How Much Does Balanced Handedness Matter for Defensive Pairs?
A fresh look at an old question
Defensive pairing handedness has become a hot topic in the NHL over the past decade. Last trade deadline, Chris Tanev was one of the most sought after defensemen available on the trade market. Tanev is a very good player, but a large part of why his demand was so high was that he shoots the puck on the right side of his body. When Tanev opted to sign with the Toronto Maple Leafs in free agency, the Stars signed Matt Dumba and Ilya Lyubushkin as replacements, giving them three right handed defensemen for the 2024-25 season along with Nils Lundkvist.
As far as I can remember, this obsession with defenseman handedness has its roots in former NHL head coach Mike Babcock’s preference for having three balanced pairings in his lineup; something that got a lot of attention ahead of the 2014 Winter Olympics. In 2016, Dominic Galamini looked at the impact of handedness on predicting relative to teammate corsi differentials and came to the conclusion that a balanced pairing was worth nearly seven additional corsi events per 60 minutes played compared to an unbalanced pairing. Two years later, Tyler Dellow mentioned Galamini’s research in an interview with MIT Sloan in which he wondered if teams had gone too far chasing balanced handedness to the detriment of their overall defense. Despite Dellow’s theory, the NHL has not grown less obsessed with right shooting defensemen in the intervening years.
In this article, I will re-examine the impact of balanced handedness on defensive pairings using modern metrics and attempt to answer the question, does balanced handedness matter for defensive pairings, and if so, how much does it matter?
Methodology
For this project, I borrowed techniques from two previous articles I’ve written. First, I used a similar method to my article on Wyatt Johnston’s performance on the Stars’ top line to identify defensive pairings. The criteria for a defensive pairing being included in this analysis was that they had played at least 200 seconds together in a game with three forwards and two defensemen on the ice for both teams. 200 seconds was chosen as the cutoff with the idea that it would cover five shifts of 40 seconds, a little bit shorter than an average shift length. This would ensure that this analysis covers defensive pairings that were intentionally put together by the coaching staff.
Next, using a similar method as my article analyzing the impact of skating speed, I constructed a game log for each pairing of total time on ice, corsi and expected goals for and against, venue, average goal differential, quality of the pairing using Evolving-Hockey’s xGAR components for even strength rates for and against and quality for and against, average quality of teammates using xGAR components, and average quality of competition using xGAR components, plus an indicator for whether the pairing had balanced handedness. With those game logs, I ran a series of multiple regressions using variously the pairing’s corsi for per 60, corsi against per 60, expected goals for per 60, and expected goals against per 60 for each game, with and without the handedness variable to compare error and impact on R-squared.
Results
First, let’s look at what percentage of defensemen play of their off-side each season, recreating a chart from Galamini’s original article looking at this question.
Galamini wrote his article near the end of the 2015-16 NHL season and included data from that nearly completed season in his analysis. Interestingly, beginning with the following season, the average time played by defensemen on their off-side has basically stabilized around 20%.1 This suggests that NHL teams had reached an equilibrium where the perceived advantage of playing a righty with a lefty was no longer greater than the disadvantage of their playing ability. This also provides a convenient cutoff point for this analysis. We can look at how balanced pairings performed both before and after reaching this talent equilibrium to see if there is a significant change in their effectiveness.
Starting with corsi rates, there is a clear difference in effectiveness between pairings with balanced between a right shot and a left shot and pairings with two defensemen who shoot the same side.
The difference is particularly stark before the 2016-17 season. Balanced pairings bother generated more corsi per 60 minutes and allowed less than pairings with a defenseman playing on their off-side. In fact, overall, off-side pairings had a negative differential per 60, while unbalanced pairings had a negative differential. This has changed a bit in the eight years since the league reach equilibrium, however. Balanced pairings still generate more offense, but after 2015-16, there’s no difference defensively between balanced and unbalanced pairings. While there still appears to be an advantage to having balanced pairings, the difference is much smaller in the current NHL.
Repeating the same exercise for expected goals reveals a similar pattern, though that is a bit deceiving.
Once again, balanced pairings appear to perform better than unbalanced pairings. However, the relative margins between the or percent share of expected goals is smaller than the corresponding corsi differentials. In fact, from 2016-17 to present, balanced pairings actually concede slightly more expected goals against than unbalanced pairings, though the offensive advantage is still enough to give them a positive differential. This means that most of the difference between pairings is driven by shot volume and that average shot quality is actually worse for balanced handed pairings than unbalanced.
Using my comparative regression method for the entire sample, I found that balanced handedness was statistically significant for explaining only for corsi for rates and expected goals against rates. However, the difference was only 0.001 R-squared for corsi for and even less than that for expected goals against. I didn’t feel like it was worth it to create a chart to show such a small difference. However, the difference between the coefficients from 2009-16 and 2016-24 displays some meaningful trends.
Prior to achieving equilibrium, having a balanced pairing had a statistically significant positive impact on a pairing’s corsi for rates, above what would be expected based on teammates, competition, score, and venue of roughly 0.7 corsi per 60 minutes of play. However, there was a corresponding statistically significant negative impact on expected goals for when accounting for all of the above, plus corsi for rates of a little less than -0.03 expected goals per 60 minutes. Over that time period, teams averaged roughly 0.04 expected goals per corsi, so the net impact was essentially a wash.
Interestingly, since 2016-17, balanced pairings have had a statistically significant impact on expected goals against as well and it’s a positive one, meaning a higher expected goals against for balanced pairings once situational factors are accounted for.
It’s worth noting for all of these coefficients, however, that they are all small values equating to less than one corsi for or against per hour and less than 0.04 expected goals. Although some are statistically significant, none of them rise to being meaningful in most cases. Last season Colton Parayko led the league in 5-on-5 time on ice with 1,615 minutes played, or about 27 hours. Over the course of the entire season, the difference between him playing on a balanced pairing versus an unbalanced pairing would be roughly eight corsi events and a little less than one expected goal.2
Discussion
So what’s going on? How did Dominic Galamini’s research show a significant impact to handedness while this analysis is much more inconclusive?
Well, one obvious thing that stands out to me is that Galamini used corsi relative to teammates as his predictor and target variables. This was a reasonable approach in 2016, but since then the hockey analytics community has developed more sophisticated measures of individual contributions and/or talent that isolate for impact of teammates, competition, zone starts, and various other situational factors. It is possible that Galamini’s analysis was affected by some of these factors, particularly teammates if team’s prioritized making their top pairings balanced over their bottom pairings meaning balanced pairings could have significantly stronger teammates.3
Another thing that Galamini’s analysis doesn’t account for is whether the defensemen played the previous two seasons on their off-side or with balanced partner. Given the lower numbers of right shot defensemen, righties almost always play on a balanced pairing, meaning they are almost always benefitting from playing on their strong side, while lefties are the only skaters affected by playing on their off-side. In other words, righties should not expect a “bump” when playing on a balanced pairing, as they had presumably already been doing that. When I attempted to replicate Galamini’s results using a pairing’s combined xGAR components as the target variable instead of relative to teammate corsi and included variables for how much previous time had been spent on a balanced pairing, I was still unable to find a significant impact for handedness in predicting their future components.
It seems likely to me that the effect of handedness is adequately captured in Evolving-Hockey’s xGAR. I’ve used the word “equilibrium” throughout this article, and it really does seem like the NHL has naturally found the balance where the benefits of balanced pairings equal the downsides to playing potentially weaker players. If teams had swung too far in chasing handedness, as Dellow suggested, I think we would have seen the trend reverse. Instead, I think a balance has been achieved and analysts can take a defenseman’s xGAR at face value, rather than making adjustments for whether they played on their strongside or off-side.
Conclusion
Balanced handedness for pairings may have some impact situationally, as some players may really struggle on their off-side,4 but in the aggregate, the effect is minimal. Even when the impact is statistically significant, the overall impact on shot quantity and quality is minimal, even over the course of a full season. Instead of focusing on handedness, it would be better for NHL teams to look at the overall impact of a player to determine their ability and value in a lineup.
There is still an overall downward trend from 2016-17 to 2023-24, particularly after the Seattle expansion for some reason, but it is not statistically significant.
Rough figures, I multiplied 27 hours by -0.3 corsi differential (sum coefficients from 2016-17 onward) and 0.035 expected goal differential.
Anecdotally this seems to be the case, but determining which pairing is the first, second, or third in a game can be difficult with line-shuffling and situational deployments. I would like to return to this as some point in the future, though.
Ryan Suter, for instance, supposedly refused to play on his off-side.
Another good one buddy!
What interests me is the fascinating ability for humans to understand what is in front of them. I can guarantee a statistical analysis was not done to determine what the optimal rate of balanced to unbalanced pairings is, the league just found it on their own. It reminds me of baseball, where players used to enjoy a benefit of 20 points of OPS by taking the first pitch, and adjusted leaguewide by swinging at the first pitch much less to the point where first pitch swing rate is unchanging and will remain unchanged unless there is a structural break (rule change, etc.). It reminds me of an experiment done in economics that showed that poor people (who must think about money every day) are just as good at being hedge fund managers as the actual hedge fund managers. Humans are just good at learning. It's fascinating.
I also believe you've touched another key point here, and another one that's constant in the field of economics, so I run into it a lot. There is a 'statistically significant' result, as in p < 0.05, and then there is a statistically significant result, as in something that would actually mean anything to anybody. Surely you know this. I was thinking about precisely that when you showed your chart showing point estimates that the benefit was less than one shot attempt per game. I was going to ask about the value of a shot attempt, but you pre-empted me by answering that question. Sign of a good article I guess.
I'm surprised that what the league has deemed the efficient amount of off-hand pairings is only 20%. I suppose I bought the propaganda. I thought there would be more non balanced pairings than that. I suppose the question I ought to ask now is that do you think being right handed is a bankable skill? Is it worth paying for? If you took the exact same defenceman, and switched which way he holds his stick, would that change his contract value to you Aaron?