topic Re: Stratification Using SAS in Statistical Procedures

Stratification Using SAS

microdire27 — Mon, 20 Jul 2020 18:50:59 GMT

Hello all,

Is it possible not to have a reference group during stratification in SAS?

As shown in the table attached to this question, my professor asked for stratification of the ADHD severity level. I am not sure of how to achieve this with SAS since there must be a comparison group (ref group) and I have only two levels. Please help

Re: Stratification Using SAS

SteveDenham — Mon, 20 Jul 2020 19:10:53 GMT

This looks like output from PROC LOGISTIC, with the response being a function of two factors - duration and severity. For a stratified analysis, you need matched observations, such as a repeated (pre/post) design or a matched design (1:1, m:n for case:control designs). Then the factor that defines the matching is used as the STRATA variable.

Without seeing more of your data, or knowing how it was collected, I can't really say what to do next. But if you can identify how the observations may be matched, then you can proceed. The example here should help:

https://documentation.sas.com/?docsetId=statug&docsetTarget=statug_logistic_examples11.htm&docsetVersion=14.3&locale=en

SteveDenham

Re: Stratification Using SAS

microdire27 — Mon, 20 Jul 2020 20:15:36 GMT

Thank you for the earlier response, Mr. Denham. The data is not a case-control data, it was a cross-sectional survey data. Basically, I want to determine if ADHD severity modifies the relationship between family reading time and sleep duration. Family reading time has 3 levels, as shown on the table, ADHD severity also have two levels. Is it possible to stratify the levels of ADHD severity without using one of the levels as the reference level?

Re: Stratification Using SAS

ballardw — Mon, 20 Jul 2020 20:29:07 GMT

@microdire27 wrote:
Thank you for the earlier response, Mr. Denham. The data is not a case-control data, it was a cross-sectional survey data. Basically, I want to determine if ADHD severity modifies the relationship between family reading time and sleep duration. Family reading time has 3 levels, as shown on the table, ADHD severity also have two levels. Is it possible to stratify the levels of ADHD severity without using one of the levels as the reference level?

What procedure(s) are you expecting to use?

It sounds like Surveyfreq might be appropriate requesting a table of ADHD vs Reading time (you don't provide variable names so can't be much more specific) or sleep duration. A chi-squared test might be appropriate to determine if there is a difference in distribution of reading levels or duration based on the ADHD level.

Re: Stratification Using SAS

microdire27 — Mon, 20 Jul 2020 20:51:49 GMT

Thanks Ballard W.

Surveylogistic was the procedure used.

Reading time was the exposure (with three levels)
Sleep duration was the outcome( binary)
ADHD severity is the third variable (potential effect modifier), it also have two levels.
I hope this additional info makes sense.

Re: Stratification Using SAS

SteveDenham — Tue, 21 Jul 2020 11:44:53 GMT

There have been a couple of threads about testing if a variable is a moderating variable in here lately, and it is something I am learning about as I go. Given that, I don't think you can use a categorical variable in a regression without specifying a reference level (at least for non-full rank parameterization). But with only two levels, it should not make a difference. You could even treat the variable as continuous.

Yet another way to look at this as stratification would be to include the interaction between severity and time spent in the model. Then using the SLICE statement, you could calculate the ORs for the simple effect of time spent at each level of severity. This would "stratify" the analysis in the sense that the ORs would be calculated within the "strata" of severity. If that is acceptable, then good on you, and this all boils down to the definition of strata that SAS uses in its logistic regression procedures being different to what you are looking for.

SteveDenham