UPDATE about tonight’s game. Everyone enjoys power rankings, which is to say everyone enjoys complaining about power rankings. How can My Favorite Team be 5th? It’s obvious that Team #25 would beat Team #20, come on. Team #3 already beat Team #2, so you’re obviously wrong (actually true in this case, except for the me being wrong part). Well, here’s another set for you to be upset about.
The methodology is the same as my power rankings in the past. I take the current stats for each team in the league and multiply them by the weights for my Luigi win percentage model. I do that separately for offensively-controlled stats (which includes fumbles and interceptions thrown) and defensively-controlled stats; it isn’t a perfect separation, but it’s kind of a division between done-by and done-to. I don’t adjust that number at all, so it’s sort of hard to interpret, by which I mean it isn’t from 0 to 100 or anything like that. Generally you want to be high on both offense and defense (closer to 0 on defense, as it turns out). I also create an average score for both for comparison; this score is the average across all teams for the current season. Teams are ordered by their total score, which is just the sum of the offense and defense scores. Behold!
Somewhat unsurprisingly, the Texans are the number 1 team in the league. They’re undefeated, have been handling their opponents pretty easily, and generally look fairly impressive (I saw an article on ESPN asking if they could go undefeated already, with a predictable answer of ‘no’). How should you interpret the numbers? Basically any team higher in the ranking would be expected to beat a team lower in the ranking. The closer two teams are in overall ranking, the closer the game would be. If two teams are close, putting the lower team at home could easily be enough to swing them ahead (hey, that might be relevant for that Pats-Ravens issue).
A couple other things to note: after four weeks the season is a quarter over, but there are still only 5 possible records (ignoring the Colts and Steelers with their early bye) teams could have, from 0 to 4 wins. As you might expect, there are a lot of 2-2 teams. But not all 2-2 records are equal. For example, the Patriots are number 2 in the rankings. Yes, they have two losses, but both were close and to good teams while their two victories were pretty convincing. Considering both losses came down to last-second field goals, the Patriots could easily also be 4-0; more importantly, the Patriots have generally produced the stats of a good team. At the other end, the Rams are also 2-2. But they have been outscored overall (their wins being by 6 and 3, their losses by 4 and 17) and tend to have trouble moving the ball and stopping opponents. Weighting their relevant stats, they come out below average on both offense and defense and are currently 29th overall.
If you’d like another set of rankings to compare to, I’d recommend Brian‘s at AdvancedNFLStats.com. We use pretty similar models, but still manage to come out with some differences.
Speaking of the Rams, let’s do the pick for tonight’s game real quick. The spread is Rams +2.5, the moneyline is +117/-127 and the over/under is 39.5. Luigi has the Rams losing by less than a point, so would take the Rams to cover. It does like the over though. Yoshi 1 has the Rams winning by 4, so would take them and the over. Yoshi 2 also likes the over and has the Rams winning by a little under a point, so would pick them to cover. The models also agree that the Rams are worth taking to win outright.
The last thing to do is look deep into the future. I can have Luigi predict the winner of each game going forward using the win probability numbers. I then add up the probabilities to get predicted wins for each team (although I grant them what happened in week 1, since I need it for data and don’t predict it). Here’s the table, and I’ll explain a bit more for the uninitiated below.
So we’ve got a list of teams and four columns. The first one is just how many games they’ve won so far. The second is how many games that team would be expected to win if they played the season over again (although, again, granting them the result from week 1). To get that, I just add up Luigi’s win probability for every game for that team. The column next to that, EV so far, is that number but only up through the current results (week 4). People of a certain statistical bent like to compare that column to the actual win column to determine how lucky a team is. The idea is, if a team has actually won a lot more games than it was expected to, it is probably taking advantage of events that aren’t likely to continue happening, like amazing special teams plays or recovering a lot of fumbles. The final column is expected wins in future games, so the last two columns add up to the predicted total. If you want a guess at a team’s final record, add the future wins to the actual current wins.
How about a few examples? Unsurprisingly for the number 1 team in the power rankings, the Texans have the highest expected wins. This doesn’t always happen because of differences in strength of schedule, but there’s a solid correspondence. Houston is 4-0, but their win probabilities say they should have only been expected to win 2 out of their last 3 games. They were favorites in every game, but favorites don’t always win. So Houston has been ‘lucky’ to the tune of 1 win so far. Atlanta and Cincy have also been fortunate. On the other side, Detroit has been fairly unfortunate. They’ve only won 1 game even though they’ve played like a team that should have 3 wins. The league’s two winless teams, the Saints and Browns, have also been a little unlucky.
The remaining wins column is obviously speculative. We’ve only seen four games from most teams, so we have some idea of how teams play but we don’t have an amazing picture. Teams could also suffer injuries or make other changes that affect the rest of the season. But if you believe them, you can eventually start making predictions about who will make the playoffs (it’s kind of fruitless to do this before week 12 or 13, and even then races are too close to be really specific). For example, Houston should get 4 (current wins) + 8.5 (future wins) = 12 or 13 wins by the end of the season. Baltimore should have about 10, given what we know right now, the Patriots should have 9, and Denver should have 9 or 10. Those are your projected AFC division winners as things currently stand. But there are obviously races involved; San Diego projects to 9 wins, Buffalo to 8 or 9, the Bengals to around 10, etc. An unexpected win or loss or a change in performance can change things quickly, especially this early in the season.
The power rankings and season predictions will come out every week from here on to the playoffs, so get your complaining out of the way now!