Reviewing Lundqvist’s Year-to-Date Stats, January Edition

I looked at Lundqvist’s year-to-date stats back in late November and found that his play this year has been statistically weaker than every season since 2007-08 (which was the first season I could get the necessary data). I have added in the last five weeks of data and re-configured my graphs to make them a bit more reader friendly:

f8ndk55

 

Above is Henrik Lundqvist’s year-to-date 5v5 save percentage since 2007. This time around, I have removed all of the data from previous seasons and condensed them into three descriptive stats: minimum, average, and maximum. Continue reading “Reviewing Lundqvist’s Year-to-Date Stats, January Edition”

Advertisements

Lundqvist’s year-to-date stats compared to previous years.

I looked at how Henrik Lundqvist’s play (a) has compared on a year-to-year basis and (b) how it tends to progress on a year-to-date basis. I used 5v5 Sv% and expected goals saved above average per 60 mins (xGSAA/60) as my two metrics. Both sets of data came from Corsica.

I chose 5v5 Sv% because it tends to eliminate a lot of team effects. All situations Sv% can be drastically affected by special teams based on the systems play that a team employs (as well as the efficacy at pulling it off). Overall, 5v5 Sv% tends to map decently year-to-year so it’s a good metric to look at as a measure of skill.

jzsquch

I chose xGSAA/60 because it helps address the issue of shot quality. xGSAA/60 basically looks at the types and distances of the shots taken and determines how many goals a hypothetical league average goalie would give up that night (ie, expected goals). By comparing this expected goals number against Lundqvist’s actual goals allowed number, we get an idea of how he performed against a league average baseline. The “per 60 mins” part helps to normalize for time in net. This stat is also only measuring 5v5 play.

y5pyssa

I wanted to emphasize Lundqvist’s play this year against previous years as a composite so I chose colors that aren’t well differentiated for previous years (ie, 2007-16). However, if there is interest in having all of the years differentiated I can fulfill that request.

My observations:

  1. Lundqvist isn’t doing well this year. His 5v5 Sv% is among the lowest of his career for this point in the season and his xGSAA/60 through yesterday is the worst of his career. There is some teeth to the idea that he is not playing at the same level as before.
  2. His xGSAA/60 this year is very close to 0, suggesting that he is overall playing at a league average level during 5v5 play.
  3. The idea that he is a slow starter is only true for the last few years. Earlier in his career he had solid starts to the season. There’s also a noticeable dip in almost every year for December, which is likely him cooling off from what looks like to be a lot of hot Novembers (just not this year).
  4. His two best years by both stats appear to be 2015-16 and 2012-13. One year he received no votes for the Vezina while the other year he finished 2nd in voting. The Vezina voters really aren’t good at their job.

Honestly, I think it’s very fair to say we are seeing a below average performance from Hank relative to what we expect from him. This year he is facing a pretty normal mix of low-, medium-, and high-danger shots. It’s not like last year where the very high average shot quality he faced gave him average traditional stats and amazing underlying stats. Given the fact he’s coming off such a stellar season, I have a lot of hope that he can bounce back and play at a really high level this year.

However, going by the eye test I am at least a bit worried. There definitely have been games where he has let in total stinkers. And not just one stinker in one game, but a few stinkers in a few games each. But he still has long stretches where he plays solidly and makes great saves. Nonetheless, I would say there is not anything in particular that I see as an explanation for his poor play. No apparent injury or major changes in style of play. As far as I can tell, he’s more or less just playing more sloppy than usual.

Something I might investigate in the future is his progressive win threshold % (WT%) and loss threshold % (LT%). WT% is basically a measure of how often a goalie “steals games” and LT% measures how often they let in so many goals that the team in front is very unlikely to overcome it. Both are based on per game expected goals and goals against numbers. And at least for last year, he was absolutely elite by those numbers. One of the best in the NHL for both stats.

A Detailed Analysis of the Brassard for Zibanejad Trade

by @Chris_Beardy

The Rangers and Senators made a very noteworthy trade on June 18, 2016:

In this post I will evaluate this trade from the perspective of a New York Rangers fan. I plan to evaluate the following aspects of this trade:

  1. Statistics-based performance review
  2. Statistics-based usage review
  3. Brief “eye test” statements
  4. Contract, cap, and asset considerations
  5. Concluding remarks

1. Statistics-Based Performance Review

Unless noted otherwise, all of the charts to follow were constructed using 2014-16 data for 5v5 situations with zone-, score-, and venue adjustments where applicable. The data has been sourced from Corsica. I will frequently use stats that say “/60”, which means the stat is adjusted to “per 60 minutes of time on ice.” This addresses any difference in time on ice by these players.

Here are some basic player data just to set the stage:

Brassard Zibanejad
Age 28.8 23.3
Height 6’1″ 6’2″
Weight 205 lbs 220 lbs
Shoots L R
GP 241 230 ( + 6 AHL)

Continue reading “A Detailed Analysis of the Brassard for Zibanejad Trade”

Estimating Arbitration Salaries Using JT Miller’s Contract Extension

By @Chris_Beardy

J.T. Miller became the first of four arbitration-bound New York Rangers to ink a deal. On July 13th, he signed a 2-yr, $2.75m AAV contract, leaving him just one year shy of UFA status. Many have proclaimed this contract is a steal for the Rangers, but I think it is right on point. I also believe it will provide a benchmark for many of the other players lined up for arbitration right now. Thus, I have used Miller’s contract to develop a simple cap hit prediction model for those other players.

Miller is being paid at market value.

@HockeyStatMiner wrote a great article for Blueshirt Banter in May that predicted:

Continue reading “Estimating Arbitration Salaries Using JT Miller’s Contract Extension”

Current Accrued Cap Advantage by Team

Above, in the featured image, you can see what I believe is an accurate set of data about teams with cap advantage currently on their books. All of the points represent the level of cap advantage currently accrued by the off season of any given year. For example, the Chicago Blackhawks have $31.6m in cap advantage on their books at the time of the 2016 off season. This number will be true until the first day of the season (Oct. 12), when the AAV and salary numbers stick ticking up for the 2016-17 season. I have tried graphing the correct day-to-day changes of the contracts, but it just comes out overly complicated. It’s only really important when either trades or contract terminations occur mid-season; that makes up only about 25% of events that affect cap advantage accrual to date.

You can find an interactive version of this chart here.

If there is a problem (or praise) you’d like to share with me about this different service I am trying out, please contact me at @Chris_Beardy on Twitter, by comment below, or even on Reddit at /u/ChocolateAlmondFudge.

Do the Rangers play worse offensively in front of Lundqvist than his backups?

For at least the last two years, the idea has been floated around  that the New York Rangers play two different games: one when Lundqvist is in the net and one when he isn’t. That idea goes on to posit that the New York Rangers who show up when Lundqvist is off the ice is a much better team. Those Rangers create more offense knowing that they can’t just grind out low scoring, one goal games, and they play a tighter defensive game knowing that the King can’t just bail them out when they misstep. Overall, they play a better game of hockey without their safety net.

In this article I intend to begin looking into the notion that a team plays differently depending on who the goalie in net is. I will do so by examining something less directly tied to the goalie position. I will be looking at the GF60 and CF60 rates of three separate teams split based on whether their franchise goalie is in the net or not.

The Narrative Behind it All

rangers-v-carolina-hurricanes
Lundqvist being attended to by the Rangers’ medical trainer after being struck in the throat by a shot.

This idea isn’t without some cursory evidence. The Rangers were without Lundqvist for a sizable stretch in the 2014-15 season. A puck to the neck caused Hank to miss all of the Rangers’ games between Feb. 4th and March 26th, which was a 26-game stretch. Cam Talbot went on to shoulder much of the weight as the interim starter and ultimately he put up a great 0.926 performance. However, an early story line was the unexpected goal support from the Rangers for Talbot. From Feb. 10th to Feb. 22nd, the Rangers logged 33 goals in 7 games. And while they came back to earth shortly thereafter, the goal support still seemed to be there more for their goaltenders who aren’t named Hank. Talbot and MacKenzie stood in net for 38 games where the Rangers put up 3.04 GF60. Lundqvist received the slightly lower goal support of 2.89 GF60 over the course of 46 appearances. That came out to an extra 5.2% goals for Talbot and MacKenzie, which is an extra goal every 7 or so games.

Less Narrative, More Data and Graphs

I began by pulling 5v5 TOI, GF, and CF data for the New York Rangers’ goalies’ individual seasons from 2010-16, excluding 2012-13 (because the sample size is small for the lockout year). The data for guys besides Lundqvist were merged into a composite for each year.

5sqpjni

The differences between the totals over these five seasons ended up being much larger than I expected. The Rangers scored an extra 0.23 GF60 in 5v5 play and generated 2.4 CF60 extra in 5v5 play when Lundqvist was on the bench or otherwise not playing. And in fact, the GF60 and CF60 data on a year-by-year basis was higher for the backups in 4 of the 5 seasons. The 5-year TOI totals were 16,202 min with Lundqvist and 5,758 min without. On a year-to-year basis, the TOI ranged from 2,225 to 3,088 min with Lundqvist and 743 to 1,785 min without. So the samples are decent in size.

Now, I do think this is compelling evidence that for some reason the Rangers are generally performing slightly better without Lundqvist in net. However, it is not possible to discern a reason from this data. Perhaps the players actually are more motivated to generate offense because they do not have their safety net. Perhaps the team’s coaches have made measurable changes to their personnel choices, such as decisions that are meant to try and keep the puck in the attacking zone. Perhaps the goalies themselves have significant contributions to the production, such as through stick handling.

To try and learn more about this apparent phenomenon, I investigated whether this situation has arisen on other teams. I chose to look at Nashville with Pekka Rinne and Chicago with Corey Crawford. Nashville, I thought, would provide a close parallel to the Rangers as both teams has generally been centered on an elite workhouse goaltender with a below average offensive team to support him. Chicago was to provide a stark contrast where I felt that Crawford was only depended upon to consistently deliver acceptable results and was infrequently leaned on to steal games. The Blackhawks tend to depend more on their offensive capabilities than the Rangers or Predators do, meaning that goaltending need not be as prized in Chicago.

ac3uj9e

ptjxndl

The results were quite contrary to my expectations. The Predators, who seemingly would need to emphasize offense more without their star goaltender in net instead have mostly faltered in Rinne’s absence. Chicago, on the other hand, ramped up their production when they didn’t have their starter in net. It seems that the Blackhawks seek to capitalize more on their scoring opportunities to help support their backups.

Below is another look at the data. These graphs give direct year-by-year comparisons of the data for all three franchises with and without their starting goalie:

mwe5xuz

n1bx3fj

The GF60 data seems really noteworthy. For all three teams, 4 of the 5 dots fall on one side of the line. In the case of the Rangers and Blackhawks, they fall above the line, where offense is up when the starter is out. The Predators see a dip in production on all but one year when Rinne is off the ice.

The CF60 data mirrors the trend for the Rangers and Predators. However, the Blackhawks see their CF60 drop in 3 of the 5 years without Crawford in the net. This is starkly different than what was seen in the GF60 data.

Closing Statements

I intend to investigate this idea further with other teams that have had a consistent starter across the past six-year period (e.g., Pittsburgh, Los Angeles, Dallas, etc). I will avoid teams that have significantly switched their starters in that period to try and avoid adding yet another set of variables into this analysis. Even someone like Lehtonen will be pushing it as he has effectively gone from being the #1 goalie in Dallas to a 1A/1B goalie this past year.

I know the Rangers better than I do any other team in the league so they would be the best team for me to dig deeper on this matter. I have a suspicion that personnel choices by the coaches may be a significant driver of what was observed for the Rangers. However, failing that I can also look into “puck luck” in those games. It is possible that what we’re seeing is just an aberration by pure chance.

Ultimately I think something like this could be an important part of understanding how a major roster change could affect a team in indirect ways. There could be an argument made that moving Lundqvist for a slightly above goalie could be a better change than would be expected than just by the GA60 and salary cap impacts. There is evidence that that a slight GF60 bump could occur, which would in part mask a rise in GA60.

Quick Look: Do Diving and Embellishment Calls Correlate with Rates of Penalties Taken and Drawn?

Thanks to the 2015-16 Diving/Embellishment List on Scouting the Refs and the easily accessible data on Corsica, I made a quick graph of how 5v5 player penalty rates break down depending on whether the player had either 0, 1, 2, or 3 diving/embellishment calls in 2015-16 or had been fined by the league for the same offenses.

moqusz2

po8arov Continue reading “Quick Look: Do Diving and Embellishment Calls Correlate with Rates of Penalties Taken and Drawn?”

Penalty Effects: Brad Marchand and How He Effects the Bruins’ Special Teams Stats

Brad Marchand, despite being one of the league’s biggest pests, is a highly skilled two-way winger who can play in all situations. He is both one of the best power play weapons for the Bruins and a key part of their penalty kill. So it is interesting to consider his prowess for getting calls and putting his team on the man advantage as well as for being called and sitting in the box while the Bruins play shorthanded. I plan to estimate the net  “penalty effects” on the Bruins from Marchand’s penalty taking and drawing abilities.

Marchand and Penalties

Marchand has often been both the among the leading penalty takers and penalty drawers for the Bruins. His penalties taken (PF) numbers have been top three on the team in 2010-11 (3rd), 2013-14 (1st), 2014-15 (1st), and 2015-16 (t-1st). In the penalties drawn (PA) category, he has had top three finishes on the team in 2010-11 (1st), 2011-12 (1st), 2013-14 (2nd), 2014-15 (1st), and 2015-16 (1st). This has led to a large spread in how Marchand effects the amount of time that the Bruins special teams receive over the season: Continue reading “Penalty Effects: Brad Marchand and How He Effects the Bruins’ Special Teams Stats”

Pitfalls to Avoid When Evaluating Players with Turnover Stats

Takeaways and giveaways are two turnover stats officially recorded by the NHL at every game. However much like other “real-time stats” like hits and blocked shots, there are issues to discuss before we can consider lending any credibility to the numbers.

Defining Takeaways and Giveaways

Perhaps the most concerning thing about takeaways and giveaways is that there is no definition for them from the NHL. The stats page does nothing to describe what either of these stats actually are. Worse yet there is no glossary or list of definitions on the entire NHL.com site that describes exactly what a takeaway or giveaway is. I asked Eric Hornick, the statistician for Islanders home broadcasts since 1982, if league had ever defined what giveaways and takeaways are:

After extensively using my google-fu, I was able to turn up a lone 2013 article from Dave Mishkin, radio broadcaster for the Lightning, that offered a definition. However, this text does not appear elsewhere on NHL.com and seems to originate from a Columbus Dispatch article in 2011. Nonetheless, Mishkin states:

Giveaways/Takeaways: Here are the league definitions: A giveaway occurs when a player’s own actions and decision making results in the loss of team possession of the puck.

A takeaway occurs when a defensive player causes a turnover and takes possession of the puck or when a defensive player makes a definitive effort to intercept a pass attempt and takes possession of the puck.

I would argue that this definition does not help us at all. It’s probably what most of us intuited before reading it. And that leads us to core problem of these two stats: They are highly subjective. While there are certainly going to be giveaways that are clearly giveaways (such as passing the puck right onto an opponent’s stick) and takeaways that are obviously takeaways (like lifting someone’s stick to take the puck), there are some times where it is less clear if a turnover was a bad offensive play or a good defensive play.

While I may sound like I am splitting hairs here, the subjective nature of these stats have led to unreliable record keeping. There have been many articles over the years describing “home rink bias” for turnovers, hits, blocked shots, and even more well-defined stats like shots on goal and assists. Any use of home stats for any player is going to introduce home rink bias into the statistics and make them rather suspicious. Below is a chart of home vs. away turnover stats for Colorado during the 2013-14 and 2014-15 seasons. The stats are for all situations and only for players with 80+ total games.

col_base stats

While the giveaway stats look fairly reasonable, it is exceedingly clear that Colorado’s home rink statistician is tracking takeaways in a way that deviates from the rest of the league’s statisticians. For this reason, I would advise the use of away stats only or the use of adjustment coefficients for home data.

(Note: While it is possible that the Avalanche’s statistician is recording takeaways in a way that is biased for the home team, we cannot say that for sure from this data alone. If his record of takeaways by Colorado’s opponents is similar, then the issue is instead a deviation of how this statistician defines takeaways relative to the rest of the league. That investigation would be another post for another day.)

Situational Bias

The turnover stats reported by the NHL are for all situations – even strength, power play, and shorthanded. In addition, these stats are reported as base counts, completely ignoring the influence of varying time on ice across players. I was actually inspired to look into this topic yesterday while reading the NHL Arbitrators blog yesterday on finding comparable players to Kucherov for estimating his next contract value. It was a great read and I really like the blog concept and approach. However, the author made use of turnover base counts from the NHL when making one of his comparisons:

For instance, over the last two [full seasons], O’Reilly had a takeaway-giveaway ratio of 181-59 while Stepan’s was 96-69.

In the course of my research, I found that this difference largely disappeared when using away stats and switching to rate-based statistics:

Player Home GV / 60 Home TK / 60 Away GV / 60 Away TK / 60
Ryan O’Reilly 1.22 4.70 1.00 2.07
Derek Stepan 1.99 2.17 1.06 2.08

The two players are almost exactly identical when removing the home rink bias from each sample. The switch to rate-based statistics adjusts for the fact that O’Reilly had a higher average time on ice per game as well as the fact that Stepan was injured at the start of the 2014-15 season, causing him to play less games than O’Reilly. Below is a further breakdown of the away stats by situation for the two players:

Player 5v5 GV / 60 5v5 TK / 60 PK GV / 60 PK TK / 60 PP GV / 60 PP TK / 60
Derek Stepan 1.12 2.25 0.00 2.75 0.94 0.31
Ryan O’Reilly 1.03 2.19 0.55 3.84 1.51 1.21

Do note that the PK and PP samples are rather small (100 -200 mins) so they may not necessarily be close to what they would be over a longer stretch of time.

Ultimately, I think it was still fair for NHL Arbitrators to say that O’Reilly had more favorable turnover stats than Stepan, but I would argue that they were much closer than the base counts implied.

An Incomplete Picture

Giveaways and takeaways make implicit descriptions about possession during a game. A giveaway signals that the player had the puck before the event occurred while a takeaway signals that the player did not. From there we can ask if a team’s ability to possess the puck is having an impact on the player’s stats.

Returning to the O’Reilly vs. Stepan comparison from before we can see that there might be something to this. Despite being one of Colorado’s best possession players, O’Reilly’s away 5v5 score-adjusted Corsi for was 45.42% over the last two full seasons. Stepan on the other hand had a 50.52% Corsi for in the same situations. This 5.10 percentage point difference suggests that the Rangers had possession of the puck more often than the Avalanche did. This makes it more likely that Stepan was in situations where he could commit a giveaway than O’Reilly, while O’Reilly would have had more opportunities to commit a takeaway than Stepan.

Now, it should be noted that Corsi is a proxy for possession because all it actually measures are certain types of shot attempts. It does not quantify time of possession, zone entries and exits, pass attempts, etc. And this also does not quantify O’Reilly’s or Stepan’s actually roles and successes on the ice. It tells nothing of how much each skates with the puck on their stick, how their positioning is, etc. All of that could be another topic for another day (depending on how much data is out there for these sorts of things).

This bridges over into a criticism that Mishkin had back in his 2013 article:

You would figure a player with a high giveaway total is prone to making bad decisions while one with a lofty takeaway total is adept at reading plays. But the player’s “giveaway” total only highlights potential bad decisions. There’s no corresponding number for how many good decisions he makes with the puck. (At least in football, the quarterback offsets his interceptions with other figures, such as completion percentage, yardage and touchdowns).

Mishkin is right that we need to know a lot more to make use of these turnover statistics when comparing players. It just further enforces the need for more data to properly make use of these statistics. Perhaps the biggest indication that we need to re-evaluate the use of these statistics is the fact that 2015 Norris trophy winner Erik Karlsson led the league in giveaways in the last two full seasons combined. Behind him were PK Subban and Joe Thornton. While these three players “did the most bad things with the puck” over those two seasons, I highly suspect that some record of the “good things they did with the puck” would vindicate them. Unfortunately, no such data is tracked by the NHL.

In Conclusion

To wrap it up, it’s important to do the following when using turnover stats in player evaluation:

  1. Either stick to away stats or find/calculate an adjustment coefficient for home statistics.
  2. Use rate-based statistics, preferably ones based on time-on-ice.
  3. Consider the situational usage of the players you’re using.Either use 5v5 stats or make a note of how PP and PK stats influence the players’ composite stats.
  4. Think about how a player’s stats can be influenced by their style of play as well as that of the team around them.
  5. Add a disclaimer about the shortfalls of turnover stats and use other metrics in your comparisons.
  6. Reconsider if using turnover stats is a good idea.