It's not easy facing the opponent's top pair every game.
One of the early steps in hockey analytics was developing quantitative measures for the strength of competition a player faced. You can look at the average +/- or Corsi of all of the opponents a given player faced and get a reasonable estimate of how tough his opponents were.
What makes me nervous about this approach is that players' usage and results are linked; we're judging my opponents by their results, which are impacted by the fact that they get used against players like me.
There are ways of getting around this challenge by putting in various adjustments for zone starts, teammates, and competition, but I wanted something simpler.Last week in an article at NHL Numbers, I described using time on ice to build a quality of competition metric. So instead of asking "who plays against the opponents with the best Corsi (or +/-)", we're asking "who plays against the opponents who log the most minutes?"
Since coaches tend to give the most minutes to their best players, this should be a decent ranking of quality of competition. And since it's not explicitly based on results, it should be largely free of the biases that come from a player's own skill and usage impacting the results of his opponents.
It turns out that this TOI Qualcomp has been calculated before, but it fell by the wayside several years ago as we moved into shot-based metrics. I'm reviving it here and adding a new twist: we can separate out the average TOI of the opposing forwards and the opposing defensemen that a player faced. This goes beyond "how tough were his opponents", to "what type of opponents did he face?"
We can also make plots illustrating a team's situational usage of their forwards. Here's Nashville's, for example:
We can see that Nashville matched power versus power. Their top forwards -- Erat, Fisher, and Kostitsyn -- land in the top right, facing the best forwards and the best defensemen on the other team. A sheltered scoring line would be in the top-left (facing strong defensemen and weak forwards), and a defensive shutdown line would be in the bottom-right (facing strong forwards and weak defensemen).
Here's what we see for the Flyers last year:
A few things stand out for me here:
- Giroux's line got very heavy attention from the opposing defense. Giroux ranked second in the whole NHL in strength of opposing defensemen. Hartnell (7th) and Jagr (13th) were not too far behind.
- The other lines all faced similar defenses. This is where the Flyers' forward depth shows, as the second and fourth lines were treated almost equally by opposing teams. Contrast that with Toronto or Vancouver, who also had a top line face premium attention but still had differentiation between the lower lines.
- This metric suggests that the top line faced much tougher competition than Corsi Rel QoC indicates. This was true in general across the league. I don't yet have data to say which is more accurate, so for now I'll look at both.
- Schenn got some pretty soft minutes.
- We knew Couturier and Talbot were used in a defensive role, but this clearly shows that they didn't get much defensive attention -- just a bit more than a typical fourth line.
Will the Flyers have a strong second line step up next year? Will Schenn blossom, and command more attention from the opponents? Will Voracek thrive as he moves up to face top lines? As you piece together proposed lineups for next season, think about how each might be deployed.
Answers to some FAQs:
- This measure uses the average total TOI of the opponents a guy faced at 5v5. So if I play 12 minutes and three of those are on special teams, my TOI Qualcomp will be based on who I faced in the nine even strength minutes, and I will factor into my opponents' TOI Qualcomp scores as a 12-minute player.
- I looked at defensemen too, but they're much less interesting. Defenses don't match up with the opposing defense to a significant degree, so the corresponding plot for a defense is basically horizontal. And the rankings also aren't that different from the conventional measures for defensemen -- the correlation between defensemen's TOI Qualcomp and Corsi Rel QoC was over 0.95.
- Home/road splits are on my to-do list.
- I'm very hesitant to say that this is better or worse than the results-based measures. I can see theoretical reasons for preferring either, and we don't have empirical proof of which is more predictive yet. So for now, I'll look at and think about both.