I’m trying to do data analysis with GEE and/or Mixed models but get stuck in choosing the right SPSS syntax.

I have data from subjects (from two distinct groups) who were tested at two time points. Each assessment consisted of a number of trials (different number of trials for each session and subject). For each trial, we measured both scale variables (response latency) and binary variables (success/fail). I want to know if the outcome for these variables differs between the two groups and for each group if it differs with time. For a similar data set of an earlier experiment, a statistician set up statistical models for me: GEE models for the binary variables and mixed models for the response latencies (the reason to use these types of models was mainly to deal with missing data). He did this in SAS with the following syntax:

PROC GENMOD DATA=SUMMARY2;

CLASS Id Group Time;

MODEL Nsuccess/Ntrials = Time Group Time*Group/DIST=BINOMIAL LINK=LOGIT TYPE3;

REPEATED SUBJECT=Id/ WITHINSUBJECT=Time TYPE=EXCH;

LSMEANS Group/ DIFF ODDSRATIO CL;

LSMEANS Time*Group / DIFF ODDSRATIO CL;

RUN;

In the SUMMARY2 data set, the binary outcome variables are represented by an nSuccess variable (the number of success trials in that session of that subject) and an nTrials variable (the total number of trials in that session of that subject), with columns for Id (subject) and Time (i.e., in long data format, with one case for each combination of Id and Time).

The scale variables were modeled with PROC MIXED on the original data set TRIALDATA (also long format) with one row per trial and columns representing Id and Time, and response latency in seconds.

PROC MIXED DATA=TRIALDATA EMPIRICAL;

CLASS Id Group Time;

MODEL RespLat = Group Time Group*Time/SOLUTION DDFM=SAT CL RESIDUAL;

RANDOM INT / SUBJECT=Id;

LSMEANS Group/DIFF CL;

LSMEANS Group*Time/DIFF CL;

RUN;

Now, the statistician has taken a job elsewhere and is not available; and I don’t have SAS . So I wanted to do the statistics in SPSS. After reading SAS documentation and several statistics tests on mixed models and GEE, I have the feeling that I understand the SAS syntax, but given the enormous flexibility of mixed models and reading that they can easily be misspecified, I’m afraid of making mistakes in converting the models to SPSS syntax. I have the following questions:

1. Is SPSS MIXED comparable to SAS PROC MIXED? Or should I use GENLINMIXED? (I’m confused with regard to the different procedures in SAS (MIXED, GLIMMIX) and SPSS (MIXED, GENLINMIXED, GLMM)

2. I’ve heard people say that SPSS can’t do (or isn’t good at) mixed models. Is this true? If so, why? Are the MIXED and GENLINMIXED procedures no good?

3. I have implemented the GEE for the binary data in SPSS syntax. If GEE is good enough for the binary data (after all, I want to control for the repeated measures effect, but I’m interested in the outcomes at group level, so a marginal model should be good enough?), can I do the scale variables with GEE as well and leave the mixed models alone? Do they handle missing data equally well?

In addition, there are some differences between my earlier experiment and this one, and I suspect that this matters for the model but I’m not entirely sure. For example, in the earlier experiment, I had three Time points at which the subjects were assessed, and these Time points had actual meaningful values (the age of the subjects at the time; they were all assessed at the same 3 ages). In the current experiment, Time is a categorical variable of two time points, and to complicate matters, I should correct for the effect of Age of the subjects. This is also a reason why I would appreciate a second opinion.