Are preseason college football polls any good (part 2)?
In Part 1 of this series, we examined whether or not there was predictive power within preseason polls and found that in general, polls had a rough impact to team quality as measured by the final postseason poll outcome.
An alternate way to assess whether or not there is merit in preseason polls is to compare the preseason poll associated by the Associated Press with the ultimate number of wins by each team.
Over the past 20 season, the raw dispersion of these outcomes is depicted below, where there is an observable negative correlation: lower ranked teams are associated with lower win totals.
The Statistical Model
Statistically it becomes straight forward to then conduct some simplistic analysis. Our model will estimate the seasonal win total for each team given 1) its pre-season rank 2) its conference affiliation and 3) the number of games by the team. Conference affiliation is important to conclude as it captures a couple of different impacts and eliminates some of the bias we might otherwise see. For example, a conference which plays a weaker schedule (like the mid-major conferences) would typically be expected to be required to win more games than a P5 counterpart in order to justify their lofty rankings. Number of games is required for obvious reasons. The more games, the more expected wins.
The Statistical Results
The results are actually quite fascinating. Important note before seeing the results: the model is ‘calibrated’ to a mid-major level conference, which is why you will see their absence in the following table. It is also why the conference values are negative. It means that on average, the P5 conferences will have fewer wins than their mid-major ranked brethren at the end of the season.
In this model, all parameters are statistically significant, and the R2 is approximately 53%. Which means 47% of the win total is affected by things not included in this model (but perhaps we will seek to refine in future posts).
The Statistical Interpretation
The most important take away is that pre-season ranking IS a statistically important determinant in the number of wins achieved by a team. Every 10 pre-season ranks lower are associated with approximately 1 fewer win. Graphically we can then depict the projected win total, by conference, for a team with a 13 game schedule. The underperformance of the Pac 12 is somewhat noteworthy.
We can also identify the single best 5 and worst 5 seasons with regards to preseason expectations over the last 20 years:
In Part 3, we will examine the conference results to see which conferences traditionally move higher or lower between the opening polls and the final week of the season.