I have 2 groups (variable = condition) Control, Experimental - I have coded them as 1,2 in SPSS.

I have measured these at three time points - pre, ret1 ret2 (currently just a string variable)

I have one outcome variable I am interested in - X (I have 1-2 more that I'm also interested in)

Is it possible to run an ANOVA comparing the two groups for X at each time point? Do I need to code the time point variable also?

I would also like to compare groups seperately, so mean of Control X at pre, ret1 and ret2 and then same for Experiemntal X

Later I'm interested in the relationship between X and Y and/or Z, best method to use here?

Apologies for all the question and simple nature, I am new to stats and working my way through it. So far I ran an independent T-Test to compare control and Experimental X at one time point by using the Select Case function in SPSS - seems to me there is a better and more correct way than using multiple t-tests

Thanks for any help here! Jack


Just want to make sure I've got your design down pat.

What you're looking at is a mixed-design wherein all of your participants provide data (X) at three measurements (pre, ret1, ret2), and your participants can be split on a variable (condition).

The way to examine this is Analyze > General Linear Model > Repeated Measures.

SPSS will then ask you to input your repeated variable. This is your three measurements. You will need to give it a name, and tell SPSS how many levels to expect (3). Once you click "Add", click on "Define" to tell it what your model actually looks like.

What you will get now is a new window that wants Within-Subjects Variables and Between-Subjects Factors.

Within-Subjects Variables is where your data (X) goes. You've told SPSS that you have three levels, so it expects three things: pre, ret1, and ret2.

The Between-Subjects Factor is where you tell SPSS how you want your data split. This is where you give it condition which will make it possible to see if there is a relationship between condition and pre/ret1/ret2.

Really important note: When you set up your variables, make sure you set the "Measure" for the different variables. Specifically, your condition variable needs to be set as "Nominal" (though SPSS should infer this).

As for comparing groups separately, SPSS has a really nice function which lets you split your data. To get this, select Data > Split File. Save your data file before doing this, just in case. Make sure to select "Compare groups", and move the variable you're interested in (condition, sex, whatever) into the box. Select "Sort the file by grouping variables". When it is done, the data window in SPSS should have "Split by condition" in the bottom right.

Now, when you re-run your ANOVA, SPSS will give you two sets of output. One for when condition = 1, and one where condition = 2. To undo this, you will need to go to Data > Split File and select "Analyze all cases, do not create groups" and hit "OK". If you don't do this, SPSS will just keep working with a split file.

When you say relationship between X, Y and/or Z this makes me think of Pearson's r, but only because you used the word relationship, and because I know nothing else about X, Y and Z.


As an alternative to a standard repeated-measures ANOVA, you could also specify a mixed model and SPSS will output an ANOVA table for your fixed effects. You could specify the model by going to Analyze > Mixed Models > Linear. Put your subject ID variable in subjects and the time point variable in repeated. I would specify an unstructured covariance structure (unless you have a working theory on how the time points will be related, in which case maybe something like toeplitz?). Click continue and specify the model as you would normally (treatment group and time point), but make sure you click fixed and add the factors to the model, and include the intercept.

To compare the groups at each time point, you can click on EM Means, add treatment and time point, and check off compare main effects.

For assessing the effect of the time points within each treatment, you may want to run a simple effects model, i.e. splitting the data in half and running the model I just explained (minus the treatment group as a factor) on each half of the dataset.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.