Just not cricket: Microsoft's big data Googly called No Ball
Redmond's Duckworth Lewis Method revamp rejected by the stats wonk who runs it
Microsoft's attempt to re-write one of cricket's oddest rules has been rejected by its statistical guardian.
This story starts with India's mania for cricket and especially one-day and Twenty20 cricket*. When matches in those abbreviated forms of the game are interrupted by weather, a formula called “The Duckworth Lewis Method” (D/L)** adjusts the run-scoring targets required to win a game. Sometimes those targets rise, even though the game is shorter. That outcome seems odd to many, but is is rooted in averages of likely scores rather than weighing probabilities of a match's outcome. It also takes into account that the team batting first doesn't know what it will be chasing and so behaves differently to the team batting second, which can strategise how to chase its target.
Duckworth and Lewis were English statisticians and their Method is very robust, but occasionally turns up oddities that get fans scratching their heads and suggesting adjustments. Among the hoped-for changes is for the Method to use the most recent data, rather than historical records.
Enter Sarvashrestha Paliwal, Microsoft India's Azure business lead who has blogged about throwing machine learning in the Microsoft cloud at the problem. Paliwal explains he's found a way to feed in records of all recent matches including “which teams are playing better this season, ranking of players, etc” in the hope of making D/L calculations more accurate.
Paliwal also points out that The Duckworth Lewis Method is now known as the The Duckworth Lewis Stern Method (D/L/S) because Duckworth and Lewis retired and handed over their work to Professor Steven Stern of the Queensland University of Technology. Who kindly answered the phone when The Register called today and said that while he thinks the technology being thrown at D/L/S is remarkable, he thinks the method is just fine as is for a couple of reasons.
First up, he explained that D/L/S is revised every year. On July 1st a new year's worth of data is applied and he calculates anew using data from “a sliding four-year window because matches are played under sufficiently different conditions that older information is no longer relevant.” Even with that annual data injection, he says annual adjustments don't make much difference to the targets that D/L/S sets.
“Cricket is a very variable game,” he explained. “Short term trends are very noisy.”
His other objection is that taking into account form and players has the potential to distort the game. To illustrate this argument he points out critics of the D/L/S targets set in the New Zealand vs. South Africa semi-final of the 2015 Cricket World Cup. South Africa batted first in that encounter but the team's innings was truncated by rain when superstar batter AB De Villiers, who scores very quickly indeed, was at bat. Stern said he was asked why New Zealand's target to win the match did not reflect De Villiers' likely feats.
His response is that if individuals' performances were taken into account it would create a perverse incentive. For example, had De Villiers not been batting when South Africa saw rain approaching, the team would have had reason to lose a wicket solely in order to get De Villiers to the crease so his presence could impact the D/L/S target New Zealand would have to chase.
Which is not to say Stern is opposed to revisions. He told The Register he got the gig as guardian of the Method after correspondence with Duckworth and Lewis in which he initially disagreed with their reasoning, in a friendly-but-robust academic way. He later earned the “S” in D/L/S by adjusting the method to take into account how to model lower-scoring games. ®
* We're not even going to try to explain this stuff to readers beyond the cricket-loving world, lest your correspondent disappear down a very deep rabbit hole. Wikipedia does a decent job of explaining cricket in all its glory.
** The Duckworth Lewis Method is so well-known in cricketing circles, it even became the name of a band that only sings about cricket. Here's the band's most remarkable work, which tells the tale of Shane Warne announcing himself to the world***.
*** It's impossible to explain Shane Warne to people beyond the cricketing world. Or inside it. Do your own Googling.