Evaluating the Switch Between Starting and Relieving by garbanzo24 March 20, 2014 We know, in broad strokes, what seemingly happens when a pitcher moves to the bullpen. His velocity tends to improve, his “stuff” tends to look better, and his run prevention receives a boost. Can we quantify that effect? What about the reverse effect, moving from the bullpen to the rotation? The Process In order to attempt to quantify this effect, I looked at the 115 pitchers since 2002 who have started for at least 100 innings and relieved for at least 100 innings. There were actually 118 pitchers who fit the criteria, but the knuckleballers Dickey, Wakefield, and Sparks skewed the sample rather greatly, and were thus thrown out. It could be said that knuckleballers play a different game, with different constants (lower BABIP), and different expectations, so it is sometimes best to neglect them in a model, but that is a discussion for a different day. Besides somewhat arbitrarily throwing three data sets to the curb, this methodology necessarily creates a sample bias, that only those pitchers who were deemed worthy of the switch for an extended period of time will show up in the results, but there is simply no way to simulate how players might have done in such a situation if teams will not actually do it. As such, the data ignores those players who were too good (or too bad) in their current roles to ever merit a switch to the other. A secondary bias is that this sort of analysis will not specifically show what happens in a switch from the bullpen to the rotation, or vice versa, but rather what people who pitched in both roles did. In other words, we’re not really quantifying what happens during the switch, but just the split itself. It is possible that switching from the rotation to the bullpen is easier than the converse, but the data would not reveal that. For those pitchers who satisfied the prerequisites, I aggregated their performance as relievers and as starters in order to gauge the difference, and looked at the individual data to find the summary statistics. The Data Peripherals Stat Overall Starting Relieving s-r Min 1Q Median 3Q Max BB% 8.6% 8.3% 9.1% -0.9% -5.6% -1.7% -0.8% 0.6% 3.1% K% 16.7% 15.6% 18.6% -3.0% -15.1% -5.3% -3.0% -1.4% 3.9% LOB% 71.1% 69.9% 73.4% -3.4% -22.7% -6.0% -3.5% -0.9% 7.7% HR/FB% 10.8% 11.4% 9.8% 1.6% -7.4% 0.0% 2.2% 3.8% 15.1% GB% 44.5% 43.7% 46.1% -2.5% -13.1% -4.2% -1.5% 1.7% 8.9% FBv 90.5 90.2 91.1 -1.6 -5.0 -1.8 -1.0 -0.3 0.9 FB% (pitch) 59.9% 59.2% 61.2% -4.2% -23.2% -3.7% -0.1% 4.0% 25.1% Z-contact% 88.2% 88.8% 87.0% 1.6% -3.8% 0.3% 1.8% 3.2% 9.4% “s-r” refers to starting minus relieving, so negative values correspond to situations in which the pitchers in the sample had a higher value of that statistic while relieving, while positive values correspond to the opposite. I chose FB% to serve as a proxy for modeling the tendency of relievers to sacrifice secondary, tertiary, and quaternary pitches coming out of the bullpen. Z-contact% is meant to be a measure of “stuff” by virtue of measuring the ability to induce swings and misses on pitches in the zone. Based on this table, we can get a rough idea of the difference between starting and relieving. It appears the group had a much higher K% and a higher BB%, along with a higher velocity and Z-contact%, supporting the assertion that, all else equal, a pitcher will be able to display better stuff while relieving, likely due to the fact that a reliever does not need to conserve energy to pitch deeper into the game. GB% and HR/FB% also improve, implying weaker contact against relievers, likely due to, again, the improved “stuff”. Run Prevention Stat Overall Starting Relieving sDif rDif s-r Min 1Q Median 3Q Max ERA 4.47 4.77 3.96 0.30 -0.51 0.81 -1.71 0.23 0.87 1.45 3.45 FIP 4.45 4.65 4.11 0.20 -0.34 0.54 -1.00 0.24 0.65 1.07 3.29 xFIP 4.39 4.51 4.19 0.12 -0.20 0.32 -1.10 -0.04 0.34 0.66 2.22 Run prevention is, unsurprisingly, also seemingly improved by being in the bullpen, although FIP and xFIP seem slower to “pick up on it”. That is also not surprising; both statistics would need to assume average performance in the other excluded peripherals to model ERA extremely accurately, but the “Peripherals” chart showed that such a base assumption would be untrue in this case. Conclusions It is tough to make conclusions from relatively small sample sizes and somewhat flawed studies like these, but there are a few rather obvious pointers. It seems rather clear that the common perception that “stuff” improves while in relief rings true. This may not be definitive evidence that moving to the bullpen would explicitly improve a player’s performance by N% in different categories, but this analysis does firmly point in the direction that there is some theoretical manner by which that effect could be ascertained. Author’s note: This is the first article I’ve submitted to FanGraphs, so constructive criticism is welcomed. Ask in the comments below if you’d like me to run another stat and see if it meshes with the results I’ve posted here.