Is Adam Larsson One-Dimensional? In Analytics Context Still Matters
TSN’s resident hockey analytics guru Travis Yost recently wrote an article entitled “Why One-Dimensional Adam Larsson Wasn’t Enough For Taylor Hall”. In the article Yost posits that we can gauge Larsson’s offensive potential by looking at the Devils offensive output when Larsson is given more “favourable” usage. His chart:
This chart shows that when Larsson is given greater than 50% offensive zone starts, the Devils goal production decreases, even as their shot differential increases. It appears to be a fairly solid evaluative measure. However the chart, and Yost’s conclusions fall apart under scrutiny:
1. Yost Obfuscates His Dataset. Let’s Make It Crystal Clear.
Yost states: “… there were plenty of instances in which he saw decent or favourable offensive zone starts, and in those games, we can analyze how New Jersey performed with his role having changed.”
However he never tells us what “plenty of instances” means. So, I went digging at http://corsica.hockey. Here’s what I found: Larsson has played 274 games in the NHL, regular season and playoffs. He’s been given greater than 50% offensive zone starts in only 26 of those games. Here is the data, for anybody who’s interested.
So let’s take a look at Yost’s chart again, but this time, we’ll also include how frequently Larsson was used in a particular OZS% range:
In his career, Larsson has more games in which he had no offensive zone starts at all (10), than games where he had more than 60%. When Yost says “plenty of instances”, what he really means is, “relatively, not a lot of instances”. So when you read Yost’s graph, remember that the right side of the graph, beyond the 50% mark represents only 26 games in Larsson’s career, and give it the appropriate relevance.
But, that’s not the only way Yost’s data is deceiving.
2. Age Matters
Given Larsson’s defensive zone heavy usage in his last two seasons, you might also wonder when those heavy zone start pushes occurred. I’m glad you asked:
As you can see, when split by season, Yost’s sample of favourable OZS% contains mostly data from Larsson’s rookie season, a total of 10 games. Larsson started that season as an 18 year old.
I don’t think anybody would argue that rookie defencemen — particularly 18–19 year old rookie defencemen — represent their full potential right out of the gate.
And yet, based on Yost’s own analysis, this is the conclusion he wants us to come to.
Given the information we have, is it fair to call a player one-dimensional based on 26 games spread over 5 seasons? In my opinion, this isn’t enough to form a concrete evaluation of any player’s offensive abilities, let alone a 23 year old defenceman, who’s offensive pushes occurred primarily when he was 18 and 19 years of age.
It’s disappointing that a featured article on TSN.ca would use such poor statistical rigor to make such bold declarations. The world of advanced stats in hockey is still young enough to be significantly tarnished by poor methodology. High profile advanced stats people like Travis Yost must try harder to challenge their own preconceptions, and their own methods. The fans deserve more thoughtful analysis.
Thanks for reading!