topic Re: Two-sample proportions test in Statistical Procedures

Two-sample proportions test

SusieQ324 — Thu, 31 Mar 2011 19:14:37 GMT

I have 2 cohorts with counts of an event for each. But I've calculated 3 different relative percentages - % of the overall cohort, % of all those patients who have evidence of the event and % of patients in a similar age group. So the denominator in the % changes for each.

I've done a chi-squared test but I don't think that's entirely what I want. Basically I want to test the 2-sample proportion values for each type of proportion to test whether p(cohort1)=p(cohort2) and I'm just not thinking clearly.

What am I missing??? Or doesn't it matter?

Sue

Re: Two-sample proportions test

Reeza — Thu, 31 Mar 2011 23:53:51 GMT

I think rather than %, calculate freq/counts and then you can use a proc freq to get the difference between cohorts for the different ratios and calculate odds ratio between the two.

Hopefully someone can verify that's correct...

Re: Two-sample proportions test

Ksharp — Mon, 04 Apr 2011 04:13:36 GMT

How about:
[pre]
proc freq data=data;
tables a*b /chisq riskdiff;
run;
[/pre]
riskdiff will give difference in proportion and its limited interval.

Ksharp