Lindy's Five Essential Websites (Non-Major Media) for 2013
[+] Team Summaries

Monday, August 18, 2008

Rectifying a Stupid Conclusion - Preseason Polls

Georgia is #1, but should we care?

This time of year, we often hear about preseason polls and, in response, we hear that preseason pollsters don't know much this early and so preseason polls are just entertainment. One might point out, for example, that only 10 times since the AP started preseason polling (1950) was the final #1 in the top spot before the season (17%) and, more condemning, 6 times the eventual national champion was not ranked in the AP preseason poll.

But to use these numbers to suggest that the preseason poll doesn't mean much is premature and, well, wrong. I used a simple logistic model and data from the AP Poll Archive and found that preseason rankings are more important than you might think.

First, a team in the top spot in the preseason is 29 times more likely to win the national championship than if they weren't in the top spot. To clarify, that doesn't mean that Georgia is 29 times more likely to take it all than USC, but that Georgia is 29 times more likely to win it all than the average college football team. But that shouldn't surprise anyone--of course the Dawgs have a better shot then, say, Wyoming.

But ranking matters even for those at the top. The top dog, no pun intended, is almost 5 times more likely to be #1 at the end of the season than the average ranked team, 2.6 times more likely to achieve that result than other time top 5 teams, and 1.5 times more likely than the #2 team to be on top at season's end. And for the statistically minded, those results are statistically significant.

Finally, I present the results for the most comprehensive model I have tried:

The important numbers for our purposes are the odds ratios, in red, that detail the probability of a team with a particular rank winning the national championship relative to the average unranked team. Teams that start off on top are 200 times more likely to win the national championship than teams that start off unranked, and teams that are #2 at the beginning are 133 times more likely to win it all than the unranked teams, etc.

In other words, preseason polls matter, and they matter a lot--the numbers presented here are large and significant. It's good to be #1.

5 comments:

  1. But we would expect the #1 team to be better than the #2 team, right? We would assume the pollsters know something, however small. So the real question is, removing differences in team quality, does being ranked #1 in the preseason poll help you win the championship? I don't know how to answer that.

    ReplyDelete
  2. I wish I could edit comments.

    Perhaps you could include another variable or two that tried to proxy for team skill -- points for, points against, or pythagorean winning percentage. Then see how important the preseason ranking is.

    ReplyDelete
  3. That's a really good idea - slightly different than what I was going for here - but it would require much more complete data than I now have. Here I just looked exclusively at the predictive power of preseason rankings because I only had the preseason rankings for future national champions (and then I could assume the rankings for the non-national champions because teams would fall in the preseason poll everywhere the future champs weren't (if that makes any sense). That's why I looked only at the probability of winning a national championship.

    To do what you're talking about I would need complete rankings with names attached (and then I could match them to historical performance rankings that I have on hand). I would either need to enter it into excel as a data set by hand (which I'm not willing to do) or find a .csv file online somewhere. But I would be interested in seeing what comes out (my guess is that preseason rankings would have no effect for major conference teams and a moderate effect for mid-major teams).

    ReplyDelete
  4. Good point, and I see what you're saying that it would be slightly different than your question.

    I think what bothers people (and me) about the pre-season ranking is that the voters tend to make the top-ranked teams "sticky." So if the top-ranked teams in the pre-season end up the year with the same number of losses, the #1 team will still be #1.

    I guess I would try to model that by including a term for # of losses. Or # of losses - (least # of losses by any team).

    Not suggesting you do that, since like you said you don't have the data and it would be a pain to get it. Just interesting to think about.

    ReplyDelete
  5. As soon as I can get my hands on the data I'm going to look into it. I agree that polls can be "sticky" sometimes, and I also hate it when Notre Dame and Michigan (as good current examples) start out way too high or jump 27 spots because they win a big game. I can already envision some interesting analyses of poll stickiness and poll love.

    ReplyDelete