jump to navigation

Carlos Torres is not a long reliever August 29, 2015

Posted by tomflesher in Baseball, Sports.
Tags: ,
add a comment

Carlos Torres has been a fixture in the Mets bullpen this year, tossing 52 1/3 innings thus far in relief. The Closer Monkey had him listed as our reliable setup man for a bit, but that spot has been taken over by the World’s Worst Sports Blog‘s current favorite bullpen man, Hansel Robles.

Torres cost the Mets last night’s game against the Red Sox, with a little bit of help from Blake Swihart. Torres has seemed unpredictable this year, sometimes seeming strong and other times unreliable. For that reason, I wanted to take a look at whether that’s recency bias or something else.

Carlos Torres' ERA, season-long and on a rolling five-game basis

Carlos Torres’ ERA, season-long and on a rolling five-game basis

To do so, I generated a time series of Torres’ ERA for the season, which (predictably) spends most of its time decreasing, hovering around 4.0, and then periodically spikes upward to start another decrease. There are very few times when Torres’ season ERA increases more than one game at a time. In fact, he put up 39 scoreless appearances making up 39 2/3 innings of work this year. When he gives up runs, though, he gives up big ones.

The rolling-five-game ERA is meant to demonstrate that Torres’ performance is much spottier than we would expect – in many cases, his ERA5 spends several games at 0 before spiking up and staying high for several games. That indicates that Torres follows up runs with shorter appearances, more runs, or a combination of both, showing that he definitely has a streaky side.

In addition, the correlation between Torres’ runs allowed and his number of pitches is quite clear – it’s about .42. That indicates that Torres tires quickly. This is borne out by a back-of-the-envelope regression; estimating Carlos’ runs allowed as a function of his pitches that night and his days of rest, I found that each pitch thrown adds about .04 runs to Carlos’ total (significant above the 95% level). Days of rest doesn’t give a statistically significant estimate, nor do the quadratics of either term. (This looks pretty linear to me.) No combination of pitches thrown in the previous 3 games and batters faced in previous games give any additional information. This isn’t a great method – there’s obviously some endogeneity – but there is a strong correlation between Torres staying in the game longer and giving up more runs.