METHOD

Following the method most often used to test the predictive value of Corsi, Fenwick, and War On Ice's Scoring Chances, where data from the previous seasons is measured against Goals-For percentage from the subsequent seasons. Data is all situations and was obtained from David Johnson's site.

Save Percentage +Shooting Percentage (PDO)

The above chart shows NHL Sv%+Sh% vs future goals-for percentage from 2007 to 2016 for all situations. Teams with higher PDO have better goals-for percentage in the subsequent season. The mean team Sv%+Sh% is 100, with a standard deviation of 1.22. A 1% increase in Sh%+Sv% gives 2% increase in Future GF%.

Corsi

The above chart shows NHL Corsi-For percentage vs future goals-for percentage from 2007 to 2016 for all situations. Teams with higher for-percentages have better goals-for percentages in the subsequent season. The mean team CF% is 50, with a standard deviation of 2.88.

Discussion

Shooting percentage and save percentage combined is a better predictor than shot rate data for NHL teams. So-called PDO (a common term for Sh%+Sv%) involves the notion that shooting and save percentages are volatile and "regress" to 100. While peaks in percentages do flatten-out to team averages over multiple games (the same is true of shot-rate data) the above data confirms their predictive value. My reason for pointing towards percentages here is not to make war with Corsi. Rate data over 10, 20, or 40 games has an extremely valuable predictive power. However, the predictive power of shooting and save percentage data over entire seasons is greater.

I hope now we can dispose of the notion of "PDO" as a measure of luck or randomness while keeping the Sv%+Sh% stat for its predictive power when an appropriate number of games is involved. As a rule of thumb, I suggest looking at rate data before 40 games, while leaning to cumulative percentage and rate data after.

One of the reasons that the community has been confused about PDO for so long is that all of the numbers look like they are close together, gathering around 100. While the data set has a mean of 100 and a standard deviation of just over 1, small increases in this number are very powerful. I will discuss in future posts why those small numbers occur, and what drives them.

My hope is not to flush Corsi. My hope is not to flush the notion of regression of spiking percentages. My hope is to turn the discussion towards shot quality and the other factors that give percentages their predictive value.

OTHER POSTS BY bituman

http://www.coppernblue.com/2015/1/18/7492033/visualizing-goaltender-performance-1982-to-2014

http://www.coppernblue.com/2016/11/29/13786388/nhl-teams-decrease-their-goals-against-by-raising-their-save

http://www.coppernblue.com/2016/11/28/13762074/nhl-teams-increase-goal-count-by-raising-their-shooting-percentage

http://www.coppernblue.com/2016/11/25/13744366/testing-positive-for-corsi

http://www.coppernblue.com/2015/6/12/8764363/untenuring-the-professor-replacing-scrivens-with-goalies-whove

http://www.coppernblue.com/2015/5/7/8564545/the-schuchards-and-the-draft-of-95

http://www.coppernblue.com/2015/3/23/8276921/when-its-tied-nelsons-behind-and-it-matters

http://www.coppernblue.com/2015/3/15/8219315/under-nelson-defense-worse-but-goaltending-better

http://www.coppernblue.com/2015/3/7/8167691/scoring-chances-goal-differential-2005-present

http://www.coppernblue.com/2015/1/26/7918781/why-the-oilers-pdo-will-not-regress-it-s-their-parzona-of-course

http://www.coppernblue.com/2014/9/24/6840177/visualizing-goaltender-performance-with-save-percentage-plus-part-one

http://www.coppernblue.com/2014/10/5/6852077/visualizing-goaltender-performance-with-save-percentage-plus-sv-part

http://www.coppernblue.com/2014/10/23/6960709/visualizing-save-percentage-plus-part-three-2013-2014

http://www.coppernblue.com/2016/1/7/10727968/ryan-nugent-hopkins-2015-2016-comparables

http://www.coppernblue.com/2016/1/6/10726162/10-justin-schultz-stats-that-will-blow-your-mind

http://www.coppernblue.com/2015/12/28/10673024/oilers-lose-3-to-5-to-flames-justin-schultz-the-dialectic-of

http://www.coppernblue.com/2015/12/22/10644388/how-the-grinch-stole-christmas

http://www.coppernblue.com/2015/11/26/9803964/the-emptiness-of-victory-canes-beat-oilers-4-1

http://www.coppernblue.com/statistical-analysis/2015/11/23/9778766/the-top-8-after-20-the-state-of-scoring-in-oil-country

http://www.coppernblue.com/2015/11/19/9751952/cumulative-faceoffs-possession-r2-0-408

http://www.coppernblue.com/statistical-analysis/2015/11/13/9720672/why-justin-schultz-is-more-valuable-than-mark-fayne

http://www.coppernblue.com/2015/11/10/9705610/10-righty-defenders-under-4-million

http://www.coppernblue.com/2015/11/7/9690376/why-the-oilers-goalies-cant-save-them

http://www.coppernblue.com/analysis-5/2015/10/5/9452123/ten-years-after-the-shootout

http://www.coppernblue.com/2015/10/4/9448443/canuck-comeback-squelches-disciplined-oilers

http://www.coppernblue.com/2015/9/26/9403491/oilers-shut-the-door-in-saskatoon-beating-wild-3-0

http://www.coppernblue.com/analysis-5/2015/7/1/8879481/nhl-free-agency-oilers-defence-sekera-schultz-nikitin

http://www.coppernblue.com/analysis-5/2015/6/26/8850529/2010-nhl-draft-revisited-hall-seguin

http://www.coppernblue.com/2015/6/16/8786911/oilers-craig-ramsay-petry-marincin-schultz-ference-dcorsi

http://www.coppernblue.com/2015/4/22/8466947/battle-of-alberta-lives-on

http://www.coppernblue.com/2014/11/27/7298915/goaltending-how-low-oilers-sunk

http://www.coppernblue.com/2014/11/5/7080537/ference-by-the-numbers-2009-2014