Hockey Analytics Researcher | Devils Blogger | Insufferable Pedant

New Jersey, USA
Joined March 2014
I’ve written about hockey analytics for 15 years. In that time, I always wanted a site that dug into not only IF a player is good, but when, why, and how. Something to help writers and fans tell those stories. 3 months ago I started building it ... and today, it's live!
30
79
22
545
98,097
The kind of work that Sunny, and @bmacGTPM, and others did on Bayesian goaltending is the predecessor to our goalie talent estimate at @PostHockeyHQ. Goalies are a mess, but this one JUST clears the threshold for predicting movers to a statistically significant degree.
My 2 questions for #NJDevils GM Sunny Mehta at his press conference today and his well thought out answers. Called him out for bluffing. Got him talking about Bayesian hierarchical models. And allegedly stole @mikemorrealeNHL’s question 😂
1
19
2,021
And, for what it's worth, yes, my model is confident that Connor Hellebuyck is, in fact, better than Jake Allen to a statistically significant degree (barely). These are 80% intervals which means I'm 90% confident Helle is better than Allen.
1
4
716
Arseny Gritsyuk is one of the fastest risers in Net Rating in the leage right now. He's still working on his first skill badge but he's running the table on outcome badges -- his team is out chancing and out scoring opponents and he's involved in the goals. He's currently a borderline top line forward.
2
4
127
4,482
Josh Doan has lost his "Netfront" badge. After coming into the NHL on fire wrt High-Danger chances, he's been getting to those spots less each year, and has just 2 such chances in 11 games so far this year.
2
7
2,420
Been working on this one for a minute. There's lowkey a ton of different features that went into this. But I think it's now a thoroughly unique offering.
Wake up babe! Post Hockey just added lines to their game pages!
3
1
59
3,695
The Kraken may have lost, but Berkly Catton continued his hot start, scoring their only goal while collecting 7 shots and drawing a penalty which earned him a "Motor" badge for the first time in his career. His estimated EOY impact is now that of an above-average NHL 2nd liner.
1
9
1,254
Batherson did all the things in his 14 minutes. He and Sanderson were a nightmare for the Rangers tonight.
1
2
38
3,026
At the event level, yes. But if you want to know who tilts ice? I'd much rather a public model that I now has been adjusted for score, venue, rest, etc. than an opaque private model that may or may not incorporate tracking information responsibly.
They also presumably have access to player & puck tracking data, which should make their models significantly better than ours.
5
10
3,652
This is one of biggest differences between the public and private xG usage. Almost all public sources have one xG model and everything is layered on top of it. Not at @PostHockeyHQ, though! We have 4 different event-level xG models and we're working on even more!
Mehta mentions that public models used to evaluate goalie performance are different than the 14 (!) different expected goals models his staff uses.
5
2
1
58
5,376
We have xG(raw), xG(sh), xG(sv), and xG(sh+sv) if you want to adjust for the shooter, goalie, both or neither. Then, additionally our goal page has a ton of different combinations of xG events that accumulate turns with and without rebounds, or tips/deflections/non-OZ shots, or all "contact" events, or low/mid/high danger. At the team level we have those plus with and without blocks. Those last few aren't the kinds of things Sunny is talking about, but it goes to show that there are different ways of describing "expectation" even for the same outcome.
1
2
14
1,800
And they are substantively different! In small samples they don't matter much, but over a season they're hugely impactful. Last year, the Flyers were 11th in the NHL in raw xG±, but the were 22nd if you adjust for shooter and goalie and the Rangers improve from 26th to 17th.
5
938
Pierre LeBrun: Re Connor Hellebuyck trade/Devils: This trade has to sting...for the Hurricanes; [they] felt they had a fair offer to the Jets on the table for a number of months. The Canes remained interested right up to the end - The Athletic (10/23)
26
11
16
275
60,045
CJ Turtoro retweeted
I have compiled every frame of all PPT goal animations provided publicly by the NHL, alongside a notebook showing how to collect the dataset for yourself with the wsba_hockey package. kaggle.com/datasets/owensing…
5
6
22
2,138
With the information EDGE gives us on burst bins and top bursts, we can build a distribution of skating speeds for each player. It's reliable. It's on us to figure out what it predicts. One example? The speed quotient I use here has a 16% r^2 with next-year carries/60 for Dmen.
thoughts 1. yes, NHL edge data is not going to grade the "best skaters" in terms of skill. i would hope that that's obvious. 2. 20+ mph bursts /60 is EXTREMELY repeatable YoY. it tells you a lot about the speed a player plays at - like it or not, there's a big signal there.
2
33
3,706
Btw these viz are now available on @PostHockeyHQ's player pages. Big red dot is their "hourly speed". posthockey.com/players/84784… Wonk note: this is a reverse Gumbel distribution which tested well enough to publish, but I'm still looking at alternatives.
1
1
1,258
Jack Hughes's burst profile has actually changed a lot in recent years. He's not got the top-end speed he had in 23-25. But this season he's been getting to 18mph more often than ever (over 10x a game).
1
5
1,081
CJ Turtoro retweeted
Find a goalie who plays well for your team. Hold him close to your heart and never let him go. Love him unconditionally, for all his is and what he means to you, because he won’t be worth a dime by the time you break up.
one takeaway you get from trying to improve any goalie statistic to be remotely predictive is that it is basically impossible. any signal you find vanishes when you look at goalies changing teams. hence, is a goalie
1
2
17
1,461
CJ Turtoro retweeted
The new iPad layout of the @hockey_gamebot website turned out so awesome. A few minor tweaks to work on, but my goal is to make the site a perfect "second screen" to have up during live games for quick access to stats & other data.
1
10
1,009
Well when you say it that way....
Replying to @CJTDevil
To put it another way, the Devils traded Markstrom, Mercer, and futures for Hellebuyck, Rodrigues, and Boqvist
2
66
4,284
Also a reminder that Jacob Markstrom's contract was $6M.
3
158
14,684