By Albert Carreno
Being the head coach of a professional sports team is no easy task. It requires the ability to manage different personalities, make decisions under pressure that will be heavily scrutinized no matter what, instill discipline and better habits in players, and be the scapegoat when the team does not perform up to expectations. As a result of these difficulties, it should come as no surprise that many coaches and assistant coaches are fired throughout the course of a regular season in the MLB, NFL, NHL, NBA, and more. This is especially true in the MLB and NHL as there is no current coach in either league that has kept their job for over 10 years. In the MLB, the longest tenured coach is Craig Counsell, who had been at the helm of the Brewers since 2015 until very recently leaving them to sign with the Chicago Cubs. In the NHL, the longest tenured coach is Jon Cooper of the Tampa Bay Lightning who has been their head coach since the 2013-14 season and has coached them to two Stanley Cups. However, the NHL in particular has seen many coaches fired recently with seven being shown the door over the course of this season, Lindy Ruff of the New Jersey Devils being the most recent. Additionally, five other teams parted ways with their head coaches during the off-season, so all told, about one third of all NHL teams have changed head coaches over the last 6-8 months. It is difficult to believe that coaches are fired so often in this league, but oftentimes a team’s ownership does so for the sake of saving face and not losing money from reduced attendance. Nevertheless, it is often believed that when a new coach is hired in the NHL and most sports in general, the team plays better at first before reverting to how they were playing before the change, an effect called the “dead cat bounce.” In this article, I intend to set the record straight by finding out which coaching factors tend to have the greatest effect on a team’s performance in terms of points percentage under a new coach.
First, we’ll consider whether the offensive and defensive improvement of a team under a new coach is what leads to better results. New coaches cannot often implement changes in how they run their team’s offensive and defensive systems right away. They need to grow accustomed to the team around them and get to learn each player’s strengths and weaknesses. However, after this adjustment period for the players and coach, there is often a change in performance, whether it be for the better or worse as the graphs below show, plotting improvement in points percentage against improvement in goals scored and against per game from old to new coach.
As we can see, there is a clear correlation in offensive and defensive improvement under a new coach and the improvement in a team’s points percentage compared with the old coach. These linear regression models have R^2’s of 0.38 and 0.40, respectively, indicating they provide a decent to good fit for the data and that 38% of the variability in points percentage improvement is explained by offensive improvement, while 40% of the variability in points percentage improvement is explained by defensive improvement. The correlation coefficients for offensive and defensive improvement in these linear models also indicate that for every 0.25 increase in goals scored, there is about a 3.6% improvement in points percentage, while for every 0.25 decrease in goals allowed, there is about a 4% improvement in points percentage. They also each have p-values under a 0.05 significance level, meaning we can conclude these coefficients are statistically significant and that our two variables indeed have a notable impact on points improvement. Overall, we see that improving on both sides of the puck is very important for a new coach to try and do with a slight edge towards defense for a team to win more hockey games.
Next, we will measure the impact of goaltending and general puck possession on a team’s improvement under a new coach. There are certain defensive and offensive systems in the NHL that may improve goaltending and those that may improve how often a team has control of the puck. The systems that improve puck possession often emphasize increased pressure and movement in the offensive zone through more passing, defensemen attacking the net more than normal, and harder forechecking at the expense of being more likely to make mistakes that lead to turnovers and create scoring opportunities for the opposing team. On the other hand, systems that boost goaltending numbers emphasize more conservative offensive play and a greater focus on making safe plays as well as maintaining a consistently rigid structure that makes it hard on the other team to create offensive plays. But do either of these systems lead to greater improvement in points percentage? Let’s have a look at the graphs below, where the first regresses points improvement against Corsi% (a team’s proportion of shot attempts compared to their opposition) while the second looks at points improvement against save percentage.
These graphs seem to indicate a little bit of correlation between improved points percentage as a result of improved goaltending and/or puck possession, but certainly to a lesser extent than overall offensive and defensive performance as we saw in the previous graphs. The trend lines for both of these models appear to be flatter, and the data appears to have more variability rather than follow a distinct pattern. The R^2’s for these models confirm these suspicions coming in at 0.05 and 0.07. Very little of the variability in points percentage improvement is explained by a team’s Corsi% and save percentage under a new coach. These results may signify that having an offensive minded coach over a defensive minded coach or vice versa may not make as large of a difference as most people might expect. The coefficients for these models indicate that a 2.6% improvement in save percentage would be required on average to see a 1% improvement in points percentage, which is astounding, but also demonstrates how little save percentage seems to affect a team’s overall performance. Similarly, about a 0.8% increase in Corsi% is needed on average to see a 1% improvement in points percentage on average. 2.6% and 0.8% may not seem like large numbers, but in terms of save percentage and Corsi%, they indicate major differences that are very difficult to achieve for any team. Additionally, the p-value for the coefficient of the Corsi% model is approximately .09, or above a 0.05 significance level, indicating that we cannot even be positive that Corsi% has a statistically significant effect on points percentage improvement at all. As such, we can safely assume that puck possession and goaltending are not the most important factors to consider when thinking about what is most important in a team’s improvement under a new coach.
The last thing we’ll look at is the importance of a coach’s experience on a team’s improvement in performance. It is often said that the more experienced a coach is, the better they are likely to be simply because they possess more knowledge of the sport and the job than an inexperienced coach does. They are familiar with all aspects of the position and may understand how to handle themselves and manage others better. Does this experience translate to success though? In order to find out, I created a binary logistic regression model where every coach in the dataset I created was assigned either “Yes” or “No” under the “Experienced” column. If a coach had at least 5 years of experience coaching NHL teams before taking on the new team, they were assigned “Yes” and were assigned “No” otherwise. This logistic regression model predicts the probability that a coach is experienced, according to my definition of having experience, based on an arbitrary team’s points percentage improvement. The graph below visualizes the results.
Visually, the results this graph represents concerning the logistic regression between experience and points percentage improvement are really surprising. There seems to be no real pattern at all as the graph shows a lot of volatility rather than an easily discernible pattern we can point to. According to this graph, a coach is most likely to be experienced when their new team’s points percentage drops by 10 percent, which does not seem to make much sense. It also indicates that a coach is very unlikely to be experienced (around 10% chance) when their team’s points percentage improves by 10 percent. Between strange data points like these and the chaotic variance shown, we find that a coach’s experience does not appear to be a key factor in a team’s success and overall improvement. While observing this outcome seems to be odd, there are many examples to back it up. Most recently, Jim Montgomery has overseen a 10.8% increase in points percentage since being hired by the Boston Bruins and led them to a record setting regular season record of 65-12-5 despite being an inexperienced coach.
Still, at the end of the day, this is an admittedly somewhat limited analysis as there are a few other more complex factors to consider that could have a large say in how much a team improves under a new coach. These factors include the amount of talent on the team, the development of prospects under the new coach’s tenure, the changes in lines a coach may make, and how a coach runs their practices and what they do to improve team conditioning/build better habits. These factors are difficult to quantify and would require more work to record as they are not readily available, but could definitely be significant reasons why a team does or does not improve under a new coach. However, we were still able to clearly demonstrate that the most important factors in whether a team improves or does not under a new coach are simple: is the team scoring more and giving up less? We also disproved a long held belief that having experience is of major importance for coaches to achieve success.
Sources
MoneyPuck
Natural Stat Trick
ESPN