|
In the fledgling field of statistical hockey analysis, there are those who know how to correctly apply and extend existing methods to answer the game’s most difficult questions, and there are those whose understanding of both the sport and statistics allow them to create new ones. Tom Awad, creator of GVT, VUKOTA and Delta SOT, is one of the latter, so it was with understandable delight that I first read his ambitious attempt to objectively describe the properties of “Good Players”.
Tom’s Good Players Series (forwards only): Part 1, Part 2, Part 3
Any study such as this must ultimately have as its objective to improve our understanding of the game, in as concrete a fashion as possible. We’ll review this series to see if it’s hitting the mark.
My Personal Bias
Before delving into my review of the Good Players Series, allow me to reveal my bias up front—ever since our early days of discussing Point Allocations back in April 2004 on Iain Fyffe’s Hockey Analyst Group, I’ve been a big fan of Tom’s work. An annual lunch in downtown Montreal has become something of a tradition for us, and I’ll attest to his being one of the more likeable guys you’d ever hope to meet.
If you think all that means that I’d be more inclined to be gentle in my review, then you don’t know Tom. I’ve met few people who have a greater fondness for healthy criticism—how else do you think he’s managed to so thoroughly refine his various creations?
The Foundation
Tom’s entire study is based on two primary principles, one that I really like and one that I really question.
I really liked how he dealt with the issue of sample size, an issue that has plagued virtually every statistical hockey analysis I’ve ever seen. The solution was to look at entire groups of players rather than individuals, an approach that was very effective in diluting the impact of noise and luck.
I’m curious how much noise is left, and whether it would be advantageous to separate home and road, or to break it up by period and/or situation (leading, trailing, tied). Unfortunately, reducing the sample size to reduce the noise might cause an even greater increase in luck. Worse yet, it could just wind up as a lot of extra work for something only marginally more accurate. In the end, I strongly agree with the chosen approach.
Players were divided into four groups, which at first glance confusingly appears to be an attempt to divide players by lines, but is actually broken up by ice time—there are far fewer players in the first group than the last.
The fact that the division was made based on ice time is the foundational principle that I question. Given all the different ways that players could be categorized as Tier 1 (the best) to Tier 4 (the worst), why was even strength ice time chosen? I’m curious how closely correlated even strength ice time is to actual ability, and how much it is influenced by other factors, such as:
•Playing on a team with a particular abundance or scarcity of talent
•The coach’s ability or inability to recognize talent
•Teams that are in the lead or trailing especially often
Tom even goes on to claim that you can supplement an analysis of a player’s puck possession ability (Corsi, Delta) with even strength playing time to determine if a player is a good “finisher” or not. I have to admit having a spit-take when I read that—Tom owes me a new monitor.
I’m reminded years ago of a method devised by Iain Fyffe to measure a player’s defensive abilities by determining how much of his ice time could be reasonably explained by his offensive production (which was relatively easy to measure even back then) and concluding that the remainder must be explained by his defense.
Using ice time like this is an intriguing concept, but ultimately I’m concerned about building on such a foundation without more certainty into what factors contribute to ice time, and to what extent. After all, every single word of the forthcoming series hinges on this lone definition of a Good Player. While we can certainly proceed with this study based on this definition, I think we should circle back for a deeper exploration of the different ways to categorize players (see my example further down for some ideas), and whether such studies would produce similar results.
The Study
Having divided the league’s players into four buckets, Tom set out to discover the common characteristics, but without hypothesizing what those common characteristics would be.
On one hand the approach of proceeding without preconceived notions of what the common attributes will be allows us to keep an open mind. On the other hand, how do we know what we’ll find if we don’t know what we’re looking for and how we’re going to look for it?
For example, I would expect that the players in the Good Player bucket would: generate more offense, better prevent opposition scoring, take fewer penalties, draw more penalties, take more shots, take higher quality shots, have better puck possession, receive more ice time*, play with better linemates, play against tougher opponents, tend to be 24-26 years old, be paid more, have lower jersey numbers, have been drafted in earlier rounds, be more numerous on the better teams, and do all of this consistently. Each of these predictions can be measured and quantified to an extent.
*Assuming I was blind to how the Good Player bucket was created
Ultimately, some measurements are chosen. Initially, it’s Corsi (a measurement of puck possession) and expected goals for and against, which is defined as a function of shots and shot quality. This initial choice is excellent, as are the others we ultimately get to look at, including Zone Starts (Situation), Quality of Competition, Quality of Teammates, shooting percentage, and PDO. We could easily name a few more stats to include, but nothing that would rank higher.
The Insights
1st and 2nd tier players are roughly equal in puck possession (Corsi) and shot quality, and though they therefore are expected to have the same plus-minus, the most interesting finding is that they do not! 1st tier players enjoy far better results than 2nd tier players. On the surface, that appears to be a rejection of using puck possession and shot quality as indicators of player quality, but Tom is careful to explain that it’s only a rejection to use them by themselves.
The exploration of why Good Players generate more actual offense despite failing to rank any higher in puck possession or shot quality is one of the highlights of the series. A big part of the explanation is the higher on-ice shooting percentage enjoyed by 1st-tier players. When looking at individuals you can normally write that off as luck, but Tom reminds us that in a group this size the luck should cancel out—this link is legitimate.
Another interesting finding is the validation that forwards appear to have very little to do with goal prevention. Both PDO and Goals Against tend to be roughly the same for top-tier players as bottom-tier—partly because higher-tier players not only get higher-tier linemates, but also higher-tier opponents.
While there are other interesting discoveries, such as the rewarding validation of our previous study that 1.7 even strength points per 60 minutes is a magic number for top six forwards, the tight correlation between even strength playing time and power play time, and the lack of such a correlation for penalty killing time (nicely explained as not just lack of defensive skill among top-tier players, but rather opportunity cost), the real gem is the discoveries of puck possession, shot quality, shooting percentage and finishing—a topic that certainly warrants more study.
Conclusions
Looking at Good Players as a group is an excellent way to help reduce luck and various types of noise, but if there’s a debate to be had, it is surrounding the way in which such players are grouped. However, even alternative groupings are likely to produce the same results, including a provocative relationship between Good Players and shooting percentage, and a much weaker relationship with puck possession and shot quality.
I encourage everyone to study Tom’s work, because not only is it well worthwhile as is, but it also affords you the opportunity to share your thoughts and influence the direction of his future study.
Robert Vollman is an author of Hockey Prospectus.
You can contact Robert by clicking here or click here to see Robert's other articles.
| 
Shots On Goal (11/29)
0 comments have been left for this article. 
