topic Re: proc genmod. how to get parameter estimates and 95% CIs for categorical variable in SAS Programming

proc genmod. how to get parameter estimates and 95% CIs for categorical variable

cpeloquin — Mon, 26 Jul 2021 15:54:43 GMT

i have a 3-level categorical character variable DOSE.

values are:

1-5
6-12
13+

i would like to get estimates and 95% confidence intervals for "6-12" and 13+" where "1-5" is the reference group. what would be the syntax of the ESTIMATE statement?

many thanks for your help.

Christine

Re: proc genmod. how to get parameter estimates and 95% CIs for categorical variable

ballardw — Mon, 26 Jul 2021 16:48:42 GMT

Confidence interval has no meaning when discussing character variables. If you mean CI of a different numeric variable for each level of a character variable that is doable. Since reference an Estimate statement then perhaps you can share the current procedure code so we have some chance of seeing what other variables you are using.

Re: proc genmod. how to get parameter estimates and 95% CIs for categorical variable

Ksharp — Tue, 27 Jul 2021 12:42:03 GMT

Calling @StatDave @SteveDenham

Re: proc genmod. how to get parameter estimates and 95% CIs for categorical variable

PaigeMiller — Tue, 27 Jul 2021 14:27:14 GMT

@cpeloquin wrote:
i would like to get estimates and 95% confidence intervals for "6-12" and 13+" where "1-5" is the reference group. what would be the syntax of the ESTIMATE statement?

Assuming you really mean that you want 95% confidence intervals for the estimated model coefficients of "6-12" and 13+", these are given by default from PROC GENMOD, as can be seen in this example:

https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.4/statug/statug_genmod_examples04.htm

If you want to compare these statistically, and get confidence intervals for the difference, you can get this from the LSMEANS statement with the CL option.

If that's not what you mean, then please explain further.

Re: proc genmod. how to get parameter estimates and 95% CIs for categorical variable

StatDave — Tue, 27 Jul 2021 14:32:44 GMT

This should be posted in the Statistical Procedures Community rather than here. However, regardless of the distribution of your actual response variable, you can get the mean of it at each level of your Dose variable by specifying it the LSMEANS statement. Do not use the ESTIMATE statement. Make sure that Dose appears in the CLASS statement. Add the ILINK and CL options which will add columns in the LSMEANS table labeled "Mean" which has the mean estimate and confidence limits. For example:

lsmeans dose / ilink cl;