Not applicable
Posts: 1

Understanding Fisher's Exact Test Results

Hi,

I ran an exact test on a data set and got this:

 Statistics for Table of Testing by Years Statistic DF Value Prob Chi-Square 16 11.5326 0.7755 Likelihood Ratio Chi-Square 16 11.7345 0.762 Mantel-Haenszel Chi-Square 1 0.4955 0.4815 Phi Coefficient 0.2188 Contingency Coefficient 0.2137 Cramer's V 0.1547 WARNING: 59% of the cells have expected counts less          than 5. Chi-Square may not be a valid test. Fisher's Exact Test Table Probability (P) 2.86E-10 Pr <= P . Sample Size = 241

Does anyone know what the '.' for the Pr <= P mean? I can't find any information about this result on-line.

Thanks!

Super User
Posts: 23,754

Re: Understanding Fisher's Exact Test Results

What does the log (not output) say?

Posts: 5,535

Re: Understanding Fisher's Exact Test Results

If you specified option MAXTIME= in your EXACT statement, then the missing probability means that the exact probability computation didn't have time to complete within the limit that you set. - PG

PG
Posts: 2,655

Re: Understanding Fisher's Exact Test Results

Continuing down this road, if you did NOT specify MAXTIME=, then you are likely in the problem area outlined in the documentation under Computational Resources (see "Also for a fixed sample size, time and memory requirements increase as the marginal row and column totals beome more homogeneous.")  Looking at the asymptotic tests, the various chi-squared values are all less than the degrees of freedom, which is an indicator of this kind of homogeneity.  The documentation recommends using Monte Carlo estimation in these situations.

However, there isn't any real indication of differences, even though the chi-squared probability values may be misleading due to the number of low-count cells.  The simple ratio of chi-squared/df < 1 is a strong indicator of no difference.

Steve Denham

Discussion stats
• 3 replies
• 718 views
• 0 likes
• 4 in conversation