Fisher Exact Test Corrections

[e_miller]

Hi all, 

I am running a Fisher Exact Test and wanted to ensure that corrections were disabled. I read this piece of documentation but could not find the default corrections or how to disable them for Fisher (if they are even applied by default). 

Could someone clarify the default corrections for Fisher (or chi-square test) and how to disable them? 

Thanks! 

[Watts · Accepted solution]

Fisher's exact test doesn't include any corrections. 

Documentation is here: Fisher's Exact Test and Exact Statistics.

The CHISQ option in PROC FREQ produces several chi-square tests and statistics.  

For more information, see the documentation here: Chi-Square Tests and Statistics.

[e_miller]

Hi, 

I just wanted to post a followup and ask about Yate's corrections in chi-square tests. I see in this post and this other sas forum post that the continuity correction is included as a different value ('Continuity Adj. Chi-Square'). 

However, when I run my tests in sas 9.4, I see the all of the same chi-square values except for the continuity chi-square value. I wanted to see whether sas9.4 included Yate's correction by default or if I need an additional option to enable it. 

Thanks!

[Watts]

The CHISQ option in the TABLES statement (in PROC FREQ) produces several chi-square tests and measures, which include the continuity-adjusted chi-square for 2x2 tables. You don't need to specify a separate/additional option to request the continuity-adjusted chi-square.

PROC FREQ computes the continuity-adjusted chi-square by using the Yates (1934) correction. For details, see the subsection "Continuity-Adjusted Chi-Square Test" in the section Chi-Square Tests and Statistics. 

This correction (Yates 1934) was developed to improve the chi-square approximation for 2x2 contingency tables in which the sample sizes (cell expected frequencies) are small. Generally speaking, continuity corrections are not heavily used these days for contingency tables because exact tests are now available. 

For 2x2 tables, Fisher's exact test is automatically produced by the CHISQ option. For larger tables, you can request Fisher's exact test (and other exact chi-square tests) by using the EXACT statement.

Hope this helps. 

[e_miller]

That's very helpful, thank you. I read through the documentation link you posted and saw that the correction is more helpful with smaller sample sizes (like mine). In that case, I'm assuming that the correction is automatically enabled. Since I have a small sample size, would you recommend sticking with a Fisher test?

[ballardw]

Show the code you used, or the code generated.

[e_miller]

It's a simple proc freq statement, shown below: 

proc freq data = newSet;
	tables var1*var2 / expected fisher;
run;