Title: On Advanced Stats and Short Series Variance
Date: May 7, 2013
Original Source: Nucks Misconduct
Synopsis: Instead of doing a Game Three post-game for Sharks/Canucks, I instead penned a piece on how advanced stats can be viewed over the short-term when variance is running rampant.
There are no two ways about it – game three was ugly. Terrible, really. After games one and two were razor-close, many expected game three to be much of the same.
Now with a 3-0 deficit, the Canucks have four straight do-or-die games (or, you know, less than four…probably less than four…okay, less than four). So rather than get into too much detail about a game that people are probably tired of complaining about by now, I thought this would be a good time to explain some of the “why” about advanced statistics, and why we should care to look at them.
This series has been understandably frustrating for everyone, and advanced stats make a good target because the story of the series hasn’t matched up with how the series has played out. To wit:
|Game||Van Corsi||Van Fenwick||Score||ES Scoring||Chances|
Chances come from the fine folks at Canucks Army, who do a tally for each game. ES only shown here.
Some will look at this and decide that possession stats are useless, given that the Canucks heavily outplayed the Sharks in this regard for two games but came out behind on the scoreboard. In game three, the possession stats show “even game” but the Canucks were stomped. We also see that despite better possession indicators, the Sharks are getting more chances. ADVANCED STATS ARE STUPID.
A lot can happen in a short amount of time, and a seven game series is undoubtedly a short amount of time. Fenwick has been shown to be both consistent over time and consistent with winning in the long run, so it’s something we should care about. These stats are also usually a good proxy for scoring chances, though they haven’t been in this series. There are a few reasons why things aren’t matching up intuitively:
*three games is a small sample size (180 minutes, plus overtime, minus special teams time).
*everything we look at is for Even Strength, and the Canucks are just 1-for-7 on the power play, both a mediocre success rate and a terrible job drawing penalties.
*everything we look at is for Even Strength, and the Sharks are a robust 4-for-17 on the power play, both a strong success rate and a high frequency of penalties drawn.
(You’ll notice that at even strength, the series has been closer at 7-4, or a goal a game difference, as opposed to 11-5 and two goals per game. The man advantage has been worth about a goal-per-game swing for the Sharks so far.)
*Variance. Some people will call this luck or “puck luck” or what have you, but it basically means that things don’t always play out the way probabilities would suggest. Even in baseball, with 600 plate appearances for a player, things don’t always “balance out.” In hockey, where there are so few “outcomes” (goals) to judge, a lot of short-term performance is subject to variance.
As an example, pretend the Canucks and Sharks were complete equals in Vancouver (meaning that basically the Sharks are a bit better but the home ice advantage negates that). If both teams were evenly matched (not a poor assumption) and played relatively to their average, the following outcomes are possible:
I also included the third game, assuming the teams were completely even across three games – the Sharks would still be up 3-0 12.5% of the time.
Now, the teams are not dead even, and the teams haven’t played to a draw in terms of quality. But you can see how, even if games one and two were pretty close, the Canucks could find themselves in a hole pretty easily.
Possession indicators tell us plenty about a team in the long run, since single-game blips and variance tend to regress to expected performance levels. For example, if the teams were even as we assumed in this simple example, the odds of the Sharks winning ten straight against the Canucks would be less than one in a thousand. But they don’t play ten games, they play seven (or less).
Some may still question why we bother in a short series, if there is so much variance involved. That’s fair – for Canucks fans, perhaps a silver lining or glimmer of hope? In seriousness, it helps us understand the narrative of the games beyond just what our eyes see. If the Canucks are possessing the puck more but getting fewer chances and/or fewer goals, we can look to some potential problems areas:
*are the special teams not pulling their weight?
*is the team selling out to stop the point shot on defense, leaving odd numbers beneath the circle, leading to better chances when the puck gets through?
*are the Sharks defensemen doing a good job forcing players outside, or giving the Canuck defensemen more room to shoot to avoid letting the Sedin’s handle the puck in more dangerous areas?
And so on.
There are a lot of ways you could go with your line of questioning, but simply throwing out possession numbers (and match-up data, for that matter) because there is variance in a short series is short-sighted.
No, it may not tell us as much as we’d like in the short run, and yes, it might be added frustration to see possession numbers failing to lead to positive outcomes. But they can help us illustrate the series, better understand the intricacies of variance at play, and point us in the right direction for identifying issues (or confirming what we thought we saw with our eyes).
I’ll check back after game four, and future games if there are some. Go Canucks!