Mark Barbour (@18sktrs, patreon.com/18skaters)
In this article I’m going to look at some interesting cases where a skater’s expected goals differ from his actual goals. My approach relies on something I call individual expected goal differential, or “ixGd”. Not sure what that is? No problem – it’s all explained below.
Note: the raw data used for this article are from Natural Stat Trick and include games played on or before November 8, 2022.
What Is ixGd And Why Does it Matter?
There are a few public models that estimate the number of goals a skater is expected to have scored. Generally speaking, the models compute the probability of each shot attempt becoming a goal based on certain criteria (such as the location on the ice from which the shot was taken, type of shot, etc.). As noted above, this article uses expected goals data provided by Natural Stat Trick.A skater won’t always score the exact number of goals that he’s “expected” to score, and that’s where ixGd comes in. A skater’s ixGd compares his actual goals to his expected goals over a period of time. The calculation is simple:
ixGd = Actual Goals / Expected Goals
When a skater scores more goals than expected his ixGd will be greater than 1.0, and vice versa.
Here’s a numerical example to illustrate the point. Suppose a skater scored 50 goals during a time period when he was expected to score 40 goals. The skater’s ixGd = 50 / 40 = 1.25, and this means the skater scored 25% more goals than expected.
Now here’s why a skater’s ixGd is important: most (but not all) of the best scorers in the NHL maintain an ixGd above 1.0, meaning they regularly score more goals than expected. This is shown in the plot below. The time period for the plot is the last 150 team games and the data is filtered to include only skaters who scored at least 20 goals (at 5v5). As you can see, most of the skaters had an ixGd above 1.0 and many of them were well-above that mark.
Once we establish a skater’s baseline ixGd we can use it to “adjust” a smaller sample of his expected goals, bringing them in line with his past performance. This is also a simple calculation:
ixGd-Adjusted Expected Goals = Expected Goals * ixGd
You’ll see this in action below.
I have a bit more to say about calculating a skater’s ixGd at the end of this article. Personally, I find this stuff pretty interesting and I think that any analysis that uses expected goals would be improved by taking it into account (including analyzing a team’s xG% or a goalie’s saves above/below expected).
Using ixGd-Adjusted Expected Goals To Analyze Goal Scoring This Season
With that introduction out of the way let’s look at some goal scoring data from this season.
Skaters Who Have Scored More 5v5 Goals Than Expected
|Skater||Team||ixGd||Actual Goals||Expected Goals||ixGd-Adjusted||Over|
- This is a list of the skaters who have scored at least 2 goals above expected (on an ixGd-adjusted basis). Remember: the data includes only 5v5 play.
- Mark Scheifele has scored 8 goals this season, with 7 of them coming at 5v5. His ixGd-adjusted expected goals at 5v5 is only 3, so it isn’t surprising to see that he’s shooting at an unsustainable 29.17% at 5v5. Looking at his assists, he has only 2 in 12 games. Nobody else is scoring when Scheifele is on the ice. So what do you do with this guy? One option is to try to sell him as a point-per-game player (which he has been for years). Perhaps the sneaky move, though, is to try to acquire him right now. Sure his goal scoring is unsustainable, but his low assist rate is also unsustainable. He plays with Kyle Connor right now, and will likely get a boost when Nikolaj Ehlers returns from injury. Those are two very good goal scorers to have as linemates. If you have an opportunity to acquire Scheifele at a discount from someone who is worried about the unsustainable scoring and the low assist total, then that’s something I would consider doing.
- Nick Suzuki has scored 5 goals at 5v5, and that’s 4 more goals than his ixGd-adjusted expected goals. His on-ice shooting percentage at 5v5 is 16.3%, which is among the highest in the league. I could go on, but suffice to say that Nick Suzuki’s current production is not sustainable. If you can find someone who always looks for the upside of young skaters like Suzuki and his linemate Cole Caufield, then now would probably be a good time to sell Nick Suzuki.
Skaters Who Have Scored Less 5v5 Goals Than Expected
|Skater||Team||ixGd||Actual Goals||Expected Goals||ixGd-Adjusted||Under|
- This is a list of the skaters who are at least 3 goals below expected (on an ixGd-adjusted basis).
- There’s Kyle Connor! He has zero goals at 5v5, but his ixGd-adjusted expected goals is 5. He’s an obvious buy low, but he’ll likely be expensive to acquire given where he was drafted. As I mentioned above, acquiring Mark Scheifele might be the sneaky way to gain access to a Kyle Connor comeback.
- Austin Matthews is going to score a lot of goals. It’s coming.
- Alexis Lafrenière is an interesting case. He’s breaking into the league slowly but his ixGd shows that he can be an efficient goal scorer (he has scored 62% more goals than expected). Will that translate when he starts to see an increased role? We’re getting a glimpse of that right now as his time-on-ice has increased by roughly 3 minutes per game this season. His ixGd-adjusted expected goals would put him at roughly 6 goals in 14 games, all at 5v5. Add some powerplay goals and you’ve got a 40-goal scorer. He’s a skater to keep an eye on, for sure.
More Discussion About ixGd
You can stop reading right here if you’ve already heard enough about ixGd. Maybe think about trading Nick Suzuki for Mark Scheifele if you can, and I’ll see you back here next week.
If you’re curious, carry on!
For this article I calculated each skater’s ixGd based on data from his team’s last 150 games played. The general idea here is to balance two competing priorities: 1) use data that is relevant (i.e., recent data is better); and 2) use as much data as possible (i.e., use historical data). There is no “right” way to strike the balance, and I have used different sampling techniques in the past.
Another issue with calculating ixGd is that a portion of the sample data is used to analyze itself. For example, the data from the start of this season were included in my 150 game sample, and that full sample was in turn used to analyze the data from the start of this season. While I don’t want to exclude the most recent data, I also don’t want it to make up a large proportion of the overall sample. Otherwise the ixGd-adjusted expected goals will look a lot like the actual goals scored.
I think it’s worth pointing out that this article is not intended to be a criticism of expected goal models. They provide useful information about shot quality, and I use that data in my skater projections. I’m simply pointing out the benefits of “personalizing” the expected goals data based on each skater’s past performance.
Along the same lines, I don’t want to mislead or confuse you about the distribution of ixGds based on the plot from earlier in this article. That plot filtered out the skaters who don’t score much, which had the effect of skewing the ixGds to the plus side. The plot below shows the same data (last 150 team games) when it’s not filtered for skaters who scored at least 20 goals.
As you might hope, the unfiltered data peaks near the expected goal line where the ixGd is 1.0. When it comes to fantasy hockey though, it’s important to remember that many of the most interesting skaters maintain an ixGd on the right side of this distribution. We should not expect them to “regress” to an ixGd of 1.0. Put another way, we should not expect their actual goals to match the number of goals predicted by the model. We should expect more goals, and in some cases we should expect quite a few more goals over the course of a full season.
If you would like to see all the ixGd data generated for this article I posted it on Patreon and it’s available to everyone.
The End Of The Article
It feels good to finally look at some data from this season. I’ll be back here next week to do it again. Cheers!