Going beyond Corsi: A look at PDO, Fenwick, High Danger Chances, and Scoring Chances

After I wrote and published my last article concerning the applicability and predictability of Corsi as an indicator of success, I became interested to see if there were any advanced statistics that were a more accurate than Corsi.  Some of the statistics that I began to look at were Fenwick, High Danger Chances For, Scoring Chances For, and even PDO.  For this study I will be looking at these stats as an indicator of success in the NHL as a more effective alternative to Corsi.  While Corsi is considered the gold standard of “advanced stats” there are plenty of other advanced stats that help to provide insight into the game of hockey.  

Personal Existing Bias

I personally believe that Corsi can be somewhat indicative of a team’s success in the NHL depending on the team’s strategy and systems.  However, as my previous study showed, Corsi only had between a 60% and 67% success rate, depending on the definition of success.  Corsi is Dead…or is it?  If a singular stat is to be used as frequently as and for as many purposes as Corsi is, it must be able to predict at a much better rate. 

I personally don’t believe that any single stat can predict success at a rate greater than 80%.   However, I do believe that some, if not all, of the statistics mentioned above will provide a more accurate assessment.  I will admit that my personal bias leans towards skepticism of the APPLICATION of statistics.  However, as I mentioned multiple times in my Corsi article, more information is ALWAYS better.  No one hates statistics (ok, maybe I did hate statistics during undergrad and grad school).  NO ONE HATES CORSI. The more we can measure and understand the intricacies of the sport of hockey, the better the sport will be and the better the on-ice product will be.  It is simply the application and misuse of statistics that is an issue.  The Rangers are not a puck possession driven team.  They do not try to lead in puck possession.  They will not have a great Corsi.  Why do we choose to judge them based primarily on Corsi?  Why not judge them based on what they are trying to do?  Ok, rant over.  Back to being scientific and objective.

Abstract

As I mentioned above, this study will attempt to determine the effectiveness of four advanced stats, Fenwick, PDO, High Danger Chances, and Scoring Chances.  As with the Corsi study I will be looking at the past five years of the NHL and attempting to determine if there is and how strong of a correlation there is between the advanced stats and success in the NHL.  Correlation will be determined by comparing the strength of each team’s stats with whether or not they made the NHL playoffs and the Conference Finals.  All statistics will be taken from naturalstattrick.com unless otherwise noted.   Please remember that the 2012-2013 season was a lockout shortened season, so while there is no indications that the stats were effected by the short season, it is possible that with longer study, anomalies are very possible. 

Introduction

Since the introduction of advanced statistics in the NHL, Corsi has been the most highly regarded and utilized statistic.  A statistic that measures shot attempts per game, Corsi passes the common sense test.  The more shot attempts that you have, the more goals you will score, and the more wins and success you will have.  However, Corsi does not take into account quality of shots. 

Just because a team throws the puck at the net from every angle possible and at every opportunity, does not mean that they will be successful.  For four years in College, I watched as one of my teams’ defenseman would drill the opposing forwards shin pads, leading to a breakaway headed the other way.  Corsi views this as a positive, while a statistic like Fenwick does not and HDCF and SCF do not even count this.  In addition, for the past five years, the Rangers have generally been an average to below average Corsi team while continuing to make the NHL playoffs (Henrik Lundqvist has something to do with this) and winning more playoff games than just about anyone.  Why is this?  Did the Rangers just get extremely lucky or was there something deeper? 

Definitions
PDO

PDO is the NHL’s “luck” stat.  A statistic that adds a team’s shooting percentage with its save percentage, the NHL average will be right at 1.000.  A team that has a high PDO percentage can indicate that the team is getting lucky and is ready to regress towards the mean.   A team that has a low PDO percentage can indicate that the team is unlucky and is underperforming.  Now, there are teams that will have slightly higher or lower PDO’s due to having sharpshooters such as Alexander Ovechkin or Steven Stamkos.  However, it is still extremely rare for a team to have a PDO much higher or lower than 1.000 over the course of an 82 game season.  PDO is not a statistic that is used to determine success but is used to help diagnose outcomes that are contrary to the expected.  However, for this study I attempted to leave no stone unturned and used every category of advanced stat on naturalstattrick.com in their basic advanced stat spreadsheet.

 

Fenwick

Fenwick is a very similar statistic to Corsi.  While Corsi measures simple shot attempts, Fenwick is slightly more selective.  Fenwick does not include blocked shots when calculating the percentage. 

For example, if the Rangers have 10 shot attempts and the Devils have 10 shot attempts, the Rangers Corsi For would be 10 and their Corsi Against would be 10, while their Corsi For percentage would be 50% (10/20).  However, if two of the Ranger’s shot attempts were blocked, their Fenwick For would be 8 and their Fenwick For percentage would be 44% (8/18).  Most teams have similar Corsi and Fenwick percentage numbers, however, it would be remiss to ignore the subtle differences between the two and not study the applicability of the statistic.

 

High Danger Chances and Scoring Chances   

High Dangers Chances are slightly more complex than Corsi, Fenwick, or PDO.  While those three have easily quantifiable metrics with which to create the stat, high danger chances have some subjectivity.  Each site that tracks high danger chances have slightly different standards or definitions.  However the basic premise is that every shot attempt is broken down into different danger levels based on the zone that it was taken from, whether it was a rebound or rush shot, and whether or not it was blocked. 

https://imgur.com/1fzcaQT   

High Danger Chances are shot attempts that have a rating of 3 or higher.  If you would like to get a more detailed breakdown and definition you can find it here at https://blog.war-on-ice.com/index.html%3Fp=512.html#ref-2.  The idea is that there are certain areas on the ice that have a much higher likelihood of the puck going into the net than others.  For example a shot from between the hash marks is going to have a much higher likelihood of beating the goalie than a shot from the corner.  If a team has more high danger chances, they will score more goals, at least in theory.  Scoring Chances are extremely similar to high danger chances.  While high danger chances require a danger rating of 3 or above, a scoring chance must have a danger rating of 2 or above. 

 

Problem Statement

How does one scientifically discover which statistic is the most accurate when predicting success in the NHL?

As with Corsi, everyone has their opinions on each statistic.  However, if there is no way to quantify or “prove” success, then those opinions are worthless.  With this study I am attempting to add another piece to the puzzle that is predicting NHL success by building on my Corsi study and attempting to find the most accurate single statistic for predicting success in the NHL.

 

Hypothesis 1a

A Fenwick For percentage above 50% will positively correlate with teams making the NHL playoffs during the period from the 2012-2013 season and the 2016-2017 season.

Hypothesis 1b

A Fenwick For percentage in the top 16 will positively correlate with teams making the NHL playoffs during the period from the 2012-2013 playoffs and the 2016-2017 playoffs.

Hypothesis 1c

A team’s Fenwick For percentage will be a more accurate indicator of success in the NHL than their Corsi For percentage.

With these hypotheses I am predicting that teams with a Fenwick For percentage over 50% or in the top 16 will have a positive correlation with making the playoffs.  Between 12 and 16 out of the 16 playoff teams per year with a Fenwick percentage over 50% or top 16 will indicate a strong positive correlation.  Between 8 and 12 out of the 16 playoff teams per year with a Fenwick percentage over 50%  or in the top 16 will indicate a moderate positive correlation.  Between 4 and 8 out of the 16 playoff teams per year with a Fenwick percentage over 50% or in the top 16 will indicate a moderate negative correlation.  Between 0 and 4 out of the 16 playoff teams per year with a Fenwick percentage over 50% or in the top 16 will indicate a strong negative correlation.  In addition hypothesis 1c will be studied by comparing a teams Corsi For and Fenwick For percentages.  If a greater number of teams with a percentage over 50% or in the top 16 for either statistic make the playoffs, that will indicate that that statistic is a more accurate indicator of success in the NHL.

 

Hypothesis 2a

A team’s PDO above 1.000 will have no correlation with teams making the NHL playoffs during the period from the 2012-2013 season and the 2016-2017 season.

Hypothesis 2b

A team’s PDO in the top 16 will have no correlation with teams making the NHL playoffs during the period from the 2012-2013 season and the 2016-2017 season.

Hypothesis 2c

A team’s PDO will be a less accurate indicator of success in the NHL than their Corsi For percentage.

With these hypotheses I am predicting that teams with a PDO of 1.000 and above or in the top 16 will have no correlation with making the playoffs.  Between 12 and 16 out of the 16 playoff teams per year with a PDO over 1.000 or top 16 will indicate a strong positive correlation.  Between 8 and 12 out of the 16 playoff teams per year with a PDO over 1.000  or in the top 16 will indicate a moderate positive correlation.  Between 4 and 8 out of the 16 playoff teams per year with a PDO above 1.000 or in the top 16 will indicate a moderate negative correlation.  Between 0 and 4 out of the 16 playoff teams per year with a PDO above 1.000 or in the top 16 will indicate a strong negative correlation.  In addition hypothesis 2c will be studied by comparing a teams Corsi For and PDO percentages.  If a greater number of teams with a percentage over 50% (Corsi) or 1.000 (PDO) or in the top 16 for either statistic make the playoffs, that will indicate that that statistic is a more accurate indicator of success in the NHL.

 

Hypothesis 3a

A High Danger Chance percentage above 50% will positively correlate with teams making the NHL playoffs during the period from the 2012-2013 season and the 2016-2017 season.

Hypothesis 3b

A High Danger Chance percentage in the top 16 will positively correlate with teams making the NHL playoffs during the period from the 2012-2013 playoffs and the 2016-2017 playoffs.

Hypothesis 3c

A team’s High Danger Chance percentage will be a better indicator of success in the NHL than their Corsi For percentage.

With these hypotheses I am predicting that teams with a high danger chance percentage above 50% or in the top 16 will have a positive correlation with making the playoffs.  Between 12 and 16 out of the 16 playoff teams per year with a high danger chance percentage above 50% or top 16 will indicate a strong positive correlation.  Between 8 and 12 out of the 16 playoff teams per year with a high danger chance percentage above 50%  or in the top 16 will indicate a moderate positive correlation.  Between 4 and 8 out of the 16 playoff teams per year with a high danger chance percentage or in the top 16 will indicate a moderate negative correlation.  Between 0 and 4 out of the 16 playoff teams per year with a high danger chance percentage above 50% or in the top 16 will indicate a strong negative correlation.  In addition hypothesis 3c will be studied by comparing a teams Corsi For and high danger chance percentages.  If a greater number of teams with a percentage over 50% or in the top 16 for either statistic make the playoffs, that will indicate that that statistic is a more accurate indicator of success in the NHL.

 

Hypothesis 4a

A Scoring Chance percentage above 50% will positively correlate with teams making the NHL playoffs during the period from the 2012-2013 season and the 2016-2017 season.

Hypothesis 4b

A Scoring Chance percentage in the top 16 will positively correlate with teams making the NHL playoffs during the period from the 2012-2013 playoffs and the 2016-2017 playoffs.

Hypothesis 4c

A team’s Scoring Chance percentage will be a better indicator of success in the NHL than their Corsi For percentage.

With these hypotheses I am predicting that teams with a scoring chance percentage above 50% or in the top 16 will have a positive correlation with making the playoffs.  Between 12 and 16 out of the 16 playoff teams per year with a scoring chance percentage above 50% or top 16 will indicate a strong positive correlation.  Between 8 and 12 out of the 16 playoff teams per year with a scoring chance percentage above 50%  or in the top 16 will indicate a moderate positive correlation.  Between 4 and 8 out of the 16 playoff teams per year with a scoring chance percentage or in the top 16 will indicate a moderate negative correlation.  Between 0 and 4 out of the 16 playoff teams per year with a scoring chance percentage above 50% or in the top 16 will indicate a strong negative correlation.  In addition hypothesis 4c will be studied by comparing a teams Corsi For and scoring chance percentages.  If a greater number of teams with a percentage over 50% or in the top 16 for either statistic make the playoffs, that will indicate that that statistic is a more accurate indicator of success in the NHL.

  

Methods

For this study I will be using data from between the 2012-2013 and 2016-2017 seasons.  In order to prove or disprove the hypotheses I will only be looking statistics according to Naturalstattrick.com.  While there are a number of great statistical sites out there including Corsica, Puckalytics (gone), Hockey Reference and others, all stats will come from naturalstattrick.com, unless otherwise mentioned, in order to maintain consistency and objectivity.  This will not be overly complex as I am limiting the study to just one variable per hypothesis.  Any additional variables would exponentially increase the length and complexity of the study.

 

Results

During the five years between the 2012-2013 season and the 2016-2017 season, there were 80 different teams that made the NHL playoffs during their respective years.  Out of these 80 teams, 59 were in the top 16 of the league in their PDO percentage.  In addition 59 out of the 80 teams had a PDO percentage above 1.000.  This means that 73.75% of playoff teams met the criteria mentioned above.

Additionally, during the five years involved in the study, 58/80 (72.5%) teams that made the NHL playoffs were in the top 16 Fenwick For percentage for their respective year.  In addition 56/80 (70%) teams had a Fenwick For percentage above 50%.  

59/80 (73.75) teams that made the NHL playoffs were in the top 16 of their respective years for high danger chances for while 56/80 (70%) of those playoff teams had a high danger chance percentage above 50%.

Finally, out of the 80 teams that made the NHL playoffs, 62 (77.5%) were in the top 16 of their respective years’ scoring chances for percentage while 62/80 (77.5%) had a scoring chance for percentage above 50%.

As a reminder for those who were unable to read my study on Corsi, 54/80 playoff teams were in the top 16 and 54/80 teams had a Corsi For Percentage above 50%.

 

Hypothesis 1

According to the data provided by naturalstattrick and the parameters set above, there is a moderately positive, while somewhat inconclusive correlation between having a Fenwick For percentage above 50% and/or in the top 16 and making the NHL playoffs.  As for hypothesis 1c, when compared to the respective Corsi For percentage, a teams Fenwick For percentage was shown to be a better indicator of NHL success.

 

Hypothesis 2

According to the data provided by naturalstattrick and the parameters set above, there is a moderately positive, while somewhat inconclusive correlation between having a PDO above 1.000 and/or in the top 16 and making the NHL playoffs.  As for hypothesis 2c, when compared to the respective Corsi For percentage, a teams Fenwick For PDO was shown to be a better indicator of NHL success.

Hypothesis 3

According to the data provided by naturalstattrick and the parameters set above, there is a moderately positive, while somewhat inconclusive correlation between having a High Danger Chance For percentage above 50% and/or in the top 16 and making the NHL playoffs.  As for hypothesis 3c, when compared to the respective Corsi For percentage, a teams High Danger Chance For percentage was shown to be a better indicator of NHL success.

Hypothesis 4

According to the data provided by naturalstattrick and the parameters set above, there is a STRONG positive correlation between having a Scoring Chance For percentage above 50% and/or in the top 16 and making the NHL playoffs.  As for hypothesis 4c, when compared to the respective Corsi For percentage, a teams Scoring Chance For percentage was shown to be a better indicator of NHL success.

 

Way Forward

Hockey is one of the most complex sports in the world.  The tiniest of variables can have a great affect on the outcome of a game or season.  By studying Corsi and other advanced statistics we are able to gain a better understanding of how success in the NHL is determined.  So far I have been able to conclude that Corsi, while a somewhat accurate indicator of success in the NHL, is not the most accurate.  In fact, every statistic that I compared Corsi to was a better indicator of success in the NHL.  Even PDO, the “luck” statistic, was a better indicator of success.  However, only a team’s Scoring Chances For percentage was shown to have a STRONG positive correlation with success in the NHL.  Even then, Scoring Chances For only had a 77.5% success rate in the study.  As mentioned in my Corsi article, 80% is the widely accepted threshold for reliability in correlations (see Kronbach’s Alpha).  In order to further the study of statistical prediction of success in the NHL there are a number of options.  First, as more and more things are measured in the NHL, more and more statistics will become available.  Perhaps one of these statistics will get to the 80% threshold. (HINT: probably not with any consistency).  The second option is to study various combinations of statistics and determine which formula is the best predictor of success.  (HINT: there won’t be any singular formula)

 

Conclusion

I will close with the same statement I made at the end of my Corsi article.  With the amount of data easily available to any person with an internet connection, it is simply irresponsible to take a single statistic and try to tell a story based on that statistic.  There is so much information available.  Take some time and put in some work and don’t fall into the trap of the easy narrative.

Mentioned in this article:

More About: