Winning Streaks Exist?

[MrBoston]

I have to admit I was skeptical about the ability of win streaks to predict the subsequent performance. Using game by game data, 2000-2006 regular seasons I analyzed for each win streak what percent won the following game. This simple analysis produced some interesting results. Further investigation is warranted, but needless to say my cuiriosity has been peaked:

Streak WinNext% P
0 48% 0%
1 50% 1%
2 51% 1%
3 52% 0%
4 52% 1%
5 53% 2%
6 53% 2%
7 64% 0%
8 55% 4%
9 49% 10%
10 47% 14%
11 64% 12%
12 33% 16%
13 100% 13%
14 100% 13%
15 33% 38%
16 100% 50%
17 100% 50%
18 100% 50%
19 100% 50%
20 0% 50%

[Tango Tiger]

So you have a team that won 8 straight games have won 55% of their 9th game? And teams that won 9 straights games won 49% of their 10th game?

I would expect to see a slightly above 50% number for the long streaks since those are biased with good teams more often than not. Obviously, at 11 and more, your sample size is likely very tiny.

I don't see anything here.

What's the last column? That the "p" value? If that's the case, all you are showing, probably, is the bias I was talking about.

[MrBoston]

Agreed on longer streaks. The sample size is too small to draw strong conclusions. For instance there are only 61 streaks of 9 games in my sample and the win rate is 49% and not statistically significant.

I prefer to look at the 4 and 5 game streaks with hit rates of 54% or so (no so different from the home field advantage). For 5 game streaks I have 582 streaks and a next game win rate of 53%. This is statistically significant at the 98% confidence level (p).

I calculate the p based on the hit rate (53%) being different from the naiive expectation of 50%. I use the binomial distribution.

Thoughts on shorter streaks? I also was interested that teams that have lost (streak =0) only won 48% of their next games.

[SABR Matt]

You need to calculate p based on the (more correct) net winning percentage of all of the teams involved in the streaks.

Moreover...you need to calculate it recognizing who the home team for the "next" game is going to be.

[MrBoston]

Good point Matt. Perhaps what I am picking up is the home field effect masqueraded within streaks (extended home stands, etc). Indeed when I look at the % of next games won in streaks by whether the next game is home or away it appears as if your hunch is supported by the data.


Next GameNext Game
Home Away
Streak Win% Win%
0 53% 45%
1 53% 47%
2 55% 46%
3 56% 48%
4 57% 47%
5 57% 49%
6 57% 49%
7 64% 63%
8 63% 49%
9 45% 48%
10 64% 41%
11 75% 60%
12 100% 38%
13 100% 100%
14 100% 100%
15 33% 0%

Also I run a simple regression on dummies for home and away and for streaks of 4 or larger.

Prob of Winning next game = .48 + .08* Home Dummy + .05 * Streak >= 4 Dummy

The t's on the intercept, streak, and Home Flag are 119,3.5, and 14.6 respectively. No test for multicollinearity but the contingency table above suggest a potential issue there.

[SABR Matt]

Indeed...I would hazard a guess that winning streaks happen because the schedule is streaked more than they happen because the teams have proclivity for long streaks.

The Mariners, for example had a 2-15 stretch this year which miraculously coincided with them playing 17 of 20 games on the road.