EDITED: For binary response models, marginal effects for each observation can be obtained using the MARGINAL option in the OUTPUT statement of PROC QLIM in SAS/ETS. Note that these values are only correct for predictors that are not involved in higher-order effects in the model, like interactions. To get the average marginal effect of a predictor not involved in interactions, simply use PROC MEANS to compute the average of it's marginal effect for the desired response level. However, standard errors are not available from QLIM for the marginal effects, and not for the average marginal effect.   Alternatively, you can use the Margins macro. It can provide estimates of predictive margins, average marginal effects, and marginal effects at the means. Appropriate computations for marginal effects are done even for predictors involved in higher-order effects in the model. Standard errors for the margins and effects are provided and tests and confidence intervals are available as are pairwise comparisons and the ability to estimate and test contrasts. Several examples are provided in the Results tab at the above link.  ... View more

Effect importance is typically assessed using partial correlation and partial R-square. If the target response is binary, see this note. If the target is continuous and you are fitting an ordinary regression model, see this note which shows how to get cumulative R-square. Partial R-square for generalized models can be computed as discussed in this note.  ... View more

You could use it for a binary response like that (though you would then specify DIST=BINOMIAL and LINK=LOGIT), but then you would presumably have to fit three separate models. That could complicate testing for Treatment differences. It's usually better to fit a single model when that is possible as it is here. ... View more

I was referring to this form of the data that you mentioned earlier and which I assume you have:   Moment                 Animal             Treatment          Behaviour 0                            1                      1                        Eat 10'                         1                      1                        Sleep 20'                         1                      1                        Sleep .                              .                       .                        . .                              .                       .                        . .                              .                       .                        . (0 to 180´)            (1 to 20)       (1 to 4)                   (Eat, Sleep or Drink) ... View more

No, I mean you should use the Behaviour variable which has values Eat or Sleep or Drink and to run PROC GEE just once with Behaviour as the response variable as I showed.  The default should be the independence correlation structure, so I don't know why you got that error unless you specified the TYPE= (or CORR=) option in REPEATED with a value other than IND. To be sure, specify TYPE=IND in REPEATED:   repeated subject=animal / type=ind; ... View more

As sld mentioned, the response is multinomial - it is an unordered, categorical response. Further, you have repeated measures per animal over time. An appropriate model can be fit using PROC GEE using the DIST=MULT and LINK=GLOGIT options in the MODEL statement and including a REPEATED statement. It might look like this if you want Treatment as the only predictor. Assuming you want comparisons among the treatments, the LSMEANS statement provides that.      proc gee; class treatment animal; model behavior = treatment / dist=mult link=glogit; repeated subject=animal; lsmeans treatment / diff; run; ... View more

Not sure if this is relevant or helpful to what you are doing, but see this note.  ... View more

You don't need a macro. Simply use the PERSIST option in an ODS OUTPUT statement to accumulate tables across multiple procedure runs.  For example, these statements save two tables across both procedure runs in the two named data sets: ods output FitStatistics(persist)=fs ModelInfo(persist)=mf; proc logistic data=ingots; model r/n=heat; run; proc logistic data=ingots; model r/n=soak; run; ods output close; Find the name of the table you want to keep in the "Details: ODS Table Names" section of the PROC LOGISTIC documentation. You might also want to use the ODDSRATIO statement in PROC LOGISTIC to generate the odds ratios. Again, see the PROC LOGISTIC documentation.   Also see this note for how to easily fit a series of models in separate procedure runs using a different predictor in each model.   However, note that if you run a bunch of separate models, the odds ratios you obtain for a particular predictor are only adjusted for the other predictors in the individual model. Any predictors that you want *adjusted* odds ratios for must have all of the predictors you want to adjust for in the model.  ... View more

If those weights are survey weights, then what you might need is the SURVEYFREQ or SURVEYLOGISTIC procedure. ... View more

Note that you only need to specify a nested effect (or interaction) in the SUBJECT= option if you don't have a single variable that has a unique value for each cluster of repeated measures. See this note. The warnings are general messages indicating that the GEE model fitting algorithm did not properly converge. Changes to the model (usually simplifications) are generally necessary to avoid them. However, if your data are survey data, you should not be using GENMOD anyway since it does not have the proper variance estimators for survey data analysis. ... View more