Bill Simmons had some time off due to his suspension from ESPN, but returned with his picks for tomorrow’s games (which I’ll have in my usual picks post later). But aside from golfing, apparently Simmons spent some time thinking about if you could maybe use something better than a team’s record as a measure of how good they are. He gives an example:
For instance, Arizona battled back in Week 1 to beat San Diego 18-17 on Monday night — the Cardinals got the “win” and San Diego got the “loss,” but really, that was an “either/or” game, right? For gambling purposes, there had to be a better way to capture that.
And so what Bill does is re-categorize each game for each team as a convincing win, a convincing loss, or in between. And he titles this an ‘epiphany’. What Bill has actually finally done is take a step towards using point differential instead of win/loss outcome.
The idea works like this: wins and losses just tells you which team came down on the right side of 0 (if it’s Carolina and Cincinnati, the answer is either no one or everyone). But it drops a lot of information. Did the winning team just win by 1? Maybe they had to make a comeback or held on at the very end. Or did they win comfortably by 14? That seems like it could give you some indication of team quality. This is essentially what Simmons is doing. Did the Broncos win by a lot? Let’s give them a ‘real’ win. Did they lose by a lot? That’s a ‘real’ loss. Anything else is a close call. Going through this process changes Buffalo from a 4-3 team to a 1-2-4 team, which isn’t really that different (they look average either way), and New England from a 5-2 team to a 3-1-3 team, which is maybe a bit weaker (the Colts, who are also 5-2, are 5-1-1 under Simmons’ scoring, which is more impressive).
But is Simmons really using point differential? He’s still doing some hard categorization after all. And then his rankings don’t follow his own record very well; he has the 2-2-2 Seahawks ahead of the 4-1-2 Lions, for example, so maybe he’s also using strength of schedule or something. To check, I took Bill’s rankings (1-32) and matched them up with each team’s actual record (as win percentage), their average point differential, and pro-football reference’s simple rating system (SRS), which is a combination of point differential and strength of schedule. Then I found the correlation between each of those and Bill’s ranking.
Simmons’ ranking correlates at .888 with win percentage. So it’s a little different, but not too much. This isn’t necessarily too surprising though; the ‘adjusted’ record isn’t going to be that far off from actual records. Teams that win comfortably will have a good record, after all. But his rankings correlate at .921 with point differential. There are only the 32 teams and there isn’t a lot more room to increase from .888, but that seems to show that Simmons’ ranking does capture a bit more of point differential than it does outright record. And it correlates in the middle, at .904, with SRS, suggesting that he isn’t using strength of schedule on top of point differential (otherwise the correlation should be .921 or stronger).
Now, I never want to complain about someone, especially someone like Simmons with a big casual audience, using ‘advanced stats’. But this is, of course, ridiculous. I’ve been using point differential (how much a team has outscored its opponents on average) since I started this blog because it’s widely accepted that point differential is a better predictor of future performance than record. Simmons’ ‘illegitimate son’ Bill Barnwell uses point differential; here are two articles from Simmons’ Grantland site. This first one credits Darly Morey with bringing the Pythagorean Theorem to football in the 90s, after it has been used by Bill James in baseball even before that. Bill Simmons knows what point differential is.
I’ve complained about Simmons and his use/choice of stats before; I thought I had a whole article on the Book of Basketball, but all I can find right now is a reference to it at the end of this post. In short, Simmons justifies all of his decisions – ranking basketball players, ranking NFL teams, picking who will cover a game – on statistics. It’s just that he tends to use crappy ones, like points per game and MVP votes for NBA players or win/loss record for NFL teams. His ‘epiphany’, after a 15-ish year writing career and a solid month off with nothing to do, is that maybe it isn’t just a win or loss that tells us about a team but also how they won or lost. I hope that this was just a framing device for his picks article, because Simmons has a lot of experience with people being disingenuous recently. It strikes me that he might be doing the same thing here.