df in NLEstimate macro for risk difference

jean_kiser — Thu, 14 Sep 2023 12:21:25 GMT

The SAS note https://support.sas.com/kb/37/228.html#note1 indicates that you can use the NLEstimate macro to get the risk difference. The example dataset has 100 records. In the example code, df=100. However, the macro indicates that it would typically be the number of observations minus the number of parameters estimated. So wouldn't that be 97 (100-3)?

 proc logistic data=test;
         class a / param=glm;
         model y(event="1") = a;
         lsmeans a / ilink;
         store logfit;
         run;
      data fd; 
         length label f $32767; 
         infile datalines delimiter=',';
         input label f; 
         datalines;
      Prob Diff A1-A2,logistic(B_p1+b_p2) - logistic(B_p1+b_p3)
      Prob Diff A1-A3,logistic(B_p1+b_p2) - logistic(B_p1)
      Prob Diff A2-A3,logistic(B_p1+b_p3) - logistic(B_p1)
      ;
      %NLEstimate(instore=logfit, fdata=fd, df=100)

df=, /* Specifies the degrees of freedom to be used in the */

/* test and confidence interval computed for the */

/* estimated function(s). If omitted, large-sample */

/* Wald statistics are given. The degrees of freedom */

/* for testing a linear combination of parameters in */

/* a linear model would typically be the number of */

/* observations used in fitting the model minus the */

/* number of parameters estimated in the model – */

/* essentially, the error degrees of freedom. */

Re: df in NLEstimate macro for risk difference

StatDave — Thu, 14 Sep 2023 14:14:21 GMT

It would be. The value specified in df= is used, as is, for the test. The documentation of df= is suggesting that if you specify df= that you typically will want to specify N-p, where N is the sample size, and p is the number of estimated parameters.

topic Re: df in NLEstimate macro for risk difference in Statistical Procedures

df in NLEstimate macro for risk difference

Re: df in NLEstimate macro for risk difference