As anyone who isn't a Yankees fan can tell you, team salary has a strong effect on winning in baseball. The NFL is different, however. Its salary cap is a fairly effective means of equalizing team payroll. Yet teams can circumvent the rules to exceed the cap in various ways, at least until it catches up with them. There is also a salary minimum that teams are required to spend because there are teams that chose not to spend up to the cap limit. For these reasons there can be wide variations among teams in terms of player salary.
In 2006, the average team salary was about $100 million. The standard deviation was about $13 million. Indianapolis was the team with the highest payroll, exceeding $131 million, while Oakland was the team with the lowest, paying their players only about $72 million.
For MLB, about $2-5 million buys your team a marginal win. What about the NFL? Is there a connection between wins and payroll?
To find out, I compared team payrolls and team wins for the 2001 through 2006 seasons. Because actual team payrolls depend on complex contract structures, I used cap charge as the definition of team salary. Data was obtained from the USAToday NFL salary database.
Each year the salary cap is adjusted as a function of league revenues, so I normalized each salary by year. That way, we can compare a 2006 salary to a 2003 salary even though the cap has grown greatly over the past few years.
I ran two regressions to test the importance of team salary. The first regression used the year-to-year change in team salary to estimate the year-to-year change in team wins. I wanted to see if teams that beef up by signing free-agents to large contracts were able to convert those dollars into wins.
VARIABLE | COEFFICIENT | STDERROR | T STAT | P-VALUE |
Z DeltaSalary | 0.24 | 0.28 | 0.88 | 0.38 |
r-squared | 0.4% |
The significance of Delta Salary (change in team salary) is not significant in terms of improving a team's win-loss record.
The second regression was a more straightforward comparison. It simply compared total team salary to total wins.
VARIABLE | COEFFICIENT | STDERROR | T STAT | P-VALUE |
Z_Team_Salary | 0.33 | 0.16 | 2.10 | 0.04 |
r-squared | 2.3% |
Team salary is significant in estimating team wins. The r-squared is quite low, however.
What does the low r-squared mean when the variable is signficant? R-squared indicates the percentage of variance in the dependent variable (team wins) accounted for by the model (team salary). But keep in mind, there is a large amount of luck in team records. The r-squared of non-luck factors is probably close to 50%, so team salary could be far more important than indicated by the r-squared.
The thing that really matters is the coefficient of the significant variable--team salary. It estimates that for every standard deviation above average a team spends, it can expect 0.33 extra wins. That means that $13 million could buy a third of a win in 2006.
The Colts, who led the league in salary, were 2.4 standard deviations above average in payroll for 2006. This would buy them 2.4 * 0.33 = 0.79 wins. The Raiders were 2.1 standard deviations below average in payroll, which equates to 0.69 losses. Keep in mind how precious a single win is in the NFL, where the difference between 7-9 and 9-7 is enormous.
Football appears no different than any other professional sport. Salary can buy wins, but teams cannot sustain payroll above the cap for more than a couple years before they need to "pay back" what they borrowed through deferred bonuses other contract devices.
Next, I'll look at how well a team spreads the wealth among players--median salary. Does a team that has few stars but lots of depth win more often than a team with many stars and little depth?
excellent analysis
wonderful help alot
Could you explain why the first regression did not yield similar results? Just curious, thanks.
Not sure. I would guess it's just that overall total salary is what matters. Salary relative to previous years doesn't appear to equate to more wins. It may be that the effect is non-linear--something like: if you spend up to +$10 mil you don't get any return, but if you spend +$12 or more it will buy slightly more wins.
One thing to consider is that if a team has better players then the team needs to pay them more. A great example is with the Colts - you need to pay more for Peyton Manning but he will give your team more wins. Jamarcus Russell gets paid more than Peyton Manning this year but Peyton is the far better QB. The underlining assumption of this regression is the teams will pay the right "market value" for each player.
A better way to see the salary effect is to control for the team level of talent and then run the regression. Defining the team talent is the hard aspect. Something like number of pro bowl players or even if you just control for the QB with the season QB rating.
You need create an average of the past 10 years before running your regression. NFL team payrolls have large variance due to major signing bonuses being added to one year. That why teams can varry 30 million from one year to the next.
The NBA is easier for this analysis. I think team payroll makes much more of a differnce in that league. One great player has more of an impact in basekball.
one thing you may want to do is adjusted for schedule. Since schedule difficulty varies year to year there might be a higher correlation with schedule-adjsuted wins than wins