Maybe one point to clarify, ordinal variables are not supported in the ASSESS statement in PROC PSMATCH (since many of the balance diagnostics consider mean differences), but they can be used in the other statements. In your example you could use the variable THAanesth in the PSMODEL statement and the binary level indicators you created could be used in the ASSESS statement. However in that case assessing the balance achieved for the variable THAanesth requires judging the balance achieved for all of the binary level indicators. Another option for assessing the balance for a categorical variable is to create an output data set of the matched data and then use PROC FREQ to request a contingency table of the variable and the treatment. That approach is illustrated in Example 1 of this recent SAS Global Forum paper.      ... View more

One small comment, I believe PROC GEE does support Type III tests for generalized logit models using the Wald test statistic, and it supports Type III tests using either the generalized score statistic or the Wald test statistic for ordinal response models. To request the Wald tests you can specify the Wald option in the MODEL statement.  ... View more

@emaguin , to answer your question about how the the correlated dependent variables enter into the estimation of the coefficients, the main source would be in the "generalized Hessian" used in the update step (step 4) of the  GEE fitting algorithm described here in the PROC GENMOD documentation. That expression is something like a "weighted" X'X computation, where components of a design row are "weighted" based on the partial derivatives of the link function and also the inverse working covariance matrix for the response variable.    To your question about why the model converges with some working correlation structures and not others that is hard to say without the data. Based on your description I doubt this is the cause (I mention it here in case someone in the future comes across this thread) but sometimes a model with an exchangeable working correlation structure results in a NPD generalized Hessian due to insufficient variability in the outcome variable within subjects/clusters. This can result in an estimate for the working correlation matrix that is near singular, which then effects the computation of the generalized Hessian. Very often a give away for this type of issue is a some notes in the log about having to ridge the estimate for the working correlation matrix before the error about the NPD generalized Hessian.  ... View more

Hmm, not what I would have expected. If this issue persists, I'd recommend possibly reaching out to SAS Technical Support to see if they can help resolve the issue. For this type of problem I suspect they'd likely ask if you can share the data with them since it is tricky problem to diagnose.    That said I would also second @SteveDenham suggestion of trying to match on precomputed propensity scores as demonstrated in Example 98.8 of the PROC PSMATCH documentation. You can see if PROC LOGISTIC provides any additional information about possible issues with fitting the model or maybe if it helps resolve the issue.    I believe there was a bug  in PROC PSMATCH fixed in SAS/STAT 15.1 where observations that had missing values for variables listed in the ASSESS statement but not the PSMODEL statement were not used to fit the propensity score model. This is grasping at straws, but if there are missing values in any of those four continuous variables, this bug (assuming you're using a version of SAS before it was fixed) could lead to different propensity score models and explain why you see the error message when you only add the variables to the ASSESS statement. Matching on the precomputed propensity score values predicted by PROC LOGISTIC would be one workaround for this issue (if it is the issue).          ... View more

@SteveDenham has the right idea, that error message indicates that there is no overlap in propensity score values for the two treatment conditions so the common support region does not exist.   You might try adding the four variables only to the ASSESS statement and not the PSMODEL statement. You can use the ASSESS statement to assess the balance achieved for any continuous or binary variable in the input data set, not just the variables used in the propensity score model. It might (hopefully) be the case that even if those variables are not included in the propensity score model good balance is still achieved in the matched data.  ... View more

Yes the Spearman rank order correlation is another option. PROC FREQ also supports both of these measures of association. You can use the PLCORR option in the TABLES statement to request the polychoric correlation coefficient and the MEASURES option (as mentioned in @PGStats response) will request the Spearman coefficient.    You can Example 3.8 in the PROC FREQ documentation for an example using the MEASURES option.  ... View more

I think @PGStats suggestion to consider using the measures of association supported by PROC FREQ is a good one.    I'll just add that for the case of two ordinal variables, if you can assume their values are derived from an unobserved bivariate normal distribution you could estimate the polychoric correlation by using PROC CORR.  ... View more