# What happens when Pearson's chi square and fisher's exact give discordant results ?

##### New Member
Hello everyone;
Just a novice here learning statistics, more into medical sciences. So I have a data set of 176 patients and need to examine the significance of and intervention which has two discrete outcomes. So I run the SPSS, and get this result where Pearson chi-square is 0.049, continuity correction is 0.075 and fisher's exact test is 0.056 (2-sided) & 0.038 (1 sided). And no cells have expected counts less than 5. So do I take the Pearson chi-square and report the test results as significant or do I take the continuity correction and Fischer's exact test and report as not significant. From a biomedical point if view what would be the ideal p value to be reported here - ?Pearson chi-square ?continuity correction ?fisher's exact test Pardon if the query was too childish.

#### spunky

##### Doesn't actually exist
+1 vote for Fisher's exact.

Whether it's one-sided or two-sided depends on your research hypothesis. But I do realize people tend to frown upon one-sided tests to, on the grounds of pragmatism, 2-sided Fisher's exact test.

##### New Member
+1 vote for Fisher's exact.

Whether it's one-sided or two-sided depends on your research hypothesis. But I do realize people tend to frown upon one-sided tests to, on the grounds of pragmatism, 2-sided Fisher's exact test.
Why prefer fishers over Pearson chi-square, is there a specific reason for that?

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Regardless, they are two different calculations that can converge as sample and counts increase. However, if you go to use Fisher's and it will run on your machine even if counts are big, you should always prefer to use Fisher's over Chi-Sq. Fisher's is the exact calculation and Chi-sq is an approximation. So ignore the values of the results and your hypothesis and use Fisher's if it will run given your quantitative resources.

A side bar, people will also recommend to use unequal variance in ttests everytime as a cautionary. So regardless of if you think variances are equal or is some test fails to show the variances are different, you should use the unequal variance version of ttests over the equal variances option.