Which matters more to a team’s overall success? The strength of its penalty killing unit or its power play?
The question was posed by NLL stat expert Adam Levi on Twitter this Thursday. The comments and replies that flooded into the query offered up a series of interesting theories and possibilities. Could the number of short-handed goals a team allows be a better predictor of a team’s winning percentage than either it’s PK or PP units? What about short-handed goals scored? How about the ratio of short-handed goals scored to a team’s power play goals allowed?
Truly there are numerous interesting directions this question can lead. But instead of spending time wondering what the relationship of each might look like anecdotally, we set out to answer the question quantitatively. This post will answer the question posed by Levi.
First, we pooled all public NLL statistics from 2005 to 2020 to create a sample size of 166 teams. From there, we overlayed each team’s special teams numbers on their overall records and winning percentages. You can see the full assortment of information below.
With the data sample in hand, figuring out the relationships between various special teams categories and winning percentages was fairly simple. In this case, the relationships we were looking for would be expressed by a standard correlation coefficient. In math and science, the correlation coefficient is called an “r-value”, meaning that we would describe each correlation coefficient as “r = [coefficient]”. Below you can see the r-values of six different variables as they relate to win percentage. As a control, we’ve included goal differential, which clearly has a strong correlation with winning.
Those numbers are cool, but if you don’t have much experience with correlation coefficients, I’ll take a moment to explain how to interpret what you see. It’s actually super simple and easy to do.
An r-value of 1 is considered a “perfect correlation”, which is something that exists more in theory than in the real world. Inversely, an r-value of 0 is considered to have no correlation whatsoever. Similarly, we seldom see situations where r = 0 in real life. In the real world, correlations are almost never of the “all or nothing” variety. There is nuance to each relationship.
Generally, we consider any r-value above 0.7 as having a “strong positive correlation” whereas an r-value of 0.5 to 0.7 is considered a “moderate positive correlation”, and an r-value of 0.3 to 0.5 is considered a “weak or mild positive correlation”. An r-value from 0 to 0.3 is not considered to have a significant correlation. This same breakdown applies to negative r-values, the difference being the direction of the correlation.
Take Goal Differential as an example. The r-value above says that goal differential and win percentage have a correlation coefficient of 0.885, which we would consider to be a high correlation. As a team’s goal differential goes up, so does its win percentage. In an alternate universe where a rising goal differential translated to a lower win percentage, it would be expressed as a negative r-value (-0.885). But seeing as how only one of the r-values we found is negative, don’t spend too much bandwidth concerning yourself with negative r-values.
Looking to our r-value chart again and applying the breakdown laid out above, we see two categories bearing weak correlations, another pair bearing little to no correlation, and one bearing a moderate correlation.
These numbers suggest that a team’s power play percentage has a slightly stronger correlation to winning than its penalty kill does. Still, neither offers a particularly strong correlation in general. Additionally, we see that short-handed goals both for and against don’t carry any real correlation to winning at all.
But one area that does have a moderate-to-strong correlation is a team’s “Net Special Teams Goals”, which is described as the following basic formula:
PowerPlay Goals Scored + Short-Handed Goals Scored – PowerPlay Goals Allowed – Short-Handed Goals Allowed
A team’s Net Special Teams Goals is just the overall net of goals scored while a team is on either the power play or penalty kill. It’s an overall total of goals scored during special teams situations, excluding 6-on-5 scenarios. We see that Net STG has an r-value of 0.555, which translates to a moderate positive correlation with win percentage. This is the answer to our question. Individually, penalty killing, power plays, and short-handed goals bear no correlation with winning. It isn’t until we consider them collectively that a true correlation emerges.
When you think about it for a second, this probably shouldn’t be a surprising conclusion. A team with a positive Net STG is probably going to have a positive overall goal differential, which we know correlates strongly with winning.
So which one is it? Power play or penalty kill? Technically, the answer is that power play success (r = 0.372) correlates to winning more strongly than the penalty kill (r = 0.333) does. But really, neither carry all that much weight individually. Elite teams are good on both ends of the floor usually, and special teams is no exception to that pattern, which is why the only viable correlation belongs to Net Special Teams Goals.
One thought on “Which Matters More in the NLL: Penalty Killing or Power Plays?”