Table of Contents

PRELIS Examples .............................................................................................................. 3

1 Running PRELIS Syntax Files ............................................................................. 4
2 Exploratory Analysis of Political Survey Data .................................................... 7
2.1 Importing Data into PRELIS ................................................................................ 7
2.2 Data Screening ..................................................................................................... 9
2.3 Assigning Labels to Categories .......................................................................... 10
2.4 Define Missing Values ....................................................................................... 12
2.5 Imputation of Missing Values ............................................................................ 14
2.6 Homogeneity Tests ............................................................................................. 16
2.7 Graphical Display of Univariate Distributions................................................... 17
2.8 Import Data and Compute Polychoric Coefficients ........................................... 18
3 Exploratory Analysis of Fitness Data ................................................................ 24
3.1 Description of the Fitness/Cholesterol Data ...................................................... 25
3.2 Exploratory Data Analysis ................................................................................. 26
3.3 Graphical Displays ............................................................................................. 30
3.4 Regression Analysis ........................................................................................... 34
4 Data Manipulation and Bivariate Plots .............................................................. 36
4.1 Assigning Labels to Categories of Ordinal Variables ........................................ 37
4.2 Compute Values for a New Variable ................................................................. 38
4.3 Creating a Subset of the Data ............................................................................. 41
4.4 Bivariate Plots .................................................................................................... 43
4.5 Calculate and Save Covariances and Means for Subgroups .............................. 46
5 Normal Scores .................................................................................................... 49
5.1 Normalizing nine Psychological Variables ........................................................ 49
6 Two-Stage Least Squares ................................................................................... 52
6.1 Klein’s Model I of US Economy........................................................................ 52
7 Exploratory Factor Analysis .............................................................................. 55
7.1 Exploratory Factor Analysis of Nine Psychological Variables .......................... 55
8 Bootstrap Samples from Political Survey Data .................................................. 59
9 Monte Carlo Studies ........................................................................................... 63
9.1 New LISREL (*.lsf) Data................................................................................... 63
9.2 Inserting Variables into the Spreadsheet ............................................................ 64

PRELIS Examples Guide 1

 9.3 Inserting Cases into the Spreadsheet .................................................................. 65
9.4 Simulating Data Values ...................................................................................... 66
9.5 Graphical Displays ............................................................................................. 68
10 Multiple Imputation ........................................................................................... 70
10.1 Use of the MCMC Option to obtain more than one Imputed Data Set .......... 77

PRELIS Examples Guide 2

PRELIS Examples
PRELIS examples are contained in the prlex subfolder and are listed in the tables below.

LIST OF EXAMPLES IN THE \PRLEX FOLDER

Example 1A: Compute matrix of Example 1B: Logarithmic

polyserial and polychoric correlations transformations and recoding of
Data File: DATA.100 variables
Input File: EX1A.PRL Data File: DATA.100
Input File: EX1B.PRL
Example 1C: Select cases, estimate Example 2: Polychoric correlation
augmented moment matrix (MA=AM) matrix with all variables ordinal
Data File: DATA.100 (MA=PM)
Input File: EX1C.PRL Data File: DATA.EX2
Input File: EX2.PRL
Example 3: Censored variables Example 4: Estimated regressions
Data Files: DATA.EX3, LABELS.EX3 Data File: DATA.EX4
Input Files: EX3A.PRL, EX3B.PRL Input File: EX4.PRL
Example 5: Tetrachoric correlations, Example 6: Estimating asymptotic
with asymptotic variances estimated variances and covariances
from grouped data Data File: DATA.EX6
Data File: LSAT6.DAT Input Files: EX6A.PRL, EX6B.PRL,
Input Files: EX5A.PRL, EX5B.PRL EX6C.PRL
Example 7A: Data screening Example 7B: Assigning labels to
Data File: EX7.RAW categories
Input File: EX7A.PRL Data File: EX7.RAW
Input File: EX7B.PRL
Example 7C: Listwise deletion of Example 7D: Imputation
missing values Data File: EX7.RAW
Data File: EX7.RAW Input File: EX7D.PRL
Input File: EX7C.PRL
Example 7E: Saving the raw data for a Example 7F: Computing scale scores
subset of variables and cases or factor scores
Data File: EX7.RAW Data File: EFFICACY.RAW
Input File: EX7E.PRL Input File: EX7F.PRL
Example 8: Merging files Example 9: Estimating the asymptotic
Data Files: PANUSAl.RAW, covariance matrix by bootstrap
PANUSA2.RAW Data Files: EFFICACY.RAW,
Input File: EX8.PRL EFFICACY.LAB
Input File: EX9.PRL

PRELIS Examples Guide 3

LIST OF EXAMPLES IN THE \PRLEX FOLDER (continued)

Example 1OA: Testing homogeneity in Example 1OB: Panel data for political
panel data efficacy
Data File: PANUSA.RAW Data File: PANUSA.RAW
Input File: EX10A.PRL Input File: EX10B.PRL
Example 11: Fixed-x and probit Example 12: Bootstrapping
regression Data Files: EFFICACY.RAW,
Data File: EX11.RAW EFFICACY.LAB
Input File: EX11.PRL Input Files: BS.ST1, BS.ST2, BS.ST3
Example 13a: Generating normal Example 13b: Generating non-normal
variables with a specified covariance variables with a specified covariance
matrix matrix
Data File: None Data File: None
Input Files: SIMEX11.PRL, SIMEX12.PRL, Input File: SIMEX21.PRL
SIMEX13.PRL,
SIMEX14.LS8,SIMEX15.PRL
Example 13c: Generating variables from Example 13d: Generating ordinal
a specified model variables
Data File: None Data File: None
Input Files: SIMEX31.PRL Input Files: SIMEX4l.PRL, SIMEX42.LS8,
SIMEX43.PRL, SIMEX44.LS8,
SIMEX45.PRL, SIMEX46.PRL

1 Running PRELIS Syntax Files

From the main menu bar, select the Open option from the File menu to activate the Open
dialog box. Select the appropriate drive, folder, and file name. Select the file ex2.prl as
shown below. When done, click Open.

PRELIS Examples Guide 4

The contents of the file ex2.prl are displayed in a text editor window. This file may be
edited; portions of it may be copied to the clipboard, etc. with the same procedures used
in a text editor program.

To run PRELIS, click on the Run PRELIS icon button shown below.

PRELIS Examples Guide 5

A selection of the output is displayed below:

The file ex2.out may be converted to *.htm format. This is accomplished by selecting the
Convert Output to HTM option from the File menu.

A selection of the output contained in ex2.htm is displayed below. Compare this with the
corresponding selection in ex2.out.

PRELIS Examples Guide 6

2 Exploratory Analysis of Political Survey Data

The SPSS file dataex7.sav contains six variables on a subset of cases from the first cross-
section of a political survey. Use is made of a number of dialog boxes to
o import dataex7.sav and save it as a LISREL (*.lsf) spreadsheet file
o do data screening
o assign labels (names) for categories of the ordinal variables
o define missing values
o impute missing values and calculate polychoric coefficients and the asymptotic
covariance matrix
o perform a homogeneity test for pairs of ordinal variables
o graphical display of the univariate distributions of the variables.

2.1 Importing Data into PRELIS

From the File menu, select Import Data…

PRELIS Examples Guide 7

The user is prompted to select the drive and path where the data file to be imported is
stored.

We would like to import an SPSS for Windows (*.sav) file named dataex7.sav. From the
Files of Type drop-down menu, select SPSS for Windows (*.sav). Once dataex7.sav is
selected from the spssex folder, click Open to proceed.

Hint: Click (left mouse button) on the List Files of Type drop-down list box and type in
the letter “S” to go directly to all files types starting with the letter “S”.

Select a file name and folder in which the LISREL system data file should be saved. Note
that the extension has to be *.lsf. When done, click Save to display the system file as a
spreadsheet.

PRELIS Examples Guide 8

The file dataex7.lsf contains data for six variables, these being NOSAY, VOTING,
COMPLEX, NOCARE, TOUCH and INTEREST.

2.2 Data Screening

The simplest form of data screening is performed by selecting Data Screening from the
Statistics menu:

PRELIS Examples Guide 9

For each variable, dataex7.out lists the number and relative frequency of each data value,
and it gives a bar chart showing the distribution of the data values. For the variable
INTEREST, it is as follows:

2.3 Assigning Labels to Categories

Each of the 6 political efficacy items has four distinct numerical values. To assign
category labels AS, A, D and DS to the numerical values 1, 2, 3, and 4 (AS = Agree
strongly, A = Agree, D = Disagree, DS = Disagree Strongly) we proceed as follows.

Select the Define Variables option from the Data menu to activate the Define Variables
dialog box.

PRELIS Examples Guide

10
Select all six variables by holding down the CTRL key on the computer keyboard while
clicking on each variable (left mouse button). Alternatively, one can drag the cursor over
the six variables with the left mouse button held down. The selected variables will be
highlighted as shown below:

Click Category Labels to obtain the Labels dialog box:

PRELIS Examples Guide

11
Use this dialog box to enter a numerical value and its corresponding label, then click Add
to display the result in the string field at the bottom. Finally, when the last value and label
(4 and SD) have been entered, click Add to obtain the following dialog box.

Any of these choices can be deleted by highlighting it (left mouse button) and then
clicking on the Delete button. When done, click OK to return to the Define Variables
dialog box.

2.4 Define Missing Values

Suppose that the data set contains values 8 and 9, which are actually codes to indicate
missing responses. From the Define Variables dialog box, we select the Missing Values
option to activate the Missing Values dialog box shown below:

PRELIS Examples Guide

12
Click the Missing values radio button. Enter the values 8 and 9 in the first two string
fields as shown below. Check the Apply to all check box and the Listwise deletion radio
button. When this is done, click OK to return to the Define Variables dialog box. Click OK
to exit this dialog box and return to the PSF window.

Note:

It is important to select Save from the File menu for the changes to take effect.

PRELIS Examples Guide

13
2.5 Imputation of Missing Values

Missing values can be handled by choosing pairwise or listwise deletion. PRELIS offers
another possible way to handle missing values, namely by imputation, i.e. by substitution
of real values for missing values. The value to be substituted for the missing value for a
case is obtained from another case that has a similar response pattern over a set of
matching variables. This method is recommended for ordinal variables. Multiple
imputations are based on maximum likelihood procedures and is recommended for
continuous variables.

Select the Impute Missing Values… option from the Statistics menu. By clicking on
Impute Missing Values…, the dialog box shown below appears.

Suppose that we would like to impute the missing values on each variable using all the
other variables as matching variables. Cases with missing values will be eliminated after
imputation.

PRELIS Examples Guide

14
Hold the CTRL key down while selecting each of the 6 variables (left mouse button).
Click the top Add>> button to add the selected list of Imputed variables and then click the
bottom Add>> button to add the selected list of Matching variables.

To obtain moment matrices, click Output Options in the dialog box shown above. The
Output dialog box appears.

From the Type of Moment matrix to be analyzed drop-down list box, select Correlations.
Select Save to File and type in the name of the file.

To save the newly created data set check the Save the transformed raw data to file check
box and enter a file name. Note that if the file extension is *.lsf, the file will be saved as a
LISREL system data file. Also check the Save to file check box under Asymptotic
Covariance matrix and enter dataex7.acm in the string field. Click OK to return to the
Impute Missing Values dialog box. If satisfied with the selections made, click Run to
invoke PRELIS. On execution, an output file named dataex7.out is produced.
Alternatively, click Syntax to create a PRELIS syntax file.

PRELIS Examples Guide

15
2.6 Homogeneity Tests

The homogeneity test is a test of the hypothesis that the marginal distributions of two
categorical variables with the same number of categories (k) are the same. The  2
statistic for testing this hypothesis has k -1 degrees of freedom.

The homogeneity test is particularly useful in longitudinal studies to test the hypothesis
that the distribution of a variable has not changed from one occasion to the next.

To test for homogeneity, select Homogeneity Test from the Statistics menu to open the
Homogeneity Test dialog box.

Hold the CTRL keyboard button down while selecting two variables from the list of
Ordinal variables as shown in the dialog box below.

PRELIS Examples Guide

16
Once two variables are highlighted, the Add button is activated, and the selection can be
entered by clicking on this button. The dialog box above shows that the variables NOSAY
and INTEREST are highlighted. By clicking Add, this test will be added to the 4 tests
shown in the bottom window of the dialog box.

When all the tests are entered, click Output Options for additional output or, if not
required, the Run PRELIS icon button to start the PRELIS analysis. A summary of the test
results is given below:

2.7 Graphical Display of Univariate Distributions

A bar chart of the univariate distribution of each variable may be obtained by clicking on
a variable name to highlight the column of data values as shown below.

PRELIS Examples Guide

17
Once this is done, click the Univariate Plot icon button (third icon from the right on the
toolbar) to create the bar chart. Alternatively, one may select the Univariate option from
the Graphs menu.

The bar chart given below shows that the majority of respondents selected category 2
(Agree), whilst only a small number selected the “Disagree Strongly” category.

2.8 Import Data and Compute Polychoric Coefficients

LISREL provides the user with numerous features when importing data as an .lsf file. The
user can, for example, delete variables, select cases and compute various sample statistics
interactively by making use of a set of menus.

PRELIS Examples Guide

18
As an illustration, an SPSS data set, data100.sav, which contains two continuous and four
ordinal variables, will be imported. The interactive graphical user interface environment
will be then used to

o delete the continuous variables

o select odd numbered cases (that is cases 1, 3, ...)
o calculate a matrix of polychoric correlation coefficients
o save this and the new data set to external files.

From the File menu, select Import Data:

The user is prompted to select the drive and path where the data file to be imported is
stored. From the List Files of Type drop-down list box select SPSS for Windows (*.sav)
and then select the SPSS for Windows file named data100.sav from the spssex folder.
Click Open to go to the Save As dialog box.

PRELIS Examples Guide

19
Select a file name and folder in which the PRELIS data should be saved. For the present
example, the folder chosen is spssex and the file name selected is data100.lsf. Note that
the extension has to be *.lsf. When done, click Save.

The LISREL system data file (spreadsheet) will be displayed in the LSF window. Note that
the file data100.lsf contains data values for six variables, these being CONTIN1,
ORDINAL1, ORDINAL2, ORDINAL3, CONTIN2 and ORDINAL4.

Suppose that we are only interested in the ordinal variables, then CONTIN1 and CONTIN2
may be deleted from the data set. To delete the variable CONTIN1, click on the rectangle
containing the name CONTIN1. The column containing the data values will change in
color as shown below. This variable may now be deleted by right-clicking on the column
header CONTIN1 to select the Delete Variables option from the popup menu or,
alternatively, by selecting the Delete Variable option from the Data menu.

PRELIS Examples Guide

20
Click the OK button of the Delete Variables text box.

The variable CONTIN2 is subsequently deleted by following the procedure described

above. Note that one should use the File, Save option to ensure that these changes take
effect.

The resulting data set consists of 4 variables and 100 cases. Note that a ####### symbol
in a data field indicates a missing value, or a value that contains too many digits to be
displayed correctly. One can view such values by changing the Width and Number of
decimals defaults by using the Edit menu to open the Data Format dialog box.

The next step in the data analysis procedure is to select the odd-numbered cases. Select
the Select Variables/Cases option from the Data menu to obtain the dialog box shown
below.
PRELIS Examples Guide
21
Click the Select only those cases radio button and then click the odd radio button. This
instruction is transferred to the text box by clicking the Add button. In doing so, the
words Select those case numbers that are odd appears in the text box. Any command in this
box can be removed by highlighting the specific command before clicking Remove. Once
done, click Output Options to obtain the Output dialog box.

Enter 10 in the Width of fields string field and 1 in the Number of decimals string field.
Using the Moment matrix drop-down list box, select Correlations. Check the Save to File
check box and type in the name of the file (data100.cor). Save the data set to be created
by checking the Save the transformed data to file check box and entering a file name. For
the present example, the name data100.raw is used.
Notes:

o Tests of multivariate normality are only performed on continuous variables.

o PRELIS computes a polychoric correlation matrix if the variables are ordinal.
o When saving data under a filename with the extension *.lsf, a LISREL system
data file with that name is created.

PRELIS Examples Guide

22
Click OK when done. This action will return control to the Select Data dialog box. If
satisfied with the selections made, click OK to invoke the PRELIS program. On execution,
an output file named data100.out is produced. A selection of the output is given.

By closing the output file to return to the LSF window, bar charts can be produced for
each of the ordinal variables. Do so by clicking on the rectangle containing the name of
the variable and then on the Univariate Plot icon button.
PRELIS Examples Guide
23
A bar chart similar to the one shown next can be obtained by clicking on a vertical bar in
the graph to invoke the Bar Graph Parameters dialog box. Change the bar type from
Vertical Bars to Vertical 3D Bars to obtain the chart below.

3 Exploratory Analysis of Fitness Data

Once a *.lsf file is opened, the LSF toolbar appears and a number of additional dialog
boxes become accessible to the user. PRELIS syntax can be generated by making use of
these dialog boxes. As an illustration, we use a data set called fitchol.dat (Du Toit, Steyn
and Stumpf, 1986.)

PRELIS Examples Guide

24
3.1 Description of the Fitness/Cholesterol Data

Two of the many factors that are known to have some influence or relevance on the
condition of the human heart are physical fitness and blood cholesterol level. In a related
research project, four different homogeneous groups of adult males were considered. A
number of plasma lipid parameters were measured on each of the 66 individuals and
fitness parameters were also measured on three of the four groups.
The groups are:
Group Description

1 Weightlifter (n1 = 17)

2 Student (control; n2 = 20)
3 Marathon athlete (n3 = 20)
4 Coronary patient (n4 = 9)

The characteristics that we will consider here are:

Variable Label
X1 = Age (years) Age
X2 = Height (cm) Len
X3 = Weight (kg) Weight
X4 = Percentage fat %Fat
X5 = Strength-breast (lb) Strength
X6 = Triglycerides Trigl
X7 = Cholesterol (total) Cholest

The folder name containing the dataset is Tutorial and the file name is fitchol.dat. Locate
this file and open with Notepad. The first 5 lines of the data set are given below:

Use Notepad (or any other text editor) and insert the variable names on top of the data:

PRELIS Examples Guide

25
3.2 Exploratory Data Analysis
From the File menu, select The Import Data… option

The Open dialog box that appears will prompt the user to select the drive and path where
the raw data is stored. For the present example, the folder name is Tutorial and the file
name is fitchol.dat. Locate this file and open with Notepad:

Select Free Format Data (*.dat, *.raw) and select the file filtchol.dat from the Tutorial
folder. Click Open to proceed.

PRELIS Examples Guide

26
Select a filename and folder in which the LISREL system data should be saved. For the
present example, the folder chosen is Tutorial and the filename selected is fitchol.lsf. Note
that the extension has to be *.lsf.

Enter the number of variables in the dataset (see below) and select Variable names at the
top of file.

When done, click OK to open the LSF window in which the newly created *.lsf file is
displayed.

Since the fitness data set (fitchol.dat) contains missing values that are denoted by -9.0, we
would like to remove these values prior to doing any exploratory analyses. To do so,
select the Define Variables option from the Data menu to obtain the Define Variables
dialog box.

PRELIS Examples Guide

27
Select the variable in the first row. This action will activate the Variable Type, Category
Labels, Missing Values and Labels buttons. Click Missing Values to obtain the Missing
Values for dialog box.

On this dialog box, check the Global Missing values radio button and then type in the
value of -9.0. Click OK to return to the Define Variables dialog box.

Suppose that we would like to specify all variables (except Group) as continuous. On the
Define Variables dialog box, select the variables Age to Cholest and then click Variable

PRELIS Examples Guide

28
Type to activate the Variable Type for dialog box. Check the Continuous radio button and
then click OK to return to the Define Variables dialog box.

The dialog box may also be used to change the variable labels by clicking on a variable
and then Rename.

Before fitting an appropriate model to data, it may be wise to run a data screening
procedure. The PRELIS data screening procedure provides information on the distribution
of missing values, univariate summary statistics and test of univariate normality for
continuous variables. The procedure also provides information on the distribution of
variables over a number of class intervals.

To do data screening, select Statistics from the main menu bar and then the Data
Screening option from the Statistics menu.

A selection of the output file obtained is shown below:

PRELIS Examples Guide

29
3.3 Graphical Displays

To produce a histogram of the data in the Height column, click on the rectangle
containing the name Length. The column containing the data values will change in color
as shown below.

On the main LISREL menu bar, select the Univariate plot icon button. Note that, if you
move the mouse to this icon, the words Univariate Plot will be displayed in the proximity
of the icon button. As an alternative to producing a histogram by clicking on the
Univariate plot icon button, one may select the Graphs, Univariate option on the main

PRELIS Examples Guide

30
menu bar as shown below. Before doing so, make sure that a variable (column) in the
spreadsheet is not selected and that the variable is appropriately defined as ordinal or
continuous.

To obtain the Univariate Plots dialog shown below, make sure that if a column is selected
to click on the rectangle containing the corresponding variable label to deselect it.

A histogram of the variable Height is displayed below. The colors of the background and
the bars can be changed by double-clicking on the appropriate area. This action will
produce a dialog box in which the necessary change may be made. Likewise, one can
change the color and appearance of the axes and labels by double clicking on a specific
label, for example on Histogram. The dialog box allows one to select a font and font size
and to change the color of the text. The various dialog box controlling aspects of the
graphs are discussed in the Graphical Users Interface Guide.

PRELIS Examples Guide

31
If a variable has less than 16 distinct categories, the Draw histogram option produces a
bar chart representation of the data. This is the case for the variable Group, which has 4
distinct values, corresponding to

1 = Weightlifter
2 = Student
3 = Marathon athlete
4 = Coronary patients

The bar chart below shows that the number of individuals in the first three groups are
almost equal (n = 17, 20 and 20), but that the sample contains a relatively small number
of coronary patients (n = 9).

The default legend shows the categories of the variable Group and the values assigned to
each category. By double clicking on the Cat1:1 description, the Text Parameters dialog
box is activated.

PRELIS Examples Guide

32
Change Cat1:1 to 1 = Weightlifter and click OK. Repeat this procedure to change the
descriptions of categories 2, 3, and 4 to Student, Marathon and Coronary, respectively.

The appearance of the vertical bars can be changed by double clicking on one of the bars
until the Bar Graph Parameters dialog box is activated. Change Type to Vert. 3-D bars and
use the bar color sliders to change to the desired color. Click OK when done.

PRELIS Examples Guide

33
The bar chart shown below reflects these changes in attributes. The bar chart for the
variable Group may now be printed. This is done by selecting the Print option from the
File menu. Alternatively, the bar chart can be exported as a Windows metafile (*.wmf) by
selecting the Export as Metafile option from the File menu.

3.4 Regression Analysis

We may be interested in the prediction of Cholesterol by the variables Age, Height,
Percentage Fat and Strength. To obtain the regression equation, select the Regressions
option from the Statistics menu.

PRELIS Examples Guide

34
The Regression dialog box is shown below. This dialog box allows the user to select
variables as either Y or X-variables. Click on a variable, for example Cholest, to activate
the Add button. Click the top Add button to add this variable to the list of Y-variables.
Select the variables Age, Height, Weight, %Fat, and Strength and click the bottom Add
button to add the variable to the list of X-variables. When the selection procedure is
completed, click on Output Options to activate the Output dialog box.

PRELIS Examples Guide

35
A selection of the output is given below:

4 Data Manipulation and Bivariate Plots

In this example, use is made of the Fitness/Cholesterol data described in detail in the
previous section. The data set contains 8 variables, these being Group, Age, Height, Weight,
%Fat, Strength, Trigl and Cholest.

The categorical variable Group has 4 values, these being 1, 2, 3, and 4, denoting weight
lifters ( n1 = 17), students ( n2 = 20), marathon athletes ( n3 = 20) and coronary patients (
n4 = 9) respectively.
PRELIS Examples Guide
36
The users interface will be used to

o Assign labels to the categories of Group

o Insert a new variable called Totchol into the data set
o Assign the value Trigl + Cholest to the new variable Totchol
o Create a data subset consisting of the 20 marathon athletes
o Display bivariate plots for the marathon athlete subgroup
o Calculate covariances and means for groups 1 to 3

4.1 Assigning Labels to Categories of Ordinal Variables

To display the fitness/cholesterol data, the File, Open option is used. Select Files of type
*.lsf from the drop-down list box. The file fitness.lsf can be found in the tutorial folder.

From the Data menu, select Define Variables and highlight the variable Group as
illustrated in the next image.

Once this is done, click on the Category Labels button.

PRELIS Examples Guide

37
In the value string field, type the value 1 and in the label string field type the value wlif.
Click the Add button to obtain the code 1 = wlif in the syntax window. Proceed in a similar
fashion until 4 = coro has been entered in the code string field. When done, click OK to
return to the Define Variables dialog box.

4.2 Compute Values for a New Variable

To insert and create values for a new variable Totchol, select the Compute… option from
the Transformation menu to activate the dialog box labeled Compute.

Next click the Add button and type Totchol in the text box and, when done, click the OK
button.

PRELIS Examples Guide

38
The dialog box below shows that the variable Totchol has been added to the list. We now
wish to enter the values Totchol = Trigl + Cholest for this variable.

PRELIS Examples Guide

39
The Compute dialog box displays the new variable as a function of Trigl and Cholest. To
enter this equation, click on Totchol (left mouse button). While holding the mouse button
down, drag this item to the position shown in the Compute dialog box and release the
mouse button. Click on the “=” sign, then drag Trigl to the position next to the “=” sign.
When this is done, click on the “+” sign and finally drag Cholest to the position shown.
Alternatively, one could also double-click on a variable name to enter the variable into
the Compute window. Click OK to update the *.lsf file. The result of this computation is
shown below.

PRELIS Examples Guide

40
It is also possible to save the data under a new name, e.g. fitness2.lsf by clicking the
Output Options button on the Compute dialog box and then checking the Save the
transformed data to file check box.

4.3 Creating a Subset of the Data

Suppose that we want to focus on the subgroup of marathon athletes (Group = 3). To
create a data set consisting of this subgroup, select the Data, Select Variables/Cases
option as shown below.

The Select Variables/Cases dialog box appears.

PRELIS Examples Guide

41
Highlight the variable Group and select the Select only those cases with value and equal
to options. Type in the value of 3 and then click the Add button to obtain the words select
those cases with value equal to 3: Group shown in the window below the Add and Remove
buttons. This syntax can be removed if it is highlighted and the Remove button is clicked.
Choose the Output Options button to save the file under a new name, for example
Fitnessmara.lsf.

PRELIS Examples Guide

42
Note:

If the file is not saved under a new name, the existing *.lsf will be overwritten.

When done, click OK to start PRELIS. The data subset consisting of Group=3 values only
will be created. This spreadsheet, Fitnessmara.lsf, is shown below.

4.4 Bivariate Plots

A bivariate plot can be obtained by clicking on the Bivariate icon button shown next to
the Univariate icon button on the second tool bar box below.
PRELIS Examples Guide
43
One can also obtain a bivariate plot of two variables by selecting the Graphs, Bivariate
option to obtain the Bivariate Plots dialog box.

Click on Totchol and then click on the Select (Y variable) button. Once this is done,
highlight %Fat as shown below, and click on the Select (X variable) button. Also select the
Scatter Plot and Line Plot options. To produce a bivariate graph, click Plot.

PRELIS Examples Guide

44
A plot of Totchol on %Fat is obtained, as shown below:

Note that the graph shows the product moment correlation coefficient and the sample
size. The line plot is obtained by taking the arithmetic average of all Y-values that
correspond to the same X-value. As such, the line plot shows the general trend in the Y-
variable with increase in X.

Histograms, bar charts and bivariate plots can be exported as Windows Enhanced
Metafiles (*.emf extension). These files can be imported into any of the Microsoft Office
products, and can also be opened with various graphics software packages.

To save a bivariate plot as a *.emf graphics file, select the Export as Metafile option from
the File menu.

PRELIS Examples Guide

45
By clicking on Export as Metafile, the user is prompted to save the *.emf file under any
name and in any folder.

4.5 Calculate and Save Covariances and Means for Subgroups

Suppose that we want to calculate covariances, means and standard deviations for each of
the first 3 groups of the fitness/cholesterol data.

The procedure is illustrated by selecting Group = 2 (2 = students). Select the Select

Variables/Cases option from the Data menu to obtain the Select Data dialog box.

Select the Select Cases tab on the Select Data dialog box:

PRELIS Examples Guide

46
Select Group (click on Group with the left mouse button) and then complete the Select
only those cases with value portion of the dialog box. Since the variable Group will now
assume values of 2 only, it should not be included in the calculation of the means and
covariances.

If this variable is to be deleted, check the Delete variables after selection option. Click the
Add button to generate the necessary syntax.

Once this is done, the Output Options button should be clicked to produce the Output
dialog box given below.

PRELIS Examples Guide

47
Complete the dialog box as shown. When done, click OK to return to the Select Cases
dialog box. Click OK to start PRELIS.

A portion of the PRELIS output is given below:

PRELIS Examples Guide

48
5 Normal Scores

The analysis of continuous non-normal variables in structural equation models is

problematic in several ways. If the maximum likelihood (ML) method is used, standard
errors and  2 may be incorrect. In theory, weighted least squares (WLS or ADF) with a
correct weight matrix should produce correct estimates of standard errors and chi-
squares, but this requires a very large sample. Sometimes a reasonable compromise is to
use ML despite the non-normality and correct for the bias in standard errors, but this too
requires a large sample.

Another solution to non-normality is to normalize the variables before analysis. Normal

scores offer an effective way of normalizing a continuous variable for which the origin
and unit of measurement have no intrinsic meaning, such as test scores.

5.1 Normalizing nine Psychological Variables

To normalize the raw scores of the nine psychological variables in file npv.lsf, we
proceed as follows.

Select the File, Open option and select LISREL system file (*.lsf) from the Files of type
drop-down list box. Select the file npv.lsf from the tutorial folder. Click Open to proceed.

PRELIS Examples Guide

49
Note that normal scores may be calculated for both ordinal and continuous variables.
From the Statistics menu, select the Normal Scores option to obtain the Normal Scores
dialog box.

Select all 9 variables and click Add. A variable may be removed by clicking on the
specific variable in the bottom dialog box and then using the Remove button. When all
the variables for which normal scores are to be computed are selected, click Output
Options to save the normal scores to an external file.

PRELIS Examples Guide

50
On the Output dialog box, check the Save the transformed data to file check box and enter
the file name npv_ns.raw. The data can also be saved as npv_ns.lsf, in which case a
PRELIS system file is created.

When done, click OK and then Run on the Normal Scores dialog box. The first ten rows
of normal scores for the nine variables are shown below.

PRELIS Examples Guide

51
6 Two-Stage Least Squares

Two-stage least-squares (TSLS) is particularly useful for estimating econometric models

of the form

y  By  Γx  u,

where y  ( y1, y2 ,..., y p ) is a set of endogenous or jointly dependent variables,

x  ( x1, x2 ,..., xq ) is a set of exogenous or predetermined variables uncorrelated with the
error terms u  (u1, u2 ,..., u p ) , and B and Γ are parameter matrices.

A typical feature of the above model is that not all y-variables and not all x-variables are
included in each equation.

A necessary condition for identification of each equation is that, for every y-variable on
the right side of the equation, there must be at least one x-variable excluded from that
equation. There is also a sufficient condition for identification, the so-called rank
condition, but this is often difficult to apply in practice. For further information on
identification of interdependent systems, see, e.g., Goldberger (1964, pp. 313-318).

6.1 Klein’s Model I of US Economy

Klein’s (1950) Model I is a classical econometric model that has been used extensively as
a benchmark problem for studying econometric methods. It is an eight-equation system
based on annual data for the United States in the period between the two World Wars. It
is dynamic in the sense that elements of time play important roles in the model.

The data set, klein.lsf (tutorial folder), consists of the following 15 variables:

Ct Aggregate Consumption
Pt_1 Total Profits, previous year
Wt* Private Wage Bill
It Net Investment
PRELIS Examples Guide
52
 Kt_1 Capital Stock, previous year
Et_1 Total Production of Private Industry, previous year
Wt** Government Wage Bill
Tt Taxes
At Time in Years from 1931
Pt Total Profits
Kt End-of-year Capital Stock
Et Total Production of Private Industry
Wt Total Wage Bill
Yt Total Income
Gt Government Non-Wage Expenditure

To estimate the consumption function, we use Ct as the y-variable, Pt, Pt_1 and Wt as x-
variables and Wt**, Tt, Gt, At, Pt_1, Kt_1 and Et_1 as the z-variables. An intercept term in
the equation can be estimated by introducing a variable denoted in this example by ONE,
which is a constant equal to 1 for each year. When an intercept term is introduced, the
moment matrix (MM) is used instead of the covariance matrix (CM).

To change the variable type from the default of ordinal to continuous, select the variable
Ct and click on Variable Type on the Define Variables dialog box to go to the Variable
Type for dialog box.

Check the Apply to all check box, then click OK to return to the Define Variables dialog
box. On this dialog box, also click OK. The spreadsheet now contains the renamed
variables. Use the File, Save option to save any changes made to the .lsf file.

Next double click on the variable Intcept, then click on the equal sign and finally on the
number 1. When done, click OK to generate the new spreadsheet with the variable Intcept
added as the last column. Finally, select Two-Stage Least-Squares from the Statistics
menu to obtain the Two-Stage Least-Squares dialog box.

Enter Ct as the Y-variable, Pt, Pt_1 and Wt as the X–variables and Wt**, Tt, Gt, At, Pt_1,
Kt_1 and Et_1 as the instrumental variables.
PRELIS Examples Guide
53
Click on Output Options and select Moments about zero from the Moment Matrix drop-
down list box. At this stage one could also save the data set under a different name. If the
file extension is *.lsf, a LISREL system data file is created.

When done, click OK to return to the Two-Stage Least-Squares dialog box and then select
Run to do the analysis or Syntax to view the newly created PRELIS syntax file (klein.prl).

A portion of the output, showing the estimated equations is given next.

PRELIS Examples Guide

54
7 Exploratory Factor Analysis

In an exploratory factor analysis, one wants to explore the empirical data to discover and
detect characteristic features and interesting relationships without imposing any definite
model on the data. An exploratory factor analysis may be structure generating, model
generating, or hypothesis generating. In confirmatory factor analysis, on the other hand,
one builds a model assumed to describe, explain, or account for the empirical data in
terms of relatively few parameters.

Exploratory factor analysis is a technique often used to detect and assess latent sources of
variation and covariation in observed measurements. It is widely recognized that
exploratory factor analysis can be quite useful in the early stages of experimentation or
test development.

7.1 Exploratory Factor Analysis of Nine Psychological

Variables

To illustrate exploratory factor analysis we use a classical data set. Holzinger and
Swineford (1939) collected data on twenty-six psychological tests administered to 145
PRELIS Examples Guide
55
seventh- and eighth-grade children in the Grant-White school in Chicago. Nine of these
tests are selected for this example. The nine selected variables and their intercorrelations
are given in the table below.

Table 1: Correlation Matrix for Nine Psychological Variables

VIS PERC 1.000

CUBES 0.326 1.000
LOZENGES 0.449 0.417 1.000
PAR COMP 0.342 0.228 0.328 1.000
SEN COMP 0.309 0.159 0.287 0.719 1.000
WORDMEAN 0.317 0.195 0.347 0.714 0.685 1.000
ADDITION 0.104 0.066 0.075 0.209 0.254 0.178 1.000
COUNTDOT 0.308 0.168 0.239 0.104 0.198 0.121 0.587 1.000
S-C CAPS 0.487 0.248 0.373 0.314 0.356 0.222 0.418 0.528 1.000

Use the File Open menu, select file of the type *.lsf and choose the file npv.lsf from the
tutorial folder. This file contains 145 observations on the nine psychological
measurements.

The nine psychological variables should be treated as continuous variables. The PRELIS
default variable type, however, is ordinal. To change the default variable type, choose the
Data, Define Variables… option then select all the variables in the variable list and click
the Variable Type button.

All the variables can be selected simultaneously by clicking on the first variable and then,
with the left mouse button held down, dragging the cursor over the remaining variables.

PRELIS Examples Guide

56
Change the variable type to continuous and click OK when done. If only one variable had
been selected, use of the Apply to all check box, would have changed all the variables in
the list from ordinal to continuous variables.

Although LISREL can determine the number of factors to be extracted if this is not
specified by the user, we shall assume that the number of factors equals three.

An exploratory factor analysis is done by selecting the Exploratory Factor Analysis option
from the Statistics menu to obtain the Exploratory Factor Analysis dialog box.

Not all the variables in a data set may be suitable for a factor analysis, since the data may
contain demographic variables such as marital status, gender, etc. A subset of the list of
variables may therefore be selected. For the present example, all nine variables are used
and therefore we do not have to use the Select button. Enter 3 for the Number of Factors
and click Run to produce output or Syntax to view (and possibly modify) the syntax file
before running PRELIS. Factor scores will be saved in the file npv.fsc.

PRELIS Examples Guide

57
A portion of the factor analysis output is shown below.

PRELIS Examples Guide

58
8 Bootstrap Samples from Political Survey Data

Suppose the original data consists of N cases and we want to draw K samples of size n.
The drawing is done with replacement. The number n may be smaller than, equal to, or
larger than N. For each of the K samples, summary statistics can be computed and saved
in a file.

It is assumed that the data does not contain missing values. If it does, bootstrap examples
will be drawn from the listwise sample remaining after deletion of all cases with missing
values.

To illustrate how this can be done using the GUI, suppose 100 bootstrap samples of size
148 each are to be drawn from the LISREL system data file dataex7.lsf. This data set is
obtained by importing the SPSS *.sav file, dataex7.sav from the spssex folder. See the
section on Exploratory data analysis of political data in this guide for more information on
how to import data.

To open the file dataex7.lsf, select the Open option on the File menu to obtain the File
Open dialog box. Select the spssex folder and from the Files of type drop-down list box
PRELIS Data (*lsf). When dataex7.lsf is selected, clicked OK to display the data set.

Select the Bootstrapping option from the Statistics menu to obtain the Bootstrapping
dialog box.

PRELIS Examples Guide

59
Since the sample fraction is 50, each bootstrap sample will be 50% of the original sample
size of 297, that is, of size 148. Select the Save: all the MA-matrices option and choose a
file name for a file that will contain 100 correlation matrices.

Click Output Options on the Bootstrapping dialog box and choose Correlations for the
matrices to be computed.

PRELIS Examples Guide

60
The Output dialog box also allows you to set the random number generator seed, so that
this run can be replicated exactly. For this example, a value of 56719 was chosen. Click
OK to return to the Bootstrapping dialog box and click Run to start the PRELIS analysis.

After this run, the file bootex7.cov contains 100 correlation matrices, each with 21 items
(the lower triangular part of a 6 x 6 corrrelation matrix). The correlation matrices for the
first two bootstrap samples are:

Note that the numbers are given in so-called scientific notation, where D+01 indicates
that the number is to be multiplied by 10, D-01 indicates multiplication by 0.1, etc.
PRELIS Examples Guide
61
A 3-D bar chart of VOTING and INTEREST can be obtained by selecting the Bivariate
option from the Graphs menu.

After completing the Bivariate Plots dialog box as shown above, click Plot. The resulting
plot is shown below:

PRELIS Examples Guide

62
9 Monte Carlo Studies

The graphical users interface may be used to simulate data from uniform or normal
distributions, or from mixtures of these distributions. To illustrate, suppose we want to
simulate 50 observations on three variables, V1, V2, and V3. Suppose further that V1 has a
normal distribution with a mean of 100 and standard deviation of 15; V2 = V1 + e2, where
e2 has a normal distribution with mean zero and standard deviation of 15; and finally V3
= V1 + V2 + e3 where e3 has a uniform distribution with a mean of 10.

This may be accomplished by specifying

V1 = 100 + 15 * NRAND
V2 = V1 + 15 * NRAND
V3 = V1 + V2 + 20 * URAND,

where NRAND has a normal (0;1) distribution and URAND a uniform (0;1) distribution.

The next sections illustrate

 The creation of a new LISREL system data file

 Insertion of variables into the LISREL system file
 Insertion of cases
 Simulation of data values
 Graphical displays

9.1 New LISREL (*.lsf) Data

From the File menu, select New to obtain the New dialog box. Select LISREL Data and
click OK when done.

PRELIS Examples Guide

63
A new (but empty) LISREL system data file is displayed in the LSF window.

In order to simulate 50 observations of the form

V1 = 100 + 15 * NRAND,
V2 = V1 + 15 * NRAND, and
V3 = V1 + V2 + 20 * URAND,

we need to create an empty data matrix with 3 columns and 50 rows, where each column
represents a variable and each row an observation (case).

9.2 Inserting Variables into the Spreadsheet

We start by inserting the three variables V1 to V3. To do so, select the Data, Insert
Variables… option.

Use the Insert Variables dialog box to insert three variables and click OK.

PRELIS Examples Guide

64
9.3 Inserting Cases into the Spreadsheet
The LISREL system file will now contain 3 columns headings corresponding to the
variables var1-var3. Next, select the Data, Insert Cases… option, as shown next.

This action invokes the Insert Cases dialog. Enter the number of cases, that is, 50. Click
OK to display the spreadsheet.

At this stage, this file should be saved as a *.lsf. This is accomplished by clicking the File,
Save as option. Select a filename and folder, for example data1.lsf. Click Save.

The spreadsheet shown below now contains 50 cases.

PRELIS Examples Guide

65
9.4 Simulating Data Values

From the Transformation menu, select the Compute option. This activates the Compute
dialog box.

The three lines of syntax are entered as follows:

Click on var1 and with the left mouse button held down, drag it to the position shown in
the transformation window (or double click on the variable name). When done, click the
“=“ sign, then enter 100 + 15 * n(0,1)by clicking the corresponding buttons on the compute
pad. To start the next syntax line, click on the Next line button. Proceed as described
above until the 3 lines have been added. Note that the PRELIS symbols for n(0,1) and
u(0,1) are NRAND and URAND respectively.

PRELIS Examples Guide

66
When done, click Output Options to obtain the Output dialog box. Select Covariances
from the Moment Matrix drop-down list box. Check the Save to file check box directly
below this and enter a filename, in this case simul.cov. Optionally, change the random
seed number (in this case to 7613) to replicate the simulated dataset shown below. Click
OK to return to the Compute dialog box and OK again when done.

PRELIS Examples Guide

67
9.5 Graphical Displays

Use the Data, Define variables option to change the variable type of the three variables to
continuous. Make sure to select the File, Save option to make these changes permanent.
To obtain a histogram of the distribution of var1-values, click the var1 button if this
column has been selected on the spreadsheet to de-select it then click the Univariate Plot
icon button to produce the Univariate Plots dialog shown below.

Change the number of class intervals to eight and click the Plot button to produce a
histogram.

PRELIS Examples Guide

68
From the display, we see that the data follows a bell-shaped curve reasonably well. The
sample mean and standard deviation values of 99.433 and 14.703 are close to the
population values of 100 and 15 respectively, given the small sample size.

Finally, a bivariate plot of the distribution of var1 and var3 may be obtained by selecting
Bivariate from the Graphs menu. In the Bivariate Plot dialog box, select var 1 as the Y-
variable and var 3 as the X-variable. Also select Scatter Plot and click Plot when done.

The resulting bivariate display is shown below and it reveals a high (linear) correlation
between var 1 and var 3.

PRELIS Examples Guide

69
10 Multiple Imputation

Ryan and Joiner (1994, Table 9.1) report serum-cholesterol levels for n = 28 patients
treated for heart attacks. Cholesterol levels were measured 2 days and 4 days after the
attack. For 19 of the 28 patients, an additional measurement was taken 14 days after the
attack.

Ten rows of the data set, chollev.dat (tutorial folder), are shown below

PRELIS Examples Guide

70
Select the Import Data… option from the File menu. Select the tutorial folder in the Open
dialog box, select chollev.dat and as file type Free Format Data (*.dat,*.raw) then click
Open.

A Save As dialog is produced by clicking the Open button. Save the file as chollev.LSF.

PRELIS Examples Guide

71
Enter the number of variables in the Number string field of the Enter Number of Variables
dialog box.

Click OK to create a LISREL system data file. From the main menu bar, select the Define
Variables option from the Data menu to obtain the Define Variables dialog box.

Use the dialog box to rename VAR1, VAR2 and VAR3 to Chol_D2, Chol_D4 and Chol_D14
respectively. Click Variable Type to define these variables as continuous.

PRELIS Examples Guide

72
When done, click Missing Values to assign the value of –9.000 as the global missing value
and select pairwise deletion. Make sure that all the changes are saved by using the File,
Save option.

The first five rows of the .lsf file are shown below.

PRELIS Examples Guide

73
Select the Multiple Imputation option on the Statistics menu to obtain the Multiple
Imputation dialog box. Select all the variables from the Variables List.

To impute data using the EM Algorithm, click the EM Algorithm radio button. To impute
data using the MCMC algorithm, click the MCMC Algorithm radio button. Select the
number of iterations and the convergence criterion.

PRELIS Examples Guide

74
To save the imputed data, means, and covariances to external files, click Output Options
to obtain the Output dialog box. Enter the values shown Above. Click OK when done to
return to the Multiple Imputation dialog box.

PRELIS syntax, generated by clicking Syntax in the Multiple Imputation dialog box, is
shown below.

A portion of the PRELIS output is shown below.

PRELIS Examples Guide

75
The percentage of missing cases is 10.71 and the EM procedure converged in 4 iterations.

The first 10 records of the imputed data set, written to a file named imputed_data.lsf (the
default name), are shown below.

PRELIS Examples Guide

76
10.1 Use of the MCMC Option to obtain more than one Imputed
Data Set
In contrast to the EM algorithm, the MCMC algorithm is based on random draws from
multivariate normal and inverse Wishart distributions. This fact enables the researcher to
obtain more than one imputed data set.

This may be accomplished by selecting the MCMC algorithm on the Multiple Imputation
dialog box. Once this is done, click on Output Options to obtain the dialog box shown
below.

Enter the Number of repetitions and enter an integer value into the Set seed to field. For
this example, the number of repetitions equals 100 and the seed is set to 8735. Note that
the data set chol_new.lsf will contain 28  100 = 2800 cases.

Click OK to return to the Multiple Imputation dialog box. If the Syntax button is clicked,
the PRELIS syntax file shown below is generated.

PRELIS Examples Guide

77
The contents of the file chol_new.lsf are displayed below for cases number 2791 to 2800.

PRELIS Examples Guide

78