tag:blogger.com,1999:blog-2830084253401570472.post6134285639530336734..comments2022-05-09T18:03:13.102+12:00Comments on Offsetting Behaviour: FAQ on the WASPEric Cramptonhttp://www.blogger.com/profile/15831696523324469713noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-2830084253401570472.post-38992885811520943162015-03-22T23:21:15.564+13:002015-03-22T23:21:15.564+13:00Chris,
I've only just seen this comment, so s...Chris,<br /><br />I've only just seen this comment, so sorry about the delay. The answer is that it does. It possibly takes it too much into account. Before the most recent rule changes, the bpp had added only about 4 runs to teams scores on average. With the new rules that require no more than 4 fielders outside the circle at all times (in contrast to the previous rule of 5), the difference between the normal situation and the bpp (maximum of 3) is smaller than it used to be.Seamus Hogannoreply@blogger.comtag:blogger.com,1999:blog-2830084253401570472.post-88301843395196604532014-05-28T14:23:43.682+12:002014-05-28T14:23:43.682+12:00Following on from Scott's reply, there are two...Following on from Scott's reply, there are two sources of uncertainty. First, the prediction (which we would rather call a projection), is just the mid point in a sense of a range of things that could happen. We like to think of it as a measure of how a team has performed so far rather than a statement of what will happen. Second, though, there is model uncertainty, in which we could have mis-estimated the expected future score or probability of winning, maybe because of evolving strategies, changes to the rules and so on. If so, reporting the model to the nearest integer might look like excessive precision, but reporting to the nearest multiple of 10 in the first innings, say, would mean situations would arise where the WASP projection would keep changing by 10 every ball. <br /><br />The other thing I would say is that we should never take probabilities (in any environment) too literally. We know that the human brain is not very good at understanding probabilities intuitively. If I see a number like 1%, it doesn't matter if it means that a win from that situation would only occur once every 100 games, literally, or maybe would occur once every 50 (i.e. should be reported as 2%). What matters is that 1% is a very low number and is lower than 2%, so when I see WASP fall from 2% to 1%, I know that the team is almost certainly going to lose, and things are getting worse.Seamus Hogannoreply@blogger.comtag:blogger.com,1999:blog-2830084253401570472.post-473598927555053872014-05-28T12:13:20.893+12:002014-05-28T12:13:20.893+12:00Thanks for your comments V - I think you raise two...Thanks for your comments V - I think you raise two good points. I agree that you can't test the system by replaying the exact same game. Indeed any individual game situation occur incredibly rarely. However, you can look at a large number of games where the probably of winning was (say) 1%, even if that 1% was a result of different game situations (eg. 30 runs needed from the last over with 5 wickets in hand vs 50 needed from 20 overs with the last pair at the crease). And then you can assess whether, on average, the batting team ended up winning 1% of those matches. On your second point, yes, in the first innings we are predicting the mean outcome. It would be great to assess the projected scores vs the actual outcomes to show the likely degree of variance from the mean. We will make the effort to look more closely at this when we get the chance.Scott Brookernoreply@blogger.comtag:blogger.com,1999:blog-2830084253401570472.post-85667447802475022342014-05-28T03:41:28.469+12:002014-05-28T03:41:28.469+12:00With the 1% example, this is where I become skepti...With the 1% example, this is where I become skeptical. Surely this model assumes some underlying mathematical distribution that may or may not reflect reality (likely to be the latter). There is essentially no way to really 'test' this system given that you can't replay the exact same game with the exact same parameters. <br />So (for example) isn't the number of predicted runs an example of false precision, in that to be truly meaningful you would have to at least give some additional parameters such as the number of runs +/- a distribution parameter.Vnoreply@blogger.comtag:blogger.com,1999:blog-2830084253401570472.post-68306314164903835682014-05-27T19:39:40.752+12:002014-05-27T19:39:40.752+12:00Excellent stuff. Many thanksExcellent stuff. Many thanksThe other Neilnoreply@blogger.comtag:blogger.com,1999:blog-2830084253401570472.post-87610933468336498602014-05-27T19:34:52.937+12:002014-05-27T19:34:52.937+12:00Reading this as it links off a recent post. This ...Reading this as it links off a recent post. This is excellent, as usual.The other Neilnoreply@blogger.com