you could do this by fitting the model in PROC NLMIXED and using the PREDICT statement.  see this note. ... View more

Use the TYPE3 option in the MODEL statement. ... View more

This falls in the general area of causal modeling of observational data.  SAS tools (macros) for this have been developed by the Rand Corp.  See their TWANG project page.  ... View more

Given a data set with variables containing the sensitivity and specificity values (such as the OUTROC= data set from PROC LOGISTIC), these statements effectively reproduce the ROC graph produced by PROC LOGISTIC.  See the options in PROC SGPLOT if you want to makes alterations.   proc sgplot data=rocdata noautolegend aspect=1; xaxis values=(0 to 1 by 0.25) grid offsetmin=.05 offsetmax=.05; yaxis values=(0 to 1 by 0.25) grid offsetmin=.05 offsetmax=.05; lineparm x=0 y=0 slope=1 / transparency=.7; series x=_1mspec_ y=_sensit_; run; ... View more

You can get the estimated marginal effects and their standard errors by fitting the model in PROC NLMIXED and using the PREDICT statement as shown in this note on marginal effects.  For example, these statements use QLIM and NLMIXED to fit the same probit model to the cancer remission data shown in the first example in the LOGISTIC documentation.  The OUTNL output data set from PROC NLMIXED contains the estimated marginal effects (Pred) and the standard errors (StdErrPred).   proc qlim data=remiss; model remiss=blast smear / discrete; output out=outqlim marginal; run; proc nlmixed data=remiss; lp=int+bblast*blast+bsmear*smear; p=probnorm(lp); model remiss ~ binary(p); predict pdf('NORMAL',lp)*bblast out=outnl; run; ... View more

The Robins method is one of the weighted GEE methods in PROC GEE.  See "Weighted Generalized Estimating Equations under the MAR Assumption : Observation-Specific Weighted GEE Method" in the Details section of the GEE documentation.   ... View more

The two procedures use entirely different estimation methods.  Conditional logistic regression in PROC LOGISTIC maximizes a conditional likelihood, while PROC GENMOD uses the Generalized Estimating Equations (GEE) method which is not a likelihood-based method. These methods, and others, are compared in the book "Logistic Regression Using SAS: Theory and Application, Second Edition," (Allison, P., SAS Institute, 2012).  ... View more

If you are asking how to fit a log-linear model to a gamma-distributed response variable, you can do that in PROC GLIMMIX or PROC GENMOD.  For example, if Y is a gamma-distributed response, the following statements fit the model (with X1 and X2 as predictors) log(mean) = intercept + b1*x1 + b2*x2 : proc genmod; model y = x1 x2 / dist=gamma link=log; run; ... View more

Actually, the canonical link is the reciprocal and this is, in fact, the default link in PROC GENMOD when you specify the DIST=GAMMA option without the LINK= option.  But the log link is the one most commonly used.  The canonical link is the link function that naturally occurs when the distribution is written in the general exponential family form that is shown in the "Details: Generalized Linear Models Theory" section of the GENMOD or GLIMMIX documentation.  More explanation can be found in a reference text like McCullagh and Nelder's Generalized Linear Models book. ... View more

This note on zero-inflated models includes showing how to fit such models using PROC NLMIXED.  If you fit the model in that procedure, you could include a RANDOM statement to deal with longitudinal data. ... View more