By Kurt Bullard
After Pittsburgh Penguins starting goalie Matt Murray allowed four goals in the first two periods of Game 4 against the Tampa Bay Lightning, head coach Mike Sullivan brought in Marc-Andre Fleury to start the third period. Fleury didn’t allow a goal on the seven shots he faced in the last twenty minutes while the Penguins almost eliminated a four-goal deficit behind Phil Kessel and Evgeni Malkin, ultimately falling 4-3.
After the game, most of the post-game conversation—both in the Penguins’ presser and the NBC Sports’ post-game show—was about who Sullivan would throw in net for a pivotal Game 5 matchup. Such talk seemed like an overreaction to a below-average performance from Murray. Fleury, who does not have the most pristine playoff history himself, only made seven saves in the entire frame. Murray let up four goals—one of which was on the man-advantage—on 30 shots, which is not a horrific performance, especially against the Lightning.
Most of the rationale for proposing the switch seemed to center around Fleury, and how he would be able to carry his solid performance back to Pittsburgh. This was being said even though Murray had a higher save percentage and lower goals allowed average in the regular season than did Marc-Andre, even though Fleury did play more games.
It’s pretty simple to check whether or not there actually was a “momentum” factor for Marc-Andre this year in his goaltending performances. If the last game’s performance affected his next game, then regressing the current game’s save percentage against the previous game’s save percentage should yield a significant coefficient for the last game. That is to say, knowing how Fleury performed in the last game would affect your estimate of how he’d play in the next game more so than just using his season average save rate.
I ran the regression on the 45 games in which Fleury had played the previous game, because this”momentum” probably did not exist after a benching. Doing so yields the below regression coefficients, which shows that the previous game’s performance does not have an effect on Fleury’s next game.
I also ran another regression, with this one adding the days in between as a predictor to see if you could tease out “momentum” by looking at the days in between starts.
Again, this shows that there was no effect from the previous game for Fleury this season. While obviously one would want to look at film to supplement this decision-making, it appears that the best predictor of how Fleury performs is simply his season save percentage, which, again, was lower than Murray’s this season.
Goalies can have bad games, and there’s nothing you can really do about it. It happens. But coaches do have the power to not make rash decisions. Seven saves shouldn’t get you a starting nod in the conference finals, especially when your career playoff save percentage is a meager .906. This may have been Fleury’s team at the start of the season, but Murray is the captain now, and one bad day in the crease shouldn’t lead to a mutiny.