Forward vs Defense Player Evaluation

Today Travis Yost of TSN.ca put up an interesting post where he ranked forwards and defensemen based on their ice time with their team (1-12 for forwards and 1-6 for defensemen) and then looked at each group (1-12) average to see how the forwards performed.

To illustrate this, I took the average Corsi% for every forward who led his team in 5-on-5 ice-time, then repeated the same through the twelfth forward. You’ll notice how the first line looks great, the second and third lines look average, and the fourth line looks poor.

 

When I do this I get the following chart which is more or less the same for the purposes of the analysis but is different enough that we must have used different methodology (maybe Yost used Score Adjusted Corsi as opposed to straight 5v5).

ForwardsCFPctByTOIRank

Regardless of the differences in the charts, the overall trend is the same, first and second line forwards have better CF% than third and fourth line forwards. Yost’s conclusion from this is that coaches are good at handing out ice time to their forwards giving the best forwards the most ice time. While true, I think Yost is missing one key point which can easily be seen by splitting apart CF% into its components – CF/60 and CA/60.

ForwardsCFCA60ByTOIRank

Basically what is happening is coaches are handing out more ice time to players that generate more offense. Shot attempts against are pretty stable from forward 1 to 12.

The next obvious question is, how do the percentages look?

ForwardsShSvPctByTOIRank

The most ice time is given to the high shooting percentage players and the least to the low shooting percentage players. That said, the significantly higher than normal shooting percentages are mostly just for the top 3 or 4 forwards on a team. Beyond that the drop off is much less significant as you go down the forward ranks. The impact is significant though. The average first line forward has an on-ice shooting percentage 1.6 percentage points above the average 3rd line forward resulting in first liners scoring >23% more goals on an equal number of shots.

Another interesting observation is that save percentage steadily rises as you drop down the line up. The average #8 forward has an on-ice save percentage that is about 0.6% higher than the average #1 forward. That’s a smaller impact than shooting percentage but not completely insignificant either.

I’d probably ignore the spikes seen in the 12th best forward. Sample size might be an issue and it is an interesting combination of players, some high end younger players and some injured players like Brandon Sutter.

Now, combining all this together we can look at goal rates for and against.

ForwardsGFGA60ByTOIRank

As expected, the top line scores far more than lines 2-4 and that is largely due to their significantly better shooting percentage. Defensively there are not significant differences from lines 1 through three with a bit of a tail off at the very bottom of the line up.

ForwardsGFPctByTOIRank

The conclusion in all of this is forwards get assigned ice time based significantly on their offensive production which drives their overall goal differential.

Now for defensemen starting with 5v5 CF%.

DefenseCFPctByTOIRank

As with what Yost found there isn’t a consistent trend here where the top ice time defensemen have the best CF% and it slowly drops off as you move down the line up. Instead it is fairly flat. The conclusion here would be that coaches don’t really hand out ice time to defensemen in an efficient way but is this really true once we start looking at the data in more detail?

Let’s look at CF/60 and CA/60 first.

DefenseCFCA60ByTOIRank

So, here we start to see something. The #1 defenseman is, on average, a very good two-way defenseman generating the most offense and good at defending as well. The #2 and #3 defensemen appear to be the worst offensive defensemen but also give up a fair number of shots as well which is not a good combination. We see this in the previous chart where the #2/#3 defensemen have the worst CF%. Defensemen 4 through six are more balanced defensemen.

And the percentages?

DefenseShSvPctByTOIRank

Interesting spike in shooting percentage for the #4 defenseman. It could mean the #4 guys are more offensive specialists that get more ice time with the top lines of teams.

The #2 and #3 defensemen that seemed do poor corsi-wise now appear much better having by far the best on-ice save percentages which are on average about 0.4 percentage points higher than the rest of the defense.

How does this all wash out in terms of goals for and against?

DefenseGFGA60ByTOIRank

The highest goals against rates are by #1 defensemen with #3 defensemen having the best (lowest) goals against rates. Interestingly the best offensive defensemen are on average #4 defensemen followed by top pairing defensemen.

 

And GF% overall?

DefenseGFPctByTOIRank

Interesting that the #1 defensemen have pretty much the worst goal differential while defensemen 2 thru 4 are much better all round. It makes you think that maybe the defensemen that teams have identified as #1 guys might be over used. Makes you wonder how many teams would be better off playing their #4 defensemen in a #1 role and vice versa.

Now let’s finish with one final plot for defensemen comparing CF% and GF%.

DefenseGFPct_vs_CFPct_ByTOIRank

Wow, do the GF% and CF% lines tell dramatically different stories. Aside from how poor the #1 defensemen appear the GF% is a much more reasonable plot with defensemen 2-4 being quite good with a significant drop off to the 5th/6th guys. That makes more sense than CF% where the 5th and 6th guys are significantly better than #2 and #3.

It is probably worth looking at other years or revisiting this later this season when sample sizes are larger before we draw any significant conclusions but the one thing I want everyone to take away from this post is it that just looking at Corsi% may not be telling the whole story. There may be far more interesting and more important messages to be found looking beyond Corsi%. In this case there is more to the story than what Yost’s posts portrays. For forwards it isn’t so much Corsi% that is driving coaching decisions for forwards but rather the overall offensive ability of the forward with defensive ability having relatively little importance.  For defensemen there appears to be some ability for #2 and #3 defensemen to boost save percentage and that appears to be an important reason that they get ice time over those further down the depth charts while the #1 defensemen might be getting mis-used or over-used. Again, this is not necessarily apparent when just looking at Corsi% but is critically important to understand what is really happening from an analytics point of view.

When I first got into hockey analytics I spend a fair bit of my time trying to come up with an all-inclusive player evaluation statistic. The more I dig into the data the more I realize that all-inclusive stats or high level stats often over generalize and miss the point. In hockey players are often given very specialized roles based on their skills and their resulting statistics are a result of a combination of their skills and the roles they have been given. There is a lot of nuance in hockey that we can’t extract from a single high-level stat like Corsi%.

 

This article has 4 Comments

  1. Excellent post – I like that you separate out SH% as a component of possible skill. I too find there is a lot of analysis done that relies heavily on Corsi to come up with results but it tells only a tiny fraction of the story. As for your concept of a player evaluation model – I have created one (found here: http://xtrahockeystats.com/wordpress/?p=15). As you mention, lots of different categories of skill are required to get the true picture of a player.

  2. Thanks, great post as always.

    One question, as it wasn’t really discussed. Top forwards have the highest save percentage, top defenders have lower save percentages, wouldn’t be a logical assumption that top D are playing more against top forwards?

    I have seen a lot of other articles out there which indicate competition balances out, but this has always felt counter intuitive to the concept of line matching and wondered if the stats are just lacking in how we track competition levels.

    1. It is possible that quality of competition may come into play but generally QoC doesn’t appear to be significant. I’ll have another look this weekend using this methodology and I’ll write a post if I find anything.

      1. One typo there, intended to say top forwards have the highest shot percentage, top D have lowest save percentage. I’ve seen a number of articles on QOC not being all that important in the end, but then stats like this where it would seem to be a simple conclusion that the top D are being matched against the best shooters.

        My real question would be related to how competition is tracked. It would be interesting to see if there are discrepancies between number or quality of shot attempts versus the time against and rank of competition. Or in other words that TOI or Corsi of Competition might not be the best measures.

        I’m kind of opening a lot of questions, if you find something it would be interesting, but either way thanks for the work you do on this site.

Comments are closed.