Predicting the NHL is a fool’s errand…or is it?

As I’ve said a lot recently, there are a lot of stats that don’t tell us anything in the early part of the year–basically any individual player’s on-ice stats, even most his counting stats–you can look at them and try to squeeze some meaning out of them, but at the end of the day, the sample size is too small and to predict anything from them would be foolish. Also, we talked on a recent podcast about how PDO is only a backward-facing stat and it has zero predictive power. More on this in a second.

One stat that I do watch closely and do put stock in from the start of the season is goal differential. The shot-based side of analytics has gotten a major boom, as people have wrapped their heads around using the larger sample of all shot attempts to measure possession than just shots on goal. However, I feel like a lot of times, the goal-based metrics are ignored or sort of taken for granted. One thing that I think most people don’t know is that despite all the time and energy we put into shot-based metrics, goals are a better predictor of future success than shots. (See: here, and here, and here.)

I want to avoid getting into too much of a tangent here, but whenever we talk about the nature of statistics, it’s very easy to lose the forest for the trees. The PDO example from our HTH podcast a couple of weeks ago is a good example: when a team has a very high PDO in the early season, you can predict that they will regress to the mean. However, just because they have a high PDO doesn’t MEAN that they will regress, there’s no causality. Rather, we look for other underlying stats that would lead to the team coming back to earth, and the PDO was just an indicator of some regression. There is zero predictive power in the PDO stat itself, but it’s kind of the canary in the mine, if you will. But that gets lost because most of the population isn’t statistically literate…so when stats guys DO put out predictions, they’re often dismissed out of hand.

So, pivot back to goal metrics, and specifically the Wild. I didn’t mean to get into a heavy discussion of the predictive power of #fancystats here, I just wanted to talk a little bit about goal-based stats, and specifically, goal differential. It’s a really easy stat to find (the NHL standings page has it right there) and it’s a good gauge of how a team is doing compared to their record. Finally, we arrive at the point of the post! I looked specifically at Minnesota’s first fifteen games of each of the last three seasons and tallied up their even strength goals for and against. I wanted to shave off special teams and just look at 5v5:

Minnesota Wild 5v5 GF and GA, first fifteen games

Goals For Goals Against Diff
2014-15 35 27  8
2015-16 33 30  3
2016-17 30 22  8


We see that the team’s goal diff after fifteen games this year is at +8, the same as in 14-15. One thing that’s a little concerning is their declining GF. We saw exactly this against Calgary the other night, a game they should have won and which would have given them two points on a night when the Blackhawks got trounced. Despite allowing only one goal (and it was on the PP,) Minnesota got shut out and blew their chance at two points.

Here’s another look at the last three seasons, this time with the path of the goal diff mapped out:


To me, we sort of have a Goldilocks situation, where the 14-15 season started out too hot, the 15-16 season started too cold, and the 16-17 season has been in the middle. This year, the team was hanging right around even, and then games 7-9 were the three consecutive shutouts, and they have plateaued again.

Despite talking about the predictive power of goal stats, I’m actually not going to make predictions here. Instead, I’ll point you to and @omgitsdomi, who already do great work on the predictive side of things. You’ll have to do a little bit of digging to find out what goes into their models, but they’re there.

Here’s a peek at some predictive stats from hockeyviz:


Minnesota is projected at 95 points right now, but no one is really running away with the Central except for Chicago, but their underlying numbers suggest they might fall back to earth. The Wild are going to need to start grabbing some more points if they want to secure their spot, as the middle five teams are all within five points of each other here. Next, it’s never too early to look at playoff odds:


There’s a pretty clear picture here: ten teams competing for eight spots. Obviously, it’s not impossible for CGY, ARI, COL, or VAN to turn their season around, but let’s be real, it would take a miracle. The Wild are at a comfortable 78% right now, but anyone who has watched this team over the last five years knows their playoff odds have looked like a roller coaster, with major slumps each of the last few years. But, overall the club has looked good so we will continue to monitor these as the year goes on.

So, this post didn’t really take the direction I though it would…I intended it just as a snap shot, and ended up getting into some heavier stuff than intended. But, I’d very much like to know your thoughts about shot-based and goal-based models, predictive stats, all that good stuff. Do you think the Wild are on pace for a good season? Why or why not? Tell me in the comments or on Twitter @HTHpod. Thanks for reading!