Mstat Software

Mstat was developed to help students analyze their data using nonparametric statistical methods. This program grew out of a graduate course, "Statistical Problems in Genetics and Molecular Biology," taught by the Department of Oncology at the University of Wisconsin-Madison over the last 30 years.

This site will have the latest version of the Mstat software package for Windows, Macintosh OSX, and Linux, available from the Download link above. Help for using Mstat is also available.

As in the past, the help files include a copy of the notes for the "Statistical Problems …" course mentioned above. These course notes have been made into a book. Both paperback and ebook versions are available, as detailed on the Help page.

Mstat 6.0.2 released March, 2014

A minor update to Mstat fixes a problem when trying to run the program under Windows with some foreign keyboards. Most users will not need to apply this update. Unfortunately, migrating our website to a new server has broken the Help>Check for updates function. If you need the keyboard fix, email me and I will provide directions on installing the update. A new version of Mstat (6.1.1) that resolves the update issue will be available for download soon.

Mstat 6.0.1 released January, 2014

Version 6.0 is a major rewrite of Mstat, based on using the Qt toolkit for the user interface. The most important consequence of this revision is that a 64-bit version of Mstat is now available for OSX. In addition, this change improves integration with the operating system across all three platforms, and eliminates the requirement for Java on OSX and Linux. The most significant functional change is the inclusion of methods to adjust p-values and compute false discovery rates in the case of multiple comparisons. Other additions include the ability to explicitly set plot sizes through the Modify plot dialog and improved error-checking for user input.

Mstat 5.5.7 released April, 2013

I've implemented significant improvements in the way that p-values are computed for the Wilcoxon rank sum (WRS) test. First, a much faster algorithm for computing the exact p-value allows doing so over a much increased range of group sizes, up to a total of 200 observations (100 per group). The first time an exact p-value is computed for two groups of 100 observations each, it will take about 8 seconds to do so. Subsequent calls to this function will only take a few milliseconds. This improved algorithm also allows larger sample sizes in the exact permutation test used for the case of multiple experiments, allowing, for example, two experiments with 25 observations per group. When ties are present, or the sample sizes are larger, an improved method is used to compute the approximate p-value, which is much more accurate than the standard Normal approximation at small p-values. For details, follow the link.

A bug that resulted in the failure to import the last row for some csv files has also been fixed.