About a month ago, we looked at a breakdown of Bryz’s save percentage in two minute buckets. The intent was to measure how “clutch” Bryz was by evaluating his save percentage at key points in the game through the first 41 Flyers’ games. What we saw was, at least at a high level, a stark difference between his performance in wins and losses.
Looking back on the previous season, though, there was an improvement in Bryz’s game once he was handed the reins after the Winter Classic. Thus, there was a necessity to complete the analysis and see what the final 41 games would do to the early season data. The effect, put simply, is pretty apparent. The negative issues noted in the previous analysis are smoothed out. Adding more data to a sample size can effectively account for the smoothing, but the data itself also shows some improvement in performance. I would say his increasing comfort level in both this city and his role led to some of these improvements. To highlight these differences, I’m going to recycle a few graphs of half-season data and contrast them with the full data set. Let’s start with Bryz’s bucketed save percentage in Flyers wins.
This graph is nearly the same one as seen in the August 9th article, except the scale has been adjusted. Notice some key aspects, including the slight upward trend in save percentage as the game went on, as well as the relatively high volatility in his performance (standard deviation of 8%). The graph below is the same format, but for the entire season. It stands to simple reason that we should expect some reduction in variability in performance as Bryz better establishes his norms and we add more than a handful of data points to the analysis, but what’s interesting is that we also see a pretty solid uptick in performance, as mentioned earlier. First, the graph:
First period performance is still relatively up-and-down, showing a standard deviation of over 5%. The fluctuation for Bryz’s save percentage over the entire game is a little less and his third period performance in wins really improved over the whole year, especially in the waning minutes of the game. While not any sort of conclusive evidence, statistically, it does point to his being more reliable in tight situations where a lead needed to be held.
As noted above, Bryz’s overall performance did get better when the whole year is evaluated. The table below illustrates this:
Again, tracking the OT save percentage in wins is done for completeness, not utility, as you can’t win an OT game without saving all of the shots. Still, while his performance was somewhat inconsistent through 41 games, it really leveled out by the end of the season. His save percentage by period in wins was rather consistent.
Now to flip it and reverse it, let’s revisit his performance in losses.
Same graph as seen on August 9th, again with an adjusted scale. The key takeaways here for me were the erratic first period performance, the high variability (nearly 16% standard deviation) and the very poor end-of-game performance. Looking at the same graph over the whole year, though, shows a much more optimistic picture, at least when remembering that these were all losses.
While there’s still a good amount of fluctuation in performance, it’s nearly 5% less on a standard deviation basis. This reduced variability is, in my view, attributable to more consistent performance across the board, as his save percentage did increase in these games. The pronounced periods of poor performance, such as the end of the first and third periods, are not as pronounced with a full season’s worth of data.
Comparing the actual numbers from the first half and the full year, that aforementioned reduction in inconsistency is very apparent.
The late-game save percentage in both the third and overtime periods is much better. While an 85-88% won’t cut it, the trend was in the right direction even in losses: better, more consistent goaltending.
So, Bryz’s performance in wins is still statistically much better than in losses and he wasn’t able to completely eradicate the late-game issues, at least when looking at the season averages, but he was able to right the ship and show real, measurable improvement. This is good news, as is the fact that his ability to seal a game away late seemed to be a real thing when looking at the third period for the full-year data.
Circling all the way back to square one, though, I still feel like I haven’t fully addressed the question of how to measure “clutchness.” This evaluation is a nice visual tool for me, and it’s one that’s helped sparked some ideas. I think the next step is to step up the analysis and explicitly define the situations that make a clutch goaltender rather than looking for trends. For example, the idea of save percentage by two-minute buckets in wins and losses works at a high level, but if the Flyers are up by two or three or four goals, do I really consider that clutch? Probably not. It’s essential that it stay a two, three or four goal lead, no doubt, but my initial intent was to put some strong quantification to this idea.
Where this will be going in my next article is to look at Bryz’s bucketed save percentage as a direct function of goal differential, broken down into five categories:
- Leading by more than 1 (Lead > 1)
- Leading by 1 (Lead = 1)
- Tied (Tied)
- Losing by 1 (Trail = 1)
- Losing by more than 1 (Trail > 1)
My expectation is that, when bucketed save percentage is separated into these five categories, we’ll get a lot more direct insight that will help to make more sense of the potential trends seen above. Until that article gets posted, though, I think it is fair to reiterate that there are effectively two Bryzes in the same body. The losses aren’t all Bryz’s fault, but he’s not posting numbers that I would consider close to his winning numbers in losing efforts. While there’s certainly a team effect, I find it hard to believe that the same roster could consistently perform so vastly different between wins and losses that Bryz’s save percentage would regularly be about 5% lower in the losses. Hopefully, as the data is more usefully sliced and diced, we’ll be able to garner a little bit more each time around.
For now, thanks for reading, and please feel free to comment with any questions, issues or errors!