Petrillo: New Defensive Metric

[Frobby]

16 hours ago, Chavez Ravine said:

These data open up all sorts of interesting analyses. For instance, to get a comprehensive index of out getting skill should we really weight the component metrics equally and add them together linearly like Frobby did? Are some of the component skills more important than others? Do some age better than others? Can some be coached up more than others?

I think these three are intended to be added linearly.   That’s how they are presented by Statcast; I didn’t make that up.    

[Chavez Ravine]

1 hour ago, Frobby said:

I think these three are intended to be added linearly.   That’s how they are presented by Statcast; I didn’t make that up.    

I wasn't sure if Statcast added them up or you did. Either way, that is just one way you could do it. Adding them up equally is a simple aggregate index of how "toolsy" a guy is, but that might not be the best approach if you want to use the index as a predictor of outfielder value. For example, maybe we should weight jump twice as much as the other metrics since great initial speed seems to allow a guy to compensate for deficiencies  in the other tools? But also maybe the impact of jump is biggest down at the slow end of the scale and there are diminishing returns up at the elite end of the scale? I am sure folks have been working on how best to use these data to model things like outfielder value. It could in fact be that the simple addition works pretty darn well.

[Frobby]

37 minutes ago, Chavez Ravine said:

I wasn't sure if Statcast added them up or you did. Either way, that is just one way you could do it. Adding them up equally is a simple aggregate index of how "toolsy" a guy is, but that might not be the best approach if you want to use the index as a predictor of outfielder value. For example, maybe we should weight jump twice as much as the other metrics since great initial speed seems to allow a guy to compensate for deficiencies  in the other tools? But also maybe the impact of jump is biggest down at the slow end of the scale and there are diminishing returns up at the elite end of the scale? I am sure folks have been working on how best to use these data to model things like outfielder value. It could in fact be that the simple addition works pretty darn well.

My impression is these are real-time measurements and that it is appropriate to add them to get the overall result of how long it takes the outfielder to get after a ball, regardless of whether the cause is a slow reaction, a slow burst or a poor route.    Read the explanation in the link in the OP.   Here it is again:  https://www.mlb.com/news/new-outfield-jump-burst-route-metrics-at-baseball-savant

[eddie83]

Example from last night.  

[Chavez Ravine]

1 hour ago, Frobby said:

My impression is these are real-time measurements and that it is appropriate to add them to get the overall result of how long it takes the outfielder to get after a ball, regardless of whether the cause is a slow reaction, a slow burst or a poor route.    Read the explanation in the link in the OP.   Here it is again:  https://www.mlb.com/news/new-outfield-jump-burst-route-metrics-at-baseball-savant

Ahh, I understand now. I misread it. I thought Jump = reaction+burst and route was a separate metric. Yes, adding them all together to get a jump index makes sense. But the point I was trying to make was just that it would be interesting to understand how jump (or really its underlying components) relate to the other stats that look at the outcome such as outs above average or defensive runs saved.

[weams]

I am just glad they are putting more effort into defensive analysis.