Comparing Completion Rates, Conversion Rates, and A/B Testing 79 Using the same example from before, we get the Yates chi-square test statistic of 22 2 j11 × 5 - 1 × 5j - × 22 2 2 = yates 12 × 10 × 16 × 6 2 = 2:905 yates Looking up this value in a chi-square table or using the Excel function =CHIDIST(2.905, 1) we get the p-value of 0.0883. Using 0.05 as our threshold for significance, we would conclude there is not a statistically significant difference between designs using the Yates correction (although, as with the Fisher exact test, this outcome would probably draw our attention to the possibility of a significant difference). For this example, the p-value for the Yates correction is higher than the Fisher exact test, which is a typical result. In general, the Yates correction tends to generate p-values higher than the Fisher exact test and is therefore even more conservative, overstating the true long-term probability of a difference. For this reason and because most software programs can easily calculate the Fisher exact test, we do not recommend the use of the chi-square test with the Yates correction. N - 1 Chi-square Test