June 7, 2013
One Small Step for Kansas City, One Giant Leap for Lineup Construction
On Wednesday, Sam Miller wrote about how lineup construction in baseball tends to change very slowly, if at all. Managers mostly fill out their lineup cards according to the same principles that governed their predecessors’ decisions decades ago, with little regard for more recent research that’s revealed some of those decisions to be suboptimal. For example, although Sam found some circumstantial evidence that this might be beginning to change, the no. 2 hitter still tend to be a team’s best bat-control guy, not its best, well, batter.
But on Wednesday night in Kansas City, one team’s lineup construction changed very quickly. The Royals were 9-22 since the start of May, with an average of 3.42 runs per game scored over that span. They were shut out in the series opener against Minnesota on Tuesday. And on the season, they had a team TAv under .240, ahead of only the White Sox in the AL and close to 20 points behind the third-worst team. In other words, they were desperate to score runs, and desperation sometimes leads to experimentation and innovation. And so it came to pass that manager Ned Yost—not generally regarded as one of the majors’ most sound or sabermetrically savvy tacticians—at least temporarily surrendered his lineup-making duties to Kansas City’s statheads (more details here):
Here’s the Royals’ (presumably Yost-determined) lineup from Tuesday night (against a righty starter) compared to their “optimized” lineup from Wednesday night (also against a righty starter):
One interesting wrinkle: The original “optimized” lineup given to Yost had Cain batting fifth and Moustakas sixth, which meant that three lefties (Moustakas, Lough, and Getz) were batting back to back to back. Yost flipped Cain and Moustakas to make the lineup less vulnerable to lefty relievers (of which the Twins have three):
It’s possible that the advantage of batting Cain ahead of Moustakas was such that it counteracted the risk of lining up three lefties, but according to Colin Wyers’ gut, “a lineup optimizer that was properly handling platoon differential wouldn’t do that.” Colin has an extremely scientific gut, so Yost’s tweak might not have been meddling—it might have made sense. (The Book says “since exact lineup construction is never that important, it is rarely, if ever, correct to bat consecutive lefties in your batting order.”)
In the “optimized” lineup, Gordon remained in the leadoff spot, and Butler continued to hit cleanup. The biggest changes were Alcides Escobar moving from second to ninth, Salvador Perez moving from seventh to third, and Chris Getz replacing Elliot Johnson. Hosmer, Moustakas, and Lough all moved up by one spot. (Also, Jeff Francoeur didn’t start. He wasn’t in Yost’s lineup on Tuesday, but he was in Yost’s lineup—also against a righty starter—on Sunday.) Prior to Wednesday’s game, Escobar hadn’t hit ninth, Perez hadn’t hit third, and Lough hadn’t hit seventh all season.
The lineup wasn’t exactly an instant success—the Royals scored four runs. Then again, that’s more runs than the Royals had been scoring lately, and they won the game, which might have made Yost more likely to stick with the stathead approach. The new lineup’s worth shouldn’t be judged by the outcome of its first few games, but innovative approaches are always in danger of being abandoned if they don’t yield immediate results. (“Not yet,” Yost said when asked after the win whether the stat guys would get a bonus. “We’ll see how it works tomorrow.”) Yost kept the new order intact on Thursday and was rewarded with seven runs and another win, with the game-winning hit coming off the bat of Hosmer, the new no. 2 hitter.
So how optimized is the Royals’ optimized lineup, exactly? It’s hard to say, since how optimized you think it is depends on your assessment of the true talent of each of the hitters involved. We don’t know what criteria the Royals were using to determine true talent—it’s possible that Yost told the front office which hitters he thought were best, then asked them to order them accordingly. It’s also possible that the Royals were considering performance against particular pitcher types, or incorporating HITf/x data or inside info from the coaching staff. We can’t know exactly what they did.*
*Yost’s explanation wasn’t much help: “They liked Salvy in the three because he’s a contact guy with power, they liked Hoz in the two for the left-handed advantage behind Alex. Billy in the four, obviously.”
What we can do is consider the projected TAv for each hitter against a right-handed pitcher. First, we can check to see whether the best offensive options were in the lineup. The Royals had a choice between Chris Getz and Elliot Johnson at second, and David Lough and Jeff Francoeur in right. The stats suggest that they picked the right players:
The more complicated question is whether the players they picked are arranged in the right order. Here are the nine starters ranked by TAv vs. RHP, in descending order:
Colin directed me to a paper called “A Markov Chain Approach to Baseball” (PDF) that includes the following 10 rules of thumb for constructing an optimal lineup:
This table shows how well each lineup lines up with these rules:
According to the above guidelines, the Royals’ new lineup makes more sense than the old one, satisfying eight rules instead of five.
Your mileage may vary if you go by The Book or use on some other method; TAv is a measure of overall production, but where that production comes from—contact, patience, power—matters, too. But it doesn’t take a rigorous study to conclude that a lineup with its worst hitter batting ninth instead of second is probably an improvement.
So why does this matter, aside from the fact that the Royals might score a few more runs over the rest of the season if they stick with their new lineup? Early last year, I made the case that managers are one of baseball’s biggest inefficiencies.
At the time, I polled several front office sources, asking what the difference in wins was between an average tactical manager and an elite tactical manager (or a manager who outsourced some of his decisions to a good front office analyst), all else being equal. The consensus was roughly three wins, quite a lot considering they wouldn’t cost anything extra. An optimal lineup is worth only a fraction of that, but it’s a fraction you can count on.
Yes, Yost probably did what he did out of desperation, trying to salvage both his team’s season and his own skin. But whatever his motivations, he ceded some control of a task traditionally handled by the skipper exclusively, delegating it to (and collaborating with) a team of subject-matter specialists who were better equipped to weigh all the implications. Ideally, Yost would have embraced the benefits before his team struggled for several weeks, which might have made those struggles slightly less severe; there’s no need to wait until a roster is in extremis to seek out an edge. But his decision still sets an important public precedent. As baseball teams continue to become bigger businesses powered by bigger, better data, expect to see more and more managers asking for (and openly acknowledging) tactical input from upstairs.
Thanks to Colin Wyers for research assistance.