topic Re: PROC FREQ and non-integer frequencies in Statistical Procedures

PROC FREQ and non-integer frequencies

AnalytX — Tue, 30 Apr 2013 14:27:58 GMT

Hi,

I would like to perform Chi-square and Fisher exact test on a simple contingency table. If I use for instance the following code:

data yesno;

input gender $ numyes total;

response="Yes"; count=numyes; output;

response="No "; count=total-numyes; output;

datalines;

Men 0 100

Women 45.2 100

;

proc freq data=yesno;

weight count;

table gender * response / chisq fisher;

run;

I obtain the following "WARNING: Fisher's exact test is not computed because of non-integer frequencies."

Do you have an idea how to proceed?

Thanking you in advance for your answer.

Best,

Re: PROC FREQ and non-integer frequencies

Reeza — Tue, 30 Apr 2013 14:57:21 GMT

Well, the log is correct, you can't have a decimal in contingency tables, because they're counts.

Can you explain some more about what you're trying to do?

It looks like you have 2 percentages and want to compare them, so you've put them over 100. You could put it out of 1000 to remove the decimal but that can change the results.

Re: PROC FREQ and non-integer frequencies

AnalytX — Tue, 30 Apr 2013 15:02:10 GMT

Hi Reeza, thanks for your prompt reply. To be more precise, I try to compare two (or more) percentages. A more concrete case. I have the following percentages to compare: 7.5, 6.2, 5.3 and 9.9.

Re: PROC FREQ and non-integer frequencies

Reeza — Tue, 30 Apr 2013 15:08:36 GMT

For a true comparison you need the true denominator, not 100.

Re: PROC FREQ and non-integer frequencies

Ksharp — Tue, 30 Apr 2013 15:21:05 GMT

Did you mean

weight total ;

Ksharp

Re: PROC FREQ and non-integer frequencies

AnalytX — Fri, 03 May 2013 12:20:22 GMT

Thanks a lot for your answer!

@Ksharp: it is well "weight count";