ERA vs. xERA, 2022

[Ohfan67]

27 minutes ago, Tony-OH said:

 

Both of the "studies" are old (2013 and 2018). I'm not a huge "predictor" of future success guy when it comes to most stats because there are so many factors that go into these things.

I would like to see a study done on whether a player expected stats in statcast are a predictor for anything in season or for the next season. I like the "x"  stats, but I just don't know how much of a true predictor they are. 

 

There are a lot of studies that correlate a ton of the new (and old) stats with future success. The newer stats, on average, do a better job and in many cases a much better job of predicting future success...but it's still a far cry from a perfect crystal ball. The ability to collect so much more data on player movement, the spin of the ball, etc., has greatly advanced. I totally assume that Elias and Sig are using a ton of the new data available to them to look for stronger relationships between data and metrics that quantify what just happened and what will happen in the future. But there will always be a ton of noise in what happens in the future. 

[Tony-OH]

2 minutes ago, Ohfan67 said:

You wish what were that simple? I wrote that ERA and ERA+ were not great predictors of future success and that's why there's been a move to other metrics. Those other metrics are better correlated with future success. The relationship between those other metrics and future success is not perfect (the correlation coefficient is not 1). They are metrics correlated with future success, not perfect crystal balls. 

Here in lies my issue with any "predictor of success" stat. Players are human and because of that, no predictor will ever be a "crystal ball." But, we can surmise that using expected stats are a general way to have an idea whether a player was "lucky" or "unlucky" in his actual results. That's really about it.

I haven't done enough studying into expected stats to understand whether they are really an indicator of future outcomes or not. 

Maybe someone will go back and look at the last 3 or 4 years and see if they were with Orioles players.

[Ohfan67]

Just now, Tony-OH said:

Here in lies my issue with any "predictor of success" stat. Players are human and because of that, no predictor will ever be a "crystal ball." But, we can surmise that using expected stats are a general way to have an idea whether a player was "lucky" or "unlucky" in his actual results. That's really about it.

I haven't done enough studying into expected stats to understand whether they are really an indicator of future outcomes or not. 

Maybe someone will go back and look at the last 3 or 4 years and see if they were with Orioles players.

I don't have time to do the searches now, about to walk out the door, but the correlations between old stats/metrics like ERA and future success and correlations between the new fangled stats and future success have been calculated (and get updated on a regular basis). I can't remember what the correlation coefficients are, but I vaguely remember that the "best" predictors of future success explains something like 40% of future variation in success.  That's better than guessing or not using data, but it's not great when you are going to invest millions and millions of dollars into a product. I think correlating the data you referenced a few posts ago, the actual spin and movement of thrown baseballs, planes and angles of swings, speed of the ball off the bat, etc., with future success is where the really juicy stuff is happening now. One frustrating issue that arises is that once you start moving into a true multivariate analysis then things get more complicated to concisely communicate. ERA and ERA+ may not be super correlated with how the pitcher is going to perform next year or the year after that, but it's a simple metric to understand and communicate. Once you start using multiple variables then the statistics that you use to calculate the relationship between past and future performance get way more complicated and it gets harder and harder to communicate the results. For my day job I do a good bit of multivariate statistics and trying to communicate the outcome of those analyses even to other specialists is challenging. I think a lot of the analyses that people like Sig are using these days are unlikely to be fully integrated into baseball "journalism" because there's often not one number, one metric that tidily explains the relationship between say future OPS and five or six variables/types of data that have been collected with advanced video, etc.  It's a very interesting situation where the amount of data and powerful analytical tools to use those data are at an all time high, but as those analyses get more complicated they get harder to communicate in a simple, straightforward manner. 

[btdart20]

18 hours ago, Frobby said:

I find it interesting that Kremer’s xERA is so much worse than his actual.  I can’t say that I got the feeling that Kremer was incredibly lucky when I watched him pitch.   Part of the reason, I think, is sequencing.  xERA won’t take into account the fact that a pitcher did well in RISP situations.   Kremer had a .756 OPSA with the bases empty, but only .626 with runners on base or in RISP situations.   Remember that as a whole, the league pitchers do worse in RISP situations than non-RISP.   So, I’d say Kremer’s low ERA compared to xERA may be partially due to good outcomes on batted balls (.312 wOBA compared to .326 xwOBA), but the fact that he pitched well when runners were on base had more to do with the ERA discrepancy.   Whether Kremer can repeat the feat of pitching better with runners on remains to be seen.   

I’ll also throw out that Kremer’s FIP of 3.80 was a lot closer to his actual ERA than his xERA.   He limited the homers and walks.  
 

That is a pretty concerning gap for Kremer.  Not to continue to make excuses for math, but this feels similar to the gap for Mountcastle's wOBA and xwOBA.  I wonder if it's a function of how the formula incorporates the LA sitting pretty close to the sweet spot at 14.7 and the EV being league average?  

[Frobby]

Saw this article about the pitchers with the biggest discrepancy between their xERA and ERA (overperformers).   The article was limited to pitchers with 150+ innings.  The biggest overperformer was Alex Manoah, who’s ERA was 1.07 runs/game lower than his ERA.  Dean Kremer, who only threw 125 innings and hence didn’t qualify for the list, was 1.13 runs under his xERA.

https://fantasy.fangraphs.com/2022-review-pitcher-xera-overperformers/

Kyle Gibson was 3rd on the list of xERA underperformers.

https://fantasy.fangraphs.com/2022-review-pitcher-xera-underperformers/

 

Edited February 1, 2023 by Frobby