In my previous contribution I outlined why Mike Smith was better than Mikko Koskinen using on-ice statistics. On-ice data is not typically used to evaluate goaltending but is often chosen to discuss possession for forwards and defencemen. Measures typically include Corsi, Fenwick, Scoring Chances and by-extension goals. The argument is usually that a higher ratio of shot attempts-for is a proxy measure for possession, and that possession leads to offensive zone time. I used on-ice data to show that the puck was in the defensive end less often when Mike Smith was in net and I attributed this better puck possession to his puck moving skills. My argument was that although Mike Smith had a lower save percentage than Mikko Koskinen, his save percentage was outweighed by his possession style of play and goals-for percentage.
Some readers were not receptive of my approach. Certainly, it was unusual and I was aware that it was having previously written about goaltending. The specific criticism was that I was misattributing data associated with context to performance. Readers felt that Smith’s better possession data was a product of quality of competition, quality of teammates, score effects, as well as other variables related to scheduling. Their interpretation of the data is the standard one in goaltender analytics, and possession data is not usually discussed relative to goaltenders.
While I believe that Koskinen and Smith have faced relatively similar (although not identical) context it’s something that is very difficult to quantify. It’s difficult to quantify if Smith does really have a possession aspect to his game without using machine vision micro-stats from companies like Sportlogic. While the Oilers likely have this data, or could possibly request it, we do not. Leauge-wide measures of possession have not yet been made public and so we still rely on Corsi and its on-ice brothers. We do however, have more direct measures of shot quality.
I have taken a different approach and narrative with this piece. I will use expected goals to show that Koskinen and Smith faced a very different quality of shots, that Smith’s task was much easier than Koskinen’s and that relative to the task Smith was not as good. I will not argue here why Smith faced a much easier quality of shot than Koskinen, although certainly the performance of the goaltenders and context were contributing factors.
Measures of quality
Expected Goals (xG) weighs unblocked shots, considering the location distance, type, and if the shot was a rebound or off of a rush. It is a method of describing shot quality. When looking at expected goals against we can have an idea of the difficulty of the task facing the goaltender and can compare a goalie to the league average. The higher the expected goals against per 60 minutes of play, the better the shot quality faced by a goalie.
The table above shows expected goals for each of the Oilers goaltenders who have played more than 500 minutes 5v5 since Connor McDavid joined the team in 2015.
Mike Smith faced the lowest quality of shot of all goaltenders who have played more than a thousand minutes (xGA/60, 2.18). In contrast, Mikko Koskinen faced the most difficult quality shots this season (xGA/60, 2.58). Smith also allowed the most goals per 60 minutes of 5v5 play (GA60, 2.92) while Koksinen allowed 0.38 less (GA60, 2.54). Smith allowed 0.74 more goals per 60 minutes than expected while Koskinen allowed 0.04 goals less (Delta xGA/60 =GA/60).
Mike Smith faced a lower quality of shot, but was quite a bit worse than an average goalie, and more than a standard deviation above the expected goals-against differential of an average goalie on the Oilers (Delta xGA/60, 0.26). Mikko Koskinen faced a higher quality of shot but relative to his task, he was better than any other goalie of the last five seasons excluding Cam Talbot’s extremely good 2016-2017 marathon (Delta xGA/60, -0.23).
The chart above shows standardized xGA/60, GA/60 and delta xGA/60 measures. The farther the bubble is to right, the more difficult the shot quality faced by the goalie. The higher the bubble the worse the performance. The smaller the bubble, the lower the delta score. Zero indicates the mean while one is equal to a standard deviation. The shot quality faced by Koskinen was more than a standard deviation higher than average while that faced by Smith was more than a standard deviation less, yet Smith’s performance was worse than average, while Koskinen’s was about average.
The charts above from Micah Blake McCurdy’s site show unblocked 5v5 shot locations and expected goals data for Mikko Koskinen’s and Mike Smith’s 2019-2020 season. When compared to the league average, Mike Smith faced fewer expected goals than average (-4%) while Koskinen faced more (+8%). There is a large difference in unblocked shots surrounding the crease area, some of which are attributable to Koskinen’s higher rebound rate.
Koskinen as starter?
During the Oilers’ exhibition debut, Koskinen and Smith split the game. Most reports were that they both played well. Each had some excellent saves and Smith let one in during a power play thanks to some goalie interference from Matthew Tkachuk. That Koskinen started the exhibition game seemed to imply to some that he would be the Oilers’ starting goaltender during the play-in round. During the post-game interviews Smith and Dave Tippett seemed to suggest that both goalies could be used going forward, however, if Smith or Koskinen heats-up they will probably be started most nights.
While this analysis reveals Koskinen to be better at stopping the puck, the previous analysis suggests that the Oilers are better in front of Smith, that Smith faced easier competition, that Smith faced less challenging contextual variables, or that there is legitimately a possession aspect to his game. However one reads the circumstance this much is clear: Koskinen has been better than expected while Smith has not.