topic Re: Cochran-Armitage Trend test interpretation in Statistical Procedures

Cochran-Armitage Trend test interpretation

dcortell — Tue, 04 Sep 2018 12:51:38 GMT

Hi All

By applying the following code, I receive the following results.

I'm quite surprised by the significance of the Exact Rend test results, as the positive effect (opportunity-new) row percentage seems

pretty alternating overtime, and I would expect a no presence of any trend effect.

My question is, There could be something ongoing here which I'm missing that could make the test not reliable?

Bests

proc freq data=lead_att (where=(outcome_rev2 is not missing and outcome_rev2 not like '%Existing%') );
tables yy_event*outcome_rev2/cmh chisq nopercent nocol nopct trend ;
exact trend;
run;

Re: Cochran-Armitage Trend test interpretation

Ksharp — Tue, 04 Sep 2018 13:44:19 GMT

I think you could trust result : there is a trend due to p-value <0.05.

But the problem is you have a WARNING: count less than 5. if you could get rid of this warning that would be perfect.

Re: Cochran-Armitage Trend test interpretation

Ksharp — Tue, 04 Sep 2018 13:45:14 GMT

I think you could trust result : there is a trend due to p-value <0.05.

But the problem is you have a WARNING: count less than 5. if you could get rid of this warning that would be perfect.

Re: Cochran-Armitage Trend test interpretation

dcortell — Tue, 04 Sep 2018 14:09:34 GMT

The warning should not be a problem here, as the Exact Cochran Test is intended to manage cases like these were obs<5 are present in cells.

Re: Cochran-Armitage Trend test interpretation

Ksharp — Wed, 05 Sep 2018 12:28:45 GMT

OK. You are right. EXACT test take care of this situation.