Who will win the NHL scoring race?

I was thinking about Jakub Voracek the other day, a player who for the average fan has seemingly come out of nowhere to become the second leading scorer in the National Hockey League, and possibly even the best player on a team that features Claude Giroux. A question hockey fans, whether it’s front office people, journalists, or — maybe most importantly — fantasy hockey players always ponder is whether a player’s early point numbers are indicative of true/improved talent, or just the result of variance. The traditional procedure for the analytic-minded among us is to casually peruse the player’s personal shooting percentage, on-ice shooting percentage, possession numbers, and maybe his overall career numbers. We don’t usually, however, have more than a vague idea of how to put these numbers together to make a true assertion about the sustainability of a player’s point totals.

I think there’s a pretty basic way to come up with a decent variance-independent estimate of the number of points a player will finish a season with, and I call it Percentage-Adjusted Point Projections (PAPP). The metric adjusts a player’s current season point totals based on two very important numbers that are very prone to variance: on-ice shooting percentage and individual points percentage. Most people are familiar with on-ice shooting percentage; it is simply the team’s shooting percentage with the player on the ice in said situation. Not everyone will have heard of IPP, however, and it is in many cases just as important. IPP measures the percentage of a team’s goals with a certain player on the ice that player records a point on. It’s often ignored in the advanced stats community for the simple reason that points are often ignored. Points are generally an overrated metric when it comes to evaluating a player, but there are certainly advantages to having a good idea of how many points a player will finish the season with, and IPP is a vital number to consult when answering that question.

When you break down the components of a player’s point scoring, it is basically just their on-ice shot attempt numbers, their on-ice shooting percentage, and their IPP. By controlling for the variance that tends to skew the latter two factors in small samples, one can get a decent point projection.

How good is this projection? Well no stats site currently has partial-season IPP numbers, so I couldn’t test it for past seasons, but intuitively it makes sense that this works. How well will have to be a subject for later, but for now, I calculated the results for the current top-10 point scorers in the NHL, and here are their projected points, provided they avoid injury and play in all 82 games.

Player PAPP (December 1)

Player PAPP (December 1)
Sidney Crosby 120
Evgeni Malkin 99
Claude Giroux 98
Jakub Voracek 88
Patric Hornqvist 84
Vladimir Tarasenko 83
Steven Stamkos 82
Phil Kessel 81
Tyler Seguin 79
Rick Nash 72

The on-ice shooting percentage and IPP adjustments are based on each player’s career numbers individually at 5v5 and 5v4. Career on-ice shooting percentage numbers are courtesy of War-On-Ice.com and include playoff performance to enlarge the sample size. IPP numbers are pulled from stats.hockeyanalysis.com and, unfortunately, had to be taken as a yearly average rather than a pooled average, with seasons of less than 200 minutes of EV ice time or 100 minutes of PP ice time discarded for those particular situations. I used the regressed point projections for the remainder of the season, adding them to the player’s current points since those aren’t going anywhere.

My only reservations about this data would be the following:

A) Tarasenko’s sample is significantly smaller than the other players, and his career 5v4 on-ice-save percentage is the highest of the group, something I expect is still somewhat variance-driven

B) When you go back several years to increase sample size, you deal with changing environments. Both Seguin and Voracek have seen their PP IPP grow as they’ve gained more importance in their teams’ respective systems over the years.

C) Since the numbers aren’t readily available, I avoided regressing any numbers that weren’t 5v5 or 5v4. That means that variance impacting 4v4, 5v3, empty net, is not taken into account. Crosby and Hornqvist led the group with six non-regressed points each, so their projections may be overshot slightly. Stamkos and Tarasenko had the fewest at two a piece.

Sidney Crosby is good. Barring injury, he will likely win the scoring race.

photo via wiki commons

Leave a Reply

Your email address will not be published. Required fields are marked *