Prediction for the final
Sri Lanka's annihilation of the West Indies (and most everybody else) has put them to the top of the ratings going into the final. Their probability of victory comes to 58%.
A word too on form, and perhaps a question. The main purpose of the form variable is to smooth out the volatility of unexpected results, by "banking" them, such that the effect of that result will not carry over until later games. If a team consistently loses worse than expected their rating will shift more quickly to its new equilibrium. On the other hand, if the loss was unusual, the rating will stay relatively stable.
Which brings up a question, possibly philosophical: is form an indicator of where the rating will go, and therefore should be added to make predictions; or where the rating has been, and therefore should be deducted because that is where it may return? In other words, is form a meaningful number, worth reporting, and in what sense? If results were completely random, then a run of bad form would dip the rating more than necessary, which oddly enough will actually increase volatility (sort of like irrational exuberance in a stock market). Conversely, the relative form of the West Indies and England prior to their Carribean match-up indicated that the series would be closer than the ratings might indicate; similarly Australia's form in test cricket began to dip markedly even before their horrid run against India and South Africa.
The reason for asking this relates to Sri Lanka's current form, currently over 100. Partly, their form indicates only that they are winning. The expected margin against other top-8 sides are so narrow that almost all wins beat the margin by a lot. Similarly, South Africa were equally dominant leading into their semi-final against Pakistan. The numbers are inconclusive to date, but anyone looking for an extra reason to lean one way or the other may like to consider the implications of strong form on their expectations.
T20 Ratings - World Cup Edition (semi-finals)
|Rankings at 16th June 2009||form|
Predictions for semi-finals
Pakistan v South Africa 7 runs
West Indies v Sri Lanka 7 runs
South Africa and Sri Lanka have emerged as the clear favourites for their respective finals, having cruised into the semi-finals undefeated, and in outstanding form. However, neither side has a greater than 60% chance of victory, so keep tossing those coins.
Princes Bridge, looking out. Taken June 2009
|Rankings at 14th June 2009||form|
Predictions for match day 6
Ireland v Pakistan 51 runs
England v West Indies 8 runs
New Zealand v Sri Lanka 6 runs
India v South Africa 4 runs
As noted the day before yesterday, I took a moment to compare predictions, based on both the current rating, and current rating with form. The short answer: neither are much help. The average error in both was almost exactly 28 runs; by comparison, the average error if a tie had been predicted for every match: 24.6 runs. Looking at the scatter-plot of results, it is possible that it will converge on a predicted margin half what it currently is, with a standard deviation of 20 runs. In other words, a massive 400 point difference in rating points is needed before the win percentage approaches 65%.
On a different note, Sri Lanka's travails against Ireland has meant their game against New Zealand is now a must-win to go through, whereas the South Africa v India game is marked as "dead", in case India annihilates South Africa and still receives little gain in the ratings.
|Rankings at 12th June 2009||form|
Predictions for match day 5
South Africa v West Indies 5 runs
New Zealand v Pakistan 1 run
Ireland v Sri Lanka 54 runs
England v India Tied
Appendum - Predictions plus
David Barry asked below about how the predictions are done. The answer is badly, given the limited data set I have. As noted earlier, by the next world cup I will be able to give more precise numbers for the difference a rating makes (with a standard deviation as well). But for now, the prediction is calculated by the (disturbingly) simple: ratings difference divide 10 (the linear regression says 0.078 +/- 20). With a 100 point bonus for being at home.
As an aside, it had already occurred to me that a published prediction should account for form (future ratings change, results depending). Doing that gives the following:
South Africa v West Indies 8 runs
New Zealand v Pakistan 6 runs
Ireland v Sri Lanka 59 runs
England v India by 5 runs
Tomorrow I will try and calculate, based on past results, what gives a more accurate rating, and use that.
The other post was getting messy. One post per round of matches until the final.
|Rankings at 9th June 2009||form|
Predictions for match day 4
Ireland v New Zealand 43 runs
England v South Africa 7 runs
Pakistan v Sri Lanka 1 run
India v West Indies 13 runs
This won't be a regular thing - to the extent that the test ratings are regular. But it occurred to me that with relatively few games to consider, and most of the ratings leg work done already, converting the 300 odd T20 games to a workable ratings system was feasible.
The main difficulty was deciding on the starting ratings, given the "minnows" have no form, and moreover, were mostly playing each other prior to the start of the tournament. I settled on a modified form of the official ODI ratings at January 2005 (multiplied by 5 essentially), with the non-ranked teams set to 0, then all teams normalised to average 1000.
Weights and expected margins were adjusted to suit: the margin for a chasing team was set to:
1 + balls_remaining * (wickets_remaining / 5)
It is far too early to judge how successful that is, but perhaps by the next world cup some proper statistical analysis can be applied. The ratings prior to the start of this tournament, including all games to date (and the uncounted "warm-up" games) were as follows:
|Rankings prior to World T20||form|
I'll keep this post updated as the tournament progresses.
Update 8th June: After six games there has been a little movement, notably Australia and Pakistan moving backwards and South Africa forwards. A little tightening has occurred in the first few days.
Update 9th June: Australia crashes out, and down, but perhaps from a ratings perspective it is better they aren't still playing as their form is at -108.9 (in comparison to South Africa's +92.6). Ireland's numbers are still low, but their form and progression to the last eight will see that improve.
Update 10th June: Perhaps unsurprisingly, Pakistan got the 25 run victory they needed - the ratings predicted 49 runs. A meaningless South Africa v New Zealand game and an error in my game details provide the other changes.
Update 11th June: Two results that went largely as expected. India and Ireland move back towards an otherwise even pack leading into the second round.
|Rankings after first round||form|
Kensington. Taken May 2009