2

I'm working with longitudinal data, and am using the mixed models function ins SPSS to look for predictors (level-1) nested within individuals (level-2).

If I add the participant ID as "subject" and days as "repeated," in the initial dialogue box, and then go into "random effects" to pull the Participant IDs into the combination box (at the bottom), do I still have to "nest" the predictors within the individuals in the "fixed effects" / "random effects" boxes? --> predictor(participant)

Thanks a bunch!

6
  • Hi there, we need more information to help you. What is your data set - what variables are you using, what do they look like (e.g. show the variable names and the first couple of rows under each, you can use mock data). Second, what model are you trying to fit?
    – Michelle
    Commented Jul 14, 2020 at 1:51
  • Hi Michelle! The daily study ran for 20 days (t=20), where participants (ID) were asked to take the same 5 tests everyday and record their test scores (TEST1, TEST2, TEST3, TEST4, TEST5). These test scores were all quantitative. I'm fitting it into a 2-level linear mixed model using the mixed model function on SPSS. Truly appreciate your help on this!
    – user291225
    Commented Jul 14, 2020 at 12:25
  • And what is the hypothesis you are testing?
    – Michelle
    Commented Jul 14, 2020 at 22:03
  • Assuming you want the results separately by test, the solution is here, where you will convert your data so that test1 has 20 columns - 1 for each day, and you'll have "time" or "day" as your factor, with 20 levels. ezspss.com/…
    – Michelle
    Commented Jul 14, 2020 at 22:09
  • Hi Michelle, it's an exploratory study to look for correlations/predictors. Because of the complex nature of the study, and the different fixed/random effects associated, I was hoping to use a linear mixed model / MLM rather than ANOVA. Any chance you might be able to briefly walk me through the process for the mixed model function in SPSS? Specifically, whether or not we need to separately "nest" the predictors within the individuals if it's specified as the "subject"? Thanks again Michelle!
    – user291225
    Commented Jul 15, 2020 at 0:39

1 Answer 1

0

If you specify subjects and repeated in the initial dialog, then you're specifying use of an R or residual matrix for subjects (participants) with the specified covariance structure. For a simple two-level model like this, you would typically either do this or specify random effects, but usually not both. Predictors would likely not be nested in fixed or random specifications in either case.

In earlier versions of the program, only one test at a time could really be handled, but in more recent ones, Kronecker product covariance structures have been added, which would let you fit all five tests at the same time, using a categorical test variable in addition to the time variable in specifying the Kronecker product structure.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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