Thursday, 2 December 2010

What is WPA?

 We love cricket!

And like cricket fans everywhere, one thing we love doing is talking about the game, analysing each player's performance, establishing who played well and who played badly, debating over which player was most influential in winning a game, and who was most culpable for letting a chance of victory slowly trickle away....

And we're rarely all in complete agreement.

For example, when England won the second Ashes Test at Lord's in 2009, which player really had the greatest contribution to the final result?  Was it the official man of the match Andrew Flintoff with his 5-92 bowling figures in the second innings?  Or did Andrew Strauss put England in the commanding position with his whopping great 161 off the bat?  Or maybe Australia's Marcus North contributed most to England's victory, by neither taking any wickets nor scoring many runs?

WPA is a new stat that might just help resolve these differences of opinion.

WPA stands for "Win Probability Added".  For each player in a Test match, we aim to calculate how much he increased the chances of his team winning.  Or how much he contributed to his team losing.  Or how much he raised the probability that the match would end in a draw.

So how does it work?

Well, imagine you're standing at the crease in the first Ashes Test.  It's a sweltering day in Brisbane and Peter Siddle is ready to bowl to you.  Before he bowls, there is a certain probability that your team can win this game; let's call that probability P1.  Now he bowls - you knock it away and take a quick single.  The probability that your team will go on to win has now changed to P2.

The difference P2-P1 is a tiny amount.  But, if we add up all these tiny differences for every ball of the Test match, then we can state exactly how much each player contributed to the match.  And that contribution to the win is WPA.

We were inspired by a similar statistic that is used in baseball to describe how each player on the team contributes towards a win or a loss.  It's worth taking a look at the Fangraphs website to see their WPA statistic applied to every US Major League Baseball game.


  1. I'm loving the stats but what do you base the win, lose and draw probabilities on?

  2. As a statistics nerd I would love to hear more about your precise methodology....are you using some kind of regression model to estimate probabilities? Or is it kernel estimation?