Friday, October 4, 2013

Strength of Schedule Revisited - Part II

Earlier this week I had a comment about why strength of schedule might be statistically insignificant, that being I measure all non-FBS schools as a 126 (for this season since there are 125 FBS schools) in the strength of schedule calculation.  So, what I did was take a look at last year's (2012) NCAA FBS data and delete out all the games played by FBS schools against FCS schools and then re-ran the regression controlling for heteroskedasticity (as I have done previously).  I plan on re-doing this at the end of the regular season this year and after the post-season (bowl games).

What I found was that strength of schedule is still statistically insignificant (i.e. t-Statistic) is less than two in absolute value, meaning that I am not at least 95% confident that strength of schedule (SOS) is different from zero in a statistic sense.  In fact I am not even 20% confident that SOS is different from zero statistically.  Given the estimated coefficient on SOS is also almost zero, even if it was significant (and it is not) the amount of impact that it has on winning percent is nearly nothing anyway.

Also note that the regression performs well in terms of R-squared and Adjusted R-Squared, and the probability of the F-Statistic is also very low indicating that all the independent variables are not jointly equal to zero in a statistical sense.

Here is the regression results for the 2012 NCAA FBS season run using E-Views.


Dependent Variable: WINPCT


Method: Least Squares



Sample: 1 124



Included observations: 124



White Heteroskedasticity-Consistent Standard Errors & Covariance





Variable Coefficient Std. Error t-Statistic Prob.  





C 0.441 0.059 7.525 0.000
PF 0.002 0.000 20.137 0.000
PA -0.002 0.000 -14.501 0.000
SOS 0.000 0.001 0.204 0.839





R-squared 0.857


Adjusted R-squared 0.854


S.E. of regression 0.095


Sum squared resid 1.089


Log likelihood 117.598


Durbin-Watson stat 1.860







Mean dependent var 0.491


S.D. dependent var 0.249


Akaike info criterion -1.832


Schwarz criterion -1.741


F-statistic 240.316


Prob(F-statistic) 0.000