The WALD option in the MODEL statement causes the tests produced by the TYPE3 option to be Wald tests instead of the default score tests.  The WALD option has no effect on the tests in the GEE parameter estimates table which are always Z statistics, which are just the square root of the Wald statistic and has the same p-value.  The table that gives the Pearson and deviance statistics does not apply to GEE models since it is not a likelihood-based method.  The table is printed if you specify the PRINTMLE option in the REPEATED statement, but those statistics apply only to an initial model that is fitted by maximum likelihood to provide starting values for the GEE estimation method.  The GEE method always includes a dispersion parameter that is used in the computation of the working correlation matrix.  See "Generalized Estimating Equations" in the Details section of the GENMOD documentation. ... View more

I think you could get the standard error you want by fitting the model using PROC NLMIXED instead of PROC GLIMMIX. Use the PREDICT statement to estimate the desired statistic and obtain its standard error. ... View more

Here is an example. You may be getting the reciprocal of the odds ratio. To get the right value you typically need to use the EVENT= option in the MODEL statement and the PARAM=REF and REF= options in the CLASS statement to control exactly which response level you are modeling and the predictor level that is in the numerator of the odds ratio. Note that by default, FREQ gives you the odds of the left column value and uses the odds in the first row as the numerator odds. The LOGISTIC options below replicate that. data FatComp; input Exposure Response Count; datalines; 0 0 6 0 1 2 1 0 4 1 1 11 ; proc freq data=FatComp; weight count; tables Exposure*Response / relrisk; run; proc logistic data=FatComp; freq count; class exposure(ref="1")/param=ref; model response(event="0")=exposure; run; ... View more

A fixed effects model can be fit in PROC LOGISTIC using the STRATA statement. For example: proc logistic; strata year; model y=v1 v2 v3; run; This fits the conditional logistic model that avoids estimating parameters for the years. However, a probit model cannot be fit when the STRATA statement is used and there is no STRATA statement in PROC PROBIT. ... View more

You could estimate nonlinear (or linear) combinations of parameters using the PREDICT statement in PROC NLMIXED if you fit the model in that procedure. ... View more

The table to be analyzed here is a 5x5 table (method1 x method2) and each of the 68 plots falls into one of the 25 cells. As stated, it sounds like you want a test of marginal homogeneity. That is, whether the 5 row proportions (marginal proportions for one method) are the same as the 5 column proportions (marginals for the other method. You can test for marginal homogeneity as shown in the example titled "Repeated Measures, 4 Response Levels, 1 Population" in the PROC CATMOD documentation. You will need to create a data set with one observation per plot and with METHOD1 and METHOD2 variables containing the plots' classes under each method. Then METHOD1 and METHOD2 are used like the RIGHT and LEFT variables in the example. See this usage note for additional interpretation: http://support.sas.com/kb/39243 ... View more

As mentioned in this usage note: http://support.sas.com/kb/22596 the CTABLE results are based on bias-adjusted predicted probabilities not on the simple, unadjusted probabilities. ... View more

If you use PROC GENMOD with the REPEATED statement, a robust variance estimator is used by default. Generalized Estimating Equations (GEE) estimation is used to fit the model. Note that 20 clusters might be a little small for this method. For a logistic model, use the DIST=BIN option in the MODEL statement. In the REPEATED statement, start with simple structures if you can such as TYPE=IND or TYPE=EXCH. The most general correlation structure, TYPE=UN, is the most difficult to estimate and can result in fitting problems. See the examples in the GENMOD documentation: SAS 9.2: http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/genmod_toc.htm SAS/STAT 9.22 (beginning in SAS 9.2 TS2M3): http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/genmod_toc.htm ... View more

While logistic regression involves a binomial or multinomial response, it is still reasonable to fit a logistic curve to a continuous response that ranges from 0 to 1. This can be done by fitting a nonlinear model such as p=1/(1+exp(-xbeta)), where xbeta is a linear function of your predictors. Also, Collett (2003, pp 329-330) suggests a quasi-likelihood approach that fortunately is similar to using events/trials syntax in PROC LOGISTIC with the proportions as the events variable, with the trials variable equal to 1, and with the SCALE=DEVIANCE option in the MODEL statement. ... View more

As noted in the LOGISTIC documentation: "Asymptotic normality of the estimated area follows from U-statistic theory..." See the "Details: Receiver Operating Characteristic Curves: ROC Computations" section of the LOGISTIC documentation: http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/logistic_toc.htm ... View more

Check out this usage note: http://support.sas.com/kb/22607 ... View more

That is correct -- GENMOD does not estimate/update the correlation matrix when TYPE=FIXED or USER is specified. ... View more

The confidence limits for the predicted probabilities are computed as shown in the "Details: Linear Predictor, Predicted Probability, and Confidence Limits" section of the LOGISTIC documentation: http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/logistic_toc.htm ... View more

You can get the Cox and Snell and the Nagelkerke R-squares by simply adding the RSQUARE option in your MODEL statement. This option can be used with ordinal models. However, these R-squares don't really have the same "variance explained' interpretation as in ordinary least squares regression. They can be used to compare competing models. The ROC curve and the area under it are only defined for the binary response case and so are not available for the ordinal model. ... View more