# Fixed bugs in the Mann-Whitney test and Wilcoxon matched pairs test in GraphPad Prism versions 3-5.

**Mann-Whitney test in Prism 5 when data are entered on one row of a Grouped table**

Prism 5 does the Mann-Whitney test just fine if you enter your data on a Column data table. If you enter all the data on a single row of a Grouped table, with side-by-side replicates, Prism 5 will not do the Mann-Whitney test. But instead of telling you that, it reports that the P value is < 0.0001.

**Wilcoxon test in Prism 3 and 4.**

When Prism 3.00 to 4.02 or 4.0b computes the Wilcoxon signed rank test (as part of the Column statisitcs analysis, to test whether the median of a column differs from a theoretical value), it labels the P value "two-tailed". In fact, it is reporting a one-tailed P value. Double that value to get the correct two tail value. Fixed in 4.03 (windows) and 4.0c (Mac)

When Prism 3.00 to 4.03 or 4.0c concludes whether the Wilcoxon signed rank test is statistically significant, it bases the conclusion on the one-tailed P value, rather than the two tail test. So when the two-tail P value is between 0.05 and 0.10, it mistakenly says that the effect is statistically significant. Fixed in 5.00 (Windows) and 5.0a (Mac).

Prism 3 and 4 first calculated the probability of obtaining a sum-of-ranks less then or equal to observed value of W. This is the one-tail P value. It then doubled this value to obtain the two-sided P value, since the distribution of W is symmetrical. The probability that the sum-of-squares will exactly equals W is essentially included twice in the calculation of the two-tail P value, so that two-tail P value was sometimes reported as being greater than 1.0. This will only happen when the P value is very large. In these cases, it really doesn't matter exactly how large it is. Similarly, the one-tail P value could be reported as having a value greater than 0.5. Fixed in 5.00 and 5.0a.

Fixed bugs in these tests in Prism 6

Keywords: bug nonparametric