There's a lot of discourse surrounding "analytics" consistently being wrong about the Islanders after their series win tonight. I think that's just a bit misleading.
A quick little thread...
A quick little thread...
Since '18-19 (which is the period being discussed), the Islanders have played a total of 41 games against 8 opponents in the playoffs. That's half of a normal regular season. And 22 of those games (including 4 for their play-in round) were in '19-20.
Yes, in that time span, that's the 2nd most playoff games for a team behind BOS's 47 games. But 6 other teams since '18-19 have played at least 34 games. 
The narrative seems to be that because NYI have, generally, performed better than expectations based on public models, the data and stats etc are all wrong. In reality, we're still talking about a number of games that, in aggregate, is half of a normal regular season.
To put this in context, as a quick example let's just compare 1st half and 2nd half regular season goal for % (all situations).
For all teams since 2007, here is the correlation between games 1-41 and games 42-82 per season.
For all teams since 2007, here is the correlation between games 1-41 and games 42-82 per season.
I'm using goal for % here as a proxy for winning % (they are highly correlated, and this is a better indicator w/o shootouts).
While there is definitely a trend here, the correlation is lower than one might expect given team construction/strength within a single season.
While there is definitely a trend here, the correlation is lower than one might expect given team construction/strength within a single season.
R2 here is .152. So roughly 15% of the first 41 game GF % can be explained by the next 41 game GF %, or vice versa (this is crude, but just a quick example). You could randomly sample games within a season or any time period and likely find a similar relationship.
Playoff hockey is very random. Unlikely outcomes happen all the time. NYI "beating" the public models in 41 games in 3 years may seem like there has to be something there. In reality, the "analytics" would argue their "success" is very much a possibility.
This entire thread and really the main point here can be summed up by something @StatsbyLopez showed years ago:
statsbylopez.netlify.app/post/part-ii-r…
statsbylopez.netlify.app/post/part-ii-r…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
