WHEN ARE SH% AND CORSI/FENWICK EQUALLY PREDICTIVE
First off, shooting percentage and goal scoring rate (i.e. goals per 20 minutes) are not ‘linked’ any more than fenwick rate and goal rate are.
Goals For per 20 minutes = Fenwick For per 20 minutes * Fenwick Shooting Percentage. If I can correlate FenF20 with GF20 then I can surely correlate FenSh% with GF20 and FenSh% is much more highly correlated, whether I use 1 year of data or 4 years of data.
Shooting percentage varies widely from season to season, more so than corsi/fenwick, but that doesn’t mean it is not a ‘talent’. The variation is due to the smaller sample sizes associated with goals vs shot attempts.
The question that I answered is, how much data is needed before shooting percentage becomes the better predictor of future goal scoring? The answer I came up with was that with 2 years of data shooting percentage and fenwick rate have more or less equal predictive value.
Finally I ask, why not just use goal scoring rate as the predictor of future goal scoring rate since goal scoring rate encompasses both the ability to generate chances (fenwick rate) and the ability to capitalize on those chances (shooting percentages). I found that the predictive value of goal scoring rate on future goal scoring rate equals or exceeds that of fenwick rate with just 1 year of data. Beyond one year of data goal scoring rate is by far the better predictor. Less than 1 year of data and corsi/fenwick is probably the better predictor.
GOAL RATES BETTER THAN CORSI OR FENWICK
(May 2011)
The belief by many that support corsi and fenwick is that by looking at fenwick +/- or fenwick ratio (i.e. fenwick for /(fenwick for + fenwick against)) is an indication of which team is controlling the play and the team that controls the play more will, over time, score the most goals and thus win the most games. There is some good evidence to support this, and controlling the play does go a long way to controlling the score board. The problem I have with many corsi/fenwick enthusiasts is that they often dismiss the influence that ability to drive or suppress shooting percentage plays in the equation. Many dismiss it outright, others feel it has so little impact it isn’t worth considering except when considering outliers or special cases. In this article I am going to take an in depth look at the two and their influence on scoring goals on an individual level.
What we find is shooting percentage is more correlated with goal production than fenwick rate.
Shooting percentage is much more highly correlated with goal scoring rate than fenwick rate is which would seem to indicate that being able to drive shooting percentage is more important for scoring goals than taking a lot of shots.
Shooting Percentage is a better indicator of offensive talent than Fenwick For rates.
An organization that focuses on puck control dominates the corsi for statistic so I guess what that shows is that corsi/fenwick probably is a good measure of puck control. But, as we have seen, fenwick (i.e. puck control) doesn’t automatically translate into goals scored.
Shooting percentage matters a lot in scoring goals, but for the staunch corsi supporters they will argue that corsi is more persistent from season to season and thus is a better predictor of future performance. So which is the better predictor of future performance?
When dealing with a single season of data the correlation with GF20 is much better for fenwick rate than for fenwick shooting percentage. The gap closes when using 2 seasons as the predictor of a single season and is almost gone when using 2 seasons to predict the following 2 seasons. It seems that the benefit of using corsi over shooting percentage diminishes to near zero when we have multiple seasons of data and though I haven’t tested it shooting percentage probably has an edge in player evaluation with 3 years of data.
When looking at single seasons both GF20 and FenF20 perform similarly at predicting next seasons GF20 with fenwick shooting percentage well behind but when we have 2 years of data as the starting point, GF20 is the clear leader. This means, when we have at least a full seasons worth of data (or approximately 500 minutes ice time), goal scoring rates are as good or better than corsi rates as a predictor of future performance and beyond a years worth of data the benefits increase. When dealing with less than a full season of data, corsi/fenwick may still be the preferred stat when evaluating offensive performance.
What about the defensive side of things?
Defensively, fenwick against rate is very poorly correlated with future goals against rate and it gets worse, to the point of complete uselessness, when we consider more seasons. Past goals against rate is a far better predictor of future goals against rate
In summary, it should be clear that we cannot simply ignore the impact of a players ability to drive or suppress shooting percentage in the individual player performance evaluation and so long as you have a full year of data (or > 500 or more minutes ice time) the preferred stat for individual player performance evaluation should be goal scoring rate.
Corsi/fenwick likely only provide a benefit to individual performance evaluation when dealing with less than a full year of data.
TEAM STATS AND PLAYOFF SUCCESS
take a look at PDO and GF% in 5v5close situations to see if they translate into post season success
As HabsEyesOnThePrize.com found, 5v5close FF% is definitely an important factor in making the playoffs and enjoying success in the playoffs. That said, GF% seems to be slightly more significant.
PDO only seems marginally important, though teams that have a very good PDO do have a slightly better chance to go deeper into the playoffs. Generally speaking though, if you are trying to predict a Stanley Cup winner, looking at 5v5close GF% is probably a better metric than looking at 5v5close FF% and certainly better than PDO.
VALUE OF CORSI POSSESSION MEASURED IN GOALS
The average on-ice shooting and save percentages a player experiences tends to be influenced by their average time on ice per game. This relationship likely occurs due to a combination of factors: shooting talents of linemates and opponent, defensive talents of linemates and opponent, system and psychological effects, and an effect I like to call “streak effects” - Players on hot streaks tend to be given more ice time by coaches, while players on cold streaks tend to receive opposite treatment. This may cause an artificial inflation of a players TOI post hoc. A player who has benefited due to bounces may garner more ice time the next few games. Unless the player then receives an equal amount of poor luck with bounces, both TOI and scoring rates become inflated.
Regardless of the reasons why, these effects indicate that not all Corsi percentages are created equal in impact.
Corsi used in player evaluations can never be separated from the conditions and environments that surround said player.
These conditions are not limited to the contextual nuances in how a player is deployed -like zone deployment, linemates, and line-matching-. Other conditions include ice time, coaching system, the player’s “role” and what game state a player tends to be deployed in (leading, lagging, tied, etc.).
You can describe the value of Corsi in terms of goal differentials over a season.
FORWARD & DEFENSEMEN GOAL REGRESSION
There is a relationship between a player’s Corsi% and Goal%. Players who tend to out shot-attempt their opposition also tend to outscore their opposition.
The relationship holds true for defensemen as well. However, the defenders spread out less than their forward counterparts.
There have been some who say Corsi% is not equally relevant for defenders as forwards, but initial testing seems to indicate differently.
A player’s Corsi% between year one and year two has a far more significant relationship than Goal%. In these regressions, Goal% is moving more towards Corsi% than vice versa.
The reasons why Goal% is the stat that regresses is likely similar -if not the same- to why there is a difference in the first place. While an individual is not the sole driver of their own Corsi% due to contextual nuances (ex: usage effects), Goal% has additional outside factors. On-ice percentages are a major driver of Goal%, but players have difficulty in controlling the on-ice percentages experienced while they are on the ice. For example, there is almost no sustainability in a defenseman’s on-ice save percentage.
FORWARDS USAGE (GF% & CF%) BY RANK ON TEAM
As you move down a teams depth chart at forward both CF% and GF% drop off as one might expect. It is important to note though that the slope of the GF% line is much more significant than that of the CF% line. The result is that typically top four forwards on a team will out perform their CF%, forwards 5-7 will perform on par with their CF% and beyond that players will typically under perform their CF%. This means that while CF% may tell you which players are better than others, it will fail to fully identify the relative strength of good players over poor players.
Taking a look at CF/60 and CA/60 we see the two behave quite differently.
From top to bottom, CA60 is relatively stable but CF60 drops off as you move down the line up. We see the same thing with the percentages.
Save percentages fluctuate a bit but are mostly stable moving down the line up. Shooting percentages on the other hand generally drop off as you move down the line up.
GA/60 is relatively stable while GF/60 varies quite a lot. The conclusion is that good offensive forwards get awarded ice time and defensive play generally has no impact.
The key takeaways from this analysis are:
· Better offensive forwards get more ice time
· Being better offensively is a combination of generating more shots and higher quality shots.
· Because shot quality is a talent and varies throughout the line up, a Corsi evaluation of forwards will underestimate the value of the top forwards and over estimate the value of weaker forwards though it will typically identify which are the best forwards and which are the worst.
· PDO doesn’t regress to 100 for all players, particularly forwards and rather than suggesting regression to 100 one should view a range of 97-103 as being a typical range for forwards over large sample sizes (a season or more).
EVALUATING PLAYER EVALUATION METRICS
The two questions we ought to try and answer are:
1. How much of a players overall value does Corsi explain? (definind an upper limit)
2. At what sample size does Corsi explain a players value better than goals do and as sample sizes increase, at what point does a goals based analysis surpass a Corsi based analysis.
Answers:
1. The upper bound on CF60 explaining offensive talent of forwards is 53%. Put in different terms, using CF60 can at best explain 53% of a players offensive talent. Corsi at best can tell you half of a players overall value and that is only if you are able to isolate all other factors such as usage, quality of teammates, quality of opponents, score effects, etc. That isn’t very good.
2. At one year of data GF60 is a the better predictor of next seasons GF60. For forwards with >500 minutes of ice time you are better off using GF60 over CF60. The sample size where CF60 is better to use in evaluating a players offensive performance is between zero and something less than 500 minutes of ice time. That’s it. A partial season. And even then you can’t even explain half of what makes a good offensive player. That makes Corsi a not overly useful statistic for player evaluation in the grand scheme of things
To summarize, Corsi can at best explain approximately 50% of a players offensive talent and somewhere around or a little less than 1 year (500 minutes) is where the sample size is large enough that GF60 is a more reliable measure of a players offensive performance.
CORSI VS GOALS PREDICTING CONF FINALS
Generally speaking GF% does a better job at predicting the conference finalists than CF% and while there was some benefit to using score adjusted data it was a fairly small benefit and CF% benefited more than GF%. So while the crux of Neil Greenberg’s tweet is true, score adjusted CF% does a decent job at predicting conference finalists, what is missing is the fact that there are seemingly better stats to use for this.
GF% BETTER THAN CF%, PDO, xGF%
(May 2021)
I want to take a look at some team level statistics and see what correlates highly with a team’s record (given the weirdness of the NHL point system where the loser point exists for OT losses, a team’s record = their points %).
I will compare CF%, FF%, SF%, ‘GF%’, ‘xGF%’, ‘SCF%’, ‘SCSF%’, ‘HDCF%’,’HDSF%’, ‘SH%’, ‘SV%’, ‘PDO’.
The top metrics that correlate with team record are GF%, PDO, SH%, and then xGF%.
It’s no surprise that GF% is so highly correlated with record given that a game result is a function of goal differential of 1 game and GF% is a function of goal differential across multiple games.
PDO for the period measured of 104 and 96 accordingly, it’s probable a team that has a PDO far outside that range will regress into that range over the course of the season. PDO is normally distributed with a mean of 1. There is only a .16% probability of a team finishing the season outside the range of .96–1.04 which is 3.15 standard deviations away from the mean in both directions.
I also combined PDO and xGF% in a multiple linear regression model. It makes sense to combine these two given they measure difference components of team success and PDO has a very low correlation with xGF%, thus avoiding any multicollinearity issues.
GF% has an R² .93 explaining 93% of the variability in Points %. The PDO and xGF% model has an R² of .89, so these features together explain 89% of the variability in Points %.
Comentarios