A bad game for WP
04.13.10
The 2010 title game was pretty boring from a win probability perspective. So much so, that I questioned whether my code was working correctly at halftime of my live WP experiment. As it turned out, I almost picked the worst possible game for semi-live win probability to make its debut.
It seemed unusual for a game to played entirely within an eight-point range. Duke led by no more than 6 points and Butler by no more than two for the entire game. That got me to thinking about how that stacked up with the 5700+ other games played last season.
Theoretically, a game must have at least a three-point range (assuming there’s a made three-pointer in the game). But there’s not much room between the eight-point range seen in the title game and the theoretical minimum.
So how many games had a smaller scoring range last season? Just one, actually. This November affair between Nebraska and USC was played entirely between Nebraska +3 and USC +4. There were ten other games that stayed within a range of eight. Notable among them was this gem between Iowa State and Missouri, won by the Tigers in OT. It was the only game besides the Nebraska/USC contest where the maximum lead by either team was four.
The main point here is that win probability is not terribly useful when a solid (but not heavy) favorite is involved in a game where their lead barely changes throughout the game.
One shining e-mail
04.05.10
This will have to stand as representative of the many e-mails that my computer received in February and March from people claiming their eyes were soooo much better. A computer has to vent, sometimes, and in light of recent events, my computer finds the subject matter especially amusing.
from Somebody
to ratings@kenpom.com
date Tue, Feb 16, 2010 at 10:42 AM
subject Skewed….
Ken,
I spent many years as a sportswriter for [a newspaper] and always found your ratings useful. This season, however, is proof that your instrument needs some tweaking. Any ratings that have a middling Duke team ranked 12 slots higher than Villanova and 15 slots higher than a Georgetown team which disemboweled Duke head to head needs some adjusting. I know your matrix is supposed to make conference affiliation superfluous, but the ACC REALLY stinks this season, and your system doesn’t reflect what any set of eyes immediately tells us.
—Somebody
Semi-live win probabilities tonight
I’ve created a crude graph on which I plan to plot in-game win probabilities for the title game. (It looks best in something other than Chrome.) I’ve updated my methodology so that I can plot the win probability for the beginning of every possession during the game. I’ll attempt to update the plot at each TV timeout (hence “semi-live”). It’s probably not worth much in the last 2 minutes if the game is close. We’ll see how it goes.