The quest continues today in our three-part series looking at different ways to formulate the Senators’ lineup. Yesterday we created combos using chemistry scores, today we shift to a different concept: playing styles.

If you’ve never heard of Ryan Stimson, it’s a name you’ll surely begin to hear more often. Ryan is a hockey analytics expert, and is spearheading a massive task known as the Passing Project. The name of the project is self-explanatory: Ryan and his team track passing sequences, and then leverage the collected data for in-depth analysis. His post “Identifying Playing Styles with Clustering”, which uses passing data, will be the basis for this post.

Without rewriting his article (which I highly recommend reading), Ryan groups players with similar types of passing data into clusters, and gives each category a name. The categories are as follows:

Playing Styles

He then pairs together all possible line combinations based on their styles and calculates which groups create the highest percentage of expected goals (shot attempts weighted by location and shot type).

The following tables are produced:

How can we apply this to the Sens? Like yesterday,we’l start with the forwards.

Forwards

(Note: Clarke MacArthur is assumed to be balanced, since there’s not enough data for him.)

Although some players don’t come as a surprise (Hoffman is classified as a shooter, Brassard as a playmaker), it’s a tad surprising to see players such as Bobby Ryan classified as dependent, and Mark Stone classified as a playmaker considering he’s best known for his two-way game.

Using this information, and with the help of a handy Excel tool made by petbugs, this is how the the system configures  the Sens’ new optimal forward lineup:

Just based on the playing style combinations alone, and a bit of weighting for TOI, this group of forwards would be expected to have an xG% of 51.1%. For reference, my projected lineup from yesterday has a projected xG% of 49.6%.

Unexpectedly, the line combos index shows that playmakers tend to perform best when paired with players classified as ‘balanced’, not the shooters. This is why, for instance, the system wouldn’t slot Hoffman with Brassard, instead utilizing MacArthur with a set of playmakers.

Defencemen

If you didn’t think Erik Karlsson was special enough, you should know that he’s one of only 11 defencemen to be classified as “All Around”; a list that contains the league’s most elite.

Oddly enough, the Sens don’t have anyone classified as a ‘puck mover’, which is the second most common category in the league. Maybe this is the type of role Thomas Chabot could fill.

The biggest surprise to me is that Fredrik Claesson is categorized as ‘defensive-oriented’. As he showed with his time in Ottawa, he seemed to be confident jumping into the offensive zone while also being quick on his back-check. He recently broke the record for the Sens’ hardest shot at their annual skills competition, so maybe this is an indicator he should be using his offensive weapons more often.

Going by the combination types and their expected goals, here is the new optimized lineup:

Note that because this is solely based on the player’s style categorization, players with the same style can be interchanged and it wouldn’t make a difference.

The xG% for this lineup goes up to 52.5%. Compare that to the lineup from part 1 of this series, which lags closely behind at 52.1%.

Dion Phaneuf playing next to Erik Karlsson is an idea that has floated around for a while, but has never really been put into action by Guy Boucher. Seeing Ottawa’s top two point-getters together would create havoc for the other team in the offensive zone, with both players working the point and pressuring the other team. They still had the third highest chemistry score of all possible defence pairings from last season, so maybe this is worth a shot. With Methot gone, there’ll be ample opportunity for experimentation with the top pairing.

You can find part 1 here if you missed, and part 3 arrives tomorrow. If you want to play around with the passing data used to cluster the playing types, you can find Stimson’s viz tool here.