIM (String)
Removes trailing spaces from the string expression.

UPPER (String)
Converts the string expression into upper case. Character variables used in index expressions
should always be converted to upper case to insure correct collating sequence.

VAL (String)
Converts a string of numeric characters into its equivalent numeric value. The conversion stops at
the first non-numeric character encountered (or the end of the string).

VAL("123ABC") returns a value of 123.

YEAR (DateVariable)

Returns the year portion of an xBase date as an integer.

176 WordStat Users Manual

 Performing Analysis on Manually Entered Codes
The QDA Miner software is a computer-assisted qualitative coding tool specifically designed to manually
code documents. While WordStat was designed mainly to perform automatic content analysis of textual
data, one may also use this module to manually assign codes to text. When used for such a purpose, codes
need to be manually typed into the text itself. WordStat may then be used to extract those codes and
perform various types of analyses (frequency, cross-tabulation, co-occurrences). At least two approaches of
coding may be used to achieve this:

Unique Keywords
Unique keywords or code name may be inserted anywhere in the text. Those keywords should preferably
not be an existing dictionary word. For example, it may consist of an abbreviation of one or several words,
includes special symbols (such as #, & ^, _ etc.) or numeric digits. The retrieval of those codes can then be
achieved by adding all those keywords to the inclusion list . If special symbols have been used, they should
also be specified as valid characters on the options page.

Codes in square brackets

WordStat provides an option to process only the text found between square brackets (i.e. [ and ] ). Codes
corresponding to categories may be typed directly in the text and placed within square brackets. By
disabling all dictionaries and setting this option, the program simply ignores all text outside those brackets
and performs a frequency count of all words found inside square brackets. This method has several
advantages over the other method. First, there is no need to enter all existing codes in the inclusion
dictionary since they will be processed automatically. Misspelled keywords will always be extracted and
may thus be easily identified and changed. Also, the processing time and memory requirement can be much
lower than the other approach since WordStat won't process any text found outside those brackets.
Besides the possibility to extract manually entered codes and perform frequency and comparison on those
codes, there are several other ways in which the program may be used to assist the work of human coders.
Here are just a few examples of possible uses:
During the exploratory phase of the analysis, word frequency and crosstabulation may be used to
identify differences between subgroups in word usages, differences that may have remained
unidentified.
The KWIC list may be used to locate and visualize text associated with specific codes either to
validate the coding or identify associated themes. The KWIC list may also be used to perform a
systematic search of words frequently found along with a specific code. The examination of all cases
containing those words may then help identify instances where the code should have been used.
A dendrogram of keyword co-occurrences or a crosstabulation of manually entered codes against all
words in a text may allow identification of specific words associated with codes and permit the
development of dictionaries. Such dictionaries may then be used either to ensure a more systematic
application of manually entered codes or to develop and validate a dictionary that will later be used
for automatic content analysis.
Reliability of coding made by a single coder at different times or by several coders may also be
assessed using inter-rater agreement statistics. See Computing Inter-Rater Agreement Statistics
(page 178) for more information on how to compute those statistics.

WordStat Users Manual 177

 Computing Inter-Rater Agreement Statistics
The reliability problem in manual coding
When coding of a text is performed automatically using dictionaries, the reliability of the coding is
irrelevant since the rules used for coding are explicit and are applied systematically on the entire corpus of
texts. However, when the coding is performed manually by human coders, individual differences in
interpretation of codes between human coders often occur no matter how explicit, unambiguous and precise
the coding rules are. Even a single coder is often unable to apply the same coding rules systematically
across time. One way to ensure the reliability of the application of coding rules is to ask different raters to
code the same content or to ask a single coder to code the same content at different times. The comparison
of codes is then used to identify differences in interpretation, clarify ambiguous rules, identify ambiguity in
the text, and ultimately quantify the final level of agreement obtained by those raters.

Inter-rater agreement measures

WordStat provides eight different inter-rater agreement measures to assess the reliability in coding. The
assessment made by WordStat is based on the presence or absence of a specific code and is performed on
each code individually. WordStat cannot assess the reliability of multinomial coding. However, this type of
assessment can be achieved within SimStat by storing those codes into numeric variables and using the
TABLES | INTER-RATER command.
The simplest measure of agreement for nominal level variables is the proportion of concordant coding out
of the total number of codings made. Unfortunately, this measure often yields spuriously high values
because it does not take into account chance agreements that occur from guessing. Several adjustment
techniques have been proposed in the literature to correct for the chance factor, three of which are available
in the WordStat. The following are the assumptions made by each of these correction techniques:
Free marginal adjustment assumes that all categories on a given scale have equal probability of being
observed. It assumes that coders decisions were not influenced by information about the distribution of
the codes. This coefficient is equivalent to the Bennett, Alpert and Goldstein's (1954) S coefficient,
Jason and Vegelius's (1979) C Coefficient, and Brennan and Prediger's (1981) kn Index (Zwick, 1988).
Scott's pi adjustment does not assume that all categories have equal probability of being observed, but
does assume that the distributions of the categories observed by the coders are equal.
Cohen's kappa adjustment also does not assume that all categories have equal probability of being
observed. Contrary to the pi measure, it does not assume that the distribution of the various categories
is equal for all coders. In the computation of the chance factor, Kappa takes into account the
differential tendencies or preferences of coders.
WordStat also offers three additional adjustments for ordinal level variables. These are similar to the
previous measures except that they also take into account the ordinal nature of the scales by weighting the
various levels of agreement. They apply the same tree model of chance agreement used in the previous
measures for nominal data.
Free marginal adjustment for ordinal level variables also assumes that all categories on a given scale
have equal probability of being observed.

178 WordStat Users Manual

 Krippendorf's R-bar adjustment is the ordinal extension of Scott's pi and assumes that the
distributions of the categories are equal for the two sets of ratings.
Krippendorf's r adjustment is the ordinal extension of Cohen's Kappa in that it adjusts for the
differential tendencies of the judges in the computation of the chance factor. The following table
illustrates the difference between those six inter-rater measures:

LEVEL OF MEASUREMENT

DISTRIBUTION ASSUMPTION Nominal Level Ordinal Level

Theoretical uniform distribution Free marginal Free marginal

Observed distribution of each rater Cohen's Kappa Krippendorf's r

Observed distribution of all raters Scott's pi Krippendorf's R-bar

WARNING: When the computation is based on keyword frequencies (rather than case occurrences) and
when codes can be used more than one time per case, it is usually recommended to use ordinal or interval
level agreement measures. Otherwise a difference in frequency for a specific code will be treated as a single
disagreement. For example, if for a single case a coder assigns a code twice and another coder uses this
same code three times, nominal level agreement measures will treat this difference as a single disagreement
and will ignore the fact that the both raters may be in agreement in two instances. As a result, the overall
agreement level will be underestimated. However, nominal level measures may still be used in those
situations if the researcher wishes to treat any difference in frequency as a disagreement.

How to compute inter-rater agreement statistics

In order to compute inter-rater agreement measures within WordStat, codes assigned by each coder
should be stored in different alphanumeric variables (character or memo variables). For example, in
four different raters are assessed, four text variables should be created, each one containing the codes of
a single coder. Two different methods can be used to achieve this. The text variable where the content
to be analyzed is stored, may be replicated as many times as there are raters. Coders are then asked to
insert their codes directly in the text variable. The second method requires the creation of empty
alphanumeric variables which will contain the codes assigned by coders. Coders then read the content to
be analyzed and write down all their codes in this empty variable.

To compare the codes of different raters

In the SimStat program, use the CHOOSE X1-Y command from the STATISTICS menu to assign all
variables containing codes of different coders to the list of dependent variables.
Select the CONTENT ANALYSIS command from the STATISTICS menu.

WordStat Users Manual 179

 Set the various options in the DICTIONARIES page so that all codes are extracted from the
text variables.
Click the CROSSTAB page.
Set the WITH list box to <variables>. The AGREEMENT list box should appear.
Set the TABULATE list box to either Keyword Frequency or Case Occurrence.
Select the appropriate measure of agreement. (Tables based on keyword frequency should
preferably be assessed using ordinal or interval level inter-rater agreement measures while
tables based on case occurrence can be analyzed using either nominal or ordinal level
measures).

To identify the source of disagreement

When codes are inserted directly in the original text, it becomes possible to verify the source of
disagreement by extracting the content in which those codes appear. To perform such a task:
Click the KEYWORD-IN-CONTEXT page.
Set the LIST option to Included Keywords and set the keyword to the category you would like
to examine.
Set the SORT BY option to Case Number.
You can then use the KWIC table to identify specific contents that were associated with a code by a
specific coder but not by other coders.

180 WordStat Users Manual

 Exporting Frequency Data
Various case statistics may be appended to the existing data file or exported to disk in different file formats
including SPSS for Windows, Excel, HTML, XML, and tab or comma delimited text files, allowing those
data to be further analyzed. The resulting data consist of a matrix where each row represents a case and
where the statistics on content categories will be stored in columns along with a few additional variables.
To append or export to disk content category statistics, click the button located at the top of the
Frequencies page is used to access the following dialog box:

DESTINATION - This option allows you to choose whether the new variables should be appended to the
current data file or written into a new file. If this last option is selected, a dialog box will appear
allowing you to specify the name and location of the new file. When data are saved to a new data
file, additional variables are created to store the case number and the numerical values of each
independent variable.

DATA TO SAVE - This option allows one to choose among four different kinds of data that may be
saved:
Keyword frequencies
Case Occurrences (i.e., a dummy variable with 0 when absent or 1 when present)
Percentage of words (i.e., the frequency of the keyword divided by the total number of words in
the case)
TF*IDF (i.e., the keyword frequency weighted by inverse document frequency).

VARIABLE NAMES - This option lets you determine what method should be used by WordStat to
create new variable names. When set to KEYWORD, the program will attempt to use each keyword

WordStat Users Manual 181

 as the name of a new variable. Illegal characters are automatically removed and long names are
truncated to the first 10 characters. Duplicated variable names are distinguished by the substitution of
numerical digits at the end of the name. When this option is set to PREFIX, variable names are
created by adding successive numeric values to a user-defined prefix. For example, if the edit box at
the right of the prefix option is set to "WORD_", the variable names will be WORD_1, WORD_2,
WORD_3, etc. The order of creation of the variables corresponds to the sort order used in the
FREQUENCIES page.

VARIABLE TYPE - By default, WordStat saves keyword statistics in as many variables as there are
keywords or content categories listed on the frequency page. For example, if the frequency table
contains 100 items, then 100 variables will be necessary to store the statistics associated with each
item. When you choose to store the occurrence of codes, WordStat offers you the possibility of
storing the observed occurrences in a limited number of polynomial (or multinomial) variables. For
example, if the maximum number of different content categories per case is no more than 10, then
you may instruct WordStat to create 10 numeric variables and store, in each of those, a numeric
value representing one of the content categories. If less than 10 categories are found in a specific
case, then the remaining variables are left empty. To store values in a limited set of nominal
variables, choose the Multiple Polynomial Variables option and enter the Maximum Number of
Variables that should be used for storing the values representing the content categories. To store the
name of those categories rather than their numerical values, select Multiple String Variables
instead. If the maximum number of content categories found in a single case is higher than the
specified number of variables, then a warning message will appear to let you know that some
information has been lost and to indicate the maximum number of content categories encountered in
the project. To export occurrences as zeros and ones in as many variables as there are codes, select
the Multiple Dichotomous Variables option.
ADD VARIABLES - This drop-down checklist box may be used to add the values stored in one or more
variables to the exported data file along with the statistics.
SAVE TOTAL NUMBER OF WORDS - This option appends a numeric variable named TOTWORDS
that contains the total number of words processed in each case.
Clicking the PREVIEW button displays a grid allowing one to see what the data file will look like.

182 WordStat Users Manual

 Exporting Categorization Models
A typical text categorization process may involve any one of the following steps:
User defined text preprocessing
Automatic lemmatization
Exclusion of words and phrases
Categorization of words, word patterns, phrases and coding rules into content categories using a
categorization dictionary

Specific settings, such as the inclusion of special characters or numerical digits, may also need to be set in
order to collect relevant information.
In order to apply such a process to external documents, one should normally import the documents into a
SimStat data file or QDA Miner project file, run WordStat and replicate the exact same settings as those
originally used. An alternate solution is to export the categorization model to disk, which would include all
the relevant information and settings, and then use either the WordStat Document Classifier utility program
or functions of the Software Developer's Kit (release date Summer 2005) to retrieve the saved model and
apply it to the new documents.

To save a categorization model to disk:

Set the various analysis options on the Dictionaries and Options pages necessary to reproduce the
required categorization process.

Go to the Frequency page and click the button located at the top of the page.
Select the Export Categorization Model command. A dialog box should appear asking you for a
file name.
Enter the file name of the model you want to create and click Save.
By default, categorization model files are saved with a .wcat file extension in the \Models subfolder under
the program folder. NOTE: While the information in the exclusion list and categorization dictionary is all
stored in the categorization file, running a categorization model from outside WordStat may still require the
availability of some resource files such as language dictionaries or preprocessing libraries (EXE or DLL).
This should not cause an inconvenience when applying those models on the same computer as the one used
to create the model, since information about the original locations of those resource files is always stored
within the model file. However, when attempting to apply those categorization models on another
computer, the calling application may have some difficulty locating the needed resource files. Those files
should be stored either under paths identical to those on the original computer, in the application folder or
under specific subfolders. When an attempt is made to apply a saved categorization or classification model
for which some resource files are missing, an error message will be displayed providing the list of all
missing files, their original location and alternate locations where they might be.
For information on how to apply the saved categorization model, please refer to the WordStat Document
Classifier section or to the WordStat Software Developer's kit.

WordStat Users Manual 183

 WordStat Document Classifier
The WordStat Document Classifier utility program is a stand-alone application that may be used to perform
content analysis and automatic text classification on a text pasted from the clipboard or stored in a file. It
may also be used to analyze a collection of documents stored in a SimStat or QDA Miner data file.
Performing a content analysis or a text classification on an existing document is quite simple and involves
three easy steps: 1) loading the document in the main editing window, 2) opening the classification or
categorization model previously saved on disk and 3) applying the model. Detailed results of the content
analysis or classification are displayed in tables at the bottom of the dialog box.

Analyzing a single document

The document classifier supports several document file formats such as ASCII text files, HTML, Rich Text,
MS Word, WordPerfect and Acrobat PDF files.

Step #1 - Opening the document

To load the document to be analyzed, select the OPEN command from the DOCUMENT menu or
click the button. An Open File dialog box will be displayed. In the File of Type list box, select
the format of the file you would like to read, locate the file, select it and click the Open button.

You may also type directly in the text editing window or paste a text previously copied to the
clipboard by moving to the text editor and then selecting the PASTE command from the
DOCUMENT windows or by clicking the button.

Step #2 - Opening the model

To open the categorization or the text classification model:, select the OPEN command from the
MODEL menu or click the button. Content analysis models are stored in files with a .wcat file
extension, while document classification models are stored in files with a .wclas file extension.

Select the model you would like to use and click the Open button.

Step #3 - Applying the model

Select the APPLY command from the MODEL menu or click the button.

When a categorization dictionary is applied, a single frequency table is displayed at the bottom of the
page with the following statistics:

184 WordStat Users Manual

 FREQUENCY Number of occurrences of the keyword.

% SHOWN Percentage based on the total number of keywords displayed in the table.

% TOTAL Percent based on the total number of words that have not been explicitly
excluded.

TF*IDF Term frequency weighted by inverse document frequency. Such a weighting is

based on the assumption that the more often a term occurs in a document, the
more it is representative of its content, yet, the more documents the term
occurs in, the less discriminating it is. The inverse document frequency term
used for weighing items in the document classifier program is based on the
observed frequency and the number of documents in the original data file used
when the model was created.

When a classifier is used, a second table is shown allowing you to examine the classification decision made
by the classifier as well as the computed values associated with each class of the categorical variable. When
the k-Nearest Neighbors algorithm is used for classification and the database containing the training set can
be located, a third table is shown, displaying the "k" most similar documents, their ranking and their
similarity scores.

Analyzing a collection of documents

The Document Classifier can perform content analysis and automatic text classification on a collection of
documents stored in a SimStat or QDA Miner data file. Categorization and classification results, as well as
scores per class, may then be stored back into the data file or exported to another file.

Step #1 - Opening the collection of documents

Select the OPEN DATA FILE command from the DOCUMENT menu, or click the button. An
Open File dialog box will be displayed. Locate the data file containing the documents to analyze,
select it and click the Open button.

The content of the data file is displayed in a table while the text to be categorized or classified is
displayed on its right.

Step #2 - Opening the model

To open the categorization or the text classification model, select the OPEN command from the
MODEL menu, or click the button. Content analysis models are stored in files with a .wcat file
extension, while document classification models are stored in files with a .wclas file extension.

Select the model you would like to use and click the Open button.

WordStat Users Manual 185

Step #3 - Applying the model

Select the APPLY command from the MODEL menu, or click the button.

When a categorization dictionary is applied, a single frequency table is displayed at the bottom of the
screen with the number of occurrences of each keyword included in the model, as well as the total
number of words.
When a classifier is used, a second table is shown, allowing one to examine the classification
decision made by the classifier as well as the computed values associated with each class of the
categorical variable. This table is synchronized with the database shown at the top of the screen, so
that moving from one row to another - either in the database or this classification table - moves to the
corresponding row in the other table.
If the k-Nearest Neighbors algorithm is used for classification and the database containing the
training set can be located, a third table is shown, displaying the "k" most similar documents, their
ranking and their similarity scores.
To store in the opened data file either the predicted class or the computed score obtained for every
class, click the button. A dialog box similar to this one will appear:

Enter the variable name that will contain the predicted class.

To save the scores associated with each class and upon which the classification has been made, put a
check mark beside Save scores for all classes and enter a variable prefix (up to 7 characters).
Variable names are created by adding successive numeric values to this prefix. For example, if the
edit box at the right of the Variable Prefix option is set to "CLASS", the variable names will be
CLASS1, CLASS2, CLASS3, etc.

If any one of the specified variables does not exist, WordStat will create new ones and store the
numerical values associated with either the predicted class or the class scores. A confirmation dialog

186 WordStat Users Manual

 box will ask to confirm the creation of those new variables, as well as to overwrite any existing
variables.

To export any table to disk:

Click the button. A Save File dialog box will appear.

In the Save as type list box, select the file format under which you would like to save the table. The
following formats are supported: ASCII file (*.TXT), Tab delimited file (*.TAB), Comma delimited
file (*.CSV), HTML file (*.HTM; *.HTML), Excel spreadsheet file (*.XLS).

Type a valid file name with the proper file extension.

Click the Save button.

To print a table:

Click the button.

WordStat Users Manual 187

 WordStat Software Developers Kit
The WordStat Software Developer's Kit (SDK) consists of a library of functions that may be called from
numerous programming languages, including some database programming environments, in order to use
WordStat categorization and classification technologies for tasks like:

Automatic assignments of tags or index terms to documents.

Classification of documents for automatic classification or routing.

Identification of documents meeting specific content criterion.

For more information on this Software Developer's Kit, please contact [email protected].

188 WordStat Users Manual

 Performing Multivariate Analysis
One benefit of the integration of a content analysis module within an existing statistical program is the
ability to easily perform on numerical results of content analysis, various statistical analyses such as
frequency, crosstabulation, multiple regressions, reliability analysis, etc. Among the various multivariate
analysis techniques used in content analysis, cluster analysis, factor analysis, and, to a lesser degree,
correspondence analysis are often used to establish the relationship existing between different words or
categories of words. The table below illustrates some types of analysis that may be performed with SimStat
and, if needed, the required module to perform those analyses.

TYPE OF ANALYSIS REQUIRED MODULE

Multiple regression analysis None

n-way ANOVA/ANCOVA None
Reliability analysis None
Inter-rater agreement None
Factor Analysis (with or without varimax rotation) None
Principal Component Analysis (without rotation) MVSP
Correspondence analysis MVSP
Advanced cluster analysis MVSP

First step - Saving numerical results into a data file

In order to perform a statistical analysis on categories or words, you first need to create new numeric
variables that will contain, for each case in the data file, the occurrence or frequency of specific words or
categories. To create those variables:

Set the various options of the DICTIONARIES page.

Perform the content analysis by clicking the FREQUENCIES page.

Click the button to access the Save Data dialog box.

Activate the SAVE KEYWORD COUNT checkbox.
Set the DESTINATION option to Append to Current File.
Set the DATA TO SAVE list box to Keyword Frequency.
Set the VARIABLE NAMES option to Keyword to instruct WordStat to use the names of the
categories (or included words) as the names for the new variables.
Click the OK button to proceed to the saving of those data and return to WordStat.
Click the OK button of the WordStat window to return to SimStat.

WordStat Users Manual 189

For more information on how to store numeric or textual results into the current data file or into a new data
file see Exporting Frequency Data (page 181).

Performing a Cluster Analysis of words or categories

Select the CHOOSE X-Y command from the STATISTICS menu and assign all the newly created
variables to the list of independent or dependent variables. (The distinction between dependent and
independent variables is not relevant for this kind of analysis. However, all variables assigned to a
single category will be processed together.)
Choose the OTHER | CLUSTER ANALYSIS command from the statistics menu to display the
option dialog box.
Set the various analysis options to one's preferences. (Please take note that in order to perform a
cluster analysis on keywords rather than on cases, the Transpose Data option should be left
deactivated).
Click the OK button to perform the statistical analysis.

Performing a Factor Analysis of words or categories

Select the CHOOSE X-Y command from the STATISTICS menu and assign all the newly created
variables to the list of independent or dependent variables. (The distinction between dependent and
independent variables is not relevant for this kind of analysis. However, all variables assigned to a
single category will be processed together.)
Choose the OTHER | FACTOR ANALYSIS command from the statistics menu.

Set the various options to one's preferences.

Click the OK button to perform the statistical analysis.

190 WordStat Users Manual

 Managing Outputs wih the Report Manager
The Report Manager is a separate application that has been designed to store, edit and organize documents,
notes, quotes, tables of results, graphics and images created by QDA Miner or imported from other
applications. Items can be added to the Report Manager directly from QDA Miner without needing to run
the Report Manager.

The button, found in many locations in QDA Miner, may be used to copy entire documents, tables and
charts to the Report Manager.

Selected text segments or image areas may also be appended by clicking the button.
To access the Report Manager from QDA Miner, run the REPORT MANAGER command from the
PROJECT menu.
The program presents its information as an outline, allowing a hierarchical organization of miscellaneous
pieces of information that is ideal for project management, organizing ideas, structuring information, or
designing and writing a research report.
The workspace emulates the appearance of Windows Explorer or of a standard Help file with the Table of
Contents (TOC) on the left and the Editor on the right.

WordStat Users Manual 191

Table of Contents panel
Report Manager files are made up of items or topics that are like chapters in a book. Each item can be
thought of as a separate word processor, a table or a graphic file editor or viewer, all of which are stored
together in the QDA Miner project file. This panel provides powerful functions for organizing items and
structuring the information in a hierarchical manner.

Item Editor
The largest panel on the right of the program window is the Item Editor, which is like a built-in word
processor. This is where the item selected in the Table of Contents can be edited. Clicking a Table of
Contents item displays its contents for editing.

Toolbar
The Toolbar provides quick access to the most frequently-used functions. Just position the mouse over a
tool button and wait for the display of a brief text describing its function.

Comment panel
The Comment panel below the Item Editor allows the insertion and editing of comments related to the
selected topic. When new items are added to the Report Manager from QDA Miner, a default comment is
often already present, providing useful information about the origin of this item.

Working with the Table of Contents

To create a new item:
Select the Table of Contents entry that will be the "parent" or "sibling" of the new item.
Select the NEW command from the ITEMS menu or click the button. A dialog box like this
one will appear:

Enter the title for the new item.

If the new item should be a "child" of the selected item, click Under Current Item; if the item
should be positioned after the current item, then set it to After Current Item.
Select whether the new item will be a Document or a Folder. Folders are empty items that are used
as containers for other items.
Click OK. The new item will become the current one.

192 WordStat Users Manual

To import items from files:
Select the item under which the imported items will be stored.
Select the IMPORT FILES command from the ITEMS menu or click the toolbar button. An
Open dialog box will appear.
Select the type of data you would like to import by selecting the appropriate Files of Type list box
option. The Report Manager can import the following data types:
DOCUMENTS - Plain text (.TXT), MS Word (.DOC), WordPerfect (.WPD), Rich Text (.RTF)
or HTML files (.HTM or .HTML)
GRAPHICS - Windows Bitmap (.BMP), Windows Metafile (.WMF), JPEG files (.JPG or
.JPEG) and Portable Network Graphic files (.PNG)
CHARTS - QDA Miner or WordStat Charts (.WSX)
DELIMITED DATA - Tab delimited (.TAB) or Comma Separated Value (.CSV) data files.
Select one or several files to be imported and click the OPEN button.

To rename an item:
Select the item to be renamed.
Select the RENAME command from the ITEMS menu or click the toolbar button.
In the Item Title Dialog, change the title.
Click OK.

To delete an item:
Select the item to delete in the Table on Contents.

Select the DELETE command from the FILE menu or click the toolbar button.
You will be asked to confirm that you really want to delete the item. If you're sure, then click Yes.
NOTE: Be aware that you cannot undo this if you make a mistake.

Moving Items
As more items are created and the Report Manager hierarchy grows, it is inevitable that you will want to
move items around, either to place one item under another, or to promote one to a higher level.
The easiest way to move items in the Table of Contents is by using drag-and-drop operations. Using the
mouse, you can move an item to a different location or move a group of items stored under a "parent" item
by dragging this "parent" item to its new location.
Select the item to move by clicking and holding down the left mouse button. (Keep the mouse
pressed until the drag-and-drop operation is completed.)
Drag the item to its new location and, only then, release the mouse button.

WordStat Users Manual 193

 The dragged item will now become a "child" of the destination item.
To move the item to the same level as the item under the cursor, simply hold the ALT key while
dropping the dragged item.
You can also use menu commands and toolbar buttons to move items. To promote an item is to move it to a
higher level in the hierarchy, making the item a "sibling" to its former "parent". To demote an item is to
move it to a lower level and make it a "child" of its previous "sibling". After selecting the item you want to
move, use one of the following four commands:

To promote the selected item, click the button, or select the PROMOTE command from the
ITEMS menu.
To demote the selected item, click the button, or select the DEMOTE command from the
ITEMS menu.
To move the selected item up relative to its siblings, click the button or select the MOVE UP
command from the ITEMS menu.
To move an item down relative to its siblings, click the button or select the MOVE DOWN
command from the ITEMS menu.

Adding or Editing Item Comments

The Comment panel below the Item Editor allows one to insert a new comment or edit an existing one
related to a selected topic.
To type a new comment or to edit one, simply click in the yellow region of this panel and start to type. The
comment is automatically saved as soon as you move to another item or leave the Comment panel. While
there is no menu item or toolbar icon associated with this feature, standard clipboard operations are
supported for copy and pasting text. A popup menu with standard editing features can also be obtained by
clicking the right mouse button.
To search for text in comments, see the information on the Global Search command.

Editing Documents
The Report Manager offers many editing features to create and edit both simple text documents and
documents with complex formatting, as well as tables and graphics. When a document item is selected in
the Table of Contents, a DOCUMENT menu appears, displaying all available formatting and editing
options. A similar menu can also be obtained by right-clicking anywhere in this document. The toolbar
portion directly over the editing area also displays buttons to access the most often-used editing and
formatting functions.

Individual documents may also be printed or exported to disk in various file formats such as plain text, Rich
Text or HTML format. An IMPORT command is also available to read a document file stored in plain text,
Rich Text, MS Word, WordPerfect, HTML and a few additional formats. Executing such a command will
replace the existing content with the content of the imported file.

194 WordStat Users Manual

Editing Tables
The Report Manager offers many editing features to customize the appearance of a table, to change the text
alignment or font setting, to set the cell background color, or to delete entire rows or columns. When a table
item is selected in the Table of Contents, a TABLE menu become visible displaying all available formatting
and editing options. A similar menu can also be obtained by right-clicking anywhere in this table. The
toolbar portion above the editing area also displays buttons to access the most often-used table editing and
formatting functions.
Individual tables may also be printed or exported to disk in various file formats such as ASCII (*.TXT), tab
delimited file (*.TAB), comma delimited (*.CSV), MS Word (*.DOC), HTML (*.HTM; *.HTML), XML
(*.XML) and Excel spreadsheet file (*.XLS).

Editing Charts
Many charts saved in the Report Manager may be edited using many of the same options as those available
in QDA Miner, such as the multidimensional scaling plot obtained through the CODE CO-OCCURRENCE
analysis command, the correspondence analysis plots, and the bar charts and line charts created by the
CODING BY VARIABLES command, as well as the bar charts and pie charts produced by the CODING
FREQUENCY command. To obtain information on the display options available for those charts, see their
corresponding page in this manual. Other charts - such as dendrograms and heatmaps - are stored as image
files, so cannot be modified. However, just like other charts, they may be exported to disk in various file
formats such BMP, JPG or PNG graphic files.

Searching and Replacing Text

Two broad types of text search are available in the Report Manager. A local, item-based search-and-replace
feature allows one to perform text searches and replacements on individual documents or tables, and a
global search engine for searching text patterns in several or all documents, tables and comments in the
Report Manager.

To perform a search in a single document or a table:

Select the document or table you would like to search, by selecting its entry in the Table of Contents.
Position the editing cursor in the document or select the cell in the table where you want the search
to begin.
Select the FIND command from the SEARCH menu, or click the button.
Enter a search expression, set the desired search options and then click Find Next.
To find additional instances of the same text, continue to click Find Next.

To replace text in a single document or table:

Select the document or table in which you would like to perform the text replacement by selecting its
entry in the Table of Contents.

WordStat Users Manual 195

 Position the editing cursor in the document or select the cell in the table where you want the search
to begin.
Select the REPLACE command from the SEARCH menu, or click the button.
In Find What, type the characters or words you want to find. In Replace With, type the text you
want to replace it with. Set the desired search options and then click Find Next. Click Replace to
change the selected text. To replace all instances of the text, click Replace All.

To perform a global search:

Select the GLOBAL FIND method from the SEARCH menu. A dialog box similar to this one will
appear:

The Text to Find edit box allows you to specify the text you want to find. The Case Sensitive and Whole
Word Only options function in the same way as in a standard word processor.
The search starts at the current topic item. Select Forward to continue searching items below the current
one, or Backward to move up and search items above the current document or table in reverse order. To
search all items, select the top item in the Table of Contents before using the Global Search dialog box.
The Scope option box is used to specify what is to be searched. You can restrict the search to Documents,
Tables, or Comments attached to items, or any combination of these three.
Once the search options have been set, click the Find button to start the search as well as to continue
searching for additional instances of this text.

Exporting items to HTML or Word

Individual documents, tables, graphics and images may be exported to disk in numerous formats. Such
exportation can be achieved by clicking the toolbar button or by selecting the EXPORT command
from the associated menu.
The Report Manager also offers the possibility of exporting the entire content or selected items into a single
HTML or MS Word document. The exportation is achieved by selecting the proper command from the
FILE | EXPORT menu. For example, to export items to HTML, select the HTML command. A dialog box

196 WordStat Users Manual

similar to this one will appear.

By default, all items are marked for export. To prevent some items from being included in the exported file,
simply remove the check marks beside them. Clicking a "parent" item affects all "children" items in the
same way. To unselect all items, uncheck the project item located at the very top of the tree.
Once the selection process is completed, click the OK button. A Save File dialog box will be displayed,
allowing you to enter a file name and select the location where the file should be saved. After the file is
created, you will be asked if you want to view this file. Clicking Yes will open a web browser if the
exported file is an HTML document or either MS Word or Wordpad if the exported file is a Word
document.

WordStat Users Manual 197

 References
Introduction to Content Analysis
ALEXA, M. (1997). Computer-assisted text analysis methodology in the social sciences. ZUMA:
Mannheim, Germany.
EVANS, W. (1996). Computer-supported content analysis: Trends, tools, and techniques. Social
Science Computer Review, 14 (3), 269-279.
KRIPPENDORFF, K. (1980). Content analysis: An Introduction to its methodology. Sage Publications:
Beverly Hills, California.
LEBART, L., SALEM, A. & BERRY, L. (1998). Exploring Textual Data. Kluwer Academic
Publishers: Dordrecht, Netherlands.
WEBER, R. P. (1990). Basic Content Analysis. Second Edition. Quantitative Applications in the Social
Sciences, vol 49. Sage Publications: Beverly Hills, California.
WEBER, R. P. (1983). Measurement models for content analysis. Quality and Quantity, 17 (2), 127-
149.
WEBER, R. P. (1984): Computer-aided content analysis: A short primer. Qualitative Sociology, 7 (1-
2), 126-147.
Interrater agreement statistics
BENNETT. E.M., ALPERT, R., & GOLDSTEIN, A.C. (1954). Communications through limited response
questioning. Public Opinion Quarterly, 19, 303-308.
BRENNAN, R.L., & PREDIGER, D. (1981). Coefficient kappa: Some uses, misuses, and alternatives. Educational
and Psychological Measurement, 41, 687-699.
COHEN, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological measurement,
20, 37-46.
JASON, S., & VEGELIUS, J. (1979). On generalizations of the G index and the phi coefficient to nominal scales.
Multivariate Behavioral Research, 14, 255-269.
KRIPPENDORFF, K. (1970). Bivariate agreement coefficients for reliability of data. In E.F. Borgatta and G.W.
Bohrnstedt (Eds.). Sociological methodology: 1970. San Francisco: Jossey-Bass.
SCOTT. W.A. (1955). Reliability of content analysis: The case of nominal scale coding. Public Opinion Quarterly,
19, 321-325.
ZWICK. R. (1988). Another look at interrater agreement. Psychological Bulletin, 103 (3), 374-378.

Others
GREENACRE, M. (1984). Theory and Applications of Correspondence Analysis. Academic Press. Orlando,
Florida.
GREFENSTETTE, G. (1994). Corpus-Derived First, Second and Third-Order Word Affinities. In W. Martin,W.
Meijs, M. Moerland, E. ten Pas, P. van Sterkenburg, and P. Vossen, editors, Proceedings of EURALEX'94,
Amsterdam, The Netherlands.
SEBASTIANI, F. (1999). Machine Learning in Automated Text Categorization. ACM Computing Surveys,
34(1):147.

198 WordStat Users Manual

 Technical Support

If you have any comment or suggestion for further improvement please contact Provalis
Research:

By Phone: 514-899-1672

By FAX: 514-899-1750

By Email: [email protected]

Web site: http://www.provalisresearch.com

Mail: Provalis Research

1255 University Street, Suite #1202
Montreal, QC
H3B 3W9

WordStat Users Manual 199