Fancy Stat Summer School – PDO

What is PDO?

PDO is a statistic that’s used to determine whether a player or team’s performance over a set sample is sustainable. The way that this is accomplished is by taking two relatively stable statistics, even strength on-ice shooting percentage and even strength on-ice save percentage, and adding them together.

Before we get too deep into PDO, it’s important to define those two stats as well. On-ice is meant to refer specifically to players, but I’m using it for both just to keep things simple. The purpose of saying on-ice is to note that the shooting percentage is the team’s shooting percentage while that player is on the ice. It’s something a single player has limited control over. Same goes for on-ice save percentage, only it’s the goaltender’s save percentage while that player is on the ice. For teams, it’s just even strength shooting and save percentages.

Because shooting percentage and save percentage for the whole league totalled up will always be even, the league’s PDO number will add up to 100%, or 1.0 if you’re using decimals.

Because of this we know that the average team will have a PDO of 1.0, and we can therefore predict that a team with a PDO lower than 1.0 will likely see a performance increase, and a team with a PDO above 1.0 is likely to see a performance decrease.

This can also be applied to players, although with players it gets much more tricky because there are more variables and smaller sample sizes.

Practical example

Looking at these examples, if it were halfway through the season, we would expect the even strength play of Team A to suffer a bit, and Team B to get better.

What does PDO stand for?

This is probably the most common question about PDO, and the answer is nothing. PDO is the username of the Oilers blogger who coined the term back in the day. Why do we still use this terminology for it? Your guess is as good as mine, but it’s probably because PDO is short and sounds like it means something.

What does PDO tell us?

Predictability is the name of the game with these so-called fancy stats, and PDO is in some ways the king of predictability. Because of the relative parity in the NHL, PDO is an extremely strong indicator of a team’s future play, although there are exceptions. Over the last six seasons, there are some teams who are above the average and some that are below. Team talent is not even, and goaltending can make a huge difference, but it’s close enough that big outliers rarely ever happen.

With players it gets a lot more complicated, as some players (and their teammates) shoot at higher percentages, and some players play with better goaltenders behind them. A player’s PDO may not have an average of 1.0, it could be above or below that, which makes regression to the mean tougher to predict.

What are PDO’s limitations?

The main limitation of PDO comes from a misunderstanding. Many people believe that teams or players have to regress to a 1.0 PDO, but this is not the case. Over long samples, great teams have higher PDO numbers, and weaker teams usually have lower ones. The variation isn’t as great as it can look in small samples, with the highest PDO team in the last six years being Vancouver at 1.013, and the lowest being Long Island at .987, but there is variation.

The main focus of PDO is to predict that massive outliers will not last.

Criticisms

PDO is one of the most hotly debated statistics in the analytics community. Pretty much everyone is willing to admit that it has serious flaws; chief among them being that we don’t have the necessary information to differentiate between team prowess and some good luck. There are also no guarantees. While generally the pull to 1.0 on a team level is consistent, outperforming or underperforming that to a great extent in the first half of the season does not guarantee the opposite in the second half.

PDO is one of those statistics that is effective, but you have to be up front and cognisant of its limitations.

As always, questions are welcome in the comments.