Friday, November 19, 2010

Does Strength of Schedule Matter?

The last two blogs have looked at how to measure strength of schedule and the teams actual schedule strength as of November 13th. Now I want to turn my attention to how much effect does strength of schedule actually have on team performance. In order to do this, I will need to analyze all 120 NCAA FBS teams through the weekend of November 13th in order to give (a series) of answers as to how (if at all) does strength of schedule affect NCAA FBS team performance. So, taking the data for all 120 NCAA FBS teams up to and including the weekend of November 13, I have calculated each teams strength of schedule, found their winning percent, points scored, points surrendered and also found their total productivity using the complex invasion sport production model. So, here are the results.

First, I ran a linear regression looking at how strength of schedule impacts winning percentage only. The regression estimated is: winning percent = f(strength of schedule). The result from this statistical estimation is that estimated coefficient for strength of schedule is positive and statistically significant, with a t-statistic greater than two in absolute value, and an adjusted r-squared equal to 0.05. Thus only looking at strength of schedule I find that it does impact winning percentage and that teams that have easier schedules have higher winning percentages. The problem is that the variation in strength of schedule "explains" very little of the variation in NCAA FBS team's winning percent - only about 5%. So, let's see if we can do better.

The second linear regression looks at how strength of schedule impacts winning percentage along with the amount of points the teams scores and the amount of points the team surrenders. (I also ran the regression on point spread - points for minus points against). Now strength of schedule variable does not work out well. The results from this regression are that strength of schedule is statistically insignificant (or that statistically strength of schedule has zero impact on winning percentage) when also taking into account the number of points scored and points surrendered. The regression does rather well on the whole, with an adjusted r-squared = 0.84, and the estimated coefficient on points scored is positive and statistically significant at the 99% level of confidence, and the estimated coefficient on points surrendered is negative and statistically significant at the 99% level of confidence.

The same goes for a regression on strength of schedule and the point spread (i.e. points scored minus points surrendered). Strength of schedule is statistically insignificant and point spread is positive and statistically significant at the 99% confidence level.

Given this result, I conclude that strength of schedule does not matter when looking at teams winning percent. It did not matter in the NCAA FBS paper I wrote up about team production, and still does not. So all this talk about one teams strength of schedule better than another teams strength of schedule is rather a waste of time.

3 comments:

  1. I don't think your findings take into account the possibility that, while, for example, the worst team in the SEC might have a comparable winning percentage to the worst team in the WAC, that SEC cellar-dweller might become bowl eligible if it competed against a WAC league schedule. Ole Miss finished 1-7 in SEC play (4-8 overall) but beat Fresno State (convincingly), which finished 5-3 in the WAC (7-4 overall). Now, I understand upsets occur. But to say strength of schedule does not when looking at a team's winning percentage seems naive to the difference in overall league or conference quality, which makes up the bulk of a team's schedule and hence greatly impacts its winning percentage.

    ReplyDelete
  2. You have chosen to focus on 1 of the 1416 NCAA FBS results (looking both home and away results).

    My comment about strength of schedule takes into account all 1416 results for this year (to date – there are still 32 results scheduled, with a few more for conference championships and bowl games to be inputted). Also, in my model Fresno State is 79th (below average) so, I am not convinced that Fresno State is as good (in terms of total production) as their record would suggest, which is one of the implications of the model.

    Yet even focusing on winning percentage, when running a linear regression (with winning percentage as the dependent variable) and using total production and my measure of strength of schedule, I find that total production is statistically significant but strength of schedule is statistically insignificant (i.e. statistically it is the same as zero). The same is true if you use points scored, points surrendered and strength of schedule. Hence from a statistical perspective strength of schedule does not matter.

    ReplyDelete
  3. Can’t argue numbers! However, I think you must prepare to address the issue of statistical insignificance of strength of schedule with regards to different leagues playing at different levels and affecting the numbers. I contest that, in general, if you took a mediocre team from a powerhouse conference, such as the SEC or PAC-10, that team would have a better record playing in an inferior league, such as Conference USA. This, if true, demonstrates that the strength of a team’s opponents affects how well they will perform. Mere winning percentage or productivity may not tell you this because teams are playing the majority of their games in different leagues. If you wanted an exaggerated example, my freshman high school football team was undefeated, and our closest game was 22-6. We were not the best team in the country; we just played weak teams. While statistically insignificant in your findings, you must avoid saying strength of schedule is “overrated.” Thoughts?

    ReplyDelete