Wednesday, 1 April 2009

Nightwatchmen Discussion

Here is a post about using nightwatchmen by a real sabermetrician, Phil Birnbaum.

I haven't read it properly yet myself, but I link to it in case you've missed it.


  1. I only just came across your blog from a comment at Phil Birnbaum's. You've been hiding away from the rest of the cricket blogosphere. I see you've already discovered my blog.

    The runs/wickets analogue of base/out states is an interesting one, but the most important issue that I see is that the run environment changes drastically from game to game in a way that doesn't happen in baseball (outside of Coors Field). It shouldn't be too hard to hit everything with an overall correction factor at the end of the game though. Say if the match average was 40, rather than an overall average of 30, scale all the runs values in the runs/wickets table down by 3/4. That won't be perfect (because of declarations) but it looks all right. You could weight wickets by the average of the batsmen dismissed to improve it further. Anyway, those are just details.

    I am not sure why you looked at innings scores in isolation. The logical thing to do would be to consider the current lead (possibly negative), wickets fallen, and innings in match. Then just fit curves to the probabilities of each outcome.

  2. paul, please respond to my facebook message, if you get trhis first.