Understanding Bias in Decision Making: Managerial and Front Office

My background is in economics – it is my undergraduate degree and I am currently a master’s student in the subject – and one of the biggest components in the field is understanding decision making. How does a firm maximize profits or how does an individual maximize utility (or minimize expenditures)? These are questions studied in economics, among many others, but these are the two basic questions. In baseball, we can think of the team as the firm and wins as a profit, making the question then, how does a team maximize their chances of winning? There is plenty that goes into that, but this part of the blog post will be focused on bullpen usage, an idea that came from reading Animal Spirts by George Akerlof and Robert Shiller.

In The Numbers Game by Chris Anderson and David Sally, wrote how the expected level of a substitute will start to exceed that of the starter as the game goes on. In soccer, this would be the 11th player on the field, usually the rightbacker or leftbacker who is the weakest link and will be substituted off. Applying this to baseball and it becomes the dreaded times through order penalty (TTOP) that Mitchel Lichtman (MGL) has written about previously on multiple levels. Dr. Russell Carelton has taken a look at it and David Laurila has interviewed managers about it. Whether the TTOP comes because batters are more familiar or because of high pitch counts, we do know starters are less effective later in games than at beginning of games, which makes sense logically.

Anderson and Sally mention that anchoring is potentially one possibility for why managers (in the context of the authors, soccer managers) refuse to make substitutions. Anchoring, as the term suggests, is when one individual puts too much weight on a piece of information that is then used to make decisions later on. It is a form of a psychological bias. In the form of the manager, it would be the starter compared to the middle reliever. The Yankees spent $325 million this past offseason for Gerrit Cole to be their ace and to pitch the Yankees deep into postseason. On the other hand, pitchers such as Luis Cessa, Jonathan Holder, and Chad Green are relief pitchers, with the highest earner being Green at $1.275 million. The Yankees are paying Cole to pitch deep into games and are paying the other three to be relief pitchers to bridge the gap to Zack Britton and Aroldis Chapman. How long would it take for manager Aaron Boone to remove Cole for Cessa, Holder, or Green?

Let’s look at an example of a Yankees game this past season, one coming September 5th against the Baltimore Orioles in Baltimore. Cole had nine strikeouts through the first five innings and was pitching to the standard that Cole has set. But in the sixth inning, DJ Stewart led off with a solo home run, followed by a groundout, error on third baseman Thairo Estrada, and then a strikeout. No big deal, Cole gave up a home run and then proceeded to get two of the next three and should’ve been done with the inning. However, he walked the next two batters (pitching coach Matt Blake went out for a visit and Luis Cessa started to warm), gave up a two run single to Ryan Mountcastle on an 0-2 fastball that Cole missed his spot with, and a two run double to Rio Riuz, before getting Pat Valika to lineout. Cole allowed five runs, only one earned, in the inning and threw a total of 102 pitches in the game.

In the link above, MGL wrote that “Good and bad pitchers show around the same magnitude of TTOP. The third time through the order, all starters are expected to pitch around .35 runs per nine innings worse than they do overall.” Using STEAMER’s 2020 projections, Cole was projected a 3.48 RA9, making the third time through (the E5) an expected 3.83 RA9 compared to Cessa’s 4.91 RA9. Cole was cruising, is a much better pitcher compared to Cessa, and even the third time through, Cole is projected to be better.

But this is where decision making comes in. After the two walks, Cole missed his 0-2 spot on a 98 mph fastball that Mountcastle smoked up the middle for a single and left a slider middle-middle for Ruiz to hit for a double. Cole wasn’t that version of Cole, and was nearing the end of his night, especially with Valika likely being his final batter of the game. Cole would need to project at a 4.56 RA9 true talent level to be removed third time through for Cessa on average, but averages change based on how the pitcher looks; feel is still part of the game. In this instance, though, perhaps Boone didn’t anchor to Cole compared to Cessa but anchored to pitching roles.

With Chapman back as the closer, Britton became the setup man for, at the time, the 21-17 Yankees. Britton pitched September 3rd and 4th and threw 48 pitches, so there’s a good chance he was unavailable, but for ease of illustration, let’s assume that Britton was available. Since joining the Yankees in 2018, the left-hander has not entered the game in the sixth inning, instead pitching the late innings to get the hold or save. This would be “anchoring” in terms of roles. “Britton is my setup guy” or “Chapman is my closer” are defined roles that the manager won’t break, especially down 1-0 in a ballgame in the sixth inning.

I think this can be looked at through an economics lens, even with anchoring as a clear possible reason for why starters are not removed earlier despite the TTOP and relief pitchers having set roles. In Risk, Uncertainty, and Profit by Frank Knight, a University of Chicago Economist, he writes that “Risk is something that can be measured by mathematical probabilities. Uncertainty refers to something that can’t be measured because there are no objective standards to express probability.” While distributions of the probabilistic outcomes are a way to express probability, we can use Knight’s statement in application. The risk is removing the starter for a lesser relief pitcher or the risk is using the setup guy/closer to early in a game and needing them later. The probabilities can be explained in win probability, but the uncertainty of the managerial decision has no objective standard of being expressed.

In Animal Spirits, based on what former General Electric CEO Jack Welch wrote in his book, the authors wrote “Decisions that matter for investment are intuitive rather than analytical. That intuition is a social process that follows the laws of psychology.” It is this sentence that illustrates the bias in managerial decision making. The manager is acting intuitively based on what they know about pitchers and how the pitcher is looking. Boone knows that Cole is better than Cessa and knows that Cole is struggling in the moment. He also knows that Britton is better than Cessa and with the projected 3.75 RA9 is a better option in the moment than Cole. At the time, the Yankees offense had zero runs and the Orioles had a 74.8 percent win expectancy after Cole’s second walk.

The risk of Cole vs Cessa vs Britton can be calculated in changes of win expectancy but the uncertainty cannot – who pitches the seventh, eighth or ninth if the Yankees take the lead? There’s no structure in which Boone has anchored Britton and Chapman into designated roles. His decision for when to make the pitching change and who to use is intuitive and not analytical, even for one of the more progressive clubs in the sport. Feel for the game is important, even if it brings in a bias. Managers to a club or important, and it’s important to understand their faults. Like investors and other individuals in the economy, managers succumb to anchoring and animal spirits when deciding on pitching changes.


We can further apply the behavioral economics framework to front office decision making to understand why teams act in the manner that they do. In an article on ESPN, Kiley McDaniel wrote about how taking risks is the next inefficiency in the sport. On one end of the spectrum you have the San Diego Padres, who traded away 16 players and added 10. The Padres also previously signed Eric Hosmer to an eight-year and $144 million deal followed by a 10-year and $300 million deal to Manny Machado for a team who didn’t win more than 77 games since 2011. General Manager AJ Preller took risks on signing Hosmer and Machado before the farm system started to produce. When the club showed they were in contention in 2020, Preller took to the offensive and brought in players to help the club try to win. Meanwhile, the Pittsburgh Pirates despite having success in 2013-2015 never made the splashy move. They were connected with Jon Lester, David Price, and Ben Zobrist at trade deadlines, but never finished the deal. Their moves in 2015 were successful, but they were calculated risks that had minimal loss if they failed and the right tail outcomes were extremely fortunate. The club never took a chance on a deal where the left tailed outcome was large (until 2018).

In Misbehaving by Richard Thaler, he talks about loss aversion and how the system of rewards and punishments changes incentives. Billy Beane has his job in Oakland secured and he traded away Josh Donaldson the year prior to him winning MVP, meanwhile Oakland accumulated 6.1 bWAR from four players. Beane is able to take risks given his situation. Meanwhile, in Pittsburgh, Neal Huntington got fired for trading Austin Meadows, Tyler Glasnow, and Shane Baz to the Tampa Bay Rays for Chris Archer in what looks to the worst trade of the 2010s. The goal for Huntington was to win every year, and he took his chance in 2018 despite not doing so from 2013-2015 with a much better Major League roster. As a result, Huntington got fired a year and half later for the results that occurred post trade and Billy Beane is going to the ALCS. The system sets incentives, and for Huntington, to the point of 2018, there was no point in taking risk because a backfire would cost him the job. Perhaps the incentives changed in 2018, and needing to win to keep his job, Huntington took the big risk, but that is more circumstantial. We do know that Beane won’t get fired for his risk taking in Oakland, but the General Manager in Pittsburgh will. It’s the structure of the two organizations in terms of rewards and punishments. Oakland is more ex-ante compared to Pittsburgh’s ex-post, even with the same ultimate measure: winning baseball games.

In terms of Huntington, he was aware of what is referred to as the winner’s curse, who at Pirates Fest 2018 called free agency a “losing game” and said “When you sign a free agent, you have automatically outbid everybody else to get him 95, 99 percent of the time. You have theoretically overpaid to get that free agent.” But the winner’s curse goes along with risk taking. The Pirates, under their current structure, don’t take risks and giving out a large contract is a risk, even if you have to outbid somebody. The Yankees are dealing with asymmetric information in their deal with Cole, but signing Cole makes the team better and he will likely bring in more value than even the $324 million. The same can be said for the Padres and Manny Machado or the Phillies and Bryce Harper. Taking risks, despite likely paying some dollar amount more than the competitor, helps in constructing a team.

The final example of bias in front office decision making is the false consensus effect, which states people share their preferences. The best example is again the move that got Huntington fired. After trading Gerrit Cole for Joe Musgrove, Colin Moran, Michael Feliz, and Jason Martin, Huntington traded away a better package for Chris Archer. Archer posted an ERA’s over 4.00 for the prior two and half seasons before being dealt to Pittsburgh, but he did have better indicators. I wrote about how Archer was a calculated risk at the time. This trade illustrates the right tail the Pirates are seeking, but it also came with a longer left tailed distribution if the players traded away reached their ceilings. But why was Huntington, after all the years of not making moves, ready to give up a larger package to get Archer compared to what he received for Cole? The answer lies in the false consensus effect.

At the time of the trade, Ken Rosenthal wrote how Kyle Stark, former Assistant General Manager and Farm Director, wanted the Pirates to acquire Archer when Stark and Huntington left Cleveland for Pittsburgh. Rosenthal, quoting Huntington said:

“We obviously were not able to acquire him (in ‘07-‘08), but we have continued to track and monitor his development into a terrific major-league pitcher and person. We have also had dialogue over the years with the Rays, but it took until today until something worked for both clubs. We believe his intelligence, character, pitch arsenal, advanced analytics and a move to our league and ballpark project really good results for him and us over the next three-plus years.”

It’s known that the Pirates wanted Archer for years and valued him highly internally, especially by moving him from the American League East to the National League Central. Archer was the hot name on the market for years but was finally moved in 2018 as the Rays felt they never got the required offer to move him, until they did. Remember the winner’s curse? That is still in play, as the Pirates had to overpay to acquire a player that no other team felt as highly on. By seeing Archer as the top of the line arm, the Pirates felt that other teams viewed him the same way and in order to acquire the talent, the club had to overpay. Two biases in one trade.

By being more away of these biases that drive front office decision making, the better a team can be. Each team will fall to a bias from here and there but limiting them will lead to better decision making. Managers anchor and follow intuition, while front offices are in organizations that reward risk taking in different ways and fall to the winner’s curse and false consensus effect. Markets are not always efficient and it’s important to understand the animal spirits and biases that drive decision making on the different levels of a Major League team. Organizations rewarding front offices for taking risks ex ante and not falling to their cognitive biases is the next frontier.

2 Comments on “Understanding Bias in Decision Making: Managerial and Front Office

  1. ‘anchoring’ bias** in bullpen utilization is interesting — very much in the spirit of “the closer is the closer because he’s the closer.”

    it’d be annoying to have to control for the day-to-day availability of RPs (perhaps so annoying as to make such a study cost-prohibitive) in the course of this work, but i wonder if you could detect/measure a bias toward managers calling upon high-dollar options before equivalently-skilled, less-expensive ones. something like a discrete choice setup with (a) as-of-date, ROS projections for each relief option; (b) some accounting of each relief option’s salary; & (c) some accounting for the scenario/context (TTO, handedness, leverage, base-out state, etc. etc. etc.).

    ** (it’s been a bit since i read kahneman — i forget if ‘anchoring’ fits under his banner of ‘answering an easier question’, which is maybe more like what i’m thinking of here — manager, while navigating uncertainty, reaches for highest-paid reliever for highest-leverage appearance instead of reaching for ‘best’ reliever.)

    good stuff — i look forward to reading back in time through this blog 🙂

  2. The day-to-day availability might be hard to gather but I think FanGraphs has game logs so I could make a proxy (back-to-back days pitched, x amount of pitches in y amount of days, etc for a proxy). I think perhaps when school dies after midterms or near break, I’ll start trying to build out a data set to measure that – that’s a good idea on a way to test for bias.

    Thank you for the kind words on the blog and I hope the rest of the entries suit your interests.

Leave a Reply

Your email address will not be published. Required fields are marked *