Wednesday, November 19, 2008

Three Cheers for Cause and Effect*

Tyler at Mudcrutch posted this research in response to ongoing concerns within the Edmonton Oilers internet community that Shawn Horcoff has not produced much offense at even-strength this season. Tyler’s basic point is that he’s still getting his shots and, perhaps more importantly, the Oilers are outshooting the opposition when Horcoff is on at evens. Horcoff at the time had scored 1 goal in 37 even-strength shot attempts, so Tyler ran some sort of programming script, I imagine, from some sort of master set of data he seems to have and looked at every other 1-for-37 streak an NHL player had last season. Not surprisingly, there were a bunch. Also not surprisingly, those players started to put the puck in the net coming out of their 1-for-37 droughts. Streaks in hockey, and anything else where there’s a certain amount of randomness involved, exist. And then they end. When an average player in the NHL, where the shooting percentage is around 8%, goes 1-for-37, the odds are he’s going to do better in his next 37. Horcoff scored an even strength marker later that night, almost proving Tyler’s point res ipsa loquitur.

What I’m more interested in here is why some players seem to have “off” years, that is, where previously established scoring rates are significantly reduced for a prolonged streak that lasts an entire season, and how can we predict whether or not such an off year will occur. To the extent Tyler is saying, “Horcoff is fine, he’s getting the same number if not more shots than last season, it’s only a matter a time before his prior scoring rate returns,” I think it’s useful to look at some players who have recently had these off years and examine what drove those results, particularly in terms of shooting percentage and shooting rate. (I’ll note that I don’t think Tyler is necessarily making the quote I attribute to him, but I do think it’s a pretty quick logical jump from what he is saying).

I identified 12 players who saw their goal scoring rates drop significantly from 2006-07 to 2007-08. At least computationally, there are two measures that can affect a player’s goal scoring rate, their shot rate – how many shots they are getting over time – and their shooting percentage – how often their shots score. I looked at these two drivers for the 12 players. Ideally, I would have broken this all down to even-strength and power play situations, but as I couldn’t find any stats from 20006-07 that separated shots into even-strength or power-play situations, we’re left with only looking for total goal scoring rates and total shooting rates. As long as these players’ saw the same relative time at evens vs. the power play, it shouldn’t create too many problems, certainly not enough to muddle what I think are fairly clear results.



TotG equals the goals per 60 min of total ice time (all rates are per 60 min of total ice time). The preponderance of sub-.70 Ratio TotG signifies that there were some significant scoring rate drops in this group. The final two columns, decline in shooting percentage and shooting rate, are the independent variables I looked at separately as possible causes of the decreases in scoring rate. And the R-squared values (where the scoring rate is the dependent variable) show a pretty decent correlation between shooting percentage and scoring rate, while the correlation between shooting rate and scoring is basically non-existent. Within this small sample size, getting shots was in no way a harbinger of scoring success. While I don’t think this necessarily means you want to be not getting any shots at all rather than getting them and misfiring, it does make me think that Horcoff isn’t necessarily out of the woods just because he's getting shots. There are numerous reasons why a good player would have a bad year (injury, change in linemates, being used in different situations, confidence, loss of ability), but I think it’s certainly possible, and these numbers reflect it, that a bad scoring year will show up only as a function of shooting percentage and not shooting rate. I think this makes intuitive sense as well: if a player is struggling, he could start to get the puck off too quick, not having the patience or confidence to make an extra move or to wait the extra moment to get in a better position, he could be floating around the periphery taking shots that he merely hopes find a way. Time will tell if Horcoff matches his impressive scoring rate from last year, and I don’t think there’s any doubt it will improve dramatically from what it is now, but his numbers (probably explained mostly by bad luck and not playing with Hemsky) are nonetheless not inconsistent with predicting a down year.

* The title of a fictional feature film of James Incandenza in David Foster Wallace's Infinite Jest.

4 comments:

Jeff said...

The only true measure of a player's offensive ability is the extent to which his presence on the ice increases the likelihood that his team will produce a "scoring chance" while he is out there (a high percentage shot between the faceoff dots from no further than the top of the circle).

I would suspect that there is a direct correlation between number of scoring chances per team and number of goals scored per team ... (over the season), and that the teams who enjoy the greatest "scoring chance differential," with respect to their opponents, are the most successful teams.

I also think that it's impossible to truthfully compare a player's offensive statistics from year to year unless that player is playing with the same players in the years you are comparing.

Not sure if that adds anything at all to your analysis, but my two cents.

rananda said...

I agree that net scoring chances created are probably the best measure of offensive ability, though I think we can add that various players' ability to capitalize on such scoring chances are not created equal, and that ability matters. And I'll also agree that any offensive stat, no matter how narrowly identified, still requires a certain level of context with which to be viewed usefully. Linemates are one very effective piece of context that helps to add to the picture. The QualTeam stat helps in that regard, though I don't think it's a substitute for examining exactly who was playing with who and arriving at some conclusions. I can say that Horcoff isnt going to be getting the same chances without Hemsky with more confidence than if all I had to go on was slightly different QualTeam numbers.

John said...

i came for the DFW quote but i stayed for the top-flight* hockey analysis.

*as far as i can tell

rananda said...

I debated not asterisking the title, but then you would have been the only one of the half-dozen or so sets of eyes that got it. And that thought upset me for some reason.