An Introduction to Data Analysis using SPSS

Introduction and Aims

Given some data from an experiment or survey of some kind, an important first step is to explore
some of the basic features of the data using simple statistics and plots. Suppose, for example, that
we conduct a survey of people and ask them how often they smoke cigarettes, as well as some
demographic information such as age and sex. We might like to get an idea of the distribution of
ages of people who responded to the survey, or the overall proportions of people who smoke
regularly, occasionally and not at all, before we proceed to a more complex analysis to determine
factors which are associated with increased/decreased levels of smoking.

This introduction concentrates on using SPSS for the exploratory phase of data analysis, then briefly
discusses some commonly used statistical techniques, as follows:
Page
1. How data is input and stored in SPSS (including import from On-Line Survey and Excel) 1
2. Summary statistics and plots (for categorical data and for scale data) 4
3. Editing variables (recoding a variable and calculating a new variable) 7
4. Managing, Saving and Exporting SPSS Output 10
5. Testing for Normality (checking data prior to doing statistical tests) 11
6. Relationships between two variables (Cross-tabulation and Chi-Squared test, boxplots,
scatter diagrams, correlation coefficient) 14
7. Data sorting, grouping, transformation and selection 17
8. Comparing means (comparison and t-tests) 17
9. Resources 18
APPENDIX 1.: Summary of Useful Commands in SPSS 19
APPENDIX 2.: What Statistical Test do I need? 25

1 Data input

SPSS presents the data in two views: data and variable

Data view Looks like Excel. Each row is a “case”, e.g. person. Each column is an “attribute” or
variable, e.g. height, age, gender, place of birth.

Data Labels: switches

between codes and labels

1
Variable view describes each attribute:
• variable name (which must not contain spaces),
• type (string/numeric),
• label (a meaningful name for the variable for use in the output),
• values e.g. 1=”Maths”, 2=”English”, 3=”Physics”, (use Data Label button to see the codes)
• missing values (code(s) to denote missing data, e.g. 999 to represent missing data on age)
• measure (nominal, ordinal, scale)

Labels and values

How to enter Data: In Data view, type in the data (just as you would in Excel)
Copy and paste data e.g. from Excel, or from a table in Word
Import data from On-Line Survey (section 1.1)
Import an Excel file using File > Open > Data (section 1.2)

1.1 Importing data from On-line survey (formerly BOS)

Choose the type of file you

want to be created.

2
Possible problems and issues
• Be aware that choosing the “multiple choice (multiple answers) question” can cause problems
with analyses. Only use this question after careful consideration of how you will deal will the
data.
• If SPSS will not calculate a new variable it may be because that variable is counting 0 as the
indicator of a missing value. Change the “Missing” attribute to “none” and it should work.
• Choosing the Scale/rank question or the Grid question will lead to variables in SPSS or Excel
labelled as sub-questions, e.g. Q5_1, Q5_2 etc.
• Consider how to download the data:
o For SPSS, under “Options for coded exports” tick the box for “Combine scale/rank values into
a single column where possible”. This affects the format of the Scale/Rank questions, giving a
single score to represent the choice made; it does not affect other question types.
o For Excel, under “Options for coded exports”, tick the box for “Code responses”

1.2 Importing data from Excel

We can open an Excel data file in SPSS using File > Open > Data

Change the ‘Files of type:’ option to “Excel”. A dialogue box will ask about what part of the
spreadsheet it should import; if the first row of your Excel file is the field names then select the
option, ‘Read variable names from the first row of data’. Click ‘OK’ and the data should appear in a
new Data Editor window. Now change to the Variable View and make the necessary changes and
additions to the attributes of the variables. As a minimum, make sure that you check/change the
attributes ‘type’, ‘label’, ‘values’ and ‘measure’ for all the variables.

An alternative method is to copy the data from Excel and paste it into SPSS, making sure that you edit
the variable names, types etc.

2 Summary statistics and plots

The way we summarise data, and indeed the way we choose to analyse all data, depends strongly on
the nature of the data we are dealing with (in SPSS language, the “measure” of the variables
involved): scale, ordinal or nominal.

2.1 Categorical variables

We will use some examples based on data from a survey of secondary school children’s attitudes to
and awareness of the dangers of the sun and skin cancer. First, we consider the summary of data
associated with categorical (either nominal or ordinal) and sometimes discrete numerical variables.

Use Analyse > Descriptive Statistics > Frequencies

4
Move the required variable (e.g. ‘age’) from the list of variables in the left-hand panel to the central
panel labelled ‘Variable(s):’ Click on the Charts button, ensure that Bar Charts and Frequencies are
selected, then click Continue and then OK.

The output window now shows the frequency table and bar chart that we requested.

From the bar chart we can immediately make observations like, ‘most respondents are aged 13, with
quite a lot aged 14 and 12 also’. Using the frequency table we can quantify these observations using
the numbers or percentages SPSS has calculated.

Note that if a discrete variable takes lots of different values then then the frequency table will
become large and unwieldy and the bar chart will have lots of bars, rendering both of them much
less useful. It can be useful to recode the variable of interest e.g. into classes like ‘high’, ‘medium’
and ‘low’ (see Section 3.1). An alternative, if appropriate, is to treat the variable as a scale variable
and use the techniques described in the next section.

2.2 Scale variables

We now consider how to summarise numerical or scale variables, such as a person’s height or weight
or the number of questions they got correct in a test.

5
Simple summary statistics (minimum, maximum, mean, standard deviation) can be obtained from
Analyse > Descriptive Statistics > Descriptives

We will probably want more than this. Our example is derived from an experiment where volunteers
had their pulse rate measured, were randomly assigned to either run on the spot or sit still for 3
minutes and had their pulse rate measured again. Some other lifestyle and personal information was
also collected. We can summarise the scale variable “pulse1” using

Analyse > Descriptive Statistics > Explore

Choose this menu item and move the variable pulse1 into the ‘Dependent List:’ panel. Click on the
‘Plots’ button and deselect ‘Stem-and-leaf ’and select ‘Histogram’, then click Continue and then OK.

The table headed Descriptives in the Output window shows a huge number of summary statistics,
some of which might be incorporated into a report as part of a brief summary of this data. Next are a
histogram and box-plot of the data.

The bin locations and widths for the histogram are

chosen automatically, but can be changed through the Chart Editor which is obtained by double-
clicking on the chart you wish to edit.

Note that a box-plot shows the minimum, maximum, median and the first and third quartiles of the
data. An example is shown in Section 6.2.

6
It is possible to treat a discrete numerical variable (such as age) as though it were continuous and use
the Explore menu item rather than Frequencies. This should be done with care, but can be
appropriate when the number of values is very large and the output from frequencies would be
unwieldy. In this situation the statistics given by Explore become more meaningful than the
frequencies of all the individual values of the variable.

2.3 Discrete numerical variables

Numerical variables taking discrete values can be summarised using either “Explore” or
“Frequencies”. If the range of values is quite small then both options give sensible output, e.g. a
numerical representation of a Likert scale of 1, 2, 3, 4, 5. However, if the range of possible values is
very large then the output from Frequencies will be much less meaningful (the tables will be
enormously long), so the Explore option is preferred.

Think about what the output would be in each case for a variable recording the ages of 100 people
randomly stopped on a high street to participate in a survey. With such a variable another option is
to create your own groups using

Transform > Recode into Different Variables, (explained in 3.1 below), then analyse the new variable
using Frequencies.

3 Editing Variables

3.1 Recoding a Variable

Sometimes we want to group items together into new groups. For example, suppose we have
data on children in the different years but are particularly interested in comparing students in
school years 7 and 8 with the other children in the survey. (Perhaps they have been the target
of a recent campaign and we want to investigate whether it has been effective.)
SPSS can create a new variable for us that indicates whether the respondents are younger than
the target group, in the target group, or older than our target school years. We want to recode
the ‘yearGroup’ variable to a new variable, say ‘targetYears’, as in the table:

yearGroup 6 and below 7 and 8 9 and above

targetYears 0 1 2
We could do this manually, but SPSS can do the hard work for us through the menu command
Transform > Recode into Different Variables.
Choose this menu item, select the yearGroup variable from the list on the left hand side and give
the Output Variable the name ‘targetYears’ and a sensible label and click ‘Change’.

7
Click on ‘Old and New Values. . . ’.
In the Old Value section of the dialogue box, click on ‘Range, LOWEST through value:’ and enter ‘6’ in
the input box; then enter ‘0’ (zero) under New Value and click the ‘Add’ button. Repeat this for the
other values (1 and 2) of the new variable, using the ‘Range:’ and ‘Range: value through HIGHEST:’
options in the left-hand panel of the dialogue box. (If you make a mistake use the ‘Change’ and/or
‘Remove’ buttons.) The dialogue box should look like this:

Check the entries in the box labelled ‘Old --> New’, then click ‘Continue’ and ‘OK’; and the if new
variable with its appropriate values, will be added to the data, ready for further analysis. In
Variable View of the Data Editor give the new variable an appropriate Label and Values and
check that the Measure attribute is appropriate.
There is also a menu item ‘Recode into Same Variables’ which overwrites old values with recoded
values. Use this option carefully. It is usually wise to recode into a new variable and keep your
original data. If you are sure that won’t want to directly analyse the original variable and have
your data backed up elsewhere then, after checking that the recode has worked as you intend,
you might delete the old variable.

8
3.2 Calculating a new variable

Suppose that we want to convert a height measured in inches to metres for further analysis and
so our reporting will conform to standard practice. We could do the calculation ourselves, but we
can also get SPSS to do it for us using the

Transform > Compute Variable

menu item. You need to name the new variable you want to create (e.g. heightMetric) and then
enter the formula (there are 2.54 cm in an inch) to compute its values and click OK. (Also note the
extensive list of available functions in the list in the lower-right section of the Compute Variable
dialogue box.)

If you switch to Data View then you should see the new variable with all its values calculated.
Note the difference between the Compute and Recode functions. They do similar things, but you
should use Compute when there is a simple mathematical formula to describe the relationship
between the existing variables and the one you want to calculate; otherwise use Recode. A little
thought about which one will be simpler can save you a lot of time!

9
4 Managing, Saving and Exporting SPSS output

4.1 Exporting output

Having produced several tables and plots that are now in the SPSS output window, you will perhaps
want to incorporate some of this output into a more formal document of some kind (a short report,
paper or thesis, for example). To export tables to Word or Excel, simply select the relevant object in
the SPSS output window and copy it, then paste it into the relevant application. The same copy-paste
method works for exporting charts/plots to Word.

To export charts/plots for use in some other program (like LATEX, or just so that you have a copy of a
plot that you can then import into various other documents without having to copy-paste from
SPSS), right-click on the plot to export and select ‘Export’ from the menu that pops up. Make sure
that the Document Type is set to ‘None (Graphics only)’ and the Graphics Type is set to an
appropriate format. Before clicking OK, check under the heading ‘Root File Name:’ to see (and
possibly change) where your file will be exported to.

To export a table, right-click on it and choose Export in the same way; this time choose ‘Text - Plain’
under Document Type and check the location your output will be saved to, which is shown under the
heading ‘File Name:’

4.2 Managing output

Once SPSS has produced a number of tables and charts in the output window, it can be a bit difficult
to navigate around the output and find a particular chart or table. The separate pane on the left
hand side of the output window is very useful for navigating around the output: the items in this
pane are a sort of ‘table of contents’.

Clicking on the grey triangles near the start of the items in the left hand pane collapses their contents
so you can more easily see the remaining items in the list. Clicking on an item in the list moves the
output in the main part of the window so that you can see the element whose title you clicked on

10
(notice the red arrows that appear in both the ‘contents pane’ and the main output window showing
which entries in the left pane correspond to which parts of the output file). Output that is not
needed can be deleted by right-clicking on the appropriate entry in the left hand pane and choosing
‘Cut’ from the menu that pops up. You can also drag and drop items in this contents pane to move
them around and change the order in which they appear in the output window.

It is good practice to delete any unnecessary output promptly to reduce the amount of clutter in the
output window and make it easier to find what you are looking for!

To further help you keep a track of what you have in the output window, you can use

Insert > New Text to write short notes in the output window about what analyses you have
done and why.

4.3 Saving Output

With the output window as the front window on your computer the usual

File > Save or File > Save As will allow you to save the output.

Note that this generates a .spv file, whereas data is saved as a .sav file.

5 Testing for Normality

An important part of many statistical procedures which you will

most likely come across in the future involves checking assumptions
about data which are necessary for the procedures to be useful.
One of the most common checks you will have to do will be a
variation of the question ‘are the data (approximately) Normally
distributed?’ The Normal (or Gaussian) distribution is perhaps
familiar to you through its famous ‘bell curve’.

Parametric test What to check for normality

Independent t-test Dependent variable by group

Paired t-test Paired differences

One-way ANOVA Residuals

Repeated measures ANOVA Residuals at each time point

Pearson’s correlation coefficient Both variables are normally distributed

Simple linear regression Residuals

11
.

The definite ‘waviness’ of the Normal Q-Q plot for the “pulse1” data indicates that the data is not
exactly described by a Normal distribution; however the fact that it does not deviate very much from
the straight line indicates that a Normal distribution does provide a fairly close approximation. In
almost all applications this is good enough!

5.4 Statistical tests

Another, more objective, way of assessing whether data might be well-described by a Normal
distribution is to do a formal statistical test, a hypothesis test. The simplest of these is the
Kolmogorov-Smirnov test, which can be accessed through

Analyze > Nonparametric Tests > Legacy Dialogs > 1-Sample K-S

Other commonly used tests are the Shapiro-Wilk and Anderson-Darling tests. The important output
from such tests is the p-value (labelled ‘Sig.’ in SPSS output tables), which is a measure of how likely
the observed data are, under the assumption that they are perfectly described by a Normal
distribution.

If the p-value is small then, assuming the data is described by a Normal distribution, the observed
data is quite unlikely; we thus conclude that the observed data is most likely not perfectly described
by a Normal distribution. Conversely, if the p-value is large then, if the data is described by a Normal
distribution, the observed data is quite likely, so the data we have observed is entirely consistent
with it being described by a Normal distribution. As a guideline, a p-value larger than 0.1 would
usually be considered as giving no evidence against the Normal distribution being appropriate,
between 0.1 and 0.05 is weak/some evidence against Normality, 0.05 to 0.01 is good evidence and
less than 0.01 is strong evidence that the data is not Normally distributed.

However, caution must be exercised when using such tests for Normality! If the sample size is even
moderately large (50–100 and larger), then such tests are very sensitive even to quite small
differences between the distribution of the data and a Normal distribution. If the p-value is small but
the sample size is large then we need to interpret the results of this test with a little caution and
common sense. If the data do not exactly fit a Normal distribution, they may still be very close and

13
for all practical purposes the Normal distribution is a good enough approximation. In this situation,
there might be a statistically significant difference between the distribution of the data and the
Normal distribution that best fits it, but no practically significant difference. A Q-Q plot is very useful
to make this case, as in the example above.

Lastly, note that Q-Q plots and the Kolmogorov-Smirnov test can also be used to test if data is well
described by other specified probability distributions. You may have heard of the Poisson,
Exponential, Gamma, Pareto and Laplace distributions; SPSS allows you to compare data to these
distributions (and a few others too). The fit of data to these other distributions is assessed using the
same methods described above for Q-Q plots and Statistical tests, selecting the appropriate
distribution instead of the Normal that we used here.

6 Relationships between two variables

We now look at exploring the relationship between 2 (or more) variables; asking questions like ‘Does
changing the value of one variable affect the value(s) of the other in some way?’

6.1 Two categorical variables

Firstly, we look at the case where both variables are categorical (ordinal or nominal). The data can
most easily be explored using a table giving the frequencies of occurrence of particular combinations
of the two variables of interest. To do this we use

Analyse > Descriptive Statistics > Crosstabs

We might wish to explore a possible relationship between gender and whether someone smokes.
Enter “smokes” as the row variable and “gender” as the column. The Statistics button allows you to
get SPSS to calculate various statistics and perform some statistical tests on the data e.g. the Chi-
squared test to check for association. The Cells button allows you to alter the information displayed
in the output tables; in particular, the Percentages section can be useful. There is also an option to
display clustered bar charts.

14
Choose the options carefully: can you determine what proportion of females are smokers? And what
proportion of smokers are female? These are not the same! Other options could be to use “smokes”
as the Row variable, “activity” as the Column and “gender” as a Layer.

Also note that the Crosstabs procedure is only sensible for categorical variables, there is a separate
row/column for every distinct value the Row/Column variable takes, so the output will be
meaningless if we put a scale variable anywhere in the Crosstabs dialogue box.

6.2 One categorical and one scale variable

Now we shall see how to explore how a categorical variable might influence a continuous variable.
An example might be to see if typical levels of activity seem to affect people’s normal resting heart
rate, (measured by pulse1). To do this we use

Analyse > Descriptive Statistics > Explore

as before. Now, however, we want to separate out the pulse1 data according to their activity levels;
to do this put activity in the Factor List.

(You might want to go into the Plots options and make sure that Histograms and Stem and Leaf plots
will not be produced.) A box plot is produced which allows some comparison between the
individuals grouped according to activity level. This can also be done by comparing some of the
statistics produced in the table above the box plot.

For a more complex box plot, use

Graphs > Legacy Dialogs > Boxplot or Graphs > Chartbuilder

For example, choose a Clustered box plot with pulse1 as the Variable, activity as the Category Axis
and Define Clusters by gender. Or, using a ‘simple’ boxplot, put gender as a variable to “Panel by”
instead of using it to cluster.

15
7 Data Grouping/Selection

Sometimes it is useful to analyse only a certain subset of the data you have collected, or alternatively
to compare different subsets of your data.

The Data > Split File menu item allows you to break the data into disjoint subsets and do analyses
independently on each subset. Choosing to ‘Compare groups’ or ‘Organise output by groups’ affects
the way tables in the output are presented, but they will give the same information. Returning to the
Split File menu option and selecting ‘Analyse all cases, do not create groups’ cancels this option.

The menu item Data > Select Cases allows you to temporarily ignore some of the data and analyse
one specific subset of the data, e.g. just males. You can restrict which observations (rows of data)
SPSS uses in any analysis by changing the contents of the ‘Select’ box; the ‘Output’ box controls what
happens to the observations that are not selected. Use the SPSS Help to assist you if you’re unsure of
what the options do.

Data > Sort will sort the data into order, e.g. to order records by region.

8 Comparison of Means

8.1 Calculation of means for different groups

Example: You may want to see whether exercise increases the pulse rate.
Choose Anaylze > Compare Means

then drag variables into the boxes as shown.

The mean pulse rate appears to be higher after running, but unchanged after sitting. However, the
standard deviation is large so we may want to check that this increase is not just part of the inherent
variability of the data.

17
8.2 T-tests
These are used to compare the means of 2 groups of
Normal scale data.
You could split the data into those who ran and those who
sat (use Data > Split File), then test the hypothesis that
there is no difference between the pulse rate before and
the pulse rate after the activity for these two groups.

Use Analyze > Compare Means > Paired t test

The data shows that the pulse rate increases after running ( t = -7.435, df = 34, p < 0.001). Also, the
95% confidence interval for the change in pulse rate is -24.084 to -13.745, indicating that the pulse
rate before running was less than that after running.

9 Resources

• The SPSS Help menu: Topics search, Tutorials and Statistics Coach can be very useful.
• Books: ask colleagues for recommendations or just look in the library for something that suits
you. Two suggestions are: “SPSS for Psychologists” by Brace, Kemp & Snelgar and “Discovering
Statistics using IBM SPSS Statistics” by Andy Field.
• The internet: the web has many examples, tutorials, guides, etc. that various people have put up
with the intention of helping people in a similar position to you. They can be very useful if you’re
not sure about something related to either SPSS or statistics in general. Try
o Statstutor, http://www.statstutor.ac.uk/, which has worksheets and guides on many
commonly used statistical techniques
o Laerd Statistics: e.g. information on data input is found at https://statistics.laerd.com/spss-
tutorials/entering-data-in-spss-statistics.php

18
APPENDIX 1: Useful Commands in SPSS

1. Menu: Main options page 19

2. Data Summaries (Analyze > Descriptive Statistics.) 20
3. Comparison of means, Confidence Intervals 20
4. Some commonly used statistics including Non-parametric tests 20
5. Graphs 22
6. Other useful information and output 24

1. Menu Main Options and some useful commands

File the usual things such as Open (including opening Excel files), Save, Print

Edit Undo/Redo, Copy/Paste, Insert Cases, Find, Go to case (to find an item)

Data Sort Cases, Select cases, Transpose variables, Split File

Transform Recode (to group data) into the same or a different variable,
Compute (to calculate another value based on other variables, e.g. calculating BMI
from height and weight)

Analyze Used a lot. See sections below.

Graphs Used a lot. See below.

View, Direct Marketing, Utilities, Add-ons, Window can be ignored for now

Analyze

Analyze Main options:

• Descriptive Statistics,
• Compare means,
• General Linear Model,
• Correlate,
• Regression,
• Non-parametric tests.

Other options include

• Dimension Reduction
(cluster analysis, factor
analysis)
• Scale (Reliability analysis,
• Etc.

19
2. Data Summaries

Analyze > Descriptive Statistics > ……

Frequencies ordinal or nominal data only. Used to count the numbers in each category and
provide a frequency table (not appropriate for continuous data)
Statistics tab: allows calculation of means, quartiles, range, sd etc.
Charts tab: allows selection of barchart, piechart or histogram.
A Normal curve can be added to a histogram by double clicking on the chart and choosing Add
Distribution.

Descriptives scale data to calculate minimum, maximum, mean, standard deviation

Explore scale or ordinal data, discrete data with a large range of values
Used to get statistics for different groups e.g. summaries for people on different
courses. Options include descriptive statistics (mean, sd, median, IQ range,
confidence interval for mean etc) and box plots.

Crosstabs (categorical data) to produce contingency tables to investigate relationships

between two variables). Use this option to choose a Chi-squared test for association.

Q-Q Plot To produce a plot which lets you test for Normality. It compares the observed values
with the expected Normal values. (Note that we could also get a P-P plot, but Q-Q plots are usually
preferred as they compare the quantiles, not the proportions, so get better resolution at the tails).

3. Comparison of Means

Analyze > Compare Means

Means Calculates the means and s.d. of specified groups (e.g. Weight at start for the two
treatment groups). Other statistics can also be chosen, e.g. skewness, range,
standard error etc.

t-tests Choice of one sample, independent samples and paired samples. Options include
confidence intervals

One-way ANOVA To compare several means (see below).

4. Some Commonly-used statistical techniques

4.1 ANOVA
ANOVA tables are obtained from:
Analyse > Compare means > One way ANOVA,
Analyse > Regression > Linear regression,
Analyse > General Linear Model >

20
 Type of Analysis of Variance SPSS commands
One way ANOVA Analyse > Compare Means
ANCOVA (Analysis of covariance) Analyse > General Linear Model > Univariate
Two way ANOVA (or n-way independent Analyse > General Linear Model > Univariate
ANOVA, factorial design) (Field, page 509)
Repeated measures design Analyse > General Linear Model > Repeated
Measures
Mixed design Analyse > General Linear Model > Repeated
Measures
MANOVA (Multivariate Analysis of variance) Analyse > General Linear Model > Multivariate
Options include profile plots, post-hoc tests (e.g. Tukey, Dunnett), saving residuals (to allow testing
for Normality).

Paired and unpaired t-tests

Analyse > Compare Means > Paired-Samples T Test
(Also: ……. > One Sample T Test, Independent samples T Test, One-way ANOVA)

4.2 Relationships Between Two Variables

Categorical data Analyse > Descriptive Statistics > Crosstabs
click “Cells” to include percentages, click “Statistics” for Chi-squared test,
can also get clustered bar-charts)

Two scale variables Graphs > Legacy Dialogs > Scatter/ Dot

Boxplots Analyse > Descriptive Statistics > Explore choose boxplots

Graphs > Legacy Dialogs > Boxplot
For one categorical compared with one scale variable, put the categorical
variable in “Factor”.

Chi-Squared Test
Analyse > Descriptive Statistics > Crosstabs

Correlation Analyze > Correlate > Bivariate to calculate correlation coefficients

(Pearson, Kendall’s Tau, Spearman)

Regression Analyze > Regression > Linear to perform linear regression (also
ANOVA tables). Options include a variety of additional statistics, histogram of
residuals, normal probability plots etc.

4.3 Nonparametric tests

Analyze > Nonparametric tests > Legacy Dialogs to perform Mann-Whitney U test, Chi- Squared
Test, Kolmogorov-Smirnov test, Kruskal-Wallis
test, Wilcoxon, Friedman etc.
These can also be accessed via
Analyze > Nonparametric tests > Related samples
or Analyze > Nonparametric tests > independent samples.
SPSS will choose an appropriate test, or you can specify what test is to be done.

21
Double-clicking on the output table will produce additional charts etc. which can be very helpful.

Example Hours spent in library by mature and non-mature students

4.4 Factor Analysis and Reliabillity

Analyze > Dimension Reduction > Factor for factor analysis
Analyze > Scale > Reliability Analysis for Cronbach’s Alpha, Split-half reliability, Inter-item
correlations and covariances, Intra-class correlation
coefficient etc.

5. Graphs

22
6. Other useful information

6.1 Input
Value Labels Button (use when in data view to see the meaning of the coded data)
In Variable view, in the “Values” column, input values and labels for variables (click on the cell, then
on the blue box at the right-hand side)
File > Open > Data (enables you to open and input an Excel file, or data from appropriately
formatted text files)

6.2 Recoding a Variable

Transform > Recode into Different Variables (usually better than “into Same variables)
Transform > Compute variable (e.g. to change from Imperial to Metric units)

6.3 Initial Checks

Data > Sort Cases (to sort the data by one or more variables)
Analyse > Descriptive Statistics > Descriptives (to check the amount of data, its location, range and
spread)
Analyse > Descriptive Statistics > Frequencies (to check the distribution)
Click on Charts to get a barchart or pie chart or histogram
Analyse > Descriptive Statistics > Explore
Click on Plots to get a histogram and stem & leaf. (tick the box to get a Normal curve).
(Use Explore rather than Frequencies when summarising discrete items where the range of values is
large: Frequencies gives too many groups in this case).
Chart Editor double click on the chart
Graphs > Legacy Dialogs > Histogram (tick the box if you want to “Display Normal Curve”)

6.4 Data Grouping and selection

Data > Split File
Data > Select Cases

6.5 Testing for Normality

1. Draw a histogram and tick “Display Normal Curve”
2. Q-Q plot Analyse > Descriptive Statistics > Q-Q Plots
3. Analyze > Nonparametric Tests > Legacy Dialogs > 1-Sample K-S (for a Kolmogorov-Smirnov
Test).

6.6 Output
Copy and Paste items to Word or Excel
Or right click and use Export to set up a new file.

24
APPENDIX 2: WHAT STATISTICAL TEST DO I NEED? From www.statstutor.ac.uk

Deciding on appropriate statistical methods for your research:

• What is your research question?

• Which variables will help you answer your research question and which is the dependent
variable?

• What type of variables are they?

• Which statistical test is most appropriate? Should a parametric or non-parametric test be used?

KEYWORDS:

VARIABLE: Characteristic which varies between independent subjects.

CATEGORICAL VARIABLES: variables such as gender with limited values. They can be further
categorised into NOMINAL (naming variables where one category is no better than another e.g. hair-
colour) and ORDINAL, (where there is some order to the categories e.g. 1st, 2nd 3rd etc).

CONTINUOUS (SCALE) VARIABLES: Measurements on a proper scale such as age, height etc.

INDEPENDENT VARIABLE: The variable we think has an effect on the dependent variable.

DEPENDENT VARIABLE: The variable of interest which could be influenced by independent variables.

PARAMETRIC TESTS: there are various assumptions for parametric tests including the assumption
that continuous dependent variables are normally distributed. There are specific tests for this within
packages such as SPSS but plotting a histogram is also a good guide. As long as the histogram of the
dependent variable peaks in the middle and is roughly symmetrical about the mean, we can assume
the data is normally distributed (see examples below).

25
Table of tests Dependent Independent Parametric test Non-parametric
alternative
(outcome) (explanatory)
variable variable

Comparing means
'
The averages of two Scale Nominal/ Independent t-test Mann-Whitney test
INDEPENDENT groups binary (Wilcoxon rank sum)

The averages of 3+ Scale Nominal One-way ANOVA Kruskal-Wallis test

independent groups

The averages of 2 paired Scale Nominal Paired t-test Wilcoxon signed rank
(matched) samples e.g. Time/condition test
weight before and after a diet variable

The 3+ measurements on the Scale Nominal Repeated measures Friedman test

same subject ANOVA

Investigating relationships

Relationship between 2 Scale Scale Pearson's Spearman's Correlation

continuous variables Correlation Coefficient.
Coefficient.
Predicting the value of Scale Any number of Simple Linear Transform the data
one variable from the scale or binary Regression
value of a predictor
variable Binary Any number of Logistic regression
scale or binary

Assessing the relationship Nominal Nominal Chi-squared test

between 2 Nominal variables
Note: The table only shows the most common tests for simple analysis of data.

Examples:

Are height and weight related? Both are continuous variables so Pearson's Correlation Co-efficient
would be appropriate if the variables are both normally distributed.
Can height predict weight? You cannot determine height from weight, but you could estimate weight
given height, so height is the continuous independent variable. Simple linear regression will help
decide if weight is a good predictor of height and produce an equation to predict weight given an
individual's height.

Is Diet 1 better than Diet 2? A researcher would randomly allocate subjects to two groups with one
group following Diet 1 and the other Diet 2. Weight would be taken before and after the diet and the
mean weight lost compared for the two groups. The dependent variable 'weight lost' is continuous and
the independent variable is the group the subject is in, which is categorical. If the data is normally
distributed, use the independent t-test, if not use the Mann-Whitney test.
Are patients taking treatment A more likely to recover than those on treatment B? Both 'Treatment'
(A or B} and 'Recovery' (Yes or No} are categorical variables so the Chi-squared test is appropriate.

26
 One scale dependent and several independent variables
1st independent 2nd independent Test
Scale Scale/ binary Multiple regression
Nominal (Independent groups) Nominal (Independent groups) 2-way ANOVA
Nominal (repeated measures) Nominal (repeated measures) 2-way repeated measures ANOVA
Nominal (Independent groups) Nominal (repeated measures) Mixed ANOVA
Nominal Scale ANCOVA

Regression or ANOVA? Use regression if you have only scale or binary independent
variables. Categorical variables can be recoded to dummy binary variables but if there are
a lot of categories, ANOVA is preferable.

Index Page
ANOVA 20, 21
APPENDIX 1: Useful commands in SPSS 19
APPENDIX 2: What statistical test do I need? 25
Bar charts 5, 22
Boxplots 15, 21
Calculating a new variable 9
Categorical variables 4, 14
Chi-squared test 14, 20, 21
Cross-tabulation 14, 20, 21
Correlation 21
Data input 1, 24
Data Select Cases / Split File / Sort 17
Descriptive statistics 4 to 6, 14, 20
Discrete variables 7
Downloading data 3
Exporting output 10
Graphs 15, 16, 22
Histogram 12
Kurtosis 12
Means 17, 20
Menu: main options 19
Menu: Analyze 4, 6, 17 to 22
Non-parametric tests 13, 21
Normality, testing for 11
On-line survey (BOS) 2
Output 10
Q-Q plots 12, 20
Recoding a variable 7
Regression 21
Relationship between two variables 14
Resources 18
Scatter diagrams 16, 21
Scale variables 5
Skew 12
Statistical tests, Appendix 2 25 to 28
Summaries for categorical, scale discrete variables 4, 5, 7
Summary statistics 4, 20
Testing for normality 11 to 14
Transform > Compute variable 9
Transform > Recode 7
T- tests 18, 20
Mary Lorimer, 19/06/19

27
 Flow Chart for statistical tests

Dependent variable/ Continuous Categorical

Outcome DEPENDENT variable DEPENDENT variable

Categorical
DEPENDENT
variable, 2
Independent One CONTINUOUS Two or more outcomes
One
variable / INDEPENDENT CONTINUOUS Two or more Two or more CATEGORICAL
CATEGORICAL
predictor variable INDEPENDENT CATEGORICAL CATEGORICAL & INDEPENDENT
INDEPENDENT
variables INDEPENDENT CONTINUOUS variable
variable
variables variables

Comparing Repeated measures, Prediction of

Assessing a independent more than one dependent
Prediction of
relationship unmatched continuous observation
dependent
groups per subject

2 More than 2 More than

groups 2 groups groups 2 groups
Parametric: Simple Linear
Pearson’s Regression Parametric:
Parametric: Parametric: Parametric: Factorial
correlation repeated Multiple Contingency table, Logistic
Independent One-way Paired t-test ANOVA
(Check measures Regression Chi-Squared Regression
samples t-test ANOVA
Non-parametric: residuals for ANOVA
Spearman’s normality) Non-parametric: Factorial
Non- ANCOVA 2×2 table: Fisher’s
correlation Non-parametric: Wilcoxon signed Non- Repeated
parametric: Exact test
Mann-Whitney rank test parametric: measures
Kruskall- ANOVA
Friedman test
Wallis
Multiple
Multiple regression
regression Ordinal DEPENDENT, with lots of levels:
Factorial treat as non-parametric continuous
Mixed
ANOVA Ordinal INDEPENDENT variable will
generally imply the use of non-parametric
tests

Ordinal DEPENDENT & INDEPENDENT

with small number of groups can be treated
as nominal

28