Playoff Success Correlation - Redux

So following up on my previous blog post regarding what regular season statistics are most likely to translate into post season wins, I decided to re-run the test with some adjustments.

To measure success, I used raw postseason wins again.  For the moment, I do not have a better method.

To measure regular season performance, instead of looking at where each team finished in each category (1st to 30th), I used the direct statistic.  In addition, because average PP, PK and save percentage varies league wide from season to season, for those three statistics, I calculated the teams statistic against the league average from that particular season.

Here are the results:

Conclusions
 -In general, making these adjustments did not change the qualitative results.  Instead, in nearly every category in resulted in a small increase in the magnitude of correlation.

-Corsi and Fenwick correlation showed the most significant increase in correlation, to the point that it implies a different qualitative conclusion.  Whereas Corsi and Fenwick were no better in predicting playoff success than simple 5v5 goal differential when looking at it via league standing, when you use the more specific Corsi or Fenwick percentage (as I did in the second analysis), it does become a significantly better predictor.  Interesting.  This throws out much of my rationalization in my last post.


-When I was re-digging into the data, the Penguins of 2008 and 2009 were an anomaly.  Those teams reached the finals in both of those seasons (and winning the cup in 2009), despite very poor performance in Corsi, PK, and save percentage.  This drags down the 7 year correlation in those statistics, and in a possibly misleading fashion.

-Defaulting to the 5 year statistical window (the green bars), PK quality remains the strongest predictor by a wide margin.  After that, total goal differential, Corsi, Fenwick, and total shot differential are all about equal in correlation strength.  Save percentage lags behind, with a decent correlation.  5v5 goal difference is weaker still, power play is worthless, and I didn't even bother looking at shooting percentage in my second analysis.