topic Re: t test with surveyfreq in New SAS User

t test with surveyfreq

EddieJackson — Fri, 12 Jul 2019 01:26:36 GMT

Hi all, thanks in advance.

I've been searching for a while and haven't been able to be successful to find a solution.

I'm looking at a bunch of survey data, and running the chi-square tests for categorical variables have been straightfoward with the following code that I'm using:

PROC SURVEYFREQ DATA=xxx;

cluster yyy;

strata zzz;

weight aaa;

VAR Kidney*Gender/row chisq;

RUN;

I have been trouble coming up with the code to run the t-tests similarly for continuous variables however. Could someone give me an example? The proc t-test doesn't seem appropriate for the large weighted survey data - is there a similar way to do this with surveymeans?

Thanks.

Re: t test with surveyfreq

Reeza — Fri, 12 Jul 2019 02:08:12 GMT

Have you checked out the examples in the documentation? Does this seem relevant:
https://documentation.sas.com/?docsetId=statug&docsetVersion=15.1&docsetTarget=statug_surveymeans_examples06.htm&locale=en

In addition, you can likely use PROC SURVEYREG to get the test statistics and it will be the same results.

Re: t test with surveyfreq

EddieJackson — Sat, 13 Jul 2019 03:01:01 GMT

I looked at he documentation provided, and came up with the following code which spit out a t value.

PROC SURVEYMEANS DATA= Analysis.Analysis;
cluster HOSP_NRD;
strata NRD_STRATUM;
weight DISCWT;
VAR Age;
DOMAIN CKD/diffmeans;
RUN;

Do you think this code is correct? Why is it that the output N's are still the unweighted N's rather than weighted N's (my weighted N's should be much higher)

If I do a simple proc t-test the results are similar but again, the N's are unweighted.

proc ttest data=Analysis.Analysis;
weight DISCWT;
class CKD;
var Age;
run;

Re: t test with surveyfreq

Reeza — Sat, 13 Jul 2019 18:44:29 GMT

Are you sure? You have weight statements in both. So I assume when you run without a weight statement you get the same results, which is why you think it's unweighted?

I don't know your data so cannot know if your code is correct, I don't know the topics, variables or what you're trying to test. I would recommend making sure that the example code works on your computer then change your code to align with the example in a manner that makes sense.