APBRmetrics

Eli W · Joined: 01 Feb 2005 Posts: 402

There's some pretty interesting discussion going on springing from a research paper that suggests that it could be a good thing to be trailing by one point at the half of a college basketball game.

Here's the NY Times piece:

http://www.nytimes.com/2009/03/16/sports/ncaabasketball/16score.html?_r=2

It points to the original paper:

http://qbox.wharton.upenn.edu/documents/mktg/research/Losing_and_Winning.pdf

Some criticism of the study can be found on various blogs (the second link in particular may be of interest as it follows up with some NBA research):

http://www.insidethebook.com/ee/index.php/site/comments/being_behind_is_a_good_thing/

http://www.insidethebook.com/ee/index.php/site/comments/being_behind_is_a_good_thing_part_ii/

http://www.stat.columbia.edu/~cook/movabletype/archives/2009/03/basketball_news.html

http://freakonomics.blogs.nytimes.com/2009/03/17/when-losing-leads-to-winning/

Ryan J. Parker · Joined: 23 Mar 2007 Posts: 708 Location: Raleigh, NC

Gelman's analysis caught my eye this morning. I love this stuff. Right or wrong, this is a great learning tool when really smart people analyze this kinda thing.
_________________
I am a basketball geek.

Guy · Joined: 02 May 2007 Posts: 128

Does anyone here know of NBA data that would show if the same pattern occurs there? Ed K. posted some data on win% at various minute/score states a while back, but the link is now broken.

Ed **Küpfer** · Joined: 30 Dec 2004 Posts: 785 Location: Toronto

I wish I could help. The original data is gone, disappeared on a long lost hard drive. I now have newer, better data, which I am obviously not allowed to give out.

I can tell you this: my original dataset came from NBA.com play by plays. You can recreate this by using the data provided at basketballvalue. In fact, it will be much better since their data is much cleaner than the stuff I used. I don't know how hard it would be to do this, but if I can do it, anyone can do it in half the time.
_________________
ed

Guy · Joined: 02 May 2007 Posts: 128

Ed: Can you just tell us how often teams win when they lead by 1 point at the half? Surely that isn't a proprietary figure. :>)

Ed **Küpfer** · Joined: 30 Dec 2004 Posts: 785 Location: Toronto

I found some old data that may be helpful. Home team halftime leads and eventual wins, games from 1992 to 2002. (The 20 point leads and deficits, I'm not sure if they mean actual 20 points or 20+ points.)

Serhat Ugur (hoopseng) · Posted: Wed Mar 18, 2009 4:11 pm Post subject:

last season it is 52-56. (trailing by 1 and win the game)
this season it is 38-47
_________________
http://www.nbastuffer.com

Guy · Joined: 02 May 2007 Posts: 128

Thanks, Ed. Very helpful.

Mike G · Posted: Thu Mar 19, 2009 6:11 am Post subject:

I'm guessing Ed's data must be referring to the halftime status of the home team. A halftime tie is won 56% of the time by somebody, eh?. Likewise:

Guy · Joined: 02 May 2007 Posts: 128

Hoopseng: Is that data for all teams, or home teams only?

schtevie · Joined: 18 Apr 2005 Posts: 408

The paper is cute, but I don't find it that interesting from a basketball perspective. And apparently not at all from an NBA perspective.

Given Ed's data, and the familiar home court advantage of between 3 to 4 points - call it half that at halftime, it would seem that, lo and behold, home teams down 2 points at the half win....half their games (241 out of 479). Maybe this has something to do the attitudes of professionals vs. amateurs. Lack of rah, rah, and all that.

More generally, It would have been interesting if the study had included corroborating analysis of overtime games. According to their set-up, as I understand it, a game that ends in a tie is actually a very slight notional deficit for the away team, given the generally accepted existence of the home court advantage, and it should invoke their relatively greater effort. It would have been nice confirmation to see away teams winning more than 50% of OT games. Maybe it is so, maybe it isn't.

Guy · Joined: 02 May 2007 Posts: 128

Actually, the NBA data provides partial support. In Ed's data, and more recent data compiled by Mitchell Lichtman, it appears that HOME teams down one at the half win more than they should. They win about as frequently (slightly more) than in tie games. (Alternatively, road teams up by one at the half win less than they should.) Road teams do not show this -1 effect. See the 2nd link in Eli's original post above for discussion of this at The Book blog.

Ed **Küpfer** · Joined: 30 Dec 2004 Posts: 785 Location: Toronto

Okay, I went ahead and did some more work on this. By incorporating the home and away teams' win percentage, I calcuated the home team win probabilty (log5) for each game in the old data I have. I created a histogram of home team half time leads, binned by the probability of the home team winning the game.

For some reason, "home team down one at the half" games are overrepresented when the home team is playing a far superior opponent.

Next home team win percentages by lead at the half, again binned by log5:

Nothing looks out of line there. It appears that once you take relative team strengths into account, home teams down one at the half win about as much as they're supposed to.
_________________
ed

ecumenopolis0 · Joined: 15 Jul 2008 Posts: 22 Location: Houston

Ed **Küpfer** · Joined: 30 Dec 2004 Posts: 785 Location: Toronto

Further evidence that I shouldn't trust my eyes to see patterns in the data. I regressed probability of home team win on home and away team win percentages along with home team lead at the half, the latter variable treated as a categorical (factor in R parlance). Here are the regression results, which do show something fishy about -1, but I'm not sure what.