clock menu more-arrow no yes

Filed under:

2015 Passing Project Data Release Volume IV: Goalie Passing Stats

New, comments

The fourth chunk of data that I'm releasing has to do with how goaltenders fared against specific shot sequences. Read on for the details.

Ed Mulholland-USA TODAY Sports

Goaltenders. Voodoo.

We really don’t have great ways to analyze goaltenders. I touched on several methods in this piece when I introduced this concept back in May. I want to quickly touch on two more by Stephen Burtch and Matt Cane. At the end of March, Burtch put together a piece on how GMs overvalue goalies and much of his post is centered on War-on-Ice's HD,MD,LD save percentages. Those are High, Medium, and Low danger zones based on the league's shooting percentages. It's a good read and one that many of you are likely familiar with.

The other article by Matt Cane was written as a follow up to some graphs posted on twitter by Conor Tompkins of Null Hypothesis Hockey. Cane develops a model based on a goaltender's HD Save% from the prior season and includes the current season's LD Shot and MD shot totals. Matt does some great work and I feel he's somewhat under-the-radar, so give him a follow (@Cane_Matt) and visit his site. Same goes for Conor. Both of these articles highlight the importance of shots coming from the high danger zone in front of goal.

So, there's some of the newer pieces on analyzing goaltenders. Now, on to the data!

Through the first four posts that have released various metrics and types of data, I’ve shown that a player’s ability to pass and generate offense is highly repeatable and has significant predictive power. That’s important because it adds another layer when we evaluate players and how they contribute to offense. The second piece highlighted linkup play between players and how they combined to generate offense together. Last week I released the on-ice data for the 262 games we tracked in that fashion.

So, how to use this data for goalies, right? Okay. I spelled everything out in the column headers so this should be one of the easier "new metric" pieces to understand. Basically, you have goalies on one tab and teams on the other. It shows how many shots a goalie/team faced in seven categories: Royal Road, Scoring Chance, One-Timer, Offensive Zone non Scoring Chance Attempts (think of these as weak shots as they are the leftovers from the other categories), Transition shots, shots from Secondary Passes, and shots from a single, primary pass.

You'll then see the save percentage for each category. The total shots, goals, and save percentage (expected and actual) follow these. The expected goals aspect of this is similar to the on-ice post in that we weight each shot based on the likelihood of that particular sequence going in. Column W is where this starts to get fun. This is where you'll see the shot distribution percentage and rates for each goalie's workload. So, if a goalie faces a higher percentage of Royal Road shots than another, we know that those are more difficult shots to stop.

The idea behind using passing data to analyze goalies is simple. As the puck moves, the goalie has to track where it is, the shooting options the pass has created, how far out to come out/stay back, etc. Passing also represents a team asserting their possession and control over another, generating chances and setting up quality attempts. If passing, keeping possession, and generating chances in this manner represents skill on the offensive team, then a goalie's ability to read the play develop and get into position would, in theory, also be considered a measurable skill.

To get an idea of which goalies faced the most difficult shots, the chart below shows the rate at which goalies faced Royal Road shots per sixty minutes. I included goalies that we had at least 400 minutes on. I know this can be a small sample size for some of these goalies, but tracking projects generally deal with small and medium-sized samples. Let's have a look.

RR_Goalie

So, we can see that of the goalies we tracked the most, Karri Ramo was left out to dry more often than the others. At the other end of the spectrum, we see that Carey Price and Devan Dubnyk had an easier time than most. Of course, some of these goalies are only represented across samples of eight, ten, or twenty games.

You'll also notice that Royal Road events really aren't that frequent, what with goalies facing about 1.5 every sixty minutes or so. It's not that much different from their goals against rate every sixty minutes. Royal Road events are just one aspect of shot sequencing that we tracked. When we put it all together and come up with expected and actual save percentages, this is how it looked for each team.

Exp_Actual_PSV

Here, we get an idea of just by each team's defense and shot sequences against, what to expect in terms of save percentage and what the actual save percentage was. Again, some of these teams we don't have extensive data on, and at most we have between forty and forty-six games for the six teams we tracked all season. But, a trend does emerge and we can explain almost sixty percent of a team's save percentage simply by looking at the shot sequencing. If I were to add shot type, distance, angle, etc. to this, it would paint a more comprehensive picture.

However, seeing as how a team plays in front of the goalie has little to do with the goalie himself, and the goalie can suppress shots in a limited way, it would seem to be better to look at how a goalie performs not based on what we expected when he's in the game, but, rather, what we expect from him in relation to the team overall. This not only gives us a larger sample size to work with from the expected side of things, but it's a more honest reflection of how a team defends.

To me, this makes sense on a logical level: this is how the team plays on a consistent basis, so how does the goalie perform in that environment? Let's look at another chart illustrating which teams allowed the most shot attempts generated from secondary passing as well as the least.

Team_A2

Buffalo and Arizona come as no surprise as the opposition walking all over them with the puck. Remember, secondary passing means the team completed multiple passes prior to attempting a shot. Boston, the New York Islanders, and Vancouver round out the worst five. Calgary, Winnipeg, Florida, L.A., and Tampa Bay came out as the best. Remember the sample size (TOI is noted in the data).

Let's take a look at the six teams we have over forty-one games on.

Team_Goalie_ExpGA

Woe unto Justin Peters and Michael Neuvirth.

Here we see the team's expected GA/60 for the entire population of tracked data in orange. How each goalie performed is the blue bar. They are grouped by team, so that's why the orange bars only move six times. What can this tell us? Well, it tells us that, through little fault of his own, Jaroslav Halak had a more difficult time night in and night out based on the team in front of him. He only just performed worse off than expected, but his defense did not offer much help. His backups played a combined twelve games.

Florida, meanwhile, was the stingiest defense in terms of expected goals allowed via passes. However, each one of their goalies performed worse than expected. Chicago was the next best team in passing defense, but, as opposed to Florida, each of their goalies exceeded their expectations and performed quite well. Considering Crawford's contract, would Chicago recognize this and possibly seek to ship him out for cap savings? They let Niemi walk, plugged in Crawford, and won two more titles. Can they do it again?

Now, there is talent evident here as well. The Rangers, Devils, and Capitals were all in the same boat in terms of expected goals against per sixty minutes. Both Cam Talbot and Henrik Lundqvist exceeded expectations, but Hank still gave up nearly 0.25 goals less over sixty minutes. He also did this while the team in front of him played worse than when Talbot was in.

Cory Schneider and Keith Kinkaid

And now we come to Cory and Keith. Based on our data, we would expect the Devils to give up 1.36 goals per sixty minutes of play. The team played slightly worst when Kinkaid was in net, yet the backup gave up fewer goals from passes than Schneider (1.22 to 1.31). Now, I'm not saying having Schneider on the team isn't worth it or that all goalies are the same, but looking at all of this data and the works of others leads to me two likely conclusions on goalies.

Conclusions

1) We don't have correct data and methods for evaluating goalies. Adjusting for location, passing, shot type, etc. (which more work is being done by @DTMAboutHeart), but without data on screens, tips/deflections, and other aspects to what happens in the zone before each shot, we're still somewhat in the dark. This is why I plan on adding a few things to how we track next season, namely screened shots, tips/deflections, and a few others I'm thinking about. This should provide more accurate levels of danger to each shot and goalie skill should shine through. There's so much situational detail we don't have that the analysis feels incomplete despite whatever method you're using.

OR

2) Goalies can only move the needle so much in terms of improving upon a team's expected save percentage that a wise GM should focus more on building the team and how the team performs rather than from the net out. Goalies may be going down the road of NFL Running backs. Previously a team just needed a workhorse, and in some cases it worked. But, as teams continue to win with goalies that aren’t viewed as elite, there has been less of a burden put on the importance of goaltending in some circles. Just as Running Backs are now used in platoons by most NFL teams, teams may not feel the need to pay a goalie six or seven million dollars because they can pick up a Devan Dubynk or a Ben Bishop and not worry about whose in goal. If the team has a sound defensive system.

This can be as simple as having forwards close down on one-timers better than others as one-timers are more dangerous than standard shots from outside the house. It could be identifying defensemen and centers who are better at defending the house and Royal Road. Or, identifying players that break up passes in transition. In fact, the more work I do on this, the more I think that this level of data, coupled with zone exits and zone entries, can ultimately quantify aspects of various hockey systems and can offer more insight into that aspect of the game.

So, I still think there's work to be done in evaluating goalies, but part of does wonder if, in the long run, focusing more on the shot sequences a team gives up and improving on that aspect would have equal benefit than focusing on goaltenders. Perhaps they go hand-in-hand, but I don't believe that building from the net out is a worthwhile strategy if your ultimate goal is to win a title. I simply don't think that there's all that much that separates a goalie you can win with and a goalie you can't. This isn't new by any means, but the more work that is done on goalies with the data we have, I don't think

Up Next

There's still more data to release, but I'm calling a time out on that for a while. Next piece you see from me will be all about the Devils. It will less metric-based and more descriptive. I feel that's something isn't given enough time and analysis. How and why things happen are just as important as what happened in the end. First up will be a look at defensemen pairings and how each performed throughout the season. Fire off suggestions for things you'd like to see once the draft and free agency die down and we get into the dog days of hockey news.