Ed Küpfer
Joined: 30 Dec 2004 Posts: 130 Location: Toronto
|
Posted: Fri Jan 28, 2005 5:03 pm Post subject: Dramatic Comebacks |
|
|
A good idea for a study from Brian Connolly: Do Dramatic Wins Matter? He looked at baseball teams who had pulled off late inning comebacks to see if these wins had affected long term performance.
I did something similar, but different: thanks to Dean LaVernge, I put together a list of 74 games between 92-93 and 01-02 in which one team had come back from a 10+ point deficit at the end of the 3rd quarter. Here's how my sample looked:
Code: | Come # of
back games
10 |xxxxxxxxxxxxxxx 15
11 |xxxxxxxxxxxxx 13
12 |xxxxxxxxxxxxxxxxx 17
13 |xxxxxxxxxxxxx 13
14 |xxxx 4
15 |xxxxxxx 7
16 | 0
17 |xx 2
18 |xx 2
19 |x 1 |
(Sorry, the only thing I love more than graphs are ascii graphs.)
Now then, did these dramatic wins affect the teams' performance in the long run? I looked at each teams' record over the previous 20 games and compared it to their record over the next 20 games. Overall, these teams won 10.1 games before the comeback win, and won 10.4 games after, a difference that is not statistically significant.
But that doesn't control for the different individual strengths of those teams. If I look at the percentage of total team wins that season that came in the 20-game period before, and the % that came after, we could compare those teams on even terms. For example, if a 40-win team had won 10 games out of 20 prior to the comeback win, and then gone to win 12 of their next 20, I would express it as
Code: | Games Before After Diff p-value
20 0.25 0.30 + 0.05 0.62 |
The team had won 10 games, or 0.25 of their season total, in the 20 games before, and 12 games, or 0.30, in the 20 games after. The p-value is the probablility of the difference coming about by chance. In the previous example, there is a 62% probablity of a fluke result, so it is not statistically significant.
Here are the results for the 74 games I looked at. In addition to comparing the records for 20 games before and after the comeback win, I also looked at 10- and 5-game records.
Code: | Games Before After Diff p-value
5 0.06 0.06 + 0.00 0.70
10 0.12 0.12 + 0.00 0.63
20 0.23 0.24 + 0.01 0.62 |
As you can see, a dramatic comeback win made no difference in to the long term records of the teams in my study. I was going to study this further, controling for the strength of schedule and other elements, but such a stunningly negative result discourages me from thinking it'll be worth the bother. _________________ ed |
|