More on anecdotes, evidence and Corsi

Last night Tyler Dellow criticized me for using ‘anecdotes’ to come up with an idea (not a conclusion or a proof, more of a hypothesis) that by improving Corsi it might have a negative impact on your shooting percentage. Also last night James Mirtle retweeted an April tweet of his looking at the relationship between possession and Corsi and it was retweeted several times. Here is that tweet:


On the surface it sounds like pretty resounding evidence. Corsi is king! Corsi rules the NHL! Long live Corsi!

The problem is, this is solely backward looking and is only focused on Corsi. How can you say Corsi is the best when you aren’t comparing it to other possible factors? Furthermore he is using 5 year Corsi numbers to predict single season Stanley Cup winners which is not really a perfect scientific endeavour since just because you are good over a 5 year span doesn’t necessarily mean you are good in the particular season you won the Stanley Cup. In many ways this is anecdotal evidence but people love it. It was retweeted 275 times and liked 222 times.

Before I go any further I want to be clear. I am not bashing Corsi (it has value as you will see later) nor am I bashing Mirtle (he has been great for the analytics community). The purpose for this post is solely to point out that we need to put the value of Corsi in context. Corsi is important. Let me repeat that. Corsi is important. Let’s just put Corsi in its proper context.

The table below shows how the last eight Stanley Cup champion teams ranked in the league in CF%, Sh% and Sv% in the regular season that they won the Stanley Cup.

Stanley Cup Winner CF% Rank Sh% Rank Sv% Rank
2014-15 Chicago 2 28 2
2013-14 Los Angeles 1 29 2
2012-13 Chicago 4 5 8
2011-12 Los Angeles 2 30 4
2010-11 Boston 14 6 1
2009-10 Chicago 1 7 29
2008-09 Pittsburgh 19 1 11
2007-08 Detroit 1 22 7
Average Rank 5.5 16.0 8.0

When I see that chart I come up with the following hypothesis:

Stanley Cup winning Teams need to be good at CF% and one of Sh% or Sv%

Six of the eight Stanley Cup winning teams finished among the top four CF% teams. The two exceptions are Boston and Pittsburgh but a late season coaching change in Pittsburgh dramatically improved their CF% so they were likely playing like a top 10 (probably higher) CF% team come play off time. I’ll get to Boston in a minute.

Of the past eight Stanley Cup winners every one of them ranked no worse than 7th in either Sh% or Sv%. Four of the eight teams finished in the top 10 in Sh% and six of the eight finished in the top 10 in Sv%. Sv% seems more important than Sh% but Chicago won with the 29th worst save % in 2010.

Boston was likely the only real exception to needing to be among the league leaders in CF% but they overcome it by being one of the few teams to rank among the best teams in both Sh% and Sv%. Very few teams can manage to be good at both Sh% and Sv% which is why PDO hovers around 1000.

Stanley Cup Winner CF% Rank Best Sh% or Sv% rank
2014-15 Chicago 2 2
2013-14 Los Angeles 1 2
2012-13 Chicago 4 5
2011-12 Los Angeles 2 4
2010-11 Boston 14 1
2009-10 Chicago 1 7
2008-09 Pittsburgh 19 1
2007-08 Detroit 1 7
Average 5.5 3.6

The above chart shows where the team ranked in CF% and their best Sh% or Sv% rank. If you buy into the idea that CF% is important to winning the Stanley Cup, you should be willing to buy into the idea that to win the Stanley Cup you need to be exceptional at at least one of Sh% or Sv%. More generally you probably just need to be good at two of the CF%, Sh% or Sv% (see Boston).

This bringes me to another hypothesis I have:

CF% is in a large part coaching driven while the percentages are more talent driven.

A lot of my recent investigations has been focused on coaching and how it impacts statistics, in particular CF%. I believe there is a relationship between all three of CF%, Sh% and Sv% and focusing one one can have a negative impact on one or both of the others. But there is also another factor at play: $$. Compared to player talent coaching is cheap and it doesn’t count against the salary cap. Coaching tactics aren’t a limited resource either, it is a choice. Every team in the league could choose to play the exact same tactics and it would take no additional resources. If coaching and coaching tactics in large part drives Corsi, then it is perfectly understandable why CF% is so important to winning the Stanley Cup. It is, relative to accumulating player talent, an easy thing to do so it is near mandatory to be an elite team.

Taking it a step further, if Sh% and Sv% are more talent driven it makes sense that Cup winning teams can’t excel at both. Player talent is expensive and is a limited resource. It is really hard, if not impossible, to accumulate an over abundance of player talent to the point that you can excel at both Sh% and Sv%. But in addition to CF% you should probably excel at at least one of them.

Now, it is likely far more complex than this. As I have said, I believe there is a trade off between CF% and Sh%. The corollary of that is that there is probably a link between CF% and Sv%. There is probably also a link between Sh% and Sv% such that focusing in on one likely has a negative impact on the other (beyond if you invest $$ on Sv% players you have less $$ to spend on Sh% players). This isn’t baseball which is largely a series of head to head, individual talent vs individual talent, battles. Hockey is a tightly woven, inter-dependent game. It isn’t as simple as “Good CF% teams win the Cup so lets just become a good CF% team”. The reality is, as more and more teams focus on CF%, getting a good CF% will become more difficult to do and we may see the charts like Mirtle’s swing more towards Sh% or Sv%. Hockey analytics needs a more holistic approach. We need to understand the relationships between coaching, CF%, Sh% and Sv%. We need to better understand the relationships between different player skills. This is why I hate so much what Tyler Dellow did yesterday. I don’t care one iota whether Dellow agrees with me or respects my work (he is free to write a post with his alternate viewpoint) but his shallow, off the cuff, cheap shot retorts only serve to discourage people to come out with new ideas or hypotheses or dissenting view points not to mention it turns a lot of people off of analytics. Hockey analytics won’t advance if we all agree with what we already know and never look outside the box.