I think everyone is familiar with the Madden Curse: players who get put on the cover of the Madden video game suffer horrible injuries or generally suck the season that the game comes out. You can see some examples here. I think at this point that everyone is pretty happy to say that there isn’t really a voodoo hex placed on the guys who are put on the Madden cover; the curse actually comes from a combination of luck in the previous season and regression to the mean. To put up the kind of season that gets you on the cover of Madden, you had to be lucky: not get injured, be on a team that does well, be in a position to make some big plays and get showy numbers like interceptions or touchdowns (you also have to be good, but to be the ‘best’ you need luck on top of skill). Since much of this luck is out of your control as a player, it is unlikely that you’ll be as lucky next year, and so you’re bound to lose some production on that front and have a worse year. On top of this you have regression to the mean, which is the idea that if you play really well one year, you’re likely to play closer to your actual ability level the next year. This is especially relevant to football, where players are inconsistent; in basketball, players are more consistent and there doesn’t appear to be the NBA equivalent of the Madden curse.
What does this have to do with decision making? There are obvious applications to free agency; you should try to pick a guy with the best overall body of work over a guy who just had a great season (at least in football). For my purposes, it doomed my predictions this past weekend. I have been presenting two models, Mario and Luigi. Luigi is similar to a model I used with success last year, but I developed Mario this summer and it seemed to do better by some measures in picking against the spread. So I started this year with Mario as my go-to guy for the spread. Then he did poorly in the first few weeks, and I decided to switch back to Luigi, who was doing better. Either of these models, however, will be inconsistent; they will both likely get about 55-60% of the games correct against the spread, so they will fluctuate above and below that level in a given week and thus be profitable or not in various weeks. Since Mario had been doing poorly and Luigi well, I decided to switch to Luigi and thus doomed myself to the Madden Curse: Luigi regressed to the mean and had a bad week. Here’s the post mortem:
Over/under: 4 games went at -105 for the over. Mario took the over on 3 of them and was correct on 2; it was wrong on the one where it picked the under. So on those four games Mario is 2-2 for -$9.50. Luigi was wrong on the same under game and right on the same two correct over games, but also correctly picked under on the fourth game, so on these four it went 3-1 for $177.40. In the remaining ten games at -110, Mario went 5-5 and Luigi went 3-7. So overall this week Mario was 7-7 for -$55 and Luigi was 6-8 for -$250. On the season they are 33-40 for -$900 (Mario) and 37-36 for -$130. Obviously one of my offseason projects will be to try and see if there’s a restricted set of games where the over/under predictions do better (perhaps picking games where the prediction is a certain number of points away from the over/under).
Moneyline: Mario liked Chicago, Green Bay, Tampa, Kansas City, Denver, and Oakland. These all missed, with 3 coming in heart-breaking fashion for their teams (Green Bay in overtime, Kansas City on the Andre Johnson-fueled comeback, and Denver on the iffy pass interference call). So Mario was 0-6 for -$600. Luigi liked Seattle instead of Chicago and additionally picked Detroit, St. Louis, and Washington. The Rams and Seahawks came through to make Luigi 2-7 for -$195. On the season Mario is 11-12 for $775 and Luigi is 13-14 for $1215.
Spread: The NE-Baltimore game was a push. Mario would have skipped the Chicago-Seattle, TB-NO, and Minnesota-Dallas games for being too close, and went 2-8 in the other games. On the season that moves Mario to 29-37-2. Luigi only skipped the Minnesota-Dallas game, so it went 5-7-1. On the season, Luigi is 34-29-2. So Luigi was still the better choice of the two and is doing well overall; hopefully next week will ‘regress’ back up to a winning set of predictions.
Bill Simmons: With Bill’s lines, the Washington-Indy game was also a push. Bill went 6-6-2 against his lines to move to 46-39-5 on the season. Mario was 4-8-2 (it switched to Seattle with Bill’s line and was correct on one game it would have skipped) and is 31-40-3 on the season. Luigi was 5-7-2 and is 38-33-3 on the year. My own picks, which have become a conglomeration of a separate model for week 1, Mario for weeks 2 to 5, and Luigi for week 6, have a record of 40-45-5. Being six games behind Bill isn’t great, but I would be even if I had been using Luigi the whole way. I think I’ll catch up; Bill will likely start getting distracted by the NBA season and I don’t think he can go above average in the long run anyway (although maybe I shouldn’t be so confident; he was 17 games over last year and a similar number the year before).