Kukla's Korner

A formula for NHL team success

It’s time to get our collective geek on, and figure out which NHL teams are truly great, and which are ripe for a fall.  The tool I like to use for this is called PythagenPuck, outlined in a Win Probabilities piece over at HockeyAnalytics.com (PDF).  The basic idea with PythagenPuck is that a team’s Goals For and Goals Against provides a more consistent and accurate picture of their quality over the course of an 82-game season, similar to the Pythagorean Expectations invented by Bill James for baseball analysis.  Early in the year, a few lucky bounces here or there can make a team look stronger or weaker in the standings than they truly are, and PythagenPuck helps identify those outliers.

I’ve provided a similar predictive look at the league in prior years, and have been pretty satisfied with the results. There are always teams which genuinely turn around their play (either for good or bad) during the course of an NHL season, but with the majority of teams, this Goals For/Goals Against analysis does a good job of identifying the high-flying frauds and the potential contenders waiting just outside the spotlight.

So here are the numbers for each team, with sage commentary included:

The Google Spreadsheet is available for public view & download.
All data through games of Sunday, December 14 2008.

In order to isolate how teams are earning Wins and Losses relative to their Goals For & Against performances, I exclude a few items from the typical NHL stats before running the numbers:

1) Points earned in Shootouts, since that action is different from regular play.
2) Points earned for Overtime Losses (OTL), so that losing 4-3 in regulation is treated the same as losing 4-3 in overtime.  We’ll come back around later to point out which teams are milking the Loser Point for all its worth.
3) Empty Net goals, because they don’t so much cause wins and losses, they’re more an indicator of which teams are already going to win or lose a given game.

A consequence of items 1 and 2 above is that a game that goes to a Shootout is treated like an old-style tie (1 point for each team), which accurately reflects the result of a tied game at the end of regular action.  The less-than-obvious columns in the table above are as follows:

Adj GF: Goals For, excluding empty netters.
Adj GA: Goals Against, excluding empty netters.
Adj Pts: Standings points, less the items listed above.
Adj W %: AdjPts / Games Played times 2 (Adjusted Points earned as a portion of all possible).
Exp Win %: Expected Winning Percentage, based on Adj GF & Adj GA run through the PythagenPuck formula*.
Diff: The difference between Adj W% and Exp Win%. Positive values represent teams winning more than expected (i.e. lucky). This list is sorted in declining Diff sequence, from “luckiest” to “unluckiest”.

As is my wont, the most favorable values are shaded green, the unfavorable ones pink.

In reading these results, high positive values in the “Diff” column indicate teams that have earned more Wins than their GF/GA performance would normally indicate, and thus over the long haul of the regular season, it’s likely that those Wins and Losses will pile up accordingly.  Sometimes you do have a team which significantly improves or deteriorates (think Ottawa last year), but those are the exceptions, not the rule.

Besides the “Diff” column, you can also review the difference between regular Pts and AdjPts to see which teams are benefiting from the peculiarities of the NHL’s rules.  The Rangers (9) and Red Wings (8), for example, have feasted on Shootout Wins and Overtime Losses to pad their place atop their respective divisions, while Phoenix and Vancouver have only snared a single point apiece through those devices.

So feel free to comb through the data here and call out your own favorite over- and underachievers in the NHL!

*The basic Pythagorean Formula runs like so:  Win Pct. = GF2/[GF2 + GA2] .  PythagenPuck changes the exponent from 2, to something that more accurately reflects the scoring rate in today’s NHL. The exponent used in this analysis is 2.1838.  If anyone has an easy way to present the formula in HTML form, I’d be glad to hear about it, I know the formula above looks clunky.



Filed in: NHL Statistical Analysis, | On the Forecheck | Permalink


Be the first to comment.

Add a Comment

Please limit embedded image or media size to 575 pixels wide.

Add your own avatar by joining Kukla's Korner, or logging in and uploading one in your member control panel.

Captchas bug you? Join KK or log in and you won't have to bother.

Commenting is not available in this channel entry.