WASHINGTON - OCTOBER 13: Jon Sim #16 of the New York Islanders reaches to get the puck out of his own zone during a penalty kill in the first period against the Washington Capitals at the Verizon Center on October 13 2010 in Washington DC. (Photo by Bruce Bennett/Getty Images)
Here's Stu: "Special teams trends. NHL.com tracks season-long power play and penalty killing stats. But as on the standings page where you can find a team’s record for its past 10 games, how about its PP and PK stats for its last five and/or 10 games to show if there’s any movement away from the season-long average?"
Bill James once wrote something to the effect of "if you're going to divide A by B in your statistic, you'd better have a reason to do it." So if you're going to take a statistic that's derived in limited play, like power-play or penalty-kill percentage, and slice it even finer, it would be good to know that you're going to get a statistically-significant conclusion out of your new statistic.
I went back to 2003 and found the top 2000 and bottom 2000 PP and PK streaks over 5- and 10-game stretches. I then looked at performance over the next 5- or 10-game stretch. Here are the results for the top teams:
|Top 2000||Previous Streak||Next Streak|
|5 Game PP||33.6||18.7|
|5 Game PK||96.1||82.6|
|10 Game PP||29.3||18.8|
|10 Game PK||92.3||82.7|
League-wide averages are 17.7% for the PP (or 82.3% for the PK). As you can see, these teams that were absolutely on massive hot streaks crashed almost all the way to league average over the next five games, with the PK regressing even more than the PP. What about the bottom 2000?
|Bottom 2000||Previous Streak||Next Streak|
|5 Game PP||4.4||16.7|
|5 Game PK||65.9||81.6|
|10 Game PP||7.8||16.6|
|10 Game PK||70.8||81.5|
This doesn't look substantially different - teams that were absolutely awful over a 5- or 10-game stretch regressed upwards almost all the way to the same results as teams that had just dominated their opponents. Here's the regression to the mean overall - 100% means full regression and no evidence of talent:
|Streak||Bottom 2000||Top 2000|
|5 Game PP||92.7||93.6|
|5 Game PK||95.5||97.8|
|10 Game PP||88.6||90.2|
|10 Game PK||93.1||96.4|
So roughly 95% of what we see over a 5-game stretch is not retained (and 92% over a 10-game stretch.) If you want to know how your team is going to do going forward, you'd be well-advised to ignore these kinds of numbers.
Verdict: there is no value to the NHL keeping statistics like this.