The PSI formula might look like this:
PSI = Sum((GroupofInterest% – ComparisonGroup%) x LN(GroupofInterest%/ComparisonGroup%))
In the templates provided an example is shown providing 25 variables from thousands of credit card clients that have defaulted. Using the PSI statistic we can quickly see that variable Pay 2 had the largest distributional difference between the <$150k group and the $150k+group. This approach saves time, makes clear what is most important and surfaces the relative importance of all the variables to one another.
A general rule of thumb is a PSI > 0.25 is a significant difference. Anything less than that isn’t significant.