04 February 2016

Lucy and the Football

I'm not the most avid sports fan, but the rise of "advanced analytics" has the benefit of amusement. AA is really nothing more than descriptive stat plus a bit of population correlation. Squared differences, variously manipulated. From what I gather, it all began in baseball and Billy Bean decades ago. Now, one can be paid silly amounts of money for running simplistic correlation programs on reams of data. Such a life.

About the only sports TeeVee I'll admit addiction to is "Pardon the Interruption", and largely because Korny and Wilbon are ex-Washington Post. I lived there for rather a while, and deeply regret leaving. That's another episode. Yesterday had, as one might expect, much discussion about the Broncos and Panthers. In particular, whether the Panthers' record is legit or not. Korny noted that the Panthers had a really weak schedule (27th, meaning almost the weakest), thus their 15-1 regular season record is some part smoke and mirrors.

I certainly don't care enough (I don't bet on anything other than MegaMillions) to confirm to what, if any, degree the following scenario applies to the Panthers this year, but just consider.

The issue is strength of schedule, which gets various names in use. Consider a team's (ABC) third game of the season against team PQR. ABC wins. At the end of the season PQR has a record of 9-7, making them a so-so team, neither really good nor really bad; a middling weight. Might even be in the playoffs. From ABC's point of view, its record's SoS weighting would view the PQR third week win as a net positive. But... how well was PQR playing in the third week (I'll assume that ABC, for these purposes, played at a constant level for the year; accounting for ABC's fluctuations just means more arithmetic), which is what really counts in determining the appropriate weight? There are two extremes that lead week three to be viewed as a local minimum or local maximum, from a weighting point of view: PQR's record in the previous and latter two (or one or three or four or...) weeks. On the one hand, PQR could have been blown away in those four games by teams that had losing records both for the year, and the local five week schedule; making a win over a 9-7 team over weighted. On the other, PQR could have blown away teams with winning records, both for the year and the local five week schedule, excepting ABC which beat them; making the win more significant. Details matter. Actual SoS calculations often go to additional levels of opponent schedules, so some folks do care to be accurate.

What matters most, of course, in attempting to predict any single game outcome is standard SWOT analysis, data appropriate to the sport. That's not so much data as anecdotal assessment. There's a reason legal betting odds on sports aren't determined by experts, but by the flow of bets on the participants. The sports books change the odds to move the money flows between the opponents aiming for a 50-50 split, taking a cut from the gross.

No comments: