In Defense Of Procrastination
So I went back and looked at how my preseason rankings compared against the AP and USA Today preseason rankings when compared to the final rankings for the past two seasons. In order to justify my procrastination, I also did a correlation of the previous year's final rankings against the current year's final rankings (basically substituting last year's final rankings for preseason rankings).
Discoveries:
1) There is very little difference in the correlation values for my preseason numbers and the previous season's numbers when compared to my final rankings.
2) In general, the AP and USA Today polls do a better job of predicting the final Top 25 based on the preseason Top 25 (but only barely), but Chaos Rankings beats them (in some cases, quite handily) when predicting the Top 10 and in the overall data set.
Conclusions:
1) I still think it is worthwhile to have some sort of stabilization in the first couple weeks to prevent teams from bouncing around too much. Since I've shown that there isn't much difference between the Chaos Rankings preseason rankings and the previous year's final rankings, I'm going to use last year's numbers as the stabilization values for this year. I want to stress again that they do not affect the final numbers in any way. They are phased out over the first 3 weeks as more on-field data is accumulated.
2) The final Chaos Rankings for a given year are more predictive of future performance than the human polls. This is particularly amazing to me given how much things change year to year in college football. It also somewhat refutes the argument that preseason polls bias the final rankings by forcing teams to overcome or ruin their initial position.
This is an admittedly limited set of data that I'm working with, but I think it's shown what I needed to see regarding my own preseason numbers. Also, please note that it's the strength of the correlation coefficient that's important here, not the value. A value of 0 indicates no correlation while a value close to 1 or -1 indicates a very strong correlation. Anyway, football is only a couple weeks away, so get your excitement all queued up!
*For the overall data set, I used all 120 teams for Chaos Rankings. For AP and USA Today, I used only the teams which were listed on either the preseason or final Top 25. This amounted to roughly 60 teams (half of Div. I-A) in each case based on variance between the lists and the "also receiving votes" teams. Although I only compared half of the teams in the human polls to the entire roster of teams in Chaos Rankings, I feel like the numbers are representative enough to make some general inferences about.
| AP Preseason | 0.433213 | 0.482831 | 0.579719 |
| USA Today Preseason | 0.026721 | 0.433424 | 0.601482 |
| CR Prev. Final | -0.417168 | 0.394527 | 0.64783 |
| CR Preseason | -0.339515 | 0.380719 | 0.628232 |
| AP Preseason | 0.276336 | 0.353627 | 0.408752 |
| USA Today Preseason | 0.283388 | 0.535087 | 0.464797 |
| CR Prev. Final | 0.568829 | 0.386338 | 0.569103 |
| CR Preseason | 0.578748 | 0.366618 | 0.586947 |
Discoveries:
1) There is very little difference in the correlation values for my preseason numbers and the previous season's numbers when compared to my final rankings.
2) In general, the AP and USA Today polls do a better job of predicting the final Top 25 based on the preseason Top 25 (but only barely), but Chaos Rankings beats them (in some cases, quite handily) when predicting the Top 10 and in the overall data set.
Conclusions:
1) I still think it is worthwhile to have some sort of stabilization in the first couple weeks to prevent teams from bouncing around too much. Since I've shown that there isn't much difference between the Chaos Rankings preseason rankings and the previous year's final rankings, I'm going to use last year's numbers as the stabilization values for this year. I want to stress again that they do not affect the final numbers in any way. They are phased out over the first 3 weeks as more on-field data is accumulated.
2) The final Chaos Rankings for a given year are more predictive of future performance than the human polls. This is particularly amazing to me given how much things change year to year in college football. It also somewhat refutes the argument that preseason polls bias the final rankings by forcing teams to overcome or ruin their initial position.
This is an admittedly limited set of data that I'm working with, but I think it's shown what I needed to see regarding my own preseason numbers. Also, please note that it's the strength of the correlation coefficient that's important here, not the value. A value of 0 indicates no correlation while a value close to 1 or -1 indicates a very strong correlation. Anyway, football is only a couple weeks away, so get your excitement all queued up!
*For the overall data set, I used all 120 teams for Chaos Rankings. For AP and USA Today, I used only the teams which were listed on either the preseason or final Top 25. This amounted to roughly 60 teams (half of Div. I-A) in each case based on variance between the lists and the "also receiving votes" teams. Although I only compared half of the teams in the human polls to the entire roster of teams in Chaos Rankings, I feel like the numbers are representative enough to make some general inferences about.
posted by Rev. Adam at
8:20 PM
0 Comments:
Post a Comment
<< Home