True Performance Models

This page describes how I create my “True” models: True Economy, True Strike Rate, True Wickets per 24

The purpose of the models is to re-create standard measures of player performance but also account for the level of difficulty given the match situation. The overall approach is extremely similar to my win probability and expected run models (see here). The differences are what we are trying to predict and it doesn’t consider as many factors

Below is a list of the factors that I considered and why I decided to include / exclude each of them...


Ball number: Obvious. Teams expect to score more runs during the Powerplay than the middle overs and the run rate gradually ramps up again towards the end of an innings

Required run rate: Batsmen chasing a low total will take fewer risks. If this factor is not included, then the models would penalise batsmen who adjust their batting down to the required rate. The converse applies to a lesser extent for bowlers

Powerplay flag: This is more of a technical issue. Without this the statistical approach used (LOESS) struggles to produce a sharp distinction between Over 6 and Over 7


Wickets lost: This was a tough one. The number of wickets lost significantly affects the expected run rate. However, from a batting perspective, wickets lost is highly correlated with batting order. For the sake of simplicity in the underlying model, I prefer to treat all batting positions the same and adjust for this as a separate issue. From a bowling perspective, it should average out in the long run

Venue: Some venues are easier to score at than others. As with Runs Added my preference is to keep the basic “True” models relatively simple at first. Venue can be adjusted for later

Competition: Same as venue. The preference is to adjust for this as a separate step rather than include it as an integral part of the model. An ultimate player rating system might be based on Expected IPL True Economy. Data from other competitions would be adjusted according to the relative difficulty