xwOBA Year to Year

[Enscheff]

I finally got around to writing my own data domain for some Statcast data so I can begin playing around with it. The most interesting new stat to me is xwOBA. I think it is currently the bet way to measure a hitter's "true talent". Here is an explanation of the stat: http://m.mlb.com/glossary/statcast/expected-woba

To determine whether or not a stat measures "true talent", and is therefore predictive, we have to compare the data year to year for the same player. I imported all the xwOBA data for every player from 2015-2017. There were 123 players with 400+ ABs in 2015 and 2016, and 98 players with 400+ ABs in 2016 and 2017 (this population will increase by the end of the season). Here is a plot of every player's year to year xwOBA:

The R2 values of 0.629 and 0.508 suggests a strong correlation in year to year xwOBA.

But I think we can do better. We know young players hit better, and old players hit worse...an aging curve. What is the aging curve for xwOBA?

For every player, I calculated the amount their xwOBA increased or decreased every year, and grouped them all by age. This is the resulting plot:

What we see here is what we expect. Players increase their xwOBA in their 20s, and then in their early 30s their xwOBA starts to decline.

Now let's apply that aging curve to the sample sets we looked at above:

Incorporating the aging curve improved the R2 values to 0.6428 and 0.556.

These are the most predictive "true talent" indicators I have seen to date. For those curious, here are the 2018 projections for current Braves (min 300 ABs in 2017):

Freddie Freeman (28) , 0.406 + 0.005 = 0.411
Matt Kemp (33) , 0.356 + -0.005 = 0.351
Nick Markakis (34) , 0.332 + -0.007 = 0.325
Ender Inciarte (27) , 0.276 + 0.006 = 0.282
Dansby Swanson (24) , 0.298 + 0.009 = 0.307
Matt Adams (29) , 0.34 + 0.003 = 0.343

 

[cajunrevenge]

xwOBA? Now your just ****ing with me.

 

[nsacpi]

According to the link you gave:

xwOBA is more indicative of a player's skill than regular wOBA, as xwOBA removes defense from the equation. Hitters, and likewise pitchers, are able to influence exit velocity and launch angle but have no control over what happens to a batted ball once it is put into play.

I think there is something to consider here about a small number of players. There have over the years been a small number of hitters with good enough bat control that they can have some control (not perfect control but some) over where the ball is placed in addition to exit velocity and launch angle. I'm thinking here of guys like Carew, Gwynn, Ichiro and Boggs. He's is not in their class, but of our current hitters I think Inciarte displays some of that skill. There is always something that even the best and latest statistical wrinkle leaves out.

 

[chop2chip]

Very interesting analysis. Thanks for sharing. I have a few follow up questions.

What does your analysis show for league wide xwOBA from year to year? I would presume we should see a trend upward considering the rise of offense across the league, so I would be interested to see how Freddie's increase compares to the league wide increase. E.g., how does my 10% raise at work compare to the 8% increase of cost of living.

Any thoughts on the factors that could explain the ~50% change YoY in xwOBA? Fifty percent seems like quite a bit to assign as random, but maybe baseball really is that random. Im not much of an econometrician, but is there a way to regress the years on one another to account for leaguewide pitching decline/improvement , or perhaps factor in minor league performance to account for players like Dansby that have initial struggles that may cloud there true talent level.

Overall, this is really impressive and I'm curious to read more about your work here.

 

[nsacpi]

One more thought: the test we want to run is similar to FIP and ERA. Is FIP a better predictor of ERA than ERA itself. Similarly is xwOBA a better predictor of wOBA than wOBA itself. I think just as there are some pitchers who can consistently outperform their peripherals there will also some hitters (like Inciarte) whose wOBA will consistently be better than what their xwOBA predicts.

 

[chop2chip]

One more thought: the test we want to run is similar to FIP and ERA. Is FIP a better predictor of ERA than ERA itself. Similarly is xwOBA a better predictor of wOBA than wOBA itself. I think just as there are some pitchers who can consistently outperform their peripherals there will also some hitters (like Inciarte) whose wOBA will consistently be better than what their xwOBA predicts.

I know it's dangerous to assume, but one would assume the creators of such metrics would have proven this with statistical significance, since that's the entire premise for their existence.

What you could research is what type of characteristics make a certain pitcher/hitter more likely to outperform their FIP/xwOBA?

 

[chop2chip]

Very interesting analysis. Thanks for sharing. I have a few follow up questions.

What does your analysis show for league wide xwOBA from year to year? I would presume we should see a trend upward considering the rise of offense across the league, so I would be interested to see how Freddie's increase compares to the league wide increase. E.g., how does my 10% raise at work compare to the 8% increase of cost of living.

Any thoughts on the factors that could explain the ~50% change YoY in xwOBA? Fifty percent seems like quite a bit to assign as random, but maybe baseball really is that random. Im not much of an econometrician, but is there a way to regress the years on one another to account for leaguewide pitching decline/improvement , or perhaps factor in minor league performance to account for players like Dansby that have initial struggles that may cloud there true talent level.

Overall, this is really impressive and I'm curious to read more about your work here.

I'm also curious as to what accounts for the players baseline in your prediction formulas, I.e., the .406 for Freddie.

 

[nsacpi]

I know it's dangerous to assume, but one would assume the creators of such metrics would have proven this with statistical significance, since that's the entire premise for their existence.

What you could research is what type of characteristics make a certain pitcher/hitter more likely to outperform their FIP/xwOBA?

The first thing I would do is take two years of data for an appropriate group of hitters (say those who have had more than 400 PAs both years). Then I would run two regressions. The dependent variable is wOBA in year 2 in both cases. The independent variable in regression 1 would wOBA in year 1 and for the second regression I would substitute xwOBA in year 1. That would allow a direct comparison to see which one is the better predictor.

Then I would move on to try to identify a subset of hitters who in both years had a wOBA significantly higher than xwOBA. I suspect there will be some hints from the characteristics of those hitters about what variables might further improve xwOBA.

 

[nsacpi]

Inciarte btw in his four major league seasons has wOBA of .303, .325, .319 and .331. So it seems to me xwOBA is missing something important about him.

 

[Enscheff]

Inciarte btw in his four major league seasons has wOBA of .303, .325, .319 and .331. So it seems to me xwOBA is missing something important about him.

Inciarte does benefit from higher than expected production on grounders. Here is Sprint Speed vs xwOBAgDelta (which is xwOBA - wOBA on grounders):

R2 = 0.342

I definitely expected there to be a much strong correlation.

 

[nsacpi]

Inciarte does benefit from higher than expected production on grounders. Here is Sprint Speed vs xwOBAgDelta (which is xwOBA - wOBA on grounders):

R2 = 0.342

I definitely expected there to be a much strong correlation.

That's part of it. With Inciarte, there is also the bat control factor. Probably not possible to quantify that except as a residual.