How Long Does Success Last? Part I

In 2005, Washington and Tampa Bay were playoff teams. In 2006, they were bad teams in a bad conference. This year, both teams have bounced back and look like strong playoff contenders.

In 2005, New Orleans was the second worst team in the league. In 2006, they made it to the NFC Championship game. In 2007, they look to be one of the worst teams in the league again.

When predicting how an upcoming season will turn out, we tend to not to think 2 or 3 years back. We shape our assumptions to be based largely on the previous season. But how long does a team's success (or failure) last into the future? At what point, can we safely say that the team this year is completely different from the team X years ago? Can we add value to our regular season win projections by looking further into a team's past?


The experiment is pretty simple. For win total data from 1996-2006, I calculated the correlation of wins in year Y with wins in Y+X, where X is [1,9]. The p-value is the probability that the correlation coefficients were the results of random data (≤0.05/5% is considered to be significant).

As you might have read elsewhere, season win totals don't correlate highly with next year's win totals to begin with. For one to three years in the future, though, the decline in correlation is pretty gradual. Even three years later, season Y is still somewhat relevant. Years 4-9, however, are useless. The correlation coefficient oscillates around 0. So teams in 2007 are still what they were in 2004 and 2005 to a small extent, but the teams of 2003 are gone. In-depth analysis, I know. But next time, I'll take a hopefully more in-depth look at how we can use this information to improve win total projections.

Correlations of WinsY with WinsY+X
X	Correlation	P-value	Sample size
1	0.21786	0.00005	343
2	0.16254	0.00405	311
3	0.11528	0.05444	279
4	0.04584	0.47325	247
5	-0.03518	0.60798	215
6	0.03106	0.67637	183
7	-0.03479	0.67051	152
8	-0.01097	0.90496	121
9	0.04845	0.65019	90