You should always avoid using the CONTRAST or ESTIMATE statement when simpler statements like LSMEANS, LSMESTIMATE, or SLICE can be used. In this case, you can drop the ESTIMATE statements and replace them with a single LSMEANS statement: lsmeans age1 / diff cl exp; The Exponentiated columns in the LSMEANS table are the estimated event probabilities and in the Differences table they give the relative risk estimates.  ... View more

I don't think you need to create dummy variables - that is what the CLASS statement is for. So, it seems like your code could be simplified and might work if your CLASS is like CLASS subjid week trtpn SOO UOR; and your MODEL statement is like MODEL LOG_VALUE_N = trtpn week trtpn*week BALSFRS ... ; and then use your LSMEANS statement  lsmeans trtpn*week; ... View more

See the "Using PROC NLMIXED" sections of this note which show how you can fit a zero-inflated model using PROC NLMIXED. You could include a RANDOM statement to add a random effect for your matched pairs.   But the zero-inflated distributions, Poisson or negative binomial, are for a count response. For a continuous response you might consider using a modified gamma or Tweedie distribution, again fit using NLMIXED to which a RANDOM statement could be added. The modified gamma is shown in this note.  ... View more

This, again, is addressed in the note that I referred you to in your last post about this log binomial model. That note refers to fitting problems that are quite likely with this model due to the log link not ensuring that predicted values are restricted to the [0,1] range and links to this note that mentions the error you are seeing. The more complex the model, the more likely these fitting problems will be. So, I again suggest that you try one of the other modeling approaches discussed in that note for estimating relative risks. ... View more

One way to do this is with a model-based approach using PROC SURVEYLOGISTIC that includes sex and age along with year in the model on hypertension. Include sex, age, and year in a CLASS statement using the PARAM=GLM option. Also include year in an LSMEANS statement with the ILINK option to estimate the adjusted hypertension proportions for each year. If comparisons among the years are desired, then you can add the NLMeans macro. This can be done in the same way as is shown with PROC LOGISTIC in this note. ... View more

The Tweedie model also allows for heteroscedasticity - as with gamma, the variance is a function of the mean. It is a compound model combining Poisson and gamma - see the "Details: Tweedie Distribution for Generalized Linear Models" section of the GENMOD documentation. Also, the compound model fit by PROC HPCDM *is* a two-part model for frequency of loss and severity of loss - the frequency model can be Poisson, negative binomial, or zero-inflated model, and the loss model is automatically selected as the best fitting among many continuous distributions including gamma and others. See the Getting Started example in the PROC HPCDM documentation.   The paper you refer to doesn't seem to suggest a specific model. It suggests first fitting a logit or probit model to the binary response of zero or positive cost, and a GLM based on a chosen distribution on the positive cost responses. The logit or probit binary response, first part model can be fit in PROC LOGISTIC or PROC GENMOD. The GLM second part, continuous response model can be fit using PROC GENMOD. PROC SEVERITY in SAS/ETS can automatically select the best distribution (as is done with HPCDM). The ASSESS statement in GENMOD allows for testing the adequacy of the link function and the specified form of the model.   For count responses, it mentions the two-part hurdle model which can be fit with PROC FMM (see this note). A similar two-part count model is the zero-inflated model which can also be fit in FMM (see "Getting Started: Modeling Zero-Inflation" in the FMM documentation) and in PROC GENMOD and PROC GAMPL. See the "Details: Zero-inflated models" section in the GENMOD documentation. FMM could be used in a similar way, as shown for the hurdle and zero-inflated models, to fit a zero-inflated gamma model. ... View more

The section of the note that I referred to shows an example of obtaining and interpreting the results. ... View more

Using Poisson regression you can still estimate differences in counts, or if an offset is included. Just use the NLMeans macro after fitting the model. See this note. ... View more

Yes. Since the delta method is used to obtain the standard error, it is assumed that the statistic is asymptotically normal. ... View more

See this note on modeling continuous response data with many zeros. The Tweedie distribution is commonly used since it can accommodate positive data with many zeros. Also, as mentioned in the note, PROC HPCDM in SAS/ETS fits a compound model to loss data consisting of both probability and amount of loss. ... View more

You just need to use the LSMEANS statement with the DIFF and EXP options as shown in the note I referred to.  To use the LSMEANS statement, you need to specify any predictors that you want relative risk estimates for in the CLASS statement. In the CLASS statement, you need to specify PARAM=GLM, not PARAM=REF. You do not need the ESTIMATE statement unless you want a relative risk estimate for a 1- (or other sized) unit increase in a continuous (non-CLASS) predictor. The example in the note uses the LSMEANS statement for the CLASS predictor, A, and the ESTIMATE statement for the continuous predictor, X. You say you have predictors using a 5-point scale, so for each one you need to have a single variable with 5 possible values and you need to include it in the CLASS statement.    The following example uses the neuralgia data in the example titled "Logistic Modeling with Categorical Predictors" in the LOGISTIC documentation. The RANK procedure is used just to change the two continuous predictors into 3-level categorical predictors. The Exponentiated columns in the LSMEANS tables show the estimated probabilities at each level of each predictor. The Exponentiated columns in the Differences of LSMEANS tables show the relative risk estimates comparing each pair of levels as indicated in the very first two columns in each table.  proc rank data=neuralgia out=tmp groups=3; var duration age; run; proc genmod; class Treatment Sex Age Duration / param=glm; model Pain = Treatment Sex Age Duration / dist=bin link=log; lsmeans Treatment Sex Age Duration / diff exp cl; run; ... View more

See this note on estimating the relative risk, particularly the section on the log binomial model. As shown there, the variable specified in the ESTIMATE statement should be the predictor that you want to compute the relative risk for, not the response variable. If your predictors are binary, then it is easier to specify them in the CLASS statement which will then let you use the LSMEANS statement instead of the more difficult ESTIMATE statement. Or, you might want to avoid the problems with the log binomial model and the need to create the proper ESTIMATE statement by using the NLMeans macro approach that is shown in the earlier part of the note.  ... View more