Ethics and morality May 5, 2010

# Lies, damned lies, and Duckworth Lewis

Was Paul Collingwood right to question the target set for West Indies by the Duckworth Lewis Method

Was Paul Collingwood right to question the target set for West Indies by the Duckworth Lewis Method? I believe he was right, and it was disappointing that Frank Duckworth dismissed his concerns so readily. Statisticians, and I work with many, have their preferred methods but the best statisticians will always accept that each method has its flaws. Show me a statistician whose first answer to a statistical question is 'it depends . . .' and I'll show you a statistician with wisdom.

Here's why Collingwood has three arguments in his favour:

1 The statistical argument: The higher number of data points available, the more reliable any statistical estimate. With only 14 balls bowled by England before the rain set in, the sample was too small to reliably estimate the trajectory of the West Indian innings. Duckworth's argument that those fourteen balls dictated the target exposes the unsuitability of the Duckworth Lewis Method when only a small proportion of an innings, around 10% in this case, has been completed. In this circumstance, there are insufficient data points (balls bowled) to reasonably predict the trajectory of an innings.

A fairer approach would be to set a minimum number of overs before wickets lost are taken into consideration. For example, if that minimum number of overs in T20 were 5 overs (ie 25% of the innings completed), the wickets West Indies had lost in those 14 balls would be irrelevant. The target set would assume that no balls had been bowled. The West Indian target should then have been higher. After 5 overs, and only after 5 overs, the runs already scored and wickets already lost would be taken into account. A more suitable minimum number of overs might in fact be 8 or 10 overs.

The alternative would be to increase the weighting in favour of the side batting first in these circumstances.

2 The spirit of cricket argument: Anybody who has played cricket at any level knows that it is much easier to score around ten runs an over for 6 overs than it is for 20 overs, especially when 10 wickets are available in both situations. It is wrong for Duckworth to suggest otherwise. When the target set by the Duckworth Lewis Method feels wrong and against the spirit of the game, then the numbers thrown up by a stastical analytics package are irrelevant.

3 The regulatory argument: The pressure for a resolution by the Duckworth Lewis Method is created by limits on playing times. In major T20 tournaments, often staged at facilities with floodlights, these regulations seem absurd. Half an hour's extra play is clearly insufficient. The option of an extra hour, at the very least, is essential. I'm confident most specatators would prefer to stay an extra hour or so to watch a proper contest than witness a farce that allows them to get home earlier.

I believe the ICC needs to act and reconsider the application of the Duckworth Lewis Method in T20 cricket. What happened to England earlier this week was unfair for several reasons. Collingwood had a point, Duckworth had a Vera moment.

Note: I edited this article on 6th May to clarify the issue of what is currently considered when a revised target is set. Duckworth Lewis takes into account the overs lost and the wickets lost at the time of interruption and not the runs scored--that's a whole other issue of potential unfairness (thanks to Cricinfo's S Rajesh for clarifying this point).