Data Analysis Blog

At several neurosciences meetings, Prism users repeatedly told me that their nonparametric test of choice to compare two unpaired groups is the Kolmogorov-Smirnov test, rather than the Mann-Whitney test. We added this test to Prism 6, which now offers two nonparametric tests to compare two unpaired groups.

How should you choose between the Kolmogorov-Smirnov (KS) test and the Mann-Whitney (MW) test? Choose between the two tests based on your experimental goals and the traditions of your particular field.

• The MW test has more power to detect an overall shift in the distribution of values, so it is useful to detect differences between the two medians. 
• The KS test is more sensitive to detecting subtle differences in the shapes of two distributions. 

Note that Prism now offers two different KS tests. Don't mix them up:

• One of the KS tests (since Prism 4) is a normality test, available both in the Column statistics analysis and in nonlinear regression (to test the normality of the residuals). It compares the actual distribution of the data with a theoretical Gaussian distribution.
• The new test added to Prism 6 compares the observed distribution of two groups, and is offered as a choice in the t test dialog, as an alternative to the Mann-Whitney test.