Bring on the Quant revolution

This is the second in a series of posts on the IPL.

In those interminable Tendulkar vs Lara debates, wouldn’t it be great if we had a way to measure who has been the better batsman in all possible conditions. Rahul has been a tireless advocate of the use of advanced statistics and in general more sophisticated metrics to measure player effectiveness (than just averages and strike rates) in cricket. On the face of it I am just as hopeful and excited that private enterprises such as IPL franchises will bring about this long needed revolution even as I am disappointed that private individuals (cricket loving geeky really) haven’t already made a start with the data currently available. The Americans have already ushered in this quant revolution starting with Moneyball and since taken forward by sports loving geeks, some of whom have been employed by teams including one who has risen to GM of a franchise.

There are inherent problems with using advanced metrics in a cricketing context: the pitch and weather are critical factors that can render many comparisons across different conditions less than meaningful. But we should not fall victim to the perfect solution fallacy. Here are some ideas to consider.

Analytics

Cricket’s basic metrics (runs, wickets, catches) are much fewer in number than say basketball (points, FG%, rebounds, assists, steals, blocks, turnovers). So if you want to devise a composite measure of player effectiveness, you have very few ingredients to play with. (Of course you could argue that a player has an influence on the game only through one of these three modes. But we know the truth is more complicated than that.) Here are possible ways to get to the elusive composite measure.

• Weight each innings by two factors: the average runs/wicket scored in the match and the average runs/wicket scored in all matches in that ground in the last 4 years. This ensures that a 100 scored when most batsmen have not done as well is obviously more valuable than a 100 in a match where both teams have declared for 500+ scores. Similarly a 100 scored in a ground unconducive to batting like Durban is better than a 100 at the Oval.
• The weighting can similarly be done for every bowling performance by using strike rate (balls/wicket) both in the match and historically.
• Running between the wickets is another “intangible” quality that has not been quantified. How about adding a running weight to a batsman’s overall rating by factoring in the percentage of run runs (1s/2s/3s) in all partnerships involving him or the number of run runs per 100 balls in a partnership involving him. This is a far from perfect measure as it may punish the boundary hitters. But over a large enough sample size, it will separate the better runners from the poor ones and more importantly provide a servicable quantitative input into an overall composite measure.
• Ground fielding is a tough nut to crack and I haven’t been able to come up with an acceptable metric. I’d love to hear suggestions.

Of course none of this is perfect and I hardly have the statistical expertise to give weights and incorporate them into a composite measure. I’m just trying to set the debate rolling so that we can ultimately get better ways to measure effectiveness. And every proposed measure can be tested for by checking if teams that have players rated better (by the measure) on average win more often than not. Cricinfo has a made a promising start with their tackily named It Figures blog – they seem to have used standard deviations to measure consistency and a variation of a moving average I have suggested. But there’s ways to go yet.

Measurement and mapping

It’s great that the wagonwheel and bowler’s pitchmap have become common tools of analyses. But it’s time to free this data from the TV production companies and release this information to the public as the NBA has. Won’t it be great if we can compare Tendulkar’s cumulative wagonwheel from 2003 and 2007 to see if he’s playing any differently. Or the wagonwheel for all runs scored in Perth over the last 3 years to see if horizontal bat shots are still the way to score runs there.

But I have a slightly more radical suggestion. How about mapping every ball bowled from start to finish and annotating it. For example, you start with recording the bowler’s line and length, the height to which the ball bounced, the shot played by the batsman, the result (runs or wicket), the fielder involved and if possible even the field positioning at the time. It should then be possible to digitally retrieve say every ball bowled to Ganguly short of good length just outside off stump in the last 4 series and see if there’s a pattern to his shots. It sure does sound like a lot of work, but the Indian team’s analyst is on record as saying that they’re already doing some version of this. Of course a team would like such data on every match played by every team. So what’s the solution? Trust the Americans to find a viable one. There is a company that does precisely this for basketball; every play is mapped in every game right down to the tee. Of course this data is expensive to record and will therefore be only available for a hefty fee. With rich IPL franchises, there is an incentive to use every possible means to get ahead and so there will most certainly be demand for this service. And it sure beats having your own analyst do an inadequate job.