Feb 12, 2016
Toronto Maple Leafs
Posts per Day
<div class="quote"><div class="quote_t">Quoting: <b>Price_is_the_goat</b></div><div>stats are like bikinis. they show something, but now what we need to see.</div></div>
Agreed stats show something. They absolutely do not show everything you need, but shouldnt we consider the same thing is possible for players opinons?
To me a player's opinion on a goalie would be formed based on their experiences and what they see. They see a goalie making saves that they dont expect an average NHL goalie to make, their opinion increases. Meanwhile, just like my beer league goalie, they see them letting in a floater or two that the player thinks he should have had, their opinion of him against an average goalie decreases. More or less i think we can agree players build their opinion on goalies over time based on the saves they he he shouldn't make but does, and the saves he should have had but didnt, and then their opinion of essentially how hard it is to score against him fluctuates from there. It's impossible for players for players to see catalog, and analyze every shot though, and they dont see every shot, they dont watch every game, they dont know what the actual expectation is in each scenario, they make assumptions and unconscious decisions to build their opinion. Their opinion has merit, but it also has it's flaws.
Now an advanced stat like dSV% is trying to measure a similar thing. It looks at shots from different areas of the ice, categorizes them, and then creates an expected save % on that shot based on the available history of similar shots, and then if the goalie saves it, he get's a little tick saying he saved a shot that 98.7% of the time get's saved. The issue is the shot data, xSV% and dSV% data doesnt capture every scenario either. It doesn't necessarily see that price's view was blocked, the head fake the player threw, the speed of the shot, or the million other factors, but it's trying, it's not perfect either, but it's making similar assumptions and decisions to build the model. We get a large enough sample, and things even out a bit, out of the thousands of shots that are taken from a given area, many will face similar screens, and the expected sv% starts to inherently build in the affect the screen has, while as a goalie faces more shots, the one screen shot he let in that hurt his stats early, is evened out by the unscreened shots he faces from the same area later on. He get's a couple more of the shots he should have, and balances out the 1 shot he should have had. It's trying to build a similar opinion of him based on shot and save data.
To me both have their inherent flaws. Neither is perfect. Some of the old stats are like a one piece they dont show a lot and they leave a lot to be desired, some new advanced stats might be more of bikini, they still cover up a bit of the situation, and the players, well they might might see everything, but the suns in their eyes so they miss a bit too.
edit oh and +/- is the awkward kid in the corner of the pool with their shirt still on because it's purposely covering up what we see based on all the random rules it tosses in.