There are numerous ways and means suitable for assessing the performance of any forward line combination. One metric capable of providing significant insight is the expected goals for percentage (xGF%) metric.
In this post we’ll take a look at each and every one of the line combinations deployed by the Washington Capitals so far this season. This will initially include lines that skated for just a few shifts.
Then we will reduce the data (remove line combinations deemed irrelevant). The post will ultimately break down what is working and what isn’t working for the Capitals, and generate optimized line combinations extrapolated from expected goals for percentages.
[The statistics used in this post are courtesy of Natural Stat Trick and the NoVa Caps Advanced Analytics Model (NCAAM). If you’d like to learn more about the statistical terms used in this post, please check out our NHL Analytics Glossary]
ALL LINE COMBINATIONS DEPLOYED SO FAR THIS SEASON
We begin this assessment by presenting the full data set. The first (long) graph plots each and every line combination deployed at 5-on-5 by the Capitals so far this season. The graph also includes total time on ice (TOI/Min) for each line combination and the expected goals for percentage (xGF%) for each line. [Click to enlarge].
I told you it was long, but you made it. It’s important to include the original data set before we begin to reduce (scrub) the data and then begin analysis.
In this section we will (logically) reduce the master data set (above.). If you want to be spared the boring details, you can skip to the end for the final data set, assessment and final optimized line combinations.
If we remove all line combinations with an expected goals for percentage of 0.0, we have the following line combinations that have skated so far this season. [Click to enlarge and of the stepped data sets for a closer look].
We can further reduce the data further to remove all line combinations that have skated for less than four minutes. They may have good expected goals for percentages, but the sample size is deemed too small for this exersice.
Reducing the data from this point forward becomes a little more tricky and even somewhat subjective. We could take the top 20 or so expected goals for percentages and work from there, But what about the lines that have only been together for five minutes compared to line combinations more near 50%, but have logged 150 minutes of ice time? Who are you more confident in?
For example, the Milano-Eller-Mantha line has an expected goals for percentage of 92.77%, the best among the remaining lines after the data reduction, but they have skated for just five minutes.
For the final data set, the top 25 expected goals for percentages and the top 15 time on ice lines were kept and all other line combinations removed. Connor Brown line combinations were also removed.
FINAL DATA SET
The following graph represents the final data set based on the aforementioned reductions and was subsequently used for the final analysis. [Click to enlarge].
PROJECTED LINE COMBINATIONS BASED ON EXPECTED GOALS FOR PERCENTAGE
Final analysis and the selection of line combinations is first based on the best expected goals for percentages and weighed against total time on ice for each of the combinations. A combination of high expected goals for percentages and significant time on ice was the driving preference.
Several line combinations have worked well with Alex Ovechkin. Ovechkin-Kuznetsov-Milano has been excellent, although they have only skated for 23 minutes. It’s certainly an excellent option for coach Laviolette within games. However, the Ovechkin-Strome-Sheary line combination has worked well (52.94 xGF%) and for significant amount of time on ice.
Alex Ovechkin – Dylan Strome – Conor Sheary
The Ovechkin-Backstrom-Sheary line has also worked well (60.6 xGF%) but have only skated for a little over six minutes. Also a potential future combination to keep an eye on.
There are also several viable options for Kuznetsov centering the second line, but by far the best combination after 47 games, based on xGF%, has been the Milano-Kuznetsov-Mantha line.
Sonny Milano – Evgeny Kuznetsov – Anthony Mantha/Tom Wilson
Milano has worked well on almost every line he’s been on. The key will be to find where he can best help a line that needs it. In addition, Anthony Mantha has struggled as of late and could eventually fall anywhere in the lineup, or fall completely out of it. Tom Wilson is just starting his season, and could very well end up on the second line.
The third line has the biggest remaining questions. How Will Nicklas Backstrom and Tom Wilson’s return pan-out? Will Lars Eller be dealt at the trade deadline, etc. In addition, the sample size for Nicklas Backstrom and Tom Wilson is still extremely small. With that in mind, here are the two best options for Laviolette, depending on how things turn-out.
Marcus Johansson – Lars Eller – Mantha/Nicolas Aube-Kubel/Tom Wilson
Marcus Johansson – Nicklas Backstrom – T.J. Oshie/Tom Wilson
It should be noted that the Milano-Backstrom-Wilson line has worked extremely well, as have most line combinations with Milano. And again, Wilson could replace Mantha on the second line.
Far and away the best overall (time and xGF%) fourth line combination has been the the Protas-Dowd-Hathaway line (63.6 xGF% and 186 minutes on ice). Protas was sent down when injured players began returning, mostly because he is waivers-exempt and he easily affords the Capitals time to get a look at other players. He will be back, full time, eventually.
Aliaksei Protas – Nic Dowd- Garney Hathaway
FINAL OPTIMIZED LINES EXTRAPOLATED FROM EXPECTED GOALS FOR PERCENTAGES
Alex Ovechkin – Dylan Strome – Conor Sheary
Sonny Milano – Evgeny Kuznetsov – Tom Wilson
Marcus Johansson – Nicklas Backstrom – T.J. Oshie
Aliaksei Protas – Nic Dowd- Garnet Hathaway
Extras: Lars Eller, Anthony Mantha, Nicolas Aube-Kubel
*At least one of the “extras” will no longer be with the team following the NHL trade deadline on March 3.
CAVEATS AND CONTEXT
As noted in the introduction, the use of expected goals for percentages to assess line performance is just one of many methods of analysis.
Nicklas Backstrom and Tom Wilson have extremely small sample sizes and could change the middle six line combinations in many ways, depending on how they perform.
Removing line combinations with minimal minutes doesn’t mean that line combination didn’t work or couldn’t work in the future. It simply means the sample size is too thin for this analysis.
Finally, injuries and potential trades leave the final (playoff?) lines very much a dynamic model until at least after March 3, the NHL trade deadline.
Anyway, this was an insightful exercise for me. What do you think are the best line combinations?
By Jon Sorensen