Factorial ANOVA in SPSS In the dataset to be used for this example, there are two N-level variables (“treatment” and “problem”) for each person—“treatment” has two levels (CBT [cognitive-behavioral therapy] vs. IPT [interpersonal psychotherapy]), and “problem” has three levels (anxiety disorder vs. depression vs. eating disorder). Each person (row in the table) has one type of diagnosis, and receives one treatment. Then each person also has a “symptom” score, indicating their level of symptom severity after treatment (so this is your outcome variable, to see if the treatment was effective—lower scores on “symptom” mean the treatment helped). The goal of this two-way ANOVA is to determine whether people who receive different treatments have different outcomes, and whether the type of diagnosis that the person is being treated for moderates the effect of treatment types on symptom levels. (In other words, the question is whether there’s an interaction between treatment type and problem type, in terms of these two variables’ effects on symptom levels). To do this analysis, we start (as usual) from the “Analyze” menu. One way to get a basic F-test is to use the “Compare Means” command, but there’s another way to do it: select “General Linear Model,” and then select “Univariate.” (A “univariate” ANOVA has one dependent variable—that’s true in this case, where the only dependent variable is “symptom” level). For this example (as well as ANCOVA, MANOVA, and other procedures) it’s necessary to use the “General Linear Model” instead of the “Compare Means/ANOVA” command. This is because it’s a more generalized procedure.
7
Embed
Factorial ANOVA in SPSS - Denver, Colorado · Factorial ANOVA in SPSS In the dataset to be used for this example, there are two N-level variables (“treatment” and “problem”)
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Factorial ANOVA in SPSS
In the dataset to be used for this example, there are two N-level variables (“treatment”
and “problem”) for each person—“treatment” has two levels (CBT [cognitive-behavioral
therapy] vs. IPT [interpersonal psychotherapy]), and “problem” has three levels (anxiety
disorder vs. depression vs. eating disorder). Each person (row in the table) has one type
of diagnosis, and receives one treatment. Then each person also has a “symptom” score,
indicating their level of symptom severity after treatment (so this is your outcome
variable, to see if the treatment was effective—lower scores on “symptom” mean the
treatment helped). The goal of this two-way ANOVA is to determine whether people who
receive different treatments have different outcomes, and whether the type of diagnosis
that the person is being treated for moderates the effect of treatment types on symptom
levels. (In other words, the question is whether there’s an interaction between treatment
type and problem type, in terms of these two variables’ effects on symptom levels).
To do this analysis, we start (as usual) from the “Analyze” menu. One way to get a basic
F-test is to use the “Compare Means” command, but there’s another way to do it: select
“General Linear Model,” and then select “Univariate.” (A “univariate” ANOVA has one
dependent variable—that’s true in this case, where the only dependent variable is
“symptom” level). For this example (as well as ANCOVA, MANOVA, and other
procedures) it’s necessary to use the “General Linear Model” instead of the “Compare
Means/ANOVA” command. This is because it’s a more generalized procedure.
The dialog window asks you to select your independent and dependent variables. There
are two N-level predictors as independent variables—they go in the “Fixed Factors” box
(this term refers to any N-level predictor in an ANOVA procedure). Then your criterion
variable—“symptom”—goes in the DV spot:
Hit the “Model” button to continue.
The “model” dialog box gives you a chance to customize the list of predictors that you
want to test out. The default model is a “full factorial” model, which means that SPSS
will test all of the possible main effects, and also all of the possible interactions among
the fixed factors you selected in the previous step. If you leave the “full factorial” setting
in place for this example, you’ll get exactly the same result as I do by customizing the
model. However, I’m going to do the customization, to show you exactly what we’re
testing for.
To set up a custom model, I select the predictors that I’m interested in from the list on the
left, select the type of test that I want to run from the drop-down menu in the middle of
the screen, and then use the arrow to move the predictors to the right-hand list. First, I
select “main effects” from the drop-down menu, and use the arrow to create a “main
effect” test for each individual predictor’s effect on the criterion variable. Next, I select
“interaction” from the drop-down menu. This time, I have to select both variables that I
want an interaction for, and use the arrow with both of these predictors selected. This
gives me the interaction term, which looks like this: “treatment*problem”.
When you’ve duplicated what I have shown above (or if you just left the “full factorial”
default setting in place), click “Continue” to go on.
Note: For now, leave this other drop down menu set to “Type III”
sums of squares. Next week we’ll change this setting, and look at
the difference between “Type I” and “Type III” sums of squares,
when we start to work with ANCOVA (analysis of covariance).
Back in the main dialog box for univariate ANOVA, click on the “Plots” button to see
this dialog box:
This dialog allows you to generate the type of line graph that we saw in the lecture notes.
One of the predictor variables goes on the horizontal axis of the graph, and the other is
used to define separate lines (refer to lecture notes for an example of the line graphs that
show main effects and interaction effects). Let’s put “treatment” on the horizontal axis,
and use “problem” as the variable to define three separate lines on the graph. The y-axis
of the graph will automatically be the dependent variable (“symptom”).
Click “Continue” to go on.
Once you have put the variables into the correct
boxes in the dialog, you have to hit the “Add”
button, and get the graph to show on this list, in
order for it to appear on your output.
Back in the main dialog window, click on “Options” to see this sub-dialog:
This window lets you get various types of descriptive statistics. First, select the variables
you want summary statistics on from the left-hand list, and use the arrow to move them to
the right-hand list. Then use the check boxes down below to select the types of summary
statistics you want to see. Let’s get “descriptive statistics” and also the “observed power”
for each statistical test.
Click “Continue” to go on. Then, back in the main dialog box, we’re ready to see the
results, so click “OK” in the main window for this analysis.
Here are some highlights of the output:
This table (“Tests of Between-Subjects Effects”) gives you the results of the F-tests for the two main effects and the interaction. Tests of Between-Subjects Effects Dependent Variable: Symptom Severity (higher = worse)