topic Re: new to proc glimmix, is it the right test? in Statistical Procedures

new to proc glimmix, is it the right test?

celesteaurora — Mon, 05 Jun 2023 02:34:17 GMT

Hello,

I am re-running analyses for my thesis and am not sure proc glimmix is the right test to run, and my in house statistician is unavailable to tell me why she chose this method, nor do I have her code. I haven't performed stats in SAS in almost a decade, so forgive my ignorance.

I am trying to create a typical "Table 1" for a research paper, which describes the patient demographics as stratified by primary outcome. For example, in this study, we screen 738 people for social work needs. I want to describe those who screened yes vs those who screened no by patient age, race, marital status, insurance type, and so on.

When I do this with proc freq and chi square analyses, I get an N, % for each group (screened yes vs screened no), and a p value for the table as a whole, but not for each individual outcome. (As in, I get a p value for marital status but not married, divorced, separated, etc).

Do I need values for each individual outcome in order to say, "this variable had a statistically significantly difference among those who screened positive vs negative"? Would the answer be sufficient by performing chi square analyses for categorical variables, and t-test for continuous, or is proc glimmix necessary?

Thanks,

Re: new to proc glimmix, is it the right test?

PaigeMiller — Mon, 05 Jun 2023 09:47:28 GMT

PROC FREQ and Chi-squared testing works only on the N (or equivalently percentages). If that's all you want to compare, then that's the right test.

If you want to compare the means of some response by (screened/not screened) or compare the means of some response by demographics, or any other comparison of means, then PROC GLIMMIX would be needed.

It depends on what you want in this table.

Re: new to proc glimmix, is it the right test?

SteveDenham — Tue, 20 Jun 2023 18:19:37 GMT

Or proc GENMOD. Faster, with less likelihood of incongruous results due to missing cells.

SteveDenham