In an earlier, ahem, somewhat long (but now updated and completed) post I described an experiment to test the Duckworth-Lewis Method against the Best Scoring Overs (BSO) and Best Scoring Sequence (BSEQ) methods, using the domestic one-day competition as a test tube.
Hey – I wasn’t the only one watching the series who was looking for alternative ways to pass the time…
Reading up on DL when I was getting into all this, I predictably found my way over to Wikipedia. The DL entry lists some legitimate criticisms of the Duckworth-Lewis Method. And then there’s this little pearler: “More common informal criticism from cricket fans and journalists of the D/L method is that it is overly complex and can be misunderstood.”
I critique your critique by saying your tautology is self-reinforcing.
Of course, the fact that something can’t be understood doesn’t necessarily invalidate it. I can imagine a problem if someone invented an overly complex series of traffic signals and there was a resultant pile-up in a busy intersection. But I understand little about aeronautics, anaesthesia or curry, and I’ve been pleased to avail myself of all three (not simultaneously – at least, not yet).
Anyhow, the beauty of DL is that there’s really only one thing you need to know about it: it works. In the matches examined (in some detail, with amusing asides, and well worth reading about at the earlier post) it came out well on top. The scores show how many times each method predicted the winner at the end of each over, from over 20 to 49 (or to the end of the match):
The only disappointing thing about the experiment was that the chasing sides dominated the tournament, so when the table is split into matches won batting first and second, there’s not enough data to judge one half.
And as you can see, DL falls short of the performance of the other methods – hardly surprising when BSEQ and BSO rarely seem to endorse the chasing efforts of any team. Still, it might be worth doing a few more of these to see if there really is a long term problem with DL and chasing teams.
In any case, on the evidence so far, the 95% confidence interval for DL has it at least 22 percentage points better than BSEQ and 30 better than BSO.
When I first wrote about this, the main driver was to make a case for using DL as an on-screen graphic during match broadcasts, giving an ongoing update of what the team batting first is likely to score, or who’s “ahead” in the match, according to DL, rather than boring but obvious numbers like these.
Given the accuracy demonstrated in the experiment, DL is clearly a useful tool. Still, I wouldn’t expect any change from the Channel 9 broadcast team – they wouldn’t produce sensible stats even if their commentary box were full of tools. Oh, hold on…Follow @newstatsman