It can be done a little bit shorter.  For example: bpart = b0+b1*(MIG=1)+b2*(MIG=2)+ ... Using copy and paste to add similar terms, this isn't too bad as long as there aren't a large number of levels to deal with.  If so, the best way to produce a coded set of design variables is (odd as it may seem) in PROC LOGISTIC via the OUTDESIGN= and OUTDESIGNONLY options.  You can then use reference (or other full-rank) coding to just produce k-1 design variables for a predictor with k levels.  And the created variables are named using the variable name and level.  See this note that describes it. ... View more

That message occurs when all of the predicted probabilities from the model are very close together as described in the "Rank Correlation of Observed Responses and Predicted Probabilities" section of the LOGISTIC documentation.  You can save the predicted probabilities to a data set by using the P= option in the OUTPUT statement.  If your model gives all observations very similar predicted probabilities then it clearly is not effective at distinguishing the response levels.  Improving the model may require alteration such as treating numeric, ordinal categorical predictors as continuous or adding other predictors.  If one of your parameter estimates is extremely large, then the computation of predicted probabilities may be dominated by it.  This can result in all predicted probabilities being almost identical, causing this note.  If some parameter estimates (or their standard errors) are very large, then you may need to remove them from the model. ... View more

The score type3 test is a test of the predictor's association with the response.  This is generally what you would use to assess the value of having the predictor in the model.  It is not a test of its effect on the correlation among the measurements.  But adding or removing predictors may effect the estimates of the correlation.  The QIC statistics don't have a measure of variability, so it is not possible to say if adding or removing a predictor has a "significant" effect on QIC.  ... View more

Just a simple application of PROC FREQ: proc freq; tables provider facility storage; run; or, if you just need the number of levels in each variable and not the counts of observations in each level: proc freq nlevels; tables provider facility storage / noprint; run; ... View more

Look at the SAS log to see if there are any errors or warnings.  If so, tell us what they are. ... View more

There is no SAS procedure for survey analysis of count data similar to PROC SURVEYLOGISTIC for binary data.  However, you may be able to use the trick described in this paper to use logistic regression to fit a Poisson model.  It may be possible to use this trick along with PROC SURVEYLOGISTIC to do what you want. ... View more

Rates are typically modeled using Poisson or negative binomial regression.  This can be done in PROC GENMOD.  See the discussion and example in this note. ... View more

You'll need to clarify what you mean by "comparing logits".  But note that if you fit a logistic model in PROC GENMOD using the DIST=BINOMIAL and LINK=LOGIT options, the estimated logits (that is, log odds) are available using the XBETA= option in the OUTPUT statement. ... View more

Yes, assuming your observations (households, I assume) are independent, you could look at the association of each variable (ignoring all other variables) with the clusters by using PROC FREQ.  For example: proc freq; table location*cluster / chisq; run; However, to determine the partial effect of each variable after taking account of the effects of the other variables, you would need to either use stratification or a model-based approach.  The model-based approach would be a nominal (generalized logit) logistic model.  It sounds like you already tried this and it resulted in separation issues, probably due to the data being too sparse to support the model.  A possible alternative is stratification using the CMH option in PROC FREQ.  For example, this tests the effect of location after stratifying on habitat.  You may want the NOPRINT option to avoid printing all the separate location*cluster tables for the various habitat levels. proc freq; table habitat*location*cluster / cmh noprint; run; ... View more

Yes, adding 0 would be an easy way to convert all character variables (whose values are always numbers) to numeric.  Another way uses the INPUT function as illustrated in this sample program: 40700 - How to convert all character variables to numeric and use the same variable names in the output data set ... View more

If your data are counts, then you should consider using a discrete distribution rather than a continuous one.  Count data are typically modeled using Poisson, negative binomial, or zero-inflated versions of those distributions.  Those distributions allow zero values.  Models using these distributions can be fit using PROC GENMOD. ... View more