1
$\begingroup$

Hello I have trouble understanding SPSS way of making repeated measurements with a general linear model. I have data of the form

id cov1 cov2 between1 between2 time1 time2
 :   :    :     :         :      :     :

With cov1 and cov2 being covariates between1/2 is the between factors both with 2 levels eg gender and drug/placebo and two repeated measures time1 and time2. I fit the full factorial model with general linear model -> repeated measures with the cov1/2 as covariates between1/2 and between subjects.

I would expect there to be 6 parameters (intercept, cov1, cov2, between1, between2, between1 * between2) but looking at parameter estimates there is 2*6 ie 6 for each level of time.

What is the interpretation of the model in terms of

y=b_0 + b_1cov1 + b_2cov2 +... etc

In R I would fit the full factorial as lm(score ~ cov1 + cov2 + between1*between2) where score is the value of time1 and time2 in long format. This gives 6 parameters as expected.

edit: syntax

GLM Time1 Time2 BY between1 between2 WITH cov1 cov2
  /WSFACTOR=time 2 Polynomial 
  /METHOD=SSTYPE(3)
  /PRINT=PARAMETER 
  /CRITERIA=ALPHA(.05)
  /WSDESIGN=time 
  /DESIGN=cov1 cov2 between1 between2 between1*between2.
$\endgroup$
6
  • $\begingroup$ Print the syntax of the GLM command in your question. $\endgroup$
    – ttnphns
    Nov 23, 2021 at 11:34
  • $\begingroup$ added the syntax now, the design looks like how I would make it in R $\endgroup$ Nov 23, 2021 at 11:45
  • $\begingroup$ When you use SPSS General Linear Model - Repeated Mesures menu or use the syntax it generates, it produces and runs the syntax in the form designed for the "wide format" of the data. Under this approach, SPSS always adds interactions of the RM-factors with the other effects, as if implied interactions. If you want a model without these interactions, you have to restructure the data into the "long format" and use the corresponding syntax (like it is a split-plot design). That is what R does and SPSS also does. "Long format". $\endgroup$
    – ttnphns
    Nov 23, 2021 at 12:13
  • $\begingroup$ so because /WSDESIGN=time and time has two levels it calculates parameters for cov1 cov2 between1 between2 between1*between2 for the first level of time and the second level of time? What would the y= b_0 + b_1*cov1+... interpretation be? $\endgroup$ Nov 23, 2021 at 12:22
  • $\begingroup$ If you run the analysis by the syntax you show, you will see in the table Tests of Within-Subjects Effects not only the main effect of Time, but also its interactions with every effect specified on the DESIGN subcommand (which is for between-subject predictors). You cannot omit any of these added interactions if you are using the "wide" format; use more general/flexible "long" format. (You can Restructure wide into long by Restructure - Variables into cases menu in Data menu. Then go to do your GLM RM analysis in GLM Univariate, not GLM Repeated Measures). $\endgroup$
    – ttnphns
    Nov 23, 2021 at 12:50

0

Your Answer

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