How Much Would An Opener Help The 2019 Pittsburgh Pirates?

With the 2019 baseball season upon us, the Pittsburgh Pirates don ?t seem to have a reliable fifth starter. Jordan Lyles, Nick Kingham and Steven Brault all competed for the last spot in what projects out to be an otherwise strong starting rotation; a spot Lyles won because he was the least bad option.

When asked about the hole in the Pirates rotation, Pirates ? GM Neal Huntington said, ?Jordan Lyles is Plan A for the 5th starter job. Nick Kingham and Steven Brault are plans B and C. Plan D would be to use an opener. ? The last line of that quote, about the Pirates ? ?Plan D ? being the use of an opener, drew ire for much of the analytics community. If a team as cutting edge as the Tampa Bay Rays employ this strategy, why shouldn ?t the Pirates?

The opener, for those unaware, is a strategy for using a relief pitcher in the first inning or two, when the opposing team has the top of their lineup up at the plate, then bringing in the more traditional innings-eating ?starter ? for the second or third inning to pitch their usual five or so innings. The game is then finished out using the back end of the bullpen. The theory goes that a relief pitcher, who only has to go an inning or two, can gas it out when going after the top of the order, something a traditional starter could not do, meaning the opener would be more capable of getting those batters out. Then, as a result of already being through the top of the order when the ?starter ? comes in, the starter has to face the top of the lineup fewer times in their innings of work, facing lower quality hitters should mean better pitching outcomes for the starter.

The evidence for the top of the order doing more damage is clear in the data. In 2018 NL pitchers posed a 4.44 ERA/.754 OPS in the first inning, the time we ?re certain the top of the order is up, and a 3.99 ERA/.713 OPS in every other inning, when there is some arbitrary mixture of batters up. The bigger question is could a reliever do any better? While we have a theory as to why an opener might be more effective than a starter, the opener still has to face the top of the order, a spot that we know from the aforementioned ERA/OPS differences, does not favor pitchers.

Analyzing this is relatively straightforward. If we can get a measure of how pitchers perform against the top of the order, and compare it to what they do against the rest of the order, we can fairly easily estimate who would be best served by an opener, and who would best serve as an opener on that basis. As a note, we ?ll be using OPS rather than ERA in this analysis as relief pitchers are certain to face the top of the order at some point, but not necessarily all in the same inning, so we need to understand batting outcomes overall using OPS, rather than directly relating pitching against the top of the order to runs using ERA.

I took event level data provided by MLBAM for the 2018 season, assigned an ID for each spot in the batting order, and calculated the OBP and SLG value for each plate appearance. Then I summed up the OBP value, SLG value, PAs, and ABs that each pitcher had against each spot in the batting order (1 through 9) and then calculated each pitcher ?s OPS overall, and split between the top 3 in the batting order and the bottom 6. Below is the list of bullpen arms you ?re likely to see in the Majors this season (who had significant playing time last season), along with the three guys most likely to pitch in the 5^th starter spot at some point this season, who are sorted by worst OPS against 1-3 in the order.

Name	PA	OPS 1-3	OPS 4-9	OPS Diff
Michael Feliz	217	1.025	0.644	0.381
Dovydas Neverauskas	119	0.931	0.948	-0.017
Nick Kingham	342	0.895	0.781	0.114
Edgar Santana	270	0.864	0.535	0.329
Jordan Lyles	371	0.781	0.684	0.097
Steven Brault	414	0.769	0.732	0.036
Francisco Liriano	586	0.760	0.778	-0.018
Richard Rodriguez	279	0.664	0.550	0.114
Keone Kela	212	0.647	0.580	0.067
Kyle Crick	255	0.518	0.587	-0.070
Felipe Vazquez	296	0.488	0.671	-0.183

The quick way to read this table is that, in order for the Opener strategy to be effective, we ?d have to use a reliever that is further down the list than the starter (or at least the opener would have advantaged them last season). For instance, Kingham could have as his opener any pitcher other than Feliz and Neverauskas and the Pirates would gain some edge. For Lyles and Brault, the list is limited to Liriano and the names at the back end of the bullpen.

How big of an effect would this be? To determine that, let ?s take the most extreme example from the data and figure the runs difference.

If we take the potential 5^th starter with the worst OPS against 1-3 in the order, Kingham, and, in a best case scenario, assume that we have access to the best 4 relievers in the bullpen, Vazquez, Crick, Kela, and Rodriguez. We can choose an opener from those 4 bullpen arms then calculate out the average OPS that that game would have with and without an opener. Since this is the most extreme and best case example, we can get a sense of how big an effect the opener actually has on the team ?s ability to win. By making the assumptions that 1) a pitcher ?s OPS 1-3 and 4-9 are constant regardless of whether the opener is employed or not and 2) The pitcher ?s average PA per inning and per game are constant regardless of the opener strategy we can make this educated guess at the effect of the opener.

For this particular example, we ?d like to see Kingham pitch 5 innings and the relievers each pitch one inning a piece. We ?ll use Kyle Crick as the opener, since he had the second best OPS 1-3 on the team and Hurdle would be very unlikely to move Vazquez from the closer role. This is a difference of .377 points of OPS against the top 3 in the order between Kingham and Crick.

In the traditional pitcher usage case, with Kingham starting and the remainder of the bullpen used traditionally, we can estimate the overall OPS we ?d expect for a game by taking the average number of PA per inning by each pitchers times the number of innings we expect out of each pitcher and multiplying that by their overall OPS to get the OPS value, then divide the sum of the OPS values by the sum of the expected PAs and we end up with our approximation of the game OPS. (Yes I know OPS isn ?t divided by PA ?s but it is a close enough approximation). What we get for this traditional-role game is an OPS of .726.

Alternatively, if we slot Crick in the opener role and Kingham after him we have to augment their respective OPS ?s that we multiply by in the above formula. In Crick ?s case, we ?ll take his OPS 1-3 as the approximation of what his new OPS would be. In Kingham ?s case we need to re-weight his OPS based on the fact that he ?d be seeing 3 fewer PAs against the top 3 and 3 more PAs against the bottom 6. The remainder of the pen arms would have their OPS remain the same as the batters they face would be in no particular order and thus we wouldn ?t change how they were calculated. Using this ?Opener Adjusted OPS ?, in the above formula we get a Game OPS of .712, an improvement, but not all that significant of one.

If we turn these respective OPS numbers into runs, using a linear regression of OPS to runs over the last 5 seasons, we get 4.35 runs per game using the traditional starter role and 4.18 runs per game with an opener. Multiplying that out over 40 games (the number of games started by Brault, Kingham, Kuhl, and Holmes in 2018), we ?d expect to see the Pirates ? total runs yielded drop by 7 runs over the entire season. If we use the 10 runs per win rule, that ?s 0.7 additional wins on the season, or not quite 2% in additional win probability per game with the opener.

Keep in mind: this is using the Pirates ? most extreme case, Kingham is the starter with the greatest potential to benefit from the opener. If we run these calculations using Jordan Lyles numbers we get a 40 game run improvement of 6.5 runs. With Steven Brault, the effect is even smaller at just 4 runs over a full season. Moreover, if one were to run these numbers using Francisco Liriano, or some other reliever not named Vazquez or Crick as the opener, the effect would be even smaller.

This has also largely been an exercise in the abstract, very little can be said about how this might actually effect the players involved. Baseball players are largely creatures of habit, breaking these habits up by bringing your starter out of the bullpen, or having a reliever take the ball to start the game rather than run on in the pressure of the late innings may have negative impacts on performance that we ?d never observe from merely moving numbers on a spreadsheet. Moreover, some starters have vocally opposed the use of an opener (here and here). It is in this author ?s opinion that having starting pitchers upset with their team or coaching staff has the potential to be significantly more damaging to the team than the fractions of a win that the opener could ever possibly generate.

Ultimately, while the opener may be a good trick for optimizing a team ?s pitching staff performance that is all it is, an optimization. If the Pirates want to get high caliber performance out of their 5^th starter, they ?re going to have to get a higher quality 5^th starter.

Postscript: In the interest of sharing data and allowing readers to ponder this idea a bit more themselves, below is the OPS splits for all the Pirates pitchers listed on the 40 man roster as of writing. Enjoy.

Name	PA	OPS	OPS 1-3	OPS 4-9	OPS Diff
Michael Feliz	217	0.776	1.025	0.644	0.381
Dovydas Neverauskas	119	0.943	0.931	0.948	-0.017
Nick Kingham	342	0.824	0.895	0.781	0.114
Chad Kuhl	373	0.801	0.871	0.760	0.111
Edgar Santana	270	0.656	0.864	0.535	0.329
Clay Holmes	129	0.797	0.811	0.789	0.022
Jordan Lyles	371	0.716	0.781	0.684	0.097
Chris Archer	638	0.766	0.775	0.760	0.015
Steven Brault	414	0.746	0.769	0.732	0.036
Francisco Liriano	586	0.771	0.760	0.778	-0.018
Jameson Taillon	785	0.679	0.718	0.657	0.062
Nick Burdi	10	1.250	0.667	1.571	-0.905
Richard Rodriguez	279	0.584	0.664	0.550	0.114
Joe Musgrove	486	0.686	0.655	0.702	-0.047
Keone Kela	212	0.604	0.647	0.580	0.067
Trevor Williams	701	0.656	0.642	0.665	-0.023
Aaron Slegers	60	0.877	0.625	1.036	-0.411
Jake Barrett	30	0.833	0.550	0.950	-0.400
Kyle Crick	255	0.568	0.518	0.587	-0.070
Felipe Vazquez	296	0.618	0.488	0.671	-0.183