Sunday, March 11, 2012
A better ERA projection
Posted by Brian CartwrightThe THT Forecasts have been as good or better than any at projecting the rates of home runs, walks and strikeouts allowed by pitchers. However, recent third-party tests have shown that we have fallen short in projecting pitcher's runs allowed, which is the most important number. After conducting a THT roundtable discussing possible solutions, and long hours coding, most modifications are now in effect.
I had been calculating runs allowed based on a non-linear formula that uses the wOBA allowed by each pitcher. This has shown to be accurate at all scoring levels, and I am still confident in its ability to project the number of runs created by the actions of the batters against each pitcher. However, that is not the only way that a pitcher can control run scoring.
The extent to which a pitcher can or cannot control base stealing is the largest factor. I have tabulated the steals, caught stealing by catchers, pickoffs, pickoff caught stealings and pickoff errors by each pitcher, assigning an extra 0.25 runs for each steal and 0.50 runs for each pick error more than expected, and a reduction of 0.5 runs and an extra out for each additional pickoff and caught stealing.
Likewise, a pitcher will reduce his projected runs allowed by 0.25 with every steal less than expected. Soon, wild pitches and balks above or below expected will be handled in the same manner.
Groundball pitchers and those with fewer walks and strikeouts will get more grounded into double plays per opportunity. GIDP is not included in the wOBA calculation, but I now assign a 0.5 run change and an out for each GIDP more or less than expcted. Pitchers with a higher groundball rate also suffer more batters reached on error, but those additional errors lead to a higher percentage of unearned runs. Soon, I will adjust the unearned run percentage for the pitcher's groundball rate.
The ERA baselines also have been adjusted. Previously, I was using a single value of the MLB averages from the beginning of Gameday in 2005 until the current date, but the accuracy of this has been called into question as the offense production in the major leagues has dropped over each of the past two seasons. The baseline MLB wOBA and runs scored per plate appearance are now based on a weighted mean of the same past three seasons that the individual projections are constructed from.
The last major item is whether the pitcher is performing as a starter or reliever. My comparison of the performances of pitchers who worked in both roles confirm research by Tango that showed that pitchers have a strikeout rate about 17 percent higher and a home run rate 17 percent lower when in relief, while the walk rate does not change.
Tango also found a reduction of the batting average on balls in play of about 17 percent in relief, compared to starting, but my research found a difference about half of that. When each of those components are adjusted, depending on their mix of skills, pitchers had starting and relief ERA projections that varied from about 0.40 to 0.80 runs.
These adjustments enable me to convert statistics compiled in relief to an equavalent starting pitcher performance before league conversion factors are calculated.
After the projections are completed—and customized to each pitcher's parent club's home ballpark—a final projection is created that is a weighted mean of the starting and relieving projections. For pitchers who work 100 percent in either role, there will be trivial changes in their projections. However, this gives us the flexibility to customize projections when pitchers are transitioning to a new role.
For example, Daniel Bard has had a 2.88 ERA over three years in the majors, 3.33 last year, working strictly as a reliever, but his 2012 projection as a starting pitcher is 3.96, reflecting the lower expected level of performance.
Brian got his start in amateur baseball way back in the 1970's as the statistician for his local college summer league in Johnstown, Pa, which also hosts the annual All-American Amateur Baseball Association. A longtime APBA and Strat-o-Matic player, he still tends to look at everything as a simulation. He has also written for StatSpeak and Fangraphs, was runnerup in the Baseball Prospectus Idol competition, and has consulted for a major league team. You can contact him at .(JavaScript must be enabled to view this email address).
The 0.5 runs INCLUDES the impact of the out. Same for the GIDP. If you are suggesting you are going to add to the IP total or something, then you will be double-counting.
That’s actually 17 POINTS, not percent. So, a .300 BABIP as a starter would be .283 as a reliever.