First off: thanks for the feedback. I'd be lying if I said I felt confident about this top 25, but even with that said, I'm not sure I have any better ideas at this point. It's tough to make sweeping generalizations about a season absent any data from said season, so let's just do this and hope for the best, right?
With that being said: let's talk stats! I'm thinking we'll go with a largely manual ranking until Week 3 or so, possibly Week 4. That's partially due to not having enough data in the first couple of weeks to do much beyond a bunch of manual overrides (plus we keep Will's duty as the Head-to-Head Police Chief to a minimum), and that's partially due to the compressed schedule. (Also, that gives me a few weeks to figure out how to use this thing.)
In previous years, BlogPoll was submitted on Wednesdays, so we'd have a few days to hash things out, confirm manual overrides, and generally speaking take care of things. This year, the schedule's compressed; the final version needs to be in by Monday morning at 8 - which for me means that I basically want the final version by about 9 PM on Sunday. I'll do my best to get that out in time, but to do that I'll need your help.
We'll have a minimum of time to figure things out after the preliminary poll (which will be out on Sunday now), so overall tweaks to the BP mechanism need to happen during the week. That's where y'all come in - below the jump, I'll outline our components, or at least the ones we used last year. As always, comments and suggestions are welcome (and probably needed).
- Win/Loss (weight: 30/81, or 37%). Relatively straightforward; in a perfect world I'd like to find a balance that doesn't use W/L, but that's the "pitcher wins are a horrible statistic" part of me. W/L is probably a pretty good predictor of team success, no?
- Strength of Schedule (weight: 10/81, or 12%). Right now this is a universally applied addition; I'd rather this be incorporated as a scaling device with all the other metrics, but I don't have a good way to do that yet. (Hence, MATLAB.) Until I figure that out, this can be left as is.
- Pass Efficiency Defense (weight: 9/81, or 11%). And now we're into the correlation metrics. This is a measure of how good your pass defense is; it incorporates opponent completion percentage, interception percentage, and total passing yards allowed per game. I'd like to add deflections to this, or normalize for opponents' passing in other games (i.e., holding last year's Georgia Tech team to a 50% completion percentage wasn't an accomplishment).
- Rushing Defense (weight: 8/81, or 10%). See above. However, fumbles forced aren't factored into this, which I'd like to do if possible.
- 3rd Down Offense (weight: 7/81, or 9%). This is strictly a raw conversion metric; I don't know if 3rd down and distance data is even available, but in a perfect world, I'd like to weight conversions on 3rd and 3 or less more, and penalize teams for getting stuck in 3rd and long constantly. I'm not sure how to do this.
- Total Defense (weight: 6/81, or 7%). If we have pass and rush D, why not have total D? This seems like it's already incorporated; I'd be open to dropping this out in place of something (rush offense?) at a lower weight, and bumping everything else up.
- Pass Efficiency Offense (weight: 5/81, or 6%). See PED above, but turn it on its head. Same things apply.
- Opponent Points Per Game (weight: 3/81, or 4%). Again, we're back in the defensive metrics. I'd almost rather incorporate an average margin of victory type of stat, or something like that in addition to this, but in essence this is an adjusted scoring average.
- Turnovers (weight: 2/81, or 2%). Not to put too fine a point on it, but this isn't really an indicator of anything that's a repeatable skill (especially with fumbles recovered, aka a 22-person coin flip). I'd rather roll this into the running / passing metrics or just jettison it entirely.
- 3rd Down Defense (weight: 1/81, or 1%). See the 3rd down offense metrics, except I'd like to give a bonus for forcing 3rd and long.
Those numbers add up to 100%, I hope. If we want to experiment with lower-weighted different metrics, I'd be up for that, provided I can figure out how to do it. Some things I'm thinking of:
- Rush Offense. Self-explanatory, with a heavy focus on average yards per rush.
- Percent Score. This is similar to MOV-type metrics, but can be a good aggregate, I think - provided there are controls to avoid 3-2 being more dominant than 41-28.
- Net Punting. I don't know what this looks like quite yet, but some kind of field position metric should be useful, I'd think, even with a low weight. We can call this the Tressel Factor (pre-2010).
- Big Plays. Not sure how I feel about this, but I think this is more of a retrodictive indicator than a prospective indicator. Low weight if it goes in.
Any additional thoughts on those metrics would be greatly appreciate, or if there's another metric that's worth including.