Written by: Sasank Vishnubhatla
On November 17th, Matt Murray was pulled in the first period against the Ottawa Senators. Giving up three goals on ten shots, Murray seemed gassed and not like himself. Shortly after the game, he was placed on injured reserve. His return on December 15th proved to be a valiant one as the Penguins beat the Los Angeles Kings 4-3 in OT. Murray stopped 38 of 41 shots that game which helped spark the Penguins. After a loss to the Anaheim Ducks, the Penguins changed direction and have won their past seven games, Murray being in net for five of them. Many analysts have begun asking: is the Matt Murray who won back-to-back Stanley Cups back? However, I raise a different question: Are both Matt Murray and Casey DeSmith performing at the starting caliber level?
To start, let’s revisit the machine I made in early November. For reference, the code is available on GitHub. The data used in both the previous analysis and this analysis was taken from Corsica Hockey and Hockey Reference.
Before Murray was put on injured reserve, the neural network model determined that Matt Murray was not a starting caliber goaltender. With a save percentage of 89.34% and a goals saved against average (GSAA) of -1.67, the model determined that Matt Murray had a 19.48% chance of being a starting goaltender. Comparing that to Casey DeSmith (93.24% save percentage and 2.23 GSAA), the model stated that he had a 31.00% chance of being a starting goaltender on any given day. When that analysis was done, Murray had started 6 games and DeSmith started 5 games. Now, with this revisit, Murray has started 17 games and DeSmith started 25. This analysis, like the previous one, still has a caveat: Murray hasn’t started 20 games yet. For there to be a statistically significant analysis, both players need to have started at least 30 games. Though that is difficult in the modern NHL, both goalies have a good chance of reaching the 30 game plateau.
With the caveat being said, let’s see how Murray and DeSmith have been playing this season. This season, Murray is 10-5-1 with a 90.8% save percentage and 3.07 goals against average (GAA). DeSmith boasts a 21-12-6 with a 92.6% save percentage and 2.40 GAA. DeSmith’s exceptional play is not just luck. His goals saved against average is a prominent 13.17 while Murray is lacking with a -0.44. However, Murray has started 8 fewer games and was speculatively playing with an injury during the beginning of the season. Murray, after his return against Los Angeles in December has been phenomenal: 6-0-0 with a 95.5% save percentage, a 1.5 GAA and a 39-save shutout. It is safe to say, Matt Murray has been putting on a goaltending clinic everytime he is between the pipes. However, will this increase be seen by the machine learning model?
Before we immediately dive into the machine learning model, I want to do a quick comparison since November. The table below shows Murray’s advanced stats from November and currently. GSAA is goals saved against average, dSv% is delta (adjusted) save percentage, LDSv% is low-danger save percentage, MDSv% is mid-danger save percentage, and HDSv% is high-danger save percentage.
|Matt Murray (November)||-1.67||-0.85||98.39%||90.48%||76.47%|
|Matt Murray (January)||-0.29||-0.06||99.46%||90.09%||78.51%|
|Difference from November to January||+1.38||+0.79||+1.07%||-0.39%||+2.04|
Murray has made major improvement in almost all of his advanced stats. Now let’s take a look at Casey DeSmith’s change in advanced stats.
|Casey DeSmith (November)||2.23||1.51||100.00%||90.00%||83.33%|
|Casey DeSmith (January)||8.76||1.2||97.13%||92.28%||84.38%|
|Difference from November to January||+6.53||-0.31||-2.87%||+2.28%||+1.05%|
Though DeSmith has slightly regressed in his delta (adjusted) save percentage and low-danger save percentage, it would be unfair to not give him credit for performing significantly above expectations. Both Murray and DeSmith have seen sizable growth and an increasing trend in performance.
Now, let us address the neural network model. When initially run in November, the model deemed both Murray and DeSmith backups. However, when we re-run the model with updated January data, these are the results:
|Player||Starter (November)||Starter (January)||Percent Chance as Starter (November)||Percent Chance as Starter (January)|
Both Murray and DeSmith have seen improvement in the model. However, Murray’s success since his return has not “convinced” the model that he has redeemed his starter-hood. That being said, DeSmith’s consistency and stellar play has continued and the model has determined that he is a starter. So, to answer our initial question - no, only Casey DeSmith is performing at starter caliber, but Matt Murray seems to be on the comeback trail.
That being said, there is still the caveat, both Murray and DeSmith have not played 30 games. So, the analysis, though valid, must be taken with slight hesitation due to the fact that both goalies have not played enough. In the end, the Pittsburgh Penguins have determined that Matt Murray is still the starting goaltender, but will happily accept that Casey DeSmith is over-performing, which in turn has pushed Murray to continue improving.