FOF: field goal data-mining and value-generation

[QuikSand]

So, I have an idea. I'm kinda sick of looking at kickers who hit on 26 out of 31 attempts for a given season, and not really knowing what to do with that.

The idea would require some data collection, and then some implementation. I'm hoping that by laying out the idea here, we can refine it, and then maybe someone(s) with the requisite access/skills can help advance the ball from here.

-Core idea: develop a "net points from kick attempts" number, generated from any given kicker's various attempt distances

-Implementation: Well, we'd need a baseline to start from, presumably drawn from some data set of fairly recent FOF outcomes. But if we knew that a kicker taking a 42 yard attempt has, on balance, a 71% chance of making that kick... then the expected outcome from the play is 0.71 x 3 = 2.13 points. So, if he actually makes the kick, he netted his team +0.87 points over that baseline, and if he misses, he cost them -2.13 points. Extend that logic to all the attempts for FGs and PATs over a season/career, and we could have a reasonable metric for that guy's value, from kick attempts, over that stretch of time - the sum of those net outcomes.

-What we'd need to start with: A FOF data set we'd consider reliable, to establish those base expectation values. And then, eventually, a way to get this into add-on tools, or a web applet, or whatever you nifty coding types might come up with, to have this evaluation tool handy.



Poke holes in this, by all means.

[JonInMiddleGA]

My first thought on first reading was that sample sizes might be misleading.

The output wouldn't be "wrong" - cause they did whatever they did - but I'm not sure it'd be indicative of anything more than a guy getting a few extra kisses from the RNG.

Phrased another way (since I ain't sure how clearly I said what I was thinking): we'd know what his WAR value was, I'm not sure we'd come away with a clear understanding of why it was what it was.

Fun toy for long term universes though, nothing wrong with that afaic.

[QuikSand]

I don't even know what sample size might be required to be comfortable with a baseline, but as long as everyone knew that the baseline is subject to some noise, that seems like it wouldn't be too misleading.

The GML runs a season every two weeks, we probably have 20-30 seasons worth of data that would be "modern FOF" and that much feels like it would be enough to have some confidence in.

I suspect the best we'd get perfect data on is <40, 40-49, and 50+ as distances, and that would be disappointing... but maybe we could draw it out into a formula and apply that to each yardage distance, specifically.