Predicting NHL Hockey Games

The theme of all my social distancing posts seem to be predicting sports games. I started with NCAA college basketballContinue Reading

The theme of all my social distancing posts seem to be predicting sports games. I started with NCAA college basketball (part 1, part 2) and continue the streak with this post about NHL hockey.

College basketball and professional hockey are very different. Obviously they are different sports, but in other ways too. Two main things come to mind, especially when trying to predict games:
1) There is a major element of luck involved in hockey which college basketball does not have as much of
2) There is little disparity in hockey, even the worst teams win and the best teams lose.

Combined, these two things make it very difficult to predict NHL games when compared to college basketball.

This post is going to look at how difficult it really is by using common hockey statistics. I am using individual NHL game data since the 2008-2009 season through the (unfinished) 2019-2020 season. Teams that win are in red, while teams that lost are in blue.

The first graphic is a histogram of the number of goals for; the percent of the particular goals for amount is shown in the bar. It’s clear from below that there is no particular number of goals for that secures a team’s victory. In fact, even teams who scored 8 or more goals only won about 50% of the time.

The second graphic looks at team’s corsi and fenwick percentages. Clearly there is a lot of overlap. There were even teams with percentages around 80% that didn’t even win the game.

It’s this kind of randomness and luck that makes hockey so different. As we see below, a team could be dominating in the corsi and fenwick numbers, but still lose the game.

The third graphic compares team’s PDO, which is made up of their shooting and save percentage. Like the two prior graphics, there is so much overlap and no distinction from the winners and losers.

The fourth graphic looks at score adjusted shot attempts for and against. Score adjusted shot attempts take into account the score of the game. For example, a team that is behind and has pulled their goalie in theory would be shooting more than the other team. Rather than counting each shot as 1 shot attempt, it adjusts it to take into the fact they have an extra attacker.

This may be the messiest graphic yet – total chaos. The winning teams do not even cover the losers as we saw in the other graphic. This shows just how unreliable stats like this are for predicting who will win a hockey game.

The final graphic looks at defensive zone giveaways against by wins and losses. The weighted average line is added for reference and the size of the circle represents the number of games with that number of defensive zone giveaways.

There is almost no difference between the two – even the weighted averages are only .01 off. Again, there is no advantage in this category

As shown above in all the graphics, predicting NHL games is much harder than college basketball. There is just so much luck and little disparity between teams that trying to pick which one will win based on numbers seems almost impossible – essentially a coin flip.

A great example of this is this season’s Detroit Red Wings. The Red Wings were the worst team in the league this season yet they swept the Montreal Canadians in their 4 games, and right before the end they beat the Tampa Bay Lighting, who ended in second place in the Atlantic Division.

In hockey, even teams who are bad win, and that makes it so difficult for the model to be accurate.


0 comments on “Predicting NHL Hockey Games

Leave a Reply

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

You are commenting using your 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 )

Connecting to %s

%d bloggers like this: