I’ll echo the comments on Dom – his model perpetually underestimates the Caps (IIRC, it hasn’t been convinced the Caps would even make the playoffs the last few offseasons?), but increasingly he’s owning it. I think analytics will only ever going to get you so far in hockey, however reliable the data. There are simply too many variables, both within a game and across a season, to predict results with any meaningful value. Many commentators seem to treat every player as though they have equivalent hypothetical skill and intangibles – many of Ovi’s shots are statistically improbable, viewed across the league, but statistically likely for him. Not because he’s an elite player, but because it’s Alex Ovechkin. Often these analytics seem to
reflect players’ on-ice success or failure better than they can explain it. Trotz having a team turtle for a period will look crappy on the advanced stats, but is what wins them games.
My biggest bug-bear, though, is the idea you can take one guy on a team with a very different system, culture and team and use his analytics to forecast how he’ll perform with his new team. It’s not a problem of analytics, it’s a problem of the person using it – it suggests the team’s system, linemate chemistry, practice drills etc. will statistically be irrelevant. So that article (not by Dom, I don’t think?) had Foligno ‘rising’ in Toronto despite having not spent a minute outside of quarantine since the trade, but Mantha (who has been performing well in the small sample) ‘falling’ based on the fact they think Vrana should have been better here than he was. Vrana wasn’t scoring 5 points in the 4 post-TDL games if he’d stayed. It’s just dumb.
If Backstrom was traded, he’d be described as ‘having a pretty good shot but scarcely uses it’. But look at his shooting % and it varies a lot – 9.15%; 12.64%; 14.86%; 8.91%; 14.74%; 9.76%; 9.18%; 11.76%; 15.5%; 14.2%; 12.73%; 13.02%; 9.1%; 17.5%. How can you forecast what he’ll do on the ice if how effectively he uses his shot changes so frequently, likely as a result of coaching and linemate changes?
And then you have a guy like Dom trying to forecast teams’ seasons before the season has begun, even though so much about teams has changed without a shred of data to analyse. The Caps, for example, changed its system and coaching staff between seasons. How does taking the performance of players who all played under a different coach (and often different teams) the previous year tell you anything about this year? Analysis like that tends to be very good at identifying what we already know – teams like Tampa and Toronto have great possession game, strong talent and will win many games. But you get silly season predictions like this:
Team | Chance of making playoffs | RS points prediction |
Washington Capitals | 58% | 63.7 |
Philadelphia Flyers | 67.1% | 74 |
Florida Panthers | 39% | 61.1 |
Minnesota Wild | 34% | 62.8 |
St Louis Blues | 89% | 70.8 |
[TBODY]
[/TBODY]
At a certain point you have to accept your model doesn’t need tweaking, it’s simply too incomplete to be meaningful? Right?
I like what I see of Mantha so far. He’s a big upgrade in how we can actually use the player and how he fits, even if the sum of all his parts may be notionally similar to Vrana’s.