PDO in College Hockey

providence_hockey

If you’re not convinced by now that puck possession matters, you might as well go ahead and close this tab. Corsi is real, and it’s here to stay – yes, even in college hockey. Ryan Lambert at College Hockey News pretty much dropped the mic on the issue as the 2015 conference tournaments started, so I won’t waste any more words on it.

Though if you’d like to be spared from an RL article, I’ll just say this: of the top 20 CF% teams in the NCAA, 60% made the tournament (12/20). Of the 39 other teams, 10% made the tournament (4/39). More possession, more shots, more goals, more wins. What’s not to get?

The next thing everyone always brings up is, “yeah but shot quality.” Sorry, but shot quality is not a game plan, nor something even the best players can sustain. If your team has a high shooting percentage, it’s likely to regress the more they play. Same with save percentage. Hot goalies are usually just that – hot. Until they’re not, because .925 simply isn’t sustainable for most goalies. Eventually they’re going to have a few .795 nights.

The stat that measures all of that, of course, is PDO. Puckology has a great post on this. PDO adds the team shooting percentage and save percentage into one stat. The baseline for this stat is 100.0, because that’s the league average – always was and always will be (the average of all shot percentages in the league plus the average of all save percentages will always be 100%). So every team in a league will ultimately regress toward a PDO of 100.0. Your incredible goaltending will falter, your super-lucky “shot quality” will come down. That’s the nature of the game – regression to the mean is a statistical fact.

So here’s where things get interesting, and where I always got a little hung up: what I just said above isn’t 100% true. The baseline for every team or player probably is not 100%. The Chicago Blackhawks have more talent than the Arizona Coyotes, yes. But this is the NHL we’re talking about – each of those 30 teams is among the most talented hockey teams in the world (yes, even Buffalo). The baseline for the Blackhawks might be higher than the Coyotes, but not much. We’re talking about a thin band: PDO in the NHL ranges between 97.0-102.0. So is PDO talent? Probably not very likely in the NHL – it’s just variance from the mean, indicating likelihood of regression.

But is that still true in college hockey? You don’t have the same talent parity as in the NHL – most people wouldn’t put Boston University and Niagara in the same sentence. No one will argue there’s not a significant talent gap there. Sure enough, the PDO range in the NCAA is much larger than in the NHL – 95.7-104.5.

Therefore, PDO in college hockey might not be as indicative of unusual good/bad luck as it is in the NHL. Some teams might just have a naturally higher or lower baseline because of a higher level of quote-unquote talent.

I wanted to see if we could figure out just how much of PDO is talent-based in college hockey, if any at all. To do that, I compared it with the stat we know measures talent (well, possession, but those two things are highly correlated). That stat is Corsi.

I did this for both the NHL and NCAA. Theoretically, PDO and Corsi For % would have no relation in the NHL because the talent level on all teams is about equal. That turns out to be true:

Corsi / PDO correlation – NHL 14-15

CF% ES CF% Close CF%
PDO 0.07 0.06 0.169

There is essentially no correlation between Corsi For % and PDO in the NHL.

Now let’s look at NCAA. If talent plays a role in NCAA PDO, we’ll see some kind of correlation between PDO and Corsi For%.

Corsi / PDO correlation – NCAA 14-15

CF% ES CF% Close CF%
PDO 0.476 0.233 0.441

Here there is a significant positive correlation between Corsi and PDO, suggesting a larger slice of NCAA PDO is driven by something else other than “puck luck.”

Is talent the difference? Is coaching? Conference? Scheduling? Whatever is driving NCAA possession might also be driving PDO, in a way that isn’t happening in the NHL. And if that’s true, NCAA PDO values are less useful as indicator of unsustainable performance, good or bad.

Tournament participation is another indicator of this. Of the 16 teams that made it, 15 had PDO higher than 100! AND 14 had Corsi higher than 50%! Every single team had one or the other, meaning 13 of the 16 teams had both high Corsi AND high PDO. Of the remaining 53 teams, only six had both a high CF% and PDO, and guess what? They were 16, 17, 18, 19, 20, and 28 in Pairwise – they were your entire tournament bubble.

Furthermore, of teams in the NCAA with a CF above 50%, more than 2/3 also had high PDO (68%). In the NHL, that same criteria only amounts to 58%. So again, some evidence that PDO and possession are more related in NCAA than in the pros. And I bet if play was more round-robin and not confined to conferences, the effect would be even more profound.

Here is Corsi and PDO for all 59 teams in 2014-15:

NCAA 14-15 Corsi, PDO and PWR

Team Win% CF% ES CF% Close CF% PDO PWR
Minnesota State 0.782 59.3 57.9 58.9 102.0 1
North Dakota 0.731 51.9 51.5 52.8 102.6 2
Boston University 0.743 54.5 55.0 56.2 102.9 3
Miami 0.654 54.8 54.2 55.8 100.2 4
Denver 0.632 53.4 53.2 51.8 101.8 5
Minnesota-Duluth 0.566 54.1 55.6 54.5 100.1 6
Michigan Tech 0.750 55.2 55.5 58.0 103.4 7
Nebraska-Omaha 0.583 48.4 47.5 48.1 102.3 8
Harvard 0.625 48.5 48.2 48.7 102.6 9
Minnesota 0.645 54.4 53.6 53.8 101.5 10
Boston College 0.608 52.7 52.9 54.4 100.8 11
St. Cloud State 0.513 50.1 49.6 49.2 100.4 12
Yale 0.641 53.6 52.6 52.5 102.0 13
Quinnipiac 0.658 56.8 56.5 57.5 99.6 14
Providence 0.622 54.5 54.3 55.0 101.6 15
Bowling Green 0.654 52.5 52.6 53.7 101.4 16
Colgate 0.632 53.8 54.1 53.6 100.9 17
Mass-Lowell 0.615 54.0 53.9 54.4 102.3 18
Michigan 0.595 54.6 53.8 54.0 101.4 19
Vermont 0.585 51.7 52.4 51.7 100.6 20
St. Lawrence 0.581 47.0 48.0 50.6 104.5 21
Dartmouth 0.576 53.6 53.7 53.5 99.8 22
Northeastern 0.500 49.1 50.3 49.9 99.9 23
Alaska 0.588 54.5 54.6 54.7 98.7 24
Robert Morris 0.716 48.8 48.5 50.0 102.4 25
Western Michigan 0.446 49.6 51.0 52.7 99.0 26
Bemidji State 0.487 49.7 4.3 48.6 100.6 27
Union 0.513 51.8 50.5 49.4 100.9 28
New Hampshire 0.500 50.8 49.7 49.1 99.6 29
Merrimack 0.474 47.0 48.2 47.1 100.0 30
Michigan State 0.514 48.8 48.9 49.3 100.4 31
Penn State 0.541 54.9 55.1 56.7 98.8 32
Notre Dame 0.488 48.9 48.7 49.2 101.5 33
Ferris State 0.475 51.5 50.9 50.0 98.9 34
Northern Michigan 0.447 46.9 46.4 25.8 100.2 35
Cornell 0.452 49.9 48.8 47.8 99.2 36
Ohio State 0.431 48.6 47.2 45.4 98.2 37
RIT 0.566 52.9 53.8 54.7 100.3 38
Mercyhurst 0.539 46.3 45.5 44.2 101.7 39
Canisius 0.581 49.3 49.4 50.6 101.1 40
Bentley 0.527 49.9 48.9 50.8 100.8 41
Maine 0.397 50.4 50.1 51.0 98.8 42
Connecticut 0.375 43.4 42.7 41.8 97.7 43
Clarkson 0.392 55.1 56.1 55.3 97.7 44
Massachusetts 0.333 44.3 43.5 42.4 97.2 45
Rensselaer 0.329 46.1 48.1 46.6 96.8 46
Brown 0.307 42.6 42.4 42.5 97.6 47
Holy Cross 0.446 49.1 49.7 50.0 99.4 48
Air Force 0.439 53.8 53.6 53.9 98.5 49
Sacred Heart 0.421 49.9 50.7 50.2 98.7 50
Colorado College 0.214 44.0 43.2 41.9 96.2 51
Anchorage 0.294 43.4 44.8 42.7 99.1 52
Ala-Huntsville 0.263 36.5 39.1 39.6 99.2 53
Lake Superior 0.237 44.2 43.9 42.9 96.8 54
Wisconsin 0.186 41.8 42.8 41.7 96.4 55
Princeton 0.183 43.9 44.7 43.3 95.7 56
Army 0.294 49.4 49.4 49.8 97.0 57
Niagara 0.231 45.9 47.1 45.3 96.5 58
American Int'l 0.208 39.2 38.8 36.8 98.7 59

Take a look at your favorite team on this chart, then consider the combined indicators to get an early idea of how your team might fare next year.

  1. If a team has a high PDO and a high CF%, yes, they are as good as their record suggests – and we should mostly expect them to continue playing well. These are likely your tournament locks, and might not necessarily regress (BU, North Dakota, Mankato, etc.).
  2. If a team has a low PDO but a high CF%, they are probably better than their record suggests and maybe they can turn it around. These are probably your tournament bubble teams (Quinnipiac, Vermont, St. Cloud, etc.).
  3. If a team has a high PDO but a low CF%, they might be lucky, or they might have some serious talent (most likely in goal), or most likely a bit of both. A few of these are probably also tournament teams on good runs (Harvard, Omaha, Robert Morris).
  4. If a team has a low PDO and a low CF%, they are probably just not very good, luck or not. (Hunstville, Wisconsin, name-that-Atlantic-Hockey-club)

Miami in 2013-14 is a good example. They lost a ton of games, and had terrible shooting and goaltending. But their possession was really high. They were a decent team having a rough year. They bounced back in 2015 and won the NCHC (their PDO regressed right back to 100.2).

Similarly, if you’re a Wisconsin fan, I’d be worried (and not just because you are a Wisconsin fan). Because possession and PDO were both atrocious. This wasn’t just an off year – Wisconsin has some serious problems. Michigan and Minnesota, on the other hand, should be fine next year (ignoring roster defections, that is).

Possession is the best and most valid indicator of NCAA teams that are good. I haven’t seen anything disproving that, and I’m not arguing against that. What I am questioning is PDO, which in the NCAA seems difficult to peg: Teams with high PDO aren’t necessarily the best teams, except that in the NCAA, they mostly are. The Frozen Four teams all had high PDO – but of those four, the team with the lowest PDO won the national championship. And the team with the higest PDO in the NCAA – St. Lawrence – also had a 47% CF and didn’t make the tournament.

Corsi and PDO may have some common denominator in college hockey that doesn’t seem to exist in the NHL. And unlike the NHL, the successful NCAA teams are likely to be high in both.

It would seem this positive correlation needs to be explored a bit more before we can fully evaluate NCAA teams by PDO as is done in the NHL. Can it be modified to account for the wider variance in NCAA talent? Is there some other way to evaluate outlier performances and understands which teams are over/under-performing?

Leave a Reply

Your email address will not be published. Required fields are marked *