In an effort to understand the arguments behind the validity of possession as an indicator of team success I decided to analyze the results of the past 5 seasons. The advanced statistics are available going back to the 2007-08 season, so I plugged in all of those numbers as well as the season Point totals, then made a scatter plot so that I could compare the R² values to determine what factors have the greatest correlation to regular season points. I also decided to look at results over 2-year, 3-year, 4-year, and 5-year periods to see how it changes with the addition of more data. The factors we are looking at are Goals For, Goals Against, Goal Differential, Shot Differential, Fenwick Differential, and Corsi Differential. So, without further ado, here are our results:
GF .2139, GA .3117, Diff .6246, Shots .3942, Fenwick .3925, Corsi .3504
GF .3814, GA .6625, Diff .7283, Shots .2456, Fenwick .2375, Corsi .2499
GF .3995, GA .4344, Diff .7407, Shots .3729, Fenwick .3891, Corsi .4304
GF .3292, GA .5049, Diff .7043, Shots .3845, Fenwick .4494, Corsi .4675
GF .0139, GA .5201, Diff .6624, Shots .1709, Fenwick .2269, Corsi .2080
So for single season point totals, the research appears to be incorrect, as we already see that goal differential has the highest correlation with regular season points. For 3 of the seasons we actually see a strong correlation (value of .700 and above). Except for one strange result in the 2007-08 season, GF and GA always gives us a moderate correlation (values from .300 to .700). All of the possession statistics range from a weak correlation (below .300) to a moderate correlation, but it varies from year to year whether Shots, Fenwick, or Corsi are better indicators. However, if we look at the average we find that Corsi has the slight edge (.3412) over Fenwick (.3391) while they are both higher than Shots (.3136). Interestingly, the numbers also indicate that defense is far more important than offense (GF .2676 vs GA .4867).
If we look at the team rankings by season, we see that in 2012 the bottom 9 teams had negative possession values while 3 of the top 5 had positive possession numbers (and 8 of the top 10). But the worst possession team (Minnesota with -794 Corsi) was #24 and the best possession teams (Los Angeles with a +657 Corsi and Detroit with a +492 Fenwick) were #13 and #8 respectively. So leading the pack in possession does not necessarily indicate one will finish the regular season at the top of the rankings, nor does a lack of possession indicate a tendency for failure. However, there is at least some correlation between the two. But to look at Fenwick/Corsi and say "This team is destined to succeed/fail because their possession numbers are high/low" is a little misguided.
Looking at the remaining seasons, we see that the trend doesn't not necessarily hold up year after year. In 2011 2 of the Top 3 (and 7 of the Top 10) were positive, while 2 of the Bottom 3 (and 7 of the Bottom 10) were negative. The best possession (San Jose) was #5, while the worst (Anaheim via Corsi, Minnesota via Fenwick) were #9 and #21 respectively. In 2010, all of the Top 10 had positive numbers, while 2 of the Bottom 3 (and 9 of the Bottom 10) were negative, which is what we would expect. The highest (Chicago) was #3, while the lowest (Florida via Corsi, Edmonton via Fenwick) were #28 and #30 respectively. It makes sense that this happens to have been the time when Corsi/Fenwick became popular, as the results supported the accuracy of using it as a predictor. 2009 was back to having 7 of the Top 10 with positive numbers, and 8 of the Bottom 10 with negative numbers. But the highest (Detroit) was #3 and the lowest (Islanders) were #30, so it still supported the supposition. 2009 saw 2 of the Top 3 (and 6 of the Top 10) were positive while 7 of the Bottom 10 were negative. The highest (Detroit by a WIDE margin) finished #1while the lowest (Atlanta) finished #28.
So when they started studying Corsi and Fenwick, using data obtained from the 2008-10 seasons, it made perfect sense to believe that they were the best available indicators of team success and that dominant puck possession numbers led to finishing at the top of the regular season standings. However, the recent data available in the 2010-12 seasons starts to show us that maybe this is not always the case. So with that in mind, let us have a look at multiple season data.
GF .4604, GA .5632, Diff .8130, Shots .3880, Fenwick .3857, Corsi .3774
GF .5054, GA .6344, Diff .8015, Shots .4788, Fenwick .4895, Corsi .5477
GF .3365, GA .6105, Diff .7979, Shots .4972, Fenwick .5504, Corsi .5692
GF .1613, GA .6073, Diff .7784, Shots .2943, Fenwick .3730, Corsi .3752
We see a continuing trend using the 2-year numbers. The Goal Differential is still the most accurate indicator of team success, having a strong correlation every time. Goals For and Goals Against has moderate correlations, except for the one off result due to the abnormally low 2007-08 numbers. Possession increases to a moderate correlation in all but 1 instance (again due to that abnormally low 2007-08 season). However, like the single season results we see that the viability of Shots vs Fenwick vs Corsi varies from year to year. But again, looking at average results we find Corsi (.4674) to have the edge over Fenwick (.4497) and Shots (.4146). In recent seasons we see the usefulness of possession to be declining, and continued importance of defense over offense.
Looking again at overall rankings, we see 2010-12 has 8 of the Top 10 with positive possession numbers, while 9 of the Bottom 10 are negative. The highest (Detroit) is #4 in points, while the lowest (Minnesota) is #22. 2009-11 we see all of the Top 10 positive and 8 of the Bottom 10 negative, with the highest (Chicago) at #4 and the lowest (Edmonton) at #30. For 2008-10 we see 9 of the Top 10 positive and 9 of the Bottom 10 negative, with the highest (Chicago) at #3 and the lowest (Florida) at #24. Finally 2007-09 we have 7 of the Top 10 positive and 8 of the Bottom 10 negative, with the highest (Detroit) at #1 and the lowest (Atlanta) at #27. In 2-year intervals, the possession numbers are slightly less indicative than they were in single seasons, although Detroit's impressive 2008 and 2009 cup years do much to support the use of possession numbers. So, moving on to 3-year numbers.
GF .4973, GA .6027, Diff .8740, Shots .5392, Fenwick .5559, Corsi .5743
GF .4485, GA .6986, Diff .8525, Shots .5458, Fenwick .5846, Corsi .6267
GF .2234, GA .6283, Diff .8298, Shots .3939, Fenwick .4642, Corsi .4641
At the 3-year level we start to see an indication that possession is more important that scoring, and all the numbers are now in the range of moderate correlation. However, Goals against still dwarfs the possession numbers, and Goal Differential is still the most accurate predictor of team success. The Anomalous 2007-08 season continues to drag down the results, with a weak correlation in GF for that period. Without even having to look at averages, we see Corsi remains the best possession metric, despite research that claims otherwise (although those studies were down back in 2011, so the 2007-10 numbers available at the time would have supported the fact that Fenwick was better as a long term predictor).
Again looking at team rankings, in 2009-12 we have 9 of the Top 10 with positive possession numbers and 9 of the Bottom 10 with negative numbers. The highest (Chicago) is #5 while the lowest (Minnesota) is #24. For 2008-11 we still have 9 of the Top 10 positive, but only 7 of the Bottom 10 negative. The highest (Detroit via Corsi and Chicago via Fenwick) are #4 and #5 respectively, while the lowest (Edmonton) is #30. As mentioned, this encompasses the period in which possession numbers as an indicator of team success first came into the mainstream view. Lastly, 2007-11 has 6 of the Top 10 with positive and 8 of the Bottom 10 with negative numbers. The highest (Detroit) is #2 while the lowest (Atlanta) is #27. Next we will look at the 4-year results.
GF .4264, GA .6817, Diff .8873, Shots .5631, Fenwick .6034, Corsi .6264
GF .2863, GA .7087, Diff .8453, Shots .4471, Fenwick .5112, Corsi .5319
Our continuing trend shows that possession has some correlation to team success, and has handily surpassed goal scoring. However, Goals Against continues to be more important, with it jumping to a strong correlation in one instance, and the ability to outscore the opponent still being the single most important factor. Team rankings we see 2008-12 has 8 of the Top 10 with positive possession numbers and 9 of the Bottom 10 with negative numbers. The highest (Detroit via Corsi and Chicago via Fenwick) are #4 and #5 respectively, while the lowest (Edmonton) is #30. For 2007-11 7 of the Top 10 were positive while 7 of the Bottom 10 were negative. The highest (Detroit) was #2 while the lowest (Edmonton) was #29. Then we have our 5-year results.
GF .3308, GA .7117, Diff .8840, Shots .4913, Fenwick .5531, Corsi .5654
So while possession has edged out Goal For in long term usefulness, it still manages to pale in comparison to Goals Against and Goal Differential, both of which has a strong correlation. Some would indicate that puck possession is the reason for this. That may be true for some teams (particularly Detroit and Chicago), but other teams manage to accomplish this through quality defense and goaltending despite losing the possession battle. One must always remember that correlation does not imply causation, so while we may be able to see some link between the two, it is unable to give us the big picture and proof as an indicator. A team with great possession can have great scoring differential (like Detroit) or low score differential (like New Jersey), yet both manage to be successful. Or a team with low possession numbers and highs score differential (like Pittsburgh) or middling score potential (like the Rangers), yet both of those teams manage to be successful. Then there are teams with negative possession numbers (like Philadelphia and Nashville) that have middling to low scoring potential yet still manage to succeed. At the other end of the spectrum we have teams with positive, albeit low possession numbers and massively negative scoring differential (like Columbus, Toronto, and Ottawa).
This season is not complete, and the limited data would make the results less than ideal, but let us compare our current standings to see how this year matches up.
2013 Season (48 games)
GF .3427, GA .4848, Diff .7447, Shots .2059, Fenwick .1831, Corsi .1409
It would seem that despite the research that indicates that Corsi is more effective with limited data, the opposite is actually true. The same research also indicated that Shots were worse than Fenwick, but gained a slight edge as more data was provided. It would appear that this is also untrue. But with a small amount of data, the ability to score goals is more important than possession. GF and GA both have moderate correlation, while all 3 possession statistics have a weak correlation. So to use possession numbers as a basis for determining the results of this lockout shortened season appears to be quite absurd. There is just no way to account for the luck factor. No, the most important indicator, from single season all the way up to 5-Year, is the team's Goal Differential.
Considering the results we see based on a smaller amount of data available, it would also be safe to assume that a playoff series, being just 7 games, has a lot more in common with this lockout shortened season than a full 82-games season. So in the playoffs luck, and special teams, become a much great factor than simple puck possession. We can make the same leap for evaluating individual players, as it appears with a limited amount of data Corsi is far less effective (and individual player performance has much less data available than a full team's worth would), so when evaluating players it may be wise to look at +/- and Fenwick/Shots rather than focusing on Corsi.
I suppose we could do a quick analysis of playoff performance based on possession as well, so I will attempt to run those numbers and post it as a reply. For now, I am satisfied with my results, and a quick look at the current standings leads me to believe that Chicago and Pittsburgh has the best chance, despite Pittsburgh's low possession numbers, but unless we can manage to improve our defensive ability and keep pucks out of the net Chicago will come away victorious. The most likely runners up are Montreal in the East, who will certainly give us a run for our money, and Anaheim in the West, despite their massively negative possession numbers they are one of the best teams in the league. New Jersey and Carolina are great examples of teams with big possession numbers but horrible standings. Though they are both being outscored, so with differential being a better indicator than possession this is no surprise. Toronto, like Anaheim, has one of the better chances, despite having one of the worst possession numbers int he entire league. Yet even with the #29/30 Corsi/Fenwick, they outscore their opponents.