You Can’t Stay Lucky Forever: PDO, Brian Stewart & The Dangers Of Hot Starts

You don’t have to be good to win. Most of the time you just have to be lucky”

There’s been a lot of buzz recently around the Coventry Blaze-deservedly so. After all, they’ve won all but one competitive game they’ve played so far this season, and the form of the Blaze’s netminder Brian Stewart in particular has raised excitement that the Blaze can be back up and challenging for a title this season.

Last weekend, I presented the theory that perhaps a combination of Brian Stewart’s hot start and opposition quality was, in fact, the reason Blaze were where they are, not necessarily because they were a team anywhere near the level some were throwing at them. I also argued that it was far too early to make any conclusions about just how good this Blaze team were, and statistically, their start was ridiculously hot and riding a hot goalie.

Predictably, this didn’t go down too well in Coventry. I was called “negative”, told I had “no proof” and even, in an argument familiar to anyone in hockey who’s ever dared to go near a stats table, told “WATCH THE GAMES, YOU IDIOT!”

So, I decided to take a slightly deeper look. Was the Blaze’s hot start symptomatic of an all-conquering team, or is there a statistical argument to say that the rest of the teams will catch up and they’ll be dragged back to the pack? Was this “incredible new era” down, in a large part, to luck and the right combination of circumstances?

I decided to look into this, using a tool which has been embraced by many in the hockey community-statistical analysis of given variables, or “fancy stats”. Very particularly, one stat. This stat, unlike many others in isolation such as save percentage, points, or anything else, takes everything into account that can happen in a game-every bad bounce, good night and blinding performance-to give us who like to analyse trends and try and draw conclusions about what’s happening in front of us a statistical tool that takes everything in a game into account and a tool to compare how teams are doing.

It’s called “PDO” (the name comes from the Internet name of the person who first came up with it.). Cam Charron, one of the foremost analytical hockey stat guys on the Internet, does an excellent job of explaining it in full here.

Put simply, though, it’s a number you come up with by adding the save percentage of your teams goalies to the percentage of shots your team take that hit the net. Simple, right?

Why should we care, though?

PDO works because it uses blind luck as a factor, and also by not looking too hard at the “quality” of shots a team faces on any given night. It simply assumes that there can be only two outcomes of a shot-either it goes into the net, or it doesn’t. Since “shots” in hockey are only counted as those that either hit the net or would have had the goalie not stopped it, and uses stats gathered over either a period of time or the whole season, it removes all other considerations about “chance quality” or “quality of opposition on a given night”-or at least makes them negligible.

In fact, as you’ll see in the Charron article, the “randomness” of taking a team’s shooting percentage over a period of time effectively negates any effect “shot quality” might have had, because it takes EVERY shot into account. The simple fact of luck variance negates any argument that stats are adversely affected by “always giving up quality shots” because it’s simply not true at a team level.

It also works to show when a team is on a particularly good run or a bad run, using “regression”. I’m going to quote Charron here to explain how:

‘Regression’ here is the theory that since every shot taken in the EIHL must result in a save or a shot, the mean PDO in the EIHL is 1. The longer a player or team plays, and the more their play is affected by random factors present in every hockey game, the closer its PDO will get to 1.

PDO is, essentially, a measure of blind luck, applied to hockey.

So-let’s first of all look at the AVERAGE PDO in the EIHL right now. The theory is, over time, it should make 1. (or, if you write it the way I have and express it as a percentage, 100.0). Let’s first have a look at the average, worked out by averaging the save (SV) and shooting (SHT) percentages of all ten EIHL teams thus far, to give us a baseline of what should be the “typical” EIHL squad:

AVERAGE EIHL PDO: 99.9 (90,9 SV, 9% SHT)

Pretty damn close to 100.0, that, isn’t it?

Now, let’s look at the PDOs of all 10 teams to see how they compare (for statistical purposes, a deviation of + or -2 is considered to be either “very good/lucky” or “very bad/unlucky”

Here they are. PDO first, then shot percentage. Teams are in order of standings.

COVENTRY: 105.3 (95.3 SV, 10% SHT)

BELFAST: 101.3 (92.3 SV, 9% SHT)

BRAEHEAD: 103.3 (89.2 SV, 13.1 SHT)

SHEFFIELD: 105.0 (93.6 SV, 11% SHT)

CARDIFF: 96.8 (89.2 SV, 7.6 SHT)

NOTTINGHAM: 97.5 (90.7 SV-average of Kowalski & Green, 6.8% SHT)

EDINBURGH: 100.2 (92.0 SV, 8% SHT)

FIFE: 98.3 (91.1 SV, 7.3 SHT

HULL: 97.2 (86.2 SV, 11% SHT

DUNDEE: 95.1 (88.9 SV 6.2 SHT)

So-what do we learn? We learn that Brian Stewart is currently 5% above the “average” save percentage gained by an EIHL goalie-and also that the Blaze are getting an above-average amount of pucks to hit the net. We can also see that the Blaze goalie is 2% ahead of his nearest competitor, which is a MASSIVE gap (normally, EIHL starters as a whole have a gap of between 3 or 4 percent between the best and worst).

We can also see that Braehead have the most deadly attack in the EIHL currently-finishing an incredible 13% of their chances, or nearly 1.5 times the average, and that Hull are letting in goals like they’re going out of fashion at the moment-David Brown is letting in nearly 10 more goals per 100 shots than Stewart, and 5 more than an “average” EIHL goalie.

Now, PDO works by assuming that eventually, over a season, all teams will “regress to the mean” and get closer to 100, either as their goalies’ save percentage changes, their shooting percentage becomes more in line with what luck would give, or both. If you’re like Dundee and have a low PDO, SV and SHT number, you’re neither stopping pucks well nor scoring-and things can only get better. So take heart, Stars fans.

Conversely, if your PDO is stupidly high or clearly being massively influenced by either a goalie playing out of this world or a team scoring for fun, you’d better hope that continues all year, or you’re going to get worse. Statistically, in fact, due to luck and the bounces evening themselves out, your PDO is only going to drop towards a hundred…unless you keep playing at a level that’s statistically unsustainable or have a run of luck that should probably see you head to Vegas and bet your life savings on one spin of a roulette wheel…neither of which are likely to happen.

It’s early in the year, granted, but what this PDO does show us is that Coventry and Sheffield in particular are having the kind of dream start and performance levels which are incredibly hard to keep up…and that, statistically, unless Stewart has the kind of season not seen in hockey outside of Disney films or Blaze get even better at scoring than their “third best in the EIHL” level, they will regress back to the pack-just like every other team.

Enjoy it while it lasts, Coventry fans. The numbers say that right now, the Blaze are enjoying the kind of purple patch of fortune and conjuction of factors that simply doesn’t last long. In any sport. It’s just a matter of hoping that they’ll last.

In sport, you need luck-and Coventry are using up a heck of a lot of theirs right now.

And if there’s one thing that is certain-luck always changes eventually.

 

Advertisements

One thought on “You Can’t Stay Lucky Forever: PDO, Brian Stewart & The Dangers Of Hot Starts

  1. I know i am about month late but i will write it anyway 🙂

    Although your post is accurate in general, you seem to be missing a VERY important point here, which unfortunately makes it all worthless… PDO should be calculated on even strength and even strength only, otherwise it loses most of its value.

    Here is example:
    Team A beats Team B 4:2 with both teams having 25 shots in game. However Team A scores all of their goals on PP.

    The way you would calculate PDO would show Team A that won would have 108.0 (16% Shots + 92% SVS) and team B that lost would have 92.0 (8% Shots + 84% SVS)

    While in reality when properly count PDO Team A that won should have 92.0 (0% shots and 92% SVS) and Team B that lost 108.0 (8% Shots + 100% SVS) That is considering Team B never had a single shot on PP in that game of course.

    As you see the EIHL results would be opposite to the proper PDO. Ok this is maybe slightly extreme example but shows how inaccurate this can be in EIHL.

    Further to that with EIHL way of calculating PDO is practically impossible, or very improbable (Either massive difference in shots or very high scoring game) to win a game having worse PDO (or under 100.0) unless the game ends up in a draw (decided by SO). That means teams that are winning will have high PDO, which although usually the case in NHL as well, is MUCH less obvious. I.e. SJ Sharks had a PDO of under 100.0 last season even though they won 51 games

    Not sure what point of the season this was written but if Coventry was 5-0-0 there is absolutely nothing unusual or worrying for them about their PDO being 105. And your argument that their level is unsustainable is right, but not because of PDO, but because of winning 5 games out of 5 is unsustainable.

    This is not “blind luck” stats for EIHL, this is efficiency stat. And team that won more games than lost will have it over 100 ALWAYS, while team that loses more game will have it under 100. This make the whole luck and regression theory pointless for our league.

    In fact the “blind luck” part of the stat is more for calculating individual players PDO rather than team but that is another story.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s