You put shit in, you get shit out.
Bad data is worse than no data. I'd be more inclined to rely on save percentage than GSAX.
I don't track this logic. We're not talking about minor statistical errors here - it appears to be (admittedly from less than complete analysis) significantly and consistently off in a lot of ways. The only way to tell *how* off would be to compare it to a good data set, but I would imagine all of those are proprietary.No data set is going to be 100% accurate. Assuming a normal distribution of errors is pretty common place in statistics. We're literally talking about probably around a million+ shots worth of data, the error should even out. Even then If you think think the errors are biased/skewed a certain way, if we look at EvolvingHockey's xG model, it specifically uses gradient boosting which is pretty good at determining how wrong or biased a model is and adjusting it, each iteration adjusting for errors in the last.
In your shooting example, yes the analytics models don't know preshot movement, but preshot movement is somewhat taken into account as the aggregate results are with it. Is preshot understated? For sure. But it's not ignored.
This didn’t age well tonightTIL: Hellebuyck not a top 10 goalie
Makes perfect sense.
Huh?This didn’t age well tonight
A goals a goalHuh?
There were 3 tipped goals. I legitimately don't even know how you could come up with that
Would screen goals not count either than? That’s some terrible logicLol. Okay then.