Multi-level modelling?

Question

In an instructional study, I have pretest and post-test measures of writing quality--no control condition. There are 110 students nested in 10 classes. I have pretest measures of spelling skill and handwriting fluency as covariates. I want to test:

-Did scores increase from pretest to post-test?

-Was the level of text quality predicted by each covariate?

-Did each covariate predict gains from pretest to post-test in text quality?

-Did classes differ in gains in text quality from pretest to post-test?

-If possible, test whether any differences among classes in text quality gains were moderated by handwriting and by spelling

With only two time points, can I use multi-level modelling for this? In SPSS?

Thanks for any advice on this!

Answer 1

You can do this, but to get your last 2 bullet points, you'd need to put in class as a categorical 10-level fixed predictor and its interactions with time, handwriting and spelling, and I don't think you have enough data to estimate all that.

Technically, however, you'd do this using SPSS Mixed Model procedure. You need to have your data in long format, i.e. two rows for each student (one for pre- and one for post-score), and a column indicating the time of measurement (e.g. with values "pre" and "post"). Something like this:

MIXED testscore by time class with cov1 cov2
 /FIXED time class cov1 cov2 time*class class*cov1 class*cov2 time*cov1 time*cov2
 /RANDOM=INTERCEPT|SUBJECT(studentid) COVTYPE(CS).

Perhaps a better solution would be to leave out the class fixed effect and interactions and put in class too as random effect. That way you can at least see how much of the variance in test scores is attributable to class. That would go something like this:

MIXED testscore BY time WITH cov1 cov2
 /FIXED time cov1 cov2 time*cov1 time*cov2
 /RANDOM=INTERCEPT|SUBJECT(studentid) COVTYPE(CS).
 /RANDOM=INTERCEPT|SUBJECT(class) COVTYPE(CS).

This way you reach the goal of your 3 first bullet points at least.