For a better browsing experience, please upgrade your browser.

FiveThirtyEight

Ahead of Wednesday night’s Chicago Blackhawks-Los Angeles Kings game, I wrote about the “hottest” goalies of the Stanley Cup playoffs to date. No. 1 on the list? Chicago’s Corey Crawford, who promptly had the worst game of his playoff career, allowing five goals on 30 shots in the Kings’ runaway victory.

It probably seems self-serving not to merely chalk this up to a bad prediction (granted, I did note that “today’s hot goalie isn’t guaranteed to be a success tomorrow”) or even a jinx of some kind. But it might be more instructive if taken as an example of a few interesting philosophical concepts in sports.

First, there’s the impact of randomness on performance. Goaltending statistics are incredibly noisy in the small sample of a game. It takes 3,000 shots faced for save percentage to be half-skill and half-luck; Crawford faced 1/100th of that amount in Game 2. Even the best prediction isn’t going to be very accurate in a given game when it goes up against that kind of volatility.

Also, because of the outsize role of random chance, a goaltender’s numbers are a good illustration for the gulf between predictive and “retrodictive” metrics, which can also be framed as a tug-of-war between ability and value. A statistic that places its emphasis on value will reward past performance, regardless of whether that performance was driven by luck or skill. So when a mediocre goalie steals a game against a good team, he gets full credit for that performance in a retrodictive metric such as our “hotness” statistic — even if he’s unlikely to repeat it. But a predictive stat will not give extra credit for a fluky performance, beyond using the evidence from that performance to (slightly) update its expectations.

Finally, some superstitious FiveThirtyEight readers may think I “stat-cursed” Crawford by anointing him the hottest goaltender of the 2014 postseason. There’s a long tradition of athletes and teams sustaining declines after being singled out for achievements. But in these kinds of cases, regression to the mean is the more likely culprit. To appear on the cover of the “Madden NFL” video game or Sports Illustrated, a player had to play at an incredibly high level, and was usually aided by luck (which includes staying healthy). When that luck dissipates, it seems there’s a curse attached to the accolade.

This is more true for the hottest goalie list, because I set up that metric to find players who were playing above a level that could be explained by their previous performance baselines and even the shooting skill of the opposing team. Whatever’s left over is, by definition, going to be fueled largely by luck, and therefore primed for regression.

In fairness to Blackhawks fans, regression rarely comes as abruptly (or as far in the opposite direction) as it did for Crawford on Wednesday night. Predictors who forecasted Crawford to allow five goals (if there were any) would have been engaging in the gambler’s fallacy, thinking he was “due” for bad luck to offset his previous fortune. In reality, luck is random. And the interplay between luck and skill is what makes sports interesting, especially in the high-stakes setting of the NHL’s conference finals.

comments Add Comment

Filed under , , , , , ,

Powered by WordPress.com VIP