Finance Theory Insights

On the Magnification of Small Biases in Hiring

Shaun W. Davies, Edward D. Van Wesep, Brian Waters

Based on: Journal of Finance (Forthcoming). https://doi.org/10.1111/jofi.13374

---

 

In many hiring scenarios, a single candidate must be selected from a pool of potential applicants. Boards choose new chief executive officers (CEOs), sports teams choose new head coaches, and CEOs choose new chief financial officers (CFOs) or divisional managers. These settings differ from those in which many candidates can be chosen, because choosing one candidate inherently requires not choosing others. Any resources spent investigating a candidate who is not selected might therefore be wasted. How then should an organization choose the optimal levels of time and effort to put into selecting a new employee? We show that it is usually best to put varying levels of resources into learning about each candidate, even if candidates look equally qualified at the outset.

To see why this is the case, consider a simple example in which a board must select a CEO from two finalists, Anne and Bill. The board must choose how much time to devote to interviewing the candidates and talking to people who have worked with or for them in the past. It must decide how much money to devote to background checks and investigations of each candidate's public statements, social media posts, and possible non-public skeletons. Suppose that the board has already sunk substantial resources into learning about Anne and has spent less in learning about Bill. For example, suppose the board has already contacted ten of Anne’s previous colleagues across each of her past employers, while the board has contacted only two of Bill’s past colleagues. If the information the board acquired about Anne was fairly positive—nine out of ten colleagues the board contacted praised her work ethic and communication skills, she seems like a good fit culturally, she has relevant technical skills and experience, etc.—then Anne will probably be chosen, regardless of what the board learned about Bill. Now consider the benefit of contacting one more of Bill’s references.  Even if the interview goes very well, it probably will not convince the board that Bill is a better choice than Anne, because three good interviews won’t outweigh the much deeper knowledge that the board has acquired from her ten interviews. The same goes if information about Anne was fairly negative: given how comprehensive that information is, Anne will not be chosen even if the information about Bill is also somewhat negative. Because the information the board uncovered about Bill is less comprehensive, it is unlikely that it will be pivotal in determining the board’s choice between candidates. This makes learning about Bill less valuable.

The board also might consider interviewing another of Anne’s contacts. If the first ten interviews were fairly positive, another interview could be valuable, especially if it reveals something very negative about her leadership or background. She was likely to be chosen after the first ten interviews, but a negative eleventh interview could have significant ramifications for her candidacy. A similar argument follows if her first ten interviews were fairly negative. A glowing eleventh interview could bring her back into consideration. Putting these two cases together, the benefit of learning more about Anne is higher than the benefit of learning more about Bill because the level of learning about Anne is higher than the level of learning about Bill. The more the board learns about a candidate, the more valuable learning about that candidate becomes, relative to the value of learning about the other candidate.

This intuition has a close correspondence to real options. When the value of oil is low, the value of an oil well is low. An increase in the value of oil does not much change the value of the well because it is unlikely to be tapped either way. When the value of oil is high, further increases in that value substantially affect the value of the well because it is now likely to be tapped.

By its nature, then, the problem of selecting a single candidate for an open position will tend to favor divergent allocations of resources. Why is this important? In the example considered above, the board does not care whether it learns more about Anne and less about Bill or vice versa. The choice is arbitrary. But while the board may not care which candidate gets more attention, the candidates do: if the board puts more time, money, and effort into learning about Anne, then Anne’s likelihood of being chosen will be different than Bill’s. It may be higher or lower, but it will be different, so Anne and Bill have an interest in how the board allocates its resources.

Suppose now that we allow the board to have an infinitesimal bias in favor of one of the two candidates. For example, suppose the board would always prefer to hire the best candidate whenever possible, but if the two candidates appear equally qualified for the role of CEO, the board prefers the candidate who is a better golfer. This bias eliminates the arbitrariness of the board’s decision: it will put more effort into learning about either Bill or Anne based on which person it believes is more likely to be a good golfer. As a result, a good golfer will be selected more often than half of the time, often much more. It is critical to understand that this is not because the board considers a candidate’s golf ability to be of first-order importance. It is used only as a tie-breaking rule, and ties almost never happen.

To continue the example, suppose that the board is biased in favor of Bill because of his golf skills and that the information the board acquires as it learns about each candidate is more likely to be positive than negative. This might occur, for example, if a glowing recommendation is the default and references that raise red flags are unusual. In this case, the board invests more resources into learning about Bill, usually receives more good news about Bill’s skills than Anne’s, and chooses Bill more often. If instead information is more likely to be bad than good (as might be the case when scanning a candidate’s social media page), then the board invests more resources in learning about Anne, usually receives more bad news about Anne’s quality than Bill’s, and therefore chooses Bill more often. Heads, Bill wins; tails, Anne loses.

One might expect that for important decisions like choosing a CEO, any infinitesimal bias would be overwhelmed by the effort that goes into making the right decision. Unfortunately, this conventional wisdom is often incorrect. The magnification of a bias can be higher when the importance of choosing a high-quality candidate is higher. The reason is that higher stakes typically incentivize more detailed investigations. When the information the board receives about each candidate is more detailed, the likelihood that the board obtains positive or negative news about each candidate is more unequal. This increases the odds of the favored candidate being chosen. A glass ceiling arises in which favored candidates are more likely to be selected for more highly valued positions.

Recently, concerns over inequality in executive hiring have emerged in the popular press. As of 2021, only five CEOs of Fortune 500 firms were Black and only 41 were female. [1] Of over 500 head coaches in the history of the National Football League (NFL), only 24 have been Black. [2] While it is difficult to prove that these facts are the result of biases, former head coach Brian Flores made this claim in a 2022 lawsuit against the NFL and its teams. Our model provides policy recommendations for solving this problem, to the extent that it exists.

First, simply reducing the size of board members’ biases will not necessarily have much effect on the frequencies with which favored candidates are chosen. If policy makers want to reduce the gap in these frequencies, then educating decision makers about implicit bias may not be very useful. This is consistent with substantial research suggesting that implicit bias training does not work. Second, a rule like the NFL’s Rooney Rule, which requires interviews of at least one minority candidate for any head coaching position, is also unlikely to solve the problem. The issue arises after a short list has been created.

Third, a rigorously enforced rule that efforts toward learning about each candidate must be equal could solve the problem of the magnification of bias. Interestingly, employment attorneys typically advise clients to make equal efforts toward learning about each candidate in interviews to avoid liability for appearing biased in the hiring process. While precisely equal efforts cannot feasibly be enforced, more equal effort reduces the differences in outcomes due to bias. The non-profit Society for Human Resource Management, citing substantial academic and non-academic literature, argues that structured rather than unstructured interviews are more effective for executive recruitment. If structured interviews are interpreted as “identical efforts,” then they are a possible solution to mitigate the magnification of small biases in hiring.

 

 

 

---

[1] https://www.npr.org/2021/05/27/1000814249/a-year-after-floyds-death-you-can-still-count-the-number-of-black-ceos-on-one-ha, and https://fortune.com/2021/06/02/female-ceos-fortune-500-2021-women-ceo-list-roz-brewer-walgreens-karen-lynch-cvs-thasunda-brown-duckett-tiaa/, accessed March 15, 2022.

[2] https://www.indystar.com/story/sports/nfl/2022/02/10/first-black-nfl-coach-fritz-pollard-akron-pros/6637386001/, accessed on March 15, 2022.

---

 

---

---