Question about Chi Square and Fisher Exact Tests

[bschanding]

Hi,

I'm very much working in stats by necessity, so much of my knowledge is coming somewhat on the fly. I had a question about two non-parametric tests and I wanted to see if I could get some feedback here....

In my field researchers are suggesting the use of Fisher exact test over chi square, claiming mainly this has to do with the fact that sometimes one or more expected frequency cells will be lower than 5. I just wanted to know if Fisher can still be used (or if there's any reason to use it) if expected frequencies will be higher than 5.

Another claim I've read is that Chi square assumes randomness, but Fisher does not. Is that accurate?

I appreciate any ideas you have related to these statistics.

Thanks,
Brian

 

[marshmugger]

Ah ha!

You MUST download and read the following reference (should be able to find a free PDF download with your favourite internet search engine).

The traditional small sample size recommendations are wrong.

The Fisher exact test is designed for 2x2 contingency tables where both sets of marginal totals are fixed (which is a rare experimental setup).

Campbell, I. 2007. Chi-squared and Fisher-Irwin tests of two-by-two tables with small sample recommendations. Statistics in Medicine 26: 3661-3675.

Stephen.