The majority of hockey analytics focus on even-strength play, which makes sense considering it constitutes the majority of time on ice. But that often leads to the neglect of other seminal aspects of the game that are far from simple noise. Special teams are a good example of this, and the power play in particular strikes me as an area of the game that is ripe for further analysis. I am fascinated with power plays, in particular, because they are the aspect of the game that can most be shaped by coaching creativity. Watching the Washington Capitals up close for the last two years, I have found that when a power play clicks, there are few more beautiful aspects of the game to behold.
There are complications in comparing players because the units tend to stay fairly constant, so isolating individual player impact can take a large sample of games, and over that time the environment can shift — new systems, changing skillsets, shifting personnel — making the analysis less relevant. But on a macro level, figuring out just which power plays are good and which are not is something that is very do-able and relevant.
And this analysis has been conducted in the past. It started with Tore Purde, who found early on that Fenwick For per 60 minutes is the best predictor of future power play goal rates. Two years later, Patrick D. of Fear The Fin expanded on the analysis, with similar results. It has now been another two years, though, and thanks to War-On-Ice we have a new metric to add to the fold, Scoring Chances. While the sample size with SCF/60 will be smaller than simple shot attempts, I felt that since FF/60 was a better predictor of future success according to past studies than CF/60, scoring chances might perform even better.
Considering Patrick’s finding that the same metrics came out on top no matter whether 10, 20, or 40 games into the season, I chose to restrict my analysis to having the first half of the season predict the second half. My cut off date was January first. So the question became, from 2008-09 (the first year scoring chance statistics are reliable) to 2013-14, which metrics were most repeatable and which correlated the strongest with goals for per 60 minutes with the man advantage.
This chart shows the percentage of the variance in each metric for the second half of the season that is accounted for by that same metric in the first half (R2). For example, in 2013-14, each team’s Corsi For per 60 in the first half of the season accounted for on-average approximately 53 percent of the variance in that team’s Corsi For per 60 in the second half. It is pretty clear that, due to the large sample, a team’s shot attempt rate is the most repeatable among power play statistics.
So looking at the past five years, how do each of these statistics in the first correlate to a team’s goal scoring rate in the second half?
It looks like a pretty fun ski jump (well unless you land right on 2011-12), but these results are quite striking. I re-checked my numbers three times to ensure that the difficulties with War-On-Ice weren’t affecting these R2 values. Up until 2011-12, some combination of CF/60, FF/60 and SF/60 appeared to best predict future success, but the last two full seasons have seen no such predictive ability. Only SCF/60, last season, even accounted for 10 percent of the variance in the latter half’s goal rates. So what can we take from this? At this point, not a whole lot other than a general skepticism. Be weary of putting too much weight into power play shot rates, because for some reason they have recently not appeared as predictive as previously advertised.