A look at various Runs to Wins Converters
A quick look at some various measures
By Tangotiger
The Tango Distribution allows us to create the true expected winning percentage between any two teams, given only their runs scored and allowed. The problem is that you have to run the program to get the results. Are there runs-to-win converters out there that comes close to estimating the true model that the Tango Distribution represents?
Here is a look at a few of the popular ones: Ben91 method, Pythag from James, Clay's Pythag version, and Patriot/David's cubic version.
The blue line is the Tango win%, and you can treat this as the true actual.
Your best bet is to follow Patriot/David's lead. I used exp = RPG ^ .278 (though they recommend .287) . What you can do is run the Tango Distribution for all sorts of teams, and try to find the best fit. I don't think you will improve the accuracy much, if at all.
The green line is Clay's version, and it comes the closest when looking at the extreme cases.
The red line is Patriot/David, and it is almost as good as Clay's, and the simplest.
The yellow line is Ben91, and while it is decent, it breaks down at the extremes.
The black line is Pythag, and it is an ok version.