This is our second blog post for our Job Market Paper Series blog for 2024-2025.
Yuen Ho is a PhD candidate in Agricultural and Resource Economics at the University of California, Berkeley. Her research lies at the intersection of development economics, behavioral economics, and firms. You can find her JMP here.
Good managers matter. Individual managers have a big impact on how workers and teams perform (Lazear et al., 2015). Given their importance, how can firms select good managers? The fact that we see so many “bad” managers in practice attests to the difficulty of solving this problem. This issue is also particularly important in developing countries, where business owners often struggle to find middle managers capable of working independently and effectively managing teams. Indeed, a “missing middle” in management is often cited as a key barrier to firm growth.
One option for firms is to promote workers based on seniority or measured performance. Such methods have the benefit of being transparent and fair. However, the best workers may not make the best managers (Benson et al., 2019). For one thing, managerial roles often require soft skills, such as communication, navigating interpersonal relationships, and the ability to motivate others. In practice, many firms rely on supervisor discretion to select workers for promotion. In theory, firms may do this because supervisors have better information about which workers would be best suited for a role. However, there is remarkably little empirical evidence supporting this theoretical prediction. This is likely due to data constraints, as selection processes within firms are typically a black box. The existing papers largely conclude that discretion leads to worse personnel decisions (Hoffman et al., 2018), crowding in bias, favoritism, or misaligned incentives.
Designing experiments with real employees within a real firm
In my job market paper, we partner with one of the largest garment manufacturing firms in Tanzania to implement a series of field experiments studying promotions to managerial roles. In the status quo, our partner firm relies almost solely on supervisors to make such decisions. To study discretionary selection, we first conduct a field experiment where all supervisors are given the opportunity to refer up to two workers for promotion. During the referral process, we randomly vary whether supervisors face financial incentives tied to the quality of their referrals. The referral bonuses were worth roughly 10% of their monthly base wage. Our experimental design allows us to test whether supervisors have informational advantages and whether their choices are in the firm’s best interests.
As workers directly experience discretion, we also study how different selection methods affect workers’ decisions to apply for promotion. In particular, we randomly vary which selection criteria are emphasized on the application forms that workers receive. In the Control group, we inform workers that selection will be based on many criteria, including their performance record. In the Discretionary group, we inform workers that selection will be based on many criteria, including supervisor referrals.
To assess the quality of referrals and applicants, we then invite all workers who applied or who were referred to take a leadership test. This test allows us to measure the managerial quality of all potential candidates, not just those that are eventually promoted. For the test, the firm provided some screening questions covering technical production knowledge. The remaining questions measured soft skills that have been found to correlate with managerial performance across a variety of settings. Indeed, our measure appears to capture some dimensions of leadership that are otherwise difficult for the firm to observe. We find that leadership scores are not significantly correlated with observable measures of line-level worker performance, such as output and attendance. However, leadership scores do significantly correlate with supervisor performance on the job. Replacing a supervisor at the 10th percentile with one at the 90th percentile of leadership scores is associated with a 12% increase in worker productivity.
What are the costs and benefits of letting supervisors make promotion decisions?
Our results show that randomly assigned referral bonuses lead supervisors to refer workers who score on average 13% higher on the leadership test relative to status quo referrals. Under a simple conceptual framework, these results suggest simultaneously that 1) supervisors have private information about the managerial quality of workers and 2) supervisors have personal preferences that are in conflict with the firm’s best interests. Supervisors also show gender bias and favoritism. This matches the stated preferences of supervisors who say, on average, that they believe men generally make better leaders than women.
Our results also show that discretion is perceived by workers and is important to workers. Randomly emphasizing supervisor referrals in the selection process leads to a 12% reduction in the number of workers who apply for promotion relative to the control group. As a direct result, the firm receives relatively fewer applications from workers with high managerial ability under Discretionary selection. These effects suggest there is also misalignment in objectives between workers and supervisors. Indeed, workers strongly prefer promotion decisions to be based on objective rather than subjective criteria. In survey responses, 67% of workers think the performance record should be the most important factor considered in promotion decisions, relative to 19% who prefer the leadership test, and 13% who prefer supervisor referrals. This is not driven by workers who are disadvantaged by discretion, as the distribution of preferences among workers who were actually referred by supervisors is nearly identical.
Could firms obtain comparable private information from other sources?
Given the costs of discretion, could firms obtain comparable information from other sources? Our results suggest no. Supervisors appear to have private information beyond what the firm could infer from existing administrative data, from workers’ self-assessments, and from horizontal referrals from coworkers. On average, supervisors select workers who score 21-24% higher on the leadership test compared to selection based on seniority or performance records. A back of the envelope calculation suggests that this difference in leadership ability between workers selected by supervisors versus other methods would correlate with a difference in daily worker productivity of 2-3%.
Workers also do not seem to have accurate assessments of their own leadership ability. Workers are highly overconfident when guessing their own scores on the leadership test, even when they earn bonus payments for answering accurately. Approximately 82% of workers overestimate their score, with 60% of workers overestimating by at least 200%. This fits with new lab evidence that finds that participants who nominate themselves for leadership positions perform worse as managers than those selected by lottery, driven in part by overconfidence (Weidmann et al. 2024).
Given that peer referrals have been found to improve hiring decisions (Beaman and Magruder, 2012), we also test whether firms could gain the same informational advantages from coworker referrals. To do so, we replicate our referral experiment with a randomly selected sample of line-level workers who are asked to refer coworkers for promotion. Our results show that incentivized coworkers, despite facing the same financial incentives as incentivized supervisors, do not refer workers with higher leadership potential. If anything, they refer workers who score lower on the leadership test.
Policy Implications
In practice, almost all firms rely on discretion to make promotion decisions. Our results suggest this can be explained, at least in part, by the fact that supervisors have valuable private information about the managerial quality of workers. This is also despite the fact that discretionary decisions are not perfectly aligned with the firm’s interests and are strongly disliked by workers. From a policy perspective, our results suggest that introducing objective measures that proxy for typically unobservable qualities can equip firms with better monitoring tools and bring discretionary decisions more into alignment with firm objectives.
One limitation of our study is that we are not able to assess whether supervisors have valuable private information beyond what we can capture in our leadership measure. To do so would have required randomizing promotions conditional on selection methods, which was not feasible within our research partnership with a real firm. We think this would be a valuable direction for future research.
Feature image captured by Yuen Ho, showing workers in the factory’s sewing department.
