Point totals have a very strong influence in how we evaluate forwards. The goal and assist columns in a stat sheet are very frequently referenced, and the plays where those stats are accumulated weigh heavily on our eye test and our memo...
Point totals have a very strong influence in how we evaluate forwards. The goal and assist columns in a stat sheet are very frequently referenced, and the plays where those stats are accumulated weigh heavily on our eye test and our memory.
But we don't talk much about what drives point scoring.
I've written about how out-shooting leads to winning at the team level, but at the individual level we know some players shoot for a much higher percentage than others. So is the difference in shot quality larger than the difference in shot rate at the individual level? What really leads to high point totals?
In the plots below, we look at the correlation between a player's point scoring rate and either (top plot) his team's shot rate with him on the ice or (bottom plot) his team's shooting percentage with him on the ice. A strong correlation would mean that having a high value for one would lead to a high value for the other and the points would fall near a straight line.
It's clear from these plots that in any given season, a player's point scoring rate is much more heavily driven by the shooting percentages than the shot rates. Both have an effect, but the correlation in the lower plot is much stronger (R^2 of 0.67 vs 0.37).
But this doesn't really settle things yet. In any given season, shooting percentage might be more important, but what if shooting percentage varies a lot? Below are two plots looking at how well shooting percentage in one year predicts shooting percentage or point scoring in the following year.
It turns out that shooting percentages aren't very repeatable after all; these correlation plots look more like blobs than lines.
Essentially, the variance in shooting percentage is a much bigger factor than either the shot rate or the talent component of shooting percentage. So the team's shooting percentage with a given player on the ice is a strong driver of his point total in any given year, but it turns out to be a pretty poor predictor of what will happen next year. Even though shot rate has less impact on point scoring, its greater repeatability (R^2 of 0.36 vs 0.11) actually makes it more useful for predicting next year's point totals.
As we move to larger sample sizes, the variance will start to wash out a bit and we can get a better indication of a player's talent. Still, shot rate remains both more repeatable and more predictive of future point scoring:
Predictive of points
Shot rate (1 yr)
Shooting percentage (1 yr)
Shot rate (2 yr)
Shooting percentage (2 yr)
Shot rate (3 yr)
Shooting percentage (3 yr)
Up to now, I've framed this as an either/or choice, but it really isn't. It's interesting to note that shot rate appears to be a larger factor than shooting percentage, but both factors do contribute, and we should use both factors when trying to project a player's future scoring.
However, it is critical to bear in mind that a single year's shooting percentage data is almost entirely noise. When we set expectations for a player, we need to look at multiple years of shooting percentage data and remember that whatever happened last month or even last year is pretty close to meaningless.
Don't fret too much about how low Hartnell's on-ice shooting percentage was this year or get too excited about how high Voracek's was. If a rookie like Nazem Kadri or Jordan Eberle has one of the very highest on-ice shooting percentages in recent years, it's a better guess that they won't keep it up than that they're destined for sure-fire superstardom.
Point totals are a reasonable way to measure a player's offensive contributions. But we need to keep in mind that they're not only driven by usage (ice time, power play time, and situational deployment); they're also heavily driven by random blips in shooting percentage, blips that don't tell us much at all about what will happen in the futur