When I wrote about analytics and PDO a few days ago, I said that for a metric to be useful, it must have some level of predictive usage and/or provide a new accurate window into the past.
Like many fans of my age or older, I can recall “the glory days” of the early 1990s as a terrific few years for hockey. The game was wide open; it featured a great balance of speed, skill, and physical play; and it was at a high level. New systems were entering the NHL, as were new ways to play a position. Top-level rookies were coming in and performing well, and European players were showcasing their own unique skillsets. There was expansion and some really putrid teams, and there were some new colors on the ice. Whatever you liked most about hockey, it was visible in almost any game to an extreme level.
Official PDO only goes back to the 2007-08 season. And remember, a metric is only as useful as its ability to predict the future or to enlighten about the past. Could it hold up in possibly the most extreme example, which is the 1992-93 season?
Limitations on PDO
By its very nature, PDO can go back no further than the 1983-84 season, which is the first time that shots against were actually tracked. Without one entire half of the metric (saves, therefore save percentage), there is no way to compile the older stuff.
In the case of the pre-2007 NHL, there is either limited or non-existent totals on situational shots for and against. This means that we’re also inherently limited to all-situations PDO, and even then only if someone is willing to compile it.
In other words, every season from 1983-84 to 2006-07 will have to be compiled by hand.
Someone is willing to compile it, and that’s me.
To start with, I went back through and compiled all the information for the 1992-93 season to see if PDO has changed over time. Of course, by sticking a stake in the ground 15 years before the previous earliest year (2007-08) without filling in between, this info is only interesting instead of meaningful…for now. This will change as more seasons get filled in.
First, shot totals both for and against. Then, goals for and goals against. Seems simple.
Remember though, every shot taken by one team is a shot against another team, and every goal for a team is a goal against another. Therefore, the numbers should match perfectly.
Unfortunately, this was not the case. The shot info I found was off by a substantial margin, meaning that a significant amount of work had to be done to correct this. And even after checking and re-checking, the shots for compared to shots against ended up off by a total of 138 shots.
After getting up and pacing for some time, then re-checking my numbers again, I realized my error. In order to determine shots for, I’d broken it down by team and combined those to find the leaguewide total. In order to determine shots against, I’d broken shots against down by team and combined those. But what I’d done was compile each team’s combined shots against a goalie, totally missing the boat on empty net goals. Those are charged as both a goal against and a shot against, but don’t show up as a shot against a goalie. The result of this was that leaguewide PDO didn’t come up as 100, but rather as something like 99.975. That’s close, but not right.
So I went back and re-compiled the information on goals against each team and goals against the actual goalies. If my theory was correct, each team would have small differences; those are the empty net goals against.
Success! There were in fact 138 empty net goals scored leaguewide in 1992-93. This bridged the gap, and re-set the leaguewide PDO in 1992-93 to 100 instead of 99 and some change.
Remember when I said that researching and compiling can create a wide range of emotions in a short period of time? It did. The worst part was that my wife had to work an early shift and was already in bed, so I had to keep my initial rage and my relieved joy significantly repressed.
1992-93 and PDO
The following table shows each team’s all situation PDO in 1992-93. The key numbers are on the right: “PDO” is the original number I had without taking the empty net goals into account, and “AdjPDO” is the big number that has everything properly balanced and comes out to a leaguewide total of 100.0000000000. SV% is the total number of goals against from number of shots against; TmSV% includes the empty net shots (which had to be adjusted for). In this table, AdjPDO is simply S% plus TmSV%.
You can see the extremely small deviations between the two numbers on the right, with only Vancouver remaining unchanged. Strange fact: Vancouver did not allow an empty net goal against in 1992-93, and that’s why the numbers are the same.
Stratification and playoffs
In my original PDO post, I mentioned that an all situations PDO above 100.9% virtually guaranteed a playoff spot, while one below 99.2% all but guaranteed an early date on the golf course. This was based on the official data that exists from 2007-08 to 2015-16, but I’m naturally curious to see if this has changed over time.
- 8 teams had a PDO above 100.9%, and all 8 made the playoffs.
- 8 teams had a PDO between 100.0% and 100.9%; 6 made the playoffs while 2 (Philadelphia and Hartford) missed.
- 3 teams had a PDO between 99.0% and 99.9%; 2 made the playoffs while 1 (NY Rangers) missed.
- 5 teams had a PDO between 98.0% and 98.9%; all missed the playoffs.
- 3 teams had a PDO not just below 98.0%, but below 97%. All missed the playoffs.
Ottawa’s jaw-dropping 94.084% all-situations PDO may be the worst mark in history, but more research is needed.
Perhaps you’re noticing a couple things, particularly since Hartford missed the playoffs by such an enormous margin. I believe the explanation is a simple one: the Whalers were extremely top-heavy in scoring, had poor depth, and poor defensive play that resulted in an incredibly large shot differential. Ottawa, which finished 60 games under .500, had a shot differential of -383. San Jose, also 60 games under .500, had a shot differential of -691. Hartford, a mere 26 games under .500, had a shot differential of -665.
For those who don’t recall, Hartford had a terrific top line of Andrew Cassels flanked by Geoff Sanderson and Pat Verbeek, which combined for 256 points that season. Their second line, which I believe was Murray Craven, Terry Yake, and Patrick Poulin, had 171 combined points. Those six players also accounted for 173 of the team’s 284 goals. If memory serves me right, the third line had Michael Nylander in between Mark Janssens and Nick Kypreos.
The problem was that the defense was mostly awful in its own end and, with the exception of Zarley Zalapski, couldn’t get the puck back up the ice either. But when the top lines were on the ice with Zalapski, there was a decent chance that the puck was going in the net.
Predictive usefulness in the playoffs
Now, could PDO and the idea of regression to mean hold sway in 1992-93 compared to today? My earlier exploration showed that there is a limited predictive usefulness in the postseason, with the lower PDO team no more likely to beat a higher PDO team than vice versa.
In the first round:
- Buffalo (higher) beat Boston (lower)
- Montreal (lower) beat Quebec (higher)
- Pittsburgh (higher) beat New Jersey (lower)
- NY Islanders (higher) beat Washington (lower)
- St. Louis (higher) beat Chicago (lower)
- Toronto (lower) beat Detroit (higher)
- Vancouver (higher) beat Winnipeg (lower)
- Los Angeles (lower) beat Calgary (higher)
The lower PDO beat the higher one in three series, and it’s interesting to note how all three lower teams that won benefited from a goaltending meltdown by the higher PDO team.
In the second round:
- Montreal (lower) beat Buffalo (higher)
- NY Islanders (lower) beat Pittsburgh (higher)
- Toronto (lower) beat St. Louis (higher)
- Los Angles (lower) beat Vancouver (higher)
Interesting; all four series were won by the lower PDO team.
- Montreal (higher) beat NY Islanders (lower)
- Los Angeles (lower) beat Toronto (higher)
- Montreal (higher) beat Los Angeles (lower) to win the Stanley Cup
All told, the 15 playoff series saw 7 won by the higher PDO team and 8 by the lower. This falls right within the range from the later 2007-08 through 2015-16 study.
Nothing at the moment. Since I have not yet compiled PDO information from 1993-94 through 2006-07 – nor from 1983-84 through 1991-92 – to establish trends and link with the other study that was done, this information is currently “interesting” but not yet “meaningful”.
This may change as more research into old PDO is done. In the meantime, I’m just happy to get the information out there.