Duckworth-Lewis vs Jayadevan
The reigning champions of predictive cricket statistics come up against a new challenger.
As a vocal supporter of the Duckworth-Lewis system I was initially sceptical of any system that claimed to be better than it. However looking through this article by Jayadevan he makes a lot of sense. I also notice that one of my favourite staticians , S. Rajesh from Cricinfo is similarly swayed. Here is his article.
I am uneasy with a couple of things that he has raised in the article, and the only thing left to do is test the two systems out.
Now it is almost impossible to do a practical test with players playing matches. But it is possible to do a historical statistical test.
Here is what I propose:
I will put together a list of matches where the result was close. Then for as many as I can get the data, look at what the teams were on after 15, 30 and 40 overs, and see if the game had been called off who would have won. We can then compare that to who actually won. While this is not the perfect way to test the two systems it is better than any other comparisson that I can think of.
Initially we will assume that the method that puts the correct team ahead the most often will be the best method, but it will certainly lend itself to further examination.
In the last 5 years there have been 117 games that have either been tied, won in the last over, won with less than 3 wickets in hand or won by less than 12 runs. Some of these will have been matches where the D/L method has been used, and so they will have to be removed from the list.
My expectation is that the Jayadevan method will come out slightly ahead, but it will be interesting to see.
If anyone wants to help me with this process, I have a work file here. It's a list of most of the close matches from the past 5 years. I need to go through and remove the DL matches, and then transfer the info into this spreadsheet. (from which I will export to Excel and finish the analysis there) Also if anyone has access to the Professional Edition of Duckworth-Lewis that would be very helpful.
Ok - so it looks like this method is actually quite old, it was published in 2002. It is just only now starting to get trialed in actual matches.
ReplyDeleteIt will be very interesting to see how it goes.