11-01-2017 04:02 PM
I tried to obtain a p-value for a 2 x 2 contingency table in which one row sums to zero. I used PROC FREQ along with the exact fisher option. The following warning was generated: "Row or column sum zero. No statistics computed for this table." For a Fisher's Exact test, the p-value should be calculated as 1 here. Can anyone tell me why SAS won't produce results when it can be calculated? (I am able to get results for other contingency tables without zero sum rows, so my code is correct.)
11-01-2017 04:45 PM
0 rows are problematic for sure, especially if the row wouldn't even appear in the data set. It means a p-value cannot be calculated, what makes you think the p-value should be 1?
11-01-2017 06:40 PM
Find a book on statistics that shows how to manually calculate a chi-square test statistic for a 2x2 table.
Do it with some tables that have a row total count of 0.
Then try to find that test statistic in a Chi-square table.