Tuesday, June 14, 2005

Does Hockey Need Some PhD's

Today, my blog was noticed by a non-hockey blogger (for the first time in my knowledge). Here is the post that came up on traffick.com. He writes:

The state of statistical analysis in hockey is utterly maddening. Think about the key statistic in baseball: the batting average. Did you ever hear a ballplayer being judged by the absolute number of hits he gets (unless it's a lot)? Yet in hockey, a player who plays seven minutes a game is routinely described by his "15 goal season" or "he only has six goals in the first half." Some supposed superstars are on the ice 30 minutes a game. Shouldn't we be looking at "points per minute played"?

There is a very good reason that statistics in hockey are not as well advanced as baseball. The problem is that much of what goes on in a hockey game is undocumentable in any meaningful statistical manner. As an example, lets make up the circumstances that take place in a short period of a hockey game. For argument's sake we will make this a Calgary vs Tampa Bay game where lineups exist as they were in the 2004 Stanley Cup finals. I doubt that this actually happened in any game it is merely an example. Ruslan Fedotenko has the puck in the center ice zone and dumps it into the calgary zone high and on the boards. Miikka Kiprusoff comes out behind his net to play the dump in, but it gets by him. Mike Commodore picks the puck up for Calgary in the corner. He tries to dump it out of the zone quickly before being flattened by a bodycheck by Brad Richards. Commodore's pass out of the zone is knocked down at the blue line by Daryl Sydor. Now no statistics were recorded during this play at all. Yet, a lot happened. If that much occurred in a baseball game, plenty of players would gather some kind of statistic. This kind of play is makes up most of a hockey game. Most of what goes on in a hockey game is undocumented statistically. For the most part, statistics are only recorded when a goal is scored and this only occurs five or six times a game on average.

There have been a few attempts to gather "real time" statistics on other events in games. The NHL used to record hits, giveaways, takeaways etc. but has discontinued this process (at least publically). These numbers were not particularly useful because there was little uniformity between what constituted a "giveaway" or a "hit" in different stadiums throughout the league. Even with these stats, there is little idea what to do with that number (or even if that number correlated with being a good player - as an example to lead the league in giveaways a player must have the puck a lot and try to do creative things with it - in general only offensive stars get a lot of giveaways.)

Now the point made is that possibly we should record goals per minute of ice time as opposed to total goals. Would this be useful to do? In general, the only players who get a relatively high goal total in limited ice time are players who offer little but goal scoring ability. Such a player would be a defensive liability to use except in situations such as power plays where the team will press for goals. Does that actually prove what is intended? Probably not. Nevertheless, it is a useful exercize. It is a useful thing to look at NHL statistics such as they exist in a different light.

Hockey cannot expect to have as deep a statistical record as baseball does. Therefore it cannot have as succesful a sabermetric theory. There are questions which can be answered with some certainty (although there are error bars that must be taken into account with any result). These questions include what was the best goal scoring season of all time (which one could attempt to use the adjusted goals I discuss yesterday). This would not show us who is the best goal scorer - it may tell us who had the best linemates setting him up (for example).

But it's what a modern game needs to make the office pools worth playing, to provide the kind of white-collar, armchair-expert discourse that keeps people talking. "Real guys" can also argue about cheap shot artists and acceptable hockey hair length if they want.

People need to understand hockey and be engaged by it. Because it is a fast game and is not constantly stopping (like football or baseball) it is a harder game to catch onto. This is easiest when people grow up with hockey. I think if there is a payoff for the southern NHL expansion it will occur when the current young fans who (for example) remember the Tampa Bay Stanley Cup grow up and are old enough to buy tickets to games. This might be a 20 year (or so) time investment. It isn't a fast process.

Being engaged by hockey includes going to games. It includes watching games. It includes playing hockey. It includes playing fantasy hockey. All these need to be available. They are available (for the most part). There are many internet fantasy hockey leagues such as this one that I play in. However, I doubt that the statistics recorded in hockey are capable of answering analytical questions on the level of "What is the value of a stolen base?" in baseball. However, if smart people look at the statistical record and try to expand it, we might be suprised by what can be learned. I am very interested in the results of these studies, however, I think the possibilites are much more limited then in baseball.

I really agree with your point on production being a lot more meaningful if ice time is taken into account. After all, probably the simplest breakdown of a player's stats would be Production = Talent * Opportunity. In actuality there are really are more factors that need be taken into account, such as teammate skill, etc, etc, etc, but I think that is an accurate way of stating it at its core. A player will produce at a level that his talent provides him, along with how much opportunity he receives (ice time).

Take, for example, the Edmonton Oilers trading away of Miroslav Satan. The press, later, liked to paint it as, "Gee, how could the Oilers have known he would produce that much? He scored 18 goals in Edmonton, and 40 in Buffalo. Nobody could guess that" A subjective view of things would have told them Satan would.

In 1996, Satan scored 18-17-35 for Edmonton, and played 868 Minutes (ice time estimation system)This leads to per-minute breakdowns of:

Goals per Minute: 0.021
Assists per Minute: 0.2
Points per Minute: 0.04

Let's contrast that with some averages during Satan's time in Buffalo.

3-Year Averages
GPM: 0.022, APM: 0.016, PPM: 0.038

5-Year Averages
GPM: 0.021, APM: 0.017, PPM: 0.038

GPM: 0.20, APM: 0.20, PPM: 0.40

His overall per-minute production matches exactly what he did in Edmonton. The difference is that Satan put up superficially bigger numbers in Buffalo because of receiving greater opportunity.

Daryl Shilling
Interesting... I wrote something on this topic a few years ago...

Post a Comment

<< Home

This page is powered by Blogger. Isn't yours?