# 如何识别优秀的经理？ / How Do You Identify a Good Manager? * **作者 / Authors**：Ben Weidmann, Joseph Vecci, Farah Said, Sonia Bhalotra, Achyuta Adhvaryu, Anant Nyshadham, Jorge Tamayo, David Deming * **期刊信息 / Citation**：《经济学季刊》（*The Quarterly Journal of Economics*）, Volume 141, Issue 2, May 2026, Pages 1581–1633, [https://doi.org/10.1093/qje/qjag004](https://doi.org/10.1093/qje/qjag004) --- ## 摘要 / Abstract 本研究提出并验证了一种识别优秀经理的新方法。在一项预注册的实验室实验中，我们通过将经理随机分配到不同团队并控制其个体技能，因果识别了经理人的管理贡献。我们发现，经理的贡献对团队成功至关重要。与之相反，主动选择（自我推荐）管理岗位的人在实际工作中的带队表现，往往比随机指派的经理更为逊色。管理表现很大程度上取决于其经济决策技能，而与人口统计学特征无关。这一“经济决策技能”由分配博弈测试进行客观测量，主要评估个体在面对高度复杂的数字信息时如何战略性地分配注意力以及发掘并利用团队的比较优势。两项实地验证研究较好地支持了我们的实验室实验结果：第一，在实验室中表现出色的参与者在现实劳动力市场中获得了更多的晋升；第二，在另一项针对零售店经理的研究中，相应的技能测验指标能够显著预测门店的销售业绩。在我们的实地估算中，店长质量提高一个标准差，对应的门店年销售额大约可增长 410 万美元（增幅达 25%）。这些结果表明，基于决策技能（而非仅凭人口特征或主观的领导欲望）来选拔经理，可以显著且稳健地改善组织绩效。 We introduce and validate a novel approach to identifying good managers. In a preregistered lab experiment, we causally identify managerial contributions by randomly assigning managers to teams and controlling for individual skill. We find that manager contributions are crucial for team success, and that people who self-select into management roles perform worse than randomly assigned managers. Managerial performance is strongly predicted by economic decision-making skill but not by demographic characteristics. Two validation studies support our experimental results. Participants who succeed in the lab receive more real-world promotions and, in a separate study of retail store managers, skill measures strongly predict store sales. A one standard deviation increase in manager quality increases annual per store sales by US$4.1 million (25% increase). Selecting managers on skills rather than demographic characteristics or the desire to lead could substantially improve organizational performance. --- ## I. 引言 / I. Introduction 管理对于组织和国家的经济绩效有着举足轻重的影响（Bloom and Van Reenen 2007; Bloom, Sadun, and Van Reenen 2016; Bruhn, Karlan, and Schoar 2018; Giorcelli 2019; Gosnell, List, and Metcalfe 2020; Metcalfe, Sollaci, and Syverson 2023b）。在同一企业内部和不同国家之间，经理人在生产力方面存在着巨大且持久的差异（Lazear, Shaw, and Stanton 2015; Bender et al. 2018; Adhvaryu, Nyshadham, and Tamayo 2023; Metcalfe, Sollaci, and Syverson 2023a）。优秀的经理能通过多种途径提升生产力，包括监督绩效、培训和激励员工，以及将员工重新分派到更适合的角色或工作任务中（Fenizia 2022; Metcalfe, Sollaci, and Syverson 2023a; Minni 2023; Adhvaryu, Nyshadham, and Tamayo 2023; Dessein et al. 2024; Benson and Shaw 2025）。 Management matters greatly for economic performance (Bloom and Van Reenen 2007; Bloom, Sadun, and Van Reenen 2016; Bruhn, Karlan, and Schoar 2018; Giorcelli 2019; Gosnell, List, and Metcalfe 2020; Metcalfe, Sollaci, and Syverson 2023b). There are large, persistent productivity differences between managers within firms and across countries (Lazear, Shaw, and Stanton 2015; Bender et al. 2018; Adhvaryu, Nyshadham, and Tamayo 2023; Metcalfe, Sollaci, and Syverson 2023a). Good managers increase productivity through many channels, including monitoring performance, training and motivating staff, and reallocating workers to better roles or job tasks (Fenizia 2022; Metcalfe, Sollaci, and Syverson 2023a; Minni 2023; Adhvaryu, Nyshadham, and Tamayo 2023; Dessein et al. 2024; Benson and Shaw 2025). 现实中，企业是如何识别优秀经理的？在实际操作中，企业严重依赖现有经理的主观判断，而这往往存在众所周知的偏见（Kahneman and Klein 2009; Hoffman, Kahn, and Li 2018; Chamorro-Premuzic 2019; Feld et al. 2022）。此外，企业在选拔经理时还会参考其性格特质和认知能力，研究表明这些特征与实地工作中的经理绩效呈正相关（Judge et al. 2002; Judge, Colbert, and Ilies 2004; Javalagi, Newman, and Li 2024）。 How do firms identify good managers? In practice, firms rely heavily on the judgment of existing managers, which suffers from well-known biases (Kahneman and Klein 2009; Hoffman, Kahn, and Li 2018; Chamorro-Premuzic 2019; Feld et al. 2022). Firms can also select managers based on personality traits and cognitive ability. Research has shown that these characteristics exhibit positive associations with manager performance in the field (Judge et al. 2002; Judge, Colbert, and Ilies 2004; Javalagi, Newman, and Li 2024). 然而，现有的经理选拔研究面临两个实际局限。第一，经理在实地中并非随机分配到团队中，这导致难以在排除其他变量干扰的前提下因果识别其绩效。第二，实地中的经理群体是从总人口中进行高度非随机筛选的样本。这容易引发对“哪些特征真正能够改善绩效”的错误推断。例如，Benson, Li, and Shue (2019) 发现，销售经理往往是基于其作为普通员工时的优异表现而被选拔的，这极容易遵循**彼得原理（Peter Principle，即在层级组织中，员工往往会被提拔到他们不能胜任的位置，直到他们的能力无法支撑进一步的晋升，从而导致管理层充斥着不称职的人）**。这种晋升机制可能会导致员工的表现与其转型成为经理后的表现呈负相关，哪怕提升经理的整体业务技能从因果关系上讲是有益的（Benson, Li, and Shue 2019）。 Existing research on manager selection suffers from two practical limitations. First, managers are not randomly assigned to teams, which makes it difficult to causally identify managerial performance. Second, managers in the field are a highly nonrandom sample of the population. This may lead to incorrect inferences about the characteristics that truly improve performance. For example, Benson, Li, and Shue (2019) find that sales managers are selected based on their performance as line workers, that is, according to the Peter principle. Promotions based on the Peter principle can induce a negative correlation between a worker’s performance and their subsequent performance as a manager, even though the causal effect of increasing worker skills among managers is likely positive (Benson, Li, and Shue 2019). 本研究通过一项大规模预注册的实验室实验（$n = 555$），引入了一种因果识别高效经理的新方法。首先，我们通过将每个经理随机分配到多个不同团队，并控制团队在特定任务上的表现，从而将管理技能与单纯的解题业务技能分离开，由此对经理的贡献进行因果识别。我们设计了一项需要协调、监督和激励员工的协作生产任务，用以模拟现实世界中的团队生产活动。实验中，每个经理被随机分派至四个不同的三人团队。直观而言，优秀的经理能让其团队产生“1+1>2”的协同效应。我们发现，好管理的影响相当显著：在控制了所有成员（包括经理自身）的基础业务技能后，管理技能每提高一个标准差，能使团队绩效提升 0.22 个标准差。更具体地讲，我们发现经理的整体质量对团队业绩的影响，与普通员工的总生产技能几乎不相上下。 We introduce a novel method for identifying effective managers, using a large preregistered lab experiment ($n = 555$). First, we causally identify managerial contributions by randomly assigning each manager to multiple teams and controlling for task-specific performance of managers and workers, which isolates management skill from productive capacity. We design a collaborative task that emulates real-world team production by requiring managers to coordinate, monitor, and motivate workers. Over the course of the experiment we randomly assign each manager to four different three-person teams. Intuitively, good managers consistently cause their teams to produce more than the sum of their parts. We find that the impact of good management is relatively large: a one standard deviation increase in managerial skill improves team performance by 0.22 standard deviations conditional on the productive skills of all workers including the manager. To put these figures into context, we find that overall manager quality matters about as much to team performance as the combined productive capacity of workers. 其次，我们调查了那些主动选择并强烈渴望成为经理（自我推荐）的人是否在实际工作中表现更出色。在向参与者描述了经理和普通员工的任务与薪酬结构后，我们让其以 1-10 分形式对成为经理的意愿进行打分。一半的组别被随机分配到“自我推荐组”，即让意愿最强烈的人担任经理；另一半的“抽签组”经理则是从参与者中随机抽取的。我们发现，由自我推荐的经理领导的团队，其表现比随机抽签经理领导的团队差 0.1 个标准差，这一差距相当于分派了一位流体智力得分低一个标准差的经理。 Second, we investigate whether people who self-select into management roles typically perform well on the job. After describing the tasks and compensation structure of the managers and workers, we elicit the preference of participants to be a manager on a 1–10 scale. Half of all groups were then randomly assigned to a self-promotion treatment where participants with the strongest preferences to be a manager were given the role. The other half of managers were drawn randomly from the participant pool. We find that teams with self-promoted managers perform 0.1 standard deviations less well than teams with randomly assigned managers, a gap that is roughly equivalent to being assigned a manager with a one standard deviation lower fluid IQ score. 探索性分析表明，自我推荐的经理表现欠佳可能与过度自信有关。我们发现，表达强烈管理意愿的人往往会高估自己的社交技能——在**“眼部情绪识别测试”（Reading the Mind in the Eyes Test，简称 RMET，这是一种常用于评估认知同理心和情绪感知能力的经典微表情心理学测试）**中，他们的客观表现与自评得分存在显著偏差。在自我推荐的经理中，自评人际关系技能与客观管理表现甚至呈显著负相关。与之相反，随机抽签的经理并没有表现出过度自信，其自评人际关系技能也与绩效没有负面关联。 Exploratory analyses suggest that the underperformance of self-promoted managers may be related to overconfidence. We find that people who express a preference for managerial roles overestimate their social skills relative to an objective test of emotional perceptiveness called the Reading the Mind in the Eyes Test (RMET). Among self-promoted managers, we also find a negative relationship between self-reported people skills and managerial performance. Conversely, randomly selected managers do not overestimate their social skills, and we find no negative relationship between self-reported people skills and managerial performance. 在实验室中，哪些特质能够预测管理者的成功？为回答这一问题，我们重点分析了“抽签组”经理的样本，因为自我推荐组的特质与业绩关联会受到主动选择机制的干扰。我们发现了两个能一致预测管理绩效的指标：**流体智力（fluid intelligence，指在缺乏经验线索时的抽象逻辑推理与快速解决新颖客观问题的硬性认知能力，采用瑞文高级渐进矩阵测验进行评估）**和**经济决策技能（economic decision-making skill，通过“分配博弈”测验进行客观测量，主要评估个体在信息过载和有限时间的压力下，如何直观评估任务重要性并优化分配有限的注意力资源，以最大限度提高决策产出）**。控制了年龄、性别、教育程度、工作经验、情绪感知以及包括风险偏好在内的 Big 5 性格特征后，这两个指标的预测效果依然稳健。其他任何特征对管理表现都没有很强的预测能力。 What characteristics do predict managerial success in the lab? To answer this question, we focus on the sample of lottery managers as the associations between manager performance and manager characteristics in the self-promoted treatment arm are moderated by the filter of self-selection. We identify two consistent predictors of managerial contributions: economic decision-making skill and fluid intelligence. These relationships are robust to a wide range of controls, including age, gender, education, work experience, emotional perceptiveness, and personality traits including risk preferences. No other characteristics strongly predict managerial performance. 为了验证实验研究的外部有效性并评估其经济意义，我们开展了两个独立的实地研究。首先，我们收集了实验室参与者公开的 LinkedIn（领英）数据，以测试实验室中测得的管理表现是否能够对应现实中真实的职业晋升差异。尽管这批年轻的早期职业员工平均年龄为 27.5 岁，平均工作经验仅为 4.9 年，但我们发现实验室管理表现每提高一个标准差，现实中每年能多获得 0.16 次晋升（$p < .001, n = 73$，基准晋升率为每年 0.25 次）。在控制了流体智力、年龄、性别、种族和性格特质后，这一结果依然十分稳健，说明我们在实验室里捕捉到的管理技能在学历、智力和人口统计学因素之外，对个体的职业生涯发展有着独立且不可替代的预测价值。 We conduct two separate field studies to validate our experimental results and evaluate their economic importance. First, we use publicly available LinkedIn data on our lab experiment participants to test whether managerial performance measured in the lab corresponds to meaningful differences in actual career progression. Although our sample is limited to early-career workers (mean age = 27.5 years; mean work experience = 4.9 years) we find that a one standard deviation increase in the lab measure of managerial performance is associated with 0.16 more real-world promotions per year ($p < .001, n = 73$; base rate of 0.25 promotions per year). The association is robust to controls for fluid intelligence, age, gender, ethnicity, and personality, suggesting that our experimental measure of managerial skill captures something predictive of career progression above and beyond traditional predictors, such as education, intelligence, or demographic factors. 其次，我们深入到具体的企业实地场景中，研究长期人际互动背景下的经理人作用。我们招募了南美一家年营业额达数十亿美元的大型零售企业中的 225 名超市门店经理。每位门店经理都接受了与实验室实验相同的技能测验组合：经济决策技能、流体智力和社交情绪感知。随后，我们将他们的得分与该企业在过去 28 个月内的门店详细销售和运营数据进行了关联。 Second, we examine the role of managerial skills in a firm setting, where managers have repeated, long-term interactions with staff. We recruit a sample of grocery store managers employed by a multibillion-dollar retail firm in South America. Each store manager completed a set of three skill assessments that matched the skills assessed before the lab experiment: economic decision-making skill, fluid intelligence, and emotional perceptiveness. We link these test scores to detailed data on store performance and sales over a 28-month period. 这一实地环境与我们的实验室实验有极佳的契合性。关键是，实地超市经理的核心职责与我们实验室的管理任务非常相似：他们将前端员工分派到不同的工作站，激励员工，监督店面表现，并需要主动排查瓶颈（如货架缺货）。由于各门店采用标准化业务流程，中央总部跨门店轮换经理的做法，在形式上完美契合了我们在实验里的多轮随机重新分配。因此，实地研究环境提供了一个能够量化中层经理价值、检验在实验室中大放异彩的那些底层技能是否能在实际商业场景中同样奏效的阵地。 The field setting is analogous to our lab experiment in several ways. Importantly, the middle managers we study have similar core responsibilities to managers in our lab setting: they delegate tasks to frontline workers, motivate staff, monitor performance, and address potential bottlenecks (e.g., inventory stockouts). As the stores share standardized practices and procedures, manager mobility across stores represents a nonrandom analogue of our repeated randomization procedure. Our field setting therefore provides a nonexperimental measure of managerial performance that can be used to quantify the value of middle managers and to examine whether the broad skills that predict manager contributions in the lab also predict performance in the field. 我们利用事件研究法分析和估计了经理的增加值。结果显示，如果派驻一位优秀的经理（定义为在经理固定效应中表现出比平均水平高一个标准差的经理），可使该门店年销售额增加 410 万美元（增幅达 25%）。这与此前在其他行业的研究发现高度一致（这部分研究认为，管理质量每改善一个标准差，大约能带来 10%–22% 的业务增幅，详见 Wells 2020; Fenizia 2022; Giardili, Ramdas, and Williams 2023）。极其引人注目的是，实地研究中预测超市经理绩效的最强指标依然是经济决策技能，这与实验室实验发现不谋而合。经理的经济决策技能每提高一个标准差，其实地管理绩效就显著增加 0.19 个标准差（$p = .01$）。 Using an event-study design, we estimate that a good manager (defined as being one standard deviation better in terms of managerial fixed effect) improves annual per store sales by US$4.1 million (25%). This improvement is broadly in line with empirical estimates from other industries, which have linked a one standard deviation improvement in management quality to performance improvements of 10%–22% (Wells 2020; Fenizia 2022; Giardili, Ramdas, and Williams 2023). Notably, we find that the strongest predictor of managerial performance is economic decision-making skill, which aligns with our experimental findings. A one standard deviation increase in economic decision-making is associated with a 0.19 standard deviation improvement in the performance of retail store managers ($p = .01$). 最后，我们模拟并量化了不同选拔政策对团队和企业产出的潜在影响。在实验室中，“抽签组”的数据表明，如果基于经济决策技能来考核和提拔经理，不仅团队表现比自我推荐机制能好上 0.7 个标准差，还显著好于随机指派或直接提拔业务最强员工的“彼彼得原理”机制。在实地超市场景中，如果企业基于经济决策技能来录用和分派店长，那么各门店的年平均销售额将增长达 16%（每店增益约 260 万美元）。 We quantify the impact of different management selection regimes. Starting with evidence from the lab, results from the random assignment arm of the experiment suggest that selecting managers based on economic decision-making skill improves performance by 0.7 standard deviations relative to self-promotion. Selecting managers based on economic decision-making skill also yields significantly better manager performance than a random-assignment benchmark or selecting the most productive workers to be managers (by following the Peter principle). An analogous exercise using our field data suggests that if new managers were hired based on their economic decision-making skills, annual per store sales would increase by US$2.6 million (16%). 此外，我们探究了优秀经理是如何确保团队成功的。在实验中，优秀经理的核心举措体现在：（1）密切监督，确保团队瓶颈问题能得到及时解决；（2）精准分工，将员工放到最能发挥其个人特长（比较优势）的模块上；（3）积极维持团队的工作热情。在定量分析上，监督和激励的作用甚至大于分工。在实验室协作中，卓越经理（即管理表现比平均水平高一个标准差的经理）在整场任务中出现的“监督失误”（即有成员做无用功）的概率大约只有普通经理的一半。在实地零售店的研究中，也呈现了同样的规律：拥有较高经济决策技能的优秀店长，其店内的货架缺货率显著降低，表明他们能更高效地监督库容、预判库存缺货瓶颈并采取预防性行动。 Last, we investigate what good managers do to help their teams succeed. We find that good managers (i) monitor their workers to make sure bottlenecks are being addressed; (ii) match workers to the tasks that best fit their skills; and (iii) keep workers motivated. Quantitatively, monitoring and motivation are more important than allocation, although all three are substantively important. In the experiment, good managers—defined as those scoring one standard deviation above average in terms of their managerial skill—make monitoring errors about half as often as other managers. Similarly, in our field study we demonstrate that good managers—those with strong economic decision-making skills—are more likely to successfully monitor stock and proactively avoid stockouts. 我们的研究主要做出了三方面的学术贡献。首先，我们开发了一种全新的、可前瞻性识别和评估卓越管理潜能的评估方法。这一方法有力地补充了目前经济学和心理学界的主流手段。在经济学研究中，学者通常是通过考察团队在面临不同经理人轮换时的业绩产出变化来回顾性地核算经理的增加值（Bertrand and Schoar 2003; Lazear, Shaw, and Stanton 2015; Janke, Propper, and Sadun 2024; Wells 2020; Fenizia 2022; Giardili, Ramdas, and Williams 2023; Metcalfe, Sollaci, and Syverson 2023a; Minni 2023）。虽然此类研究贡献卓越，但因为属于回顾性评估，它们无法在前瞻性选拔时提供有效预测，且数据要求极度苛刻。另一些经济学家则采用访谈手段来抓取企业层面管理实践的整体差异（例如著名的世界管理调查，Bloom and Van Reenen 2007, 2010; Bloom, Sadun, and Van Reenen 2016）。我们的研究与之互补，展示了如何剥离企业因素来准确识别并剥离出单个经理人自带的个体特质。 Our article makes three main contributions. First, we develop a skills-based approach to assessing managerial talent that can prospectively identify people with strong managerial potential. Our method complements the measures commonly used in economics and psychology to study managers. In economics, researchers measure managerial contributions by examining variation in output when teams (or firms) are exposed to different managers (Bertrand and Schoar 2003; Lazear, Shaw, and Stanton 2015; Janke, Propper, and Sadun 2024; Wells 2020; Fenizia 2022; Giardili, Ramdas, and Williams 2023; Metcalfe, Sollaci, and Syverson 2023a; Minni 2023). Although these studies have many strengths, they cannot be used prospectively to identify strong managers and have very demanding data requirements. Economists have successfully used interviews to capture differences in management practices across firms (e.g., Bloom and Van Reenen 2007, 2010; Bloom, Sadun, and Van Reenen 2016). We complement this literature by showing how to identify the skill of individual managers, as distinct from firm practices. 在工业与组织心理学中，传统评估多依赖于三类方法：（1）自我或他人评估的经理人性格及自信心（Rotemberg and Saloner 2000; Judge et al. 2002; van den Steen 2005）；（2）自评或他评的管理行为和领导风格，如变革型、交易型、真实型或身份型领导（Bass and Avolio 2000; Walumbwa et al. 2008; Van Dick et al. 2018; Jensen et al. 2019; Beenen et al. 2021）；以及（3）情境判断测验（SJT），即向参与者提供虚拟工作情境，让他们选择最恰当的反应（Chan and Schmitt 2002; Whetzel and McDaniel 2009）。我们的方法专注于“基于具体客观工作任务表现”的评估，有力地拓宽了这一领域。 In industrial organization psychology, a related literature examines individual differences in managerial quality, relying on three main approaches to measurement: (i) self- or peer-reported managerial traits, such as personality or confidence (see Rotemberg and Saloner 2000; Judge et al. 2002; van den Steen 2005); (ii) self- or peer-reported managerial behaviors and leadership styles such as transformational, transactional, authentic, or identity leadership (Bass and Avolio 2000; Walumbwa et al. 2008; Van Dick et al. 2018; Jensen et al. 2019; Beenen et al. 2021); and (iii) situational judgment tests in which participants are presented with hypothetical workplace scenarios and asked multiple-choice questions about the best way to respond (Chan and Schmitt 2002; Whetzel and McDaniel 2009). Our approach complements this literature by focusing on performance-based measures of managerial skill. 其次，我们丰富了量化经理人价值的经验文献（Bertrand and Schoar 2003; Lazear, Shaw, and Stanton 2015; Metcalfe, Sollaci, and Syverson 2023a）。长期以来，关于中层经理（Middle Managers）这一群体如何促进一线生产效率的研究一直没有得到像大企业 CEO 和高管那样程度的关注（Metcalfe, Sollaci, and Syverson 2023a; Hoffman and Stanton 2024）。我们证明在两种情景下，经理的质量和员工的整体基础技能对于最终团队产出的贡献都是同等量级，技能指标和人口特征解释经理人绩效变异的比例也高度一致。 Second, we contribute to the empirical literature that quantifies the value of managers (Bertrand and Schoar 2003; Lazear, Shaw, and Stanton 2015; Metcalfe, Sollaci, and Syverson 2023a) by providing new evidence of the value of middle managers, an important group that has been relatively understudied compared to senior leaders (Metcalfe, Sollaci, and Syverson 2023a; Hoffman and Stanton 2024). Encouragingly, our lab and field studies produce similar results: in both cases the importance of manager quality and the overall quality of workers are roughly equivalent. Similarly, the extent to which variation in manager quality is explained by skill measures and demographic characteristics is very similar in the lab and the field. 第三，我们拓展了关于领导者选拔机制、高效管理者的行为及技能特质的相关研究（Berger et al. 2020; Erkal, Gangadharan, and Xiao 2022; Englmaier et al. 2024; Judge et al. 2002; Judge, Colbert, and Ilies 2004; Javalagi, Newman, and Li 2024）。我们证明，无论是在实验室协作还是实地运营中，对管理绩效最显著的预测变量不是性格、性别或任何人口特征，而是通过前沿理论构建的“经济决策技能”。这一发现为企业录用带来了重要启示：与其被动依赖自告奋勇的管理职位候选人（极可能陷入过度自信或过度自我推销陷阱的个体），企业不如拓宽候选人基本盘，并引入经济决策等技能测验来主动筛选管理人才。 Third, we extend two related literatures that examine the importance of leader selection mechanisms (Berger et al. 2020; Erkal, Gangadharan, and Xiao 2022; Englmaier et al. 2024), as well as the characteristics and behaviors of good managers (Judge et al. 2002; Judge, Colbert, and Ilies 2004; Javalagi, Newman, and Li 2024). We find that the single best predictor of managerial performance in both the lab and the field is a theoretically grounded measure of economic decision-making skill (Caplin et al. 2024). Managerial effectiveness is not strongly predicted by personality and demographic traits. Notably, we also find that people with a strong preference to be in charge perform marginally worse than managers appointed at random. These findings suggest that proactively engaging a wide set of candidates and screening them based on measures of skill could substantially improve managerial quality. 本文余下部分的结构如下。第二部分详细介绍在实验室中识别好经理的实验设计；第三部分报告实验室实验的主要发现；第四部分呈现两项旨在验证外部有效性的实地研究证据；第五部分探讨优秀经理起作用的核心机制；第六部分对全文进行总结。 The article proceeds as follows. Section II describes our approach to identifying good managers in the lab and Section III reports our experimental findings. Section IV presents two tests of external validity and provides initial quantitative evidence about whether our experimental results are economically meaningful. Section V explores the mechanisms associated with good management in the lab and the field. Section VI concludes. --- ## II. 实验室因果识别优秀的经理 / II. Experimentally Identifying Good Managers ### II.A. 识别策略 / II.A. Identification Strategy #### 1. 记号与模型设定 / 1. Notation and Setup 我们用 $i = 1, ..., n$ 表示个人，并将其随机分配到三个人的小组中，小组用 $g$ 表示。令 $I_{ig}$ 为二元指示变量，如果参与者 $i$ 属于小组 $g$，则该变量为 1，否则为 0。该实验设定了两种角色：经理和普通员工。令 $M_{ig}$ 为二元指示变量，如果参与者 $i$ 是小组 $g$ 的经理，则该变量为 1，否则为 0。类似地，令 $W_{ig}$ 为二元指示变量，表示参与者 $i$ 是否为小组 $g$ 的普通员工。最后，我们定义了一系列代表任务绩效的变量：$X_i$ 衡量受试者在单独工作（即在主实验前进行的“基准个人能力测试”）时的业务解题能力；$G_g$ 表示小组 $g$ 在新型协作团队生产任务中的最终得分。显然，有些小组表现极其优秀，可能仅仅是因为他们随机分到了单兵作战能力极强的员工。为了在估计时剔除这种由于个体业务能力差异带来的干扰，我们建立了一个关于小组 $g$ 最终产出的精简模型： Individuals are indexed by $i = 1, ..., n$ and are randomly assigned to groups of three people, with groups indexed by $g$. Let $I_{ig}$ be a binary indicator equal to one if participant $i$ is in group $g$ and zero otherwise. The experiment contains two roles: manager and worker. Let $M_{ig}$ be a binary indicator equal to one if participant $i$ is the manager for group $g$ and zero otherwise. Similarly, let $W_{ig}$ be a binary indicator of whether participant $i$ is a worker in group $g$. Last, we have a set of variables that describe task performance: $X_i$ measures individual productive capacity when working alone (assessed before the main experiment); $G_g$ denotes the performance of group $g$ on our novel group task. Some groups may perform well simply because they are randomly assigned participants with high levels of productive skill. To control for this, consider a simple model for the output of group $g$: $$G_g = \gamma_M \sum_i X_i M_{ig} + \gamma_W \sum_i X_i W_{ig} + \epsilon_g$$ $$\epsilon_g \sim N(0, \sigma_G^2)$$ 其中，$\sum_i X_i M_{ig}$ 和 $\sum_i X_i W_{ig}$ 分别代表小组 $g$ 中经理与员工的自身基准业务生产技能的总和。 The terms $\sum_i X_i M_{ig}$ and $\sum_i X_i W_{ig}$ measure the productive skill of the manager and the workers in group $g$. #### 2. 估计经理的表现 / 2. Estimating Manager Performance 公式 (1) 中的残差项 $\epsilon_g$ 可以被视为剔除了小组成员自身业务技能优势后、由于“团队协作与管理效率”带来的经调整后的小组净绩效。如果每个参与者仅分派进一个小组，那么我们根本无法断定经调整后的残差变异 $\epsilon_g$ 是源于难以直接测量的个体管理技能（经理效应），还是由于组员之间偶发的人际默契。然而，得益于我们将每个经理随机、多轮次地分配到多个不同小组，我们可以通过求取各轮次内小组净残差的平均值，因果估计出经理 $i$ 的特定固定效应 $\alpha_i$——即在完全控制了小组成员自身业务基础技能的条件下，经理 $i$ 的管理决策行为对团队绩效的平均净贡献： The residuals $\epsilon_g$ in equation (1) can be viewed as a measure of group performance that adjusts for differences in each group’s endowment of productive capacity. If participants were only assigned to one group, it would be impossible to determine whether variation in $\epsilon_g$ arises from unmeasured individual attributes such as management skill, or from idiosyncratic match effects. However, by randomly assigning managers to multiple groups we can estimate $\alpha_i$—defined as the average causal impact manager $i$ has on group performance after controlling for individual differences in productive skill—by averaging the residuals: $$\hat{a}_i = \frac{1}{\sum_g M_{ig}} \sum_g M_{ig} \hat{\epsilon}_g$$ 在我们的框架中，$\hat{a}_i$ 是经理人在控制了其所在小组的普通生产能力禀赋之后，其自身管理质量的净因果贡献估算值。由于我们仅能将每位经理随机分配到四个团队，在微观个体层面上，$\hat{a}_i$ 的估算仍带有一些偶然性。因此，我们的核心是分析小组成效在同一个经理人名下是否呈现显著的系统性关联——换言之，即便随机换了不同的组员，优秀的经理是否依然能持续稳定地让团队产生良好的绩效。为此，我们拟合了一个经典的多水平层级模型： In our framework, $\hat{a}_i$ is an estimate of the average manager’s causal contribution, conditional on each group’s endowment of productive capacity. Because we only randomly assign managers to four teams, $\hat{a}_i$ is somewhat noisy at the individual level. Thus, we focus on whether $\epsilon_{gi}$ are correlated within managers—that is, whether managers have a consistent effect on their teams, after controlling for each group’s endowment of productive skill. To do this, we fit a multilevel model: $$\hat{\epsilon}_{gi} = \alpha_i + e_{gi}$$ $$\alpha_i \sim N(0, \sigma_\alpha^2)$$ $$e_{gi} \sim N(0, \sigma^2)$$ 我们首先估计出了经理固定效应的总体标准差 $\sigma_\alpha$。在层级模型 (3) 中，$\hat{\epsilon}_{gi}$ 是经技能调整后的小组绩效表现，$\alpha_i$ 代表针对个体经理人 $i$ 拟合的随机经理效应，而 $e_{gi}$ 则是不可避免的小组内其余残差扰动。我们在下标中加入了 $i$，以便强调该分析是在经理人这一层面考察管理质量的变异程度。估算值 $\hat{\sigma}_\alpha$ 反映了管理者个人能力对团队产出带来的因果影响弹性，即当分派了一位能力高出一个标准差的经理时，能使整个团队在净绩效上改善多少。我们使用了一个类似的计量框架来独立估计普通员工自身的固定效应。 We start by estimating $\sigma_\alpha$, the standard deviation of the $\alpha_i$ estimates. In model (3), $\hat{\epsilon}_{gi}$ is a vector of skill-adjusted group performance, $\alpha_i$ is a random manager effect for individual $i$, and $e_{gi}$ is residual error. The subscript $i$ is included to indicate that this analysis examines variation at the level of individual managers. $\hat{\sigma}_\alpha$ is our estimate of the typical manager effect, that is, the impact on groups of having a manager who is one standard deviation above average in terms of their managerial performance. We use an analogous framework to estimate worker effects. 在推断和检验上，我们检验了原假设，即在排除和控制了团队自身员工拥有的业务技能优势后，管理者个人的能力对最终团队表现完全没有系统性作用（即原假设 $H_0: \sigma_\alpha = 0$）。根据我们的预注册方案，我们首要通过“随机化推断”来计算其 p 值。为了确保结果的稳健可靠，我们还一并汇报了经典的沃尔德（Wald）检验以及轮廓似然法（Profile Likelihood）的推断结果。 For inference, we test the null hypothesis that managers have no effect on their teams after controlling for each team’s endowment of productive skill (i.e., $H_0: (\sigma_\alpha) = 0$). Our preregistered inferential approach is to calculate p-values using randomization inference. For robustness, we report alternative estimates of uncertainty using a Wald estimator and profile likelihood estimates. --- ### II.B. 实验描述 / II.B. Description of Experiment 我们的实验方案在 AEA 随机对照试验注册库（AEA RCT registry）上进行了完整预注册。图 I 提供了本研究实验流程的整体图解。我们的实验核心围绕一个定制的“协作团队生产任务”展开。该任务旨在通过要求经理监督进度、精确分派工种并实时鼓舞士气，来模拟组织行为学中中层管理者的日常运作。整个实验中，每个充当经理角色的参与者都会被随机重新配对到四个不同的小组中（每组包括该经理和两个新员工），并连续参与四场协作生产任务。 Our experiment was preregistered at the AEA RCT registry. Figure I presents an overview of the experiment. Our design centers on a novel group task called the collaborative production task. The task mimics several aspects of real-world team production by requiring managers to coordinate, monitor, and motivate workers. During the experiment, each manager is randomly assigned to four different teams of three people (one manager, two workers). Every time a manager is allocated to a team, they work on a parallel version of the collaborative production task. ![Figure I / 图 I](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/qje/141/2/10.1093_qje_qjag004/1/m_qjag004fig1.jpeg?Expires=1788379800&Signature=yeafCyV78dpQu98S8JW1zmm1NpJo7hAt3WjjV5DKadEPuyawTvnO46zey3GAgPWk1jneM81jd5lYFJ06gZR8NJ3wT8QAusIDI-Si5fLd1Wpt0-r9o4Au~Abi3qzutnr~aL-jC8rMO-iKMfHYo-H2Grb22U7Cxo0YenVUbhZc3gb0pQEdPJGk0HBNjG4AsFlJ2aZWPlTcbePOg7LajTlqqjwCqWJ71DfvJ9rIQA9GifbWKKMNT-6uWEkL7DRlaXXBurGuzALNQOL8HV8cXic5omy~FGSU36xUK-ZHIfW9ZUfhtnEqFXdXxRMDrHeoigmrXPBpoZDFoTZuK-Wdv6dx~g__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA) **图 I / Figure I: Overview of Experiment / 实验流程总览** 本实验在两个独立维度上嵌入了随机化：第一，我们反复且随机地重新配对小组成员；第二，我们以实验室“场次”为单位，随机决定如何选拔经理人。在总共 39 个实验场次中，19 场被分派到“抽签组”，其中经理完全由电脑摇号随机指定；另外 20 场被分派到“自我推荐组”，其中经理角色将无条件授予那些表达了最强烈管理意愿和领导欲望的受试者。参与者在填写自评意愿表时，并不知道自己那一场次将会执行何种选拔机制。整体的随机化逻辑和流量图详见图 II： The experiment has two levels of randomization. We randomly assign participants to teams (repeatedly) and we randomly vary the way in which managers are selected. Half the managers are chosen via a lottery, and the other half are self-promoted based on their stated willingness to be in charge. The manager selection mechanism is randomized at the level of lab session: 19 sessions had managers selected by lottery and 20 sessions had self-promoted managers. The randomization scheme is summarized in Figure II. ![Figure II / 图 II](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/qje/141/2/10.1093_qje_qjag004/1/m_qjag004fig2.jpeg?Expires=1788379800&Signature=BRIU9maDO83lWFqEW2k3UFdEYSn16VavSFbn3Ctauk4CAsxKZxruH~h-BZXY0-Ki9Dy8L9R0~ZbhqOGPEiIyElZgJ39xQlzoPUBdzddC23~Npp7gOcxN7gMvgsnckhFDErC71hUcfcyXo1fF4XpsTH5Dd9jPUlbtiA636k3IkY91GzgAKTZFe5pxe9srr~NaCJEKiR3jKgTxWbNtS~Tihi8FSE3wv~N~rEuZqKD9W1bCS9KQ5zYC3uY5jDC1rg3QhSwqywv0I-uncg8-sZtrX15v1STP-B3hUAaKdmlr7ycgqXgrEpzoYMMGXM6Q1bOCguJMcWY0Z1rgOAY7qPUezg__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA) **图 II / Figure II: Randomization Scheme / 实验随机化分配设计** 实验整体由两个主要阶段构成：初始个人技能测评和团队协作测试。在第一阶段，参与者需要首先独立在电脑上完成多项广泛心理与认知测试，以此来前瞻性地探测可能对未来管理表现具有显著预测力、但在传统简历中极为隐蔽的核心素质，包括：流体智力、眼部情绪识别测试（RMET，用以精确客观测量其对他人微妙情绪与人际氛围变化的察觉能力）以及极具创新性的“经济决策测验”（Assignment Game，分配博弈测验）。团队协作测试则在英国埃塞克斯大学专门的经济学实验室开展。每个场次包含 12、15 或 18 名受试者，持续运行约两个小时。 The experiment had two main elements: initial individual testing and group testing. The individual phase measured potential predictors of manager skill, including fluid intelligence (Ravens), emotional perceptiveness (RMET), economic decision-making skills, and participants’ preference for being a manager. We also assessed participants’ individual productive capacity by asking people to solve analogous problems on their own. Group assessments were conducted at the Essex lab in the United Kingdom. Sessions involved 12, 15, or 18 participants and lasted around two hours. Every session was randomly assigned to one of two treatments governing how managers were selected (lottery; self-promotion). 在正式踏入团队任务前，每位参与者在阅读完详细的“经理核心职责、权限和可能面临的管理冲突”介绍后，会被问到：“你想在多大程度上成为本次任务的经理？”并在 1 至 10 的刻度中打分（其中 1 表示“完全不希望成为经理”，10 表示“极度强烈渴望担任经理”，平均填写时长长达一分多钟）。在自我推荐的干预分支中，这套得分就是任命决策的唯一金标准：意愿得分最高的那 1/3 的受试者会被自动任命为该场的经理人，并在随后的全部四轮团队任务中代表其团队行使指挥和协调职责。 Before group testing began, all participants rated their preference for being a manager on a scale of 1–10. Before stating their preference, participants were given detailed information about the role of the manager. In the self-promotion condition, the role of manager was assigned to participants with the strongest preference for being in charge. Participants did not know how managers would be appointed when their managerial preferences were elicited. --- ### II.C. 协作团队生产任务 / II.C. Group Task 协作团队生产任务是我们专为该研究原创开发的新颖任务。在这项高度互动的挑战中，三个人的团队（包含一个经理，两个员工）必须协力解答三大不同类型的客观学术智力测试：数学逻辑计算（numerical）、空间旋转想象（spatial）和演绎推理分析（analytical）。团队中每个成员在各自电脑屏幕前都会对应看到题目，并可以独立作答。小组的积分规则是：团队答对一道题加 1 分，但若答错会扣 0.5 分。最关键的一点在于，经理有绝对权力决定谁当前解答哪一个类别的任务。 The group task was created for this experiment. In the task, groups are required to work on three question modules: numerical, spatial, and analytical reasoning. The group receives a score for each module based on how many problems they have solved. They receive one point for correct answers and lose 0.5 points for incorrect answers. Each person in the team, including the manager, works on their own computer trying to solve a different problem. Crucially, the manager decides who will be working on each module. 我们特别将团队的合并最终成绩定义为：在数学、空间和分析这三个版块中所得的**最低分数（Minimum Module Score）**。这构成了博弈论中极为经典的**最弱环节协调博弈（weakest-link coordination game。在博弈论中，“最弱环节”博弈指一个团队或系统的整体产出完全取决于其中表现最差、效率最低的那个“短板”。例如链条的强度取决于最薄弱的一环，因此这种设计能够极佳地模拟需要严密协调、排查生产瓶颈的团队生产环境，使管理者的监督和协调角色变得至关重要）**。这种较为严苛的规则使得团队的绝对成败完全系于那块“短板”，需要极高的全局协作。 The team’s overall score is given by the minimum module score, similar to the weakest-link coordination game where success hinges on effective collaboration. 在埃塞克斯实验室的现场排布中，同组成员会排坐在一起：经理坐在中间位置，左右两侧各坐一个员工。为了排除不同小组间的干扰或抄袭，每个团队的电脑隔间都被高耸的物理隔板挡住。在开始任务前，所有人被明确告知：实验中鼓励并允许随意交流和交谈。因此，绝大多数团队在整场测试中都在不间断地讨论。整场任务持续大约 15 分钟，包含前面大量的操作规则教学与练习、随后 8 分钟的限时正赛以及中间两个长达 1 分钟的“暂停期”（供经理复盘、安抚组员、重构大局）。 Each group of three sat next to each other in the lab, with the manager in the middle and a worker on either side. Before the task began, participants were informed that they were allowed to talk to each other, and teams communicated freely throughout the task. To avoid cross-team communication or interference, each team was separated from other teams by barriers and spare computer terminals. Overall, each group worked together for around 15 minutes. In the first round of the experiment, participants received detailed instructions, including multiple comprehension checks. At the start of the task, participants were given time to introduce themselves and strategize about how they wanted to tackle the task. The timer began after the managers had entered their initial task assignments. At this point the first set of questions were shown. Groups worked on problems for eight minutes in total. This included two break periods of one minute, in which managers had time to regroup and motivate their team and/or restrategize. #### 1. 经理的角色 / 1. The Role of the Manager 我们对该任务的设计核心，就是确保经理人的角色完全独立于普通执行人员，并且对于团队的最终绩效不可或缺。本任务中经理承担了三项各不相同、但同等重要且极具商业实操性的核心中层职责：分工分派（Delegation）、监督进度（Monitoring）和维持士气（Motivation）。 We designed the task so that the manager role was both clearly differentiated from workers and essential for group success. Managers had multiple distinct responsibilities including delegation, monitoring, and motivation. 第一，经理承担了决定**“谁去做什么”**的第一决策权。经理不仅必须对属下的两位普通员工进行分派，同时也必须将自己那一台机器分派到其中的某一个模块中协同作战。任何时候（包括正赛之中），经理都可以通过点击其主控台屏幕，动态且实时地修改和调配大家的工种。为了辅助经理科学决策，在正赛前和整场测试中，经理的屏幕边缘都常驻展示着两个组员（包括自身）在第一阶段单独完成数学、空间和推理测试时的详细测验得分，相当于随时能查阅员工的专业特长。 First, managers were responsible for deciding who did what. Managers were allowed to delegate in any way they saw fit, provided everyone (including the managers) was allocated to a module. Managers were able to allocate more than one person to any given module. Allocations were dynamic, and managers could change their allocation at any time. Before the timer began, managers were presented with full information about each team member’s individual scores on the assessments of numerical, spatial, and analytical tests. Managers could review information about their teammates’ skill profile throughout the task. Managers also had constant access to a table describing the current allocation of tasks among their team. 第二，经理需要**“时刻紧盯团队进展”**。由于博弈遵循的是最低项得分决定总分，这一职责比任何环节都更为关键。在实验设计中，只有坐在中间的经理才能在其主控台实时查看团队当前三个版块的具体得分（例如，看到“数学 15 分，空间 2 分，分析 12 分”），但电脑屏幕不会告诉经理哪一个类别目前是最低的，或者现在具体是谁遇到了答题瓶颈。两翼的员工也只能在他们自己的显示屏上看到他们当前解答版块的分数。由于经理本人也有答题任务，有些不称职的经理很容易“沉浸在自己那一亩三分地里算数”，忽视了去和两侧的组员搭话交谈。要想获得良好成绩，优秀的经理必须不断向员工发出质问和确认，以统筹全局，随时把擅长相关模块的人调遣到表现最差的那个“瓶颈”上进行支援。 Second, managers monitored progress. Monitoring was especially important given the weakest-link scoring rule. Only the manager’s computer terminal showed the overall team score (i.e., the minimum module score). However, managers were not told which module had the lowest score. In contrast, workers are able to see the score of the module they are currently working on. Consequently, managers needed to talk with their teammates to find out which modules needed attention. Effective management required clear communication along with strong situational awareness as managers could, and often did, get swept up working on the module they assigned to themselves. 第三，经理需要**“时刻维持员工的士气与工作热情”**。考虑到这三类客观智力测试既单调枯燥、又极度消耗注意力，让组员始终不松劲显得极其不易。更为关键的是，由于后面详述的差异化激励结构，普通员工在团队协作测试中仅拿固定工资，其任何形式的多答题、勤思考完全没有财务层面的利益回馈。这就断绝了经理使用“金钱物质激励”的可能性，完全仰赖于经理本人的感召、施压或安抚等各种形式的言语激励和团队气氛掌控来保持成员的专注力。 Last, managers motivated their team—an important role given that the tasks are both repetitive and cognitively demanding. As noted in Section II.E, the incentive structure meant that managers could not rely on financial incentives to motivate workers. 这些核心的管理职能（分配、排查短板和组织协调），契合了现实中绝大数企业中层经理（Middle Managers）的日常职责，这部分经理在既往学术文献中较少被系统解剖。 These responsibilities are common managerial duties, particularly among middle managers. We deliberately focus on middle managers—distinct from senior leaders and CEOs—because this group is relatively neglected in the literature. --- ### II.D. 个人任务 / II.D. Individual Tasks #### 1. 衡量个人业务生产力 / 1. Measuring Individual Productivity 在团队协作测试开始前数天，所有参与者必须先独立参加这三类学术测试的限时单兵作答。这三大模块的设计特意使其各自得分在统计学上保持极低的相关系数（在 0.16 到 0.19 之间，这意味着几乎没有全能型天才，每个人都各有所长），这让经理后续面对特长各异的组员时，其分工和决策更具有分量和变数。单兵测评采用了与正赛一模一样的扣分机制以防猜题。 Before group testing began, we assessed individuals’ ability to solve problems on their own. The group task involved three types of problems (numerical, spatial, and analytical reasoning), so participants were asked to complete individual assessments in these domains beforehand. For each of the three tests, participants were given four minutes to solve as many problems as possible. They received 1 point for correct answers and lost 0.5 points for incorrect answers. Participants were aware of this scoring rule. We made this design choice to discourage guessing. #### 2. 广泛的个人技能测试 / 2. Broad Measures of Individual Skill 我们对三大底层潜在管理技能进行了深度测评。为了排除实验当天身体状态或疲劳程度带来的内生干扰，这三大测试是在前几天以线上形式由参与者自行完成后，再来埃塞克斯实验室参会： We measured three broad skills that might predict managerial performance: economic decision-making, fluid intelligence, and emotional perceptiveness. Importantly, we asked participants to complete these assessments well before they came to the lab for group testing. * **Economic decision-making (经济决策技能)**：我们采用了 Caplin 等人于 2024 年提出并建立的“分配博弈”（Assignment Game）测验。该博弈要求参与者在高注意力负荷环境里，直观并战略性地在面临瓶颈的各个选项间分配注意力这一极其宝贵且有限的稀缺资源。近期研究显示，分配博弈的得分极强地预测了美国和丹麦等国全职白领的终身劳动收入。 We measured economic decision-making skill using the assignment game (Caplin et al. 2024). The game requires participants to deal with an attentionally demanding numerical environment, understand comparative advantage intuitively, and avoid biases that undermine numerical decision-making. * **Fluid intelligence (流体智力)**：使用公认的瑞文高级渐进矩阵（Ravens Advanced Progressive Matrices）测试，评估在缺乏经验线索时的抽象逻辑推理与快速解决新颖客观问题的硬性认知能力。 Fluid intelligence was measured with a set of Ravens Advanced Progressive Matrices, a widely used assessment of the ability to solve novel problems. * **Emotional perceptiveness (社交情绪感知)**：通过眼部情绪识别测试（RMET，Baron-Cohen et al. 2001）进行测定，通过辨识眼周微表情对被测试者的认知同理心和社交感知洞察力给出度量。 We measured social skill using the RMET (Baron-Cohen et al. 2001). The RMET assesses emotional perceptiveness by presenting participants with photos of faces, cropped so that only the eyes are visible. For each set of eyes, participants are asked to choose which emotion, from four options, best describes the emotion being expressed. #### 3. 自评性格特质与工作风格 / 3. Self-Reported Measures of Personality and Working Styles 参与者还填写了 10 道题目构成的简易 Big 5（大五性格）特质自评表、缩短版的政治技能量表（Political Skill Inventory, 简称 PSI，旨在科学衡量个体是否具有能够精准探测他人心理、拉拢、说服和影响他人的社交手段，详见 Ferris et al. 2005）以及自评风险偏好刻度得分。 Participants completed the 10-item version of the Big 5 personality inventory (Gosling, Rentfrow, and Swann 2003). Big 5 personality traits typically have positive correlations with job performance and with performance in laboratory studies of small-group problem solving. Participants also completed the shortened Political Skill Inventory (PSI) described in Ferris et al. (2005). The PSI measures “the ability to effectively understand others at work, and to use such knowledge to influence others to act in ways that enhance one’s personal and/or organizational objectives”. Finally, we measured risk appetite, based on a single question asking about risk preferences. --- ### II.E. 招募、样本与激励 / II.E. Recruitment, Sample, and Incentives 我们从英国埃塞克斯大学经济学实验室的受试者数据库中招募了 555 名受试者，总共构建了 728 个小组场次样本。受试者的核心基本特征见表 I。我们发现，两组在所有人口学和客观技能得分上均通过了均值平衡性检验（Balance Test，在随机实验中，通过检验干预组和对照组在实验开始前的各项基准特征是否存在统计学上的显著差异，用以确认随机化分配是否成功。表 I 第 (4) 列显示所有 p 值均不显著，确认了平衡性良好）。 We recruited participants from the Essex University Economics Lab sample pool. Table I, column (1) reports descriptive statistics. Our lab sample comprises 555 individuals, forming a total of 728 groups. Forty-six percent of the sample was female, with an average age of 25. The sample was ethnically diverse with a majority of participants identifying as Asian (including South Asian) or Asian British. Columns (2) and (3) report sample statistics for the treatment arms (self-promoted and lottery). Column (4) presents the results of balance tests across the two arms. None of the characteristics have mean differences that are significantly different from zero at the 5% level. #### TABLE I / 表 I: Sample and Balance / 样本特征与基准平衡性 | 特征类别与变量 / Characteristics | 全样本 / Overall sample (1) | 自我推荐组 / Self-promoted arm (2) | 抽签组 / Lottery arm (3) | p值 / p-value (4) | | :--- | :---: | :---: | :---: | :---: | | **人口统计学 / Demographics** | | | | | | 女性 (%) / Female (%) | 46.3 | 49.1 | 43.3 | .17 | | 年龄均值 / Age mean (years) | 25.0 | 25.1 | 24.9 | .37 | | 工作经验均值 / Work experience mean (years) | 2.5 | 2.5 | 2.5 | .98 | | 亚洲或英籍亚裔 (%) / Asian or Asian British (%) | 53.9 | 52.6 | 55.3 | .54 | | 白人 (%) / White (%) | 18.3 | 17.8 | 18.8 | .76 | | 黑色人种或非洲裔 (%) / Black or African (%) | 15.5 | 17.0 | 13.9 | .32 | | 其他族裔 (%) / Other ethnic identity (%) | 12.3 | 12.6 | 12.0 | .10 | | 研究生比例 (%) / Graduate students (%) | 67.6 | 70.0 | 65.0 | .22 | | **客观技能测评 / Skill assessments** | | | | | | 业务解题技能 / Task skills | 0.00 | -0.01 | 0.01 | .86 | | 流体智力 (Ravens) / Fluid IQ | 0.00 | -0.01 | 0.01 | .81 | | 经济决策技能 (AG) / Economic decision-making | 0.00 | -0.04 | 0.05 | .30 | | 社交情绪感知 (RMET) / Emotional perceptiveness | 0.00 | -0.07 | 0.08 | .08 | | **自评性格特质 / Personality and working styles** | | | | | | 外倾性 (Big5) / Extraversion | 0.00 | -0.01 | 0.01 | .90 | | 开放性 (Big5) / Openness | 0.00 | -0.05 | 0.05 | .25 | | 宜人性 (Big5) / Agreeableness | 0.00 | -0.01 | 0.01 | .77 | | 神经质 (Big5，反向评分) / Neuroticism | 0.00 | 0.05 | -0.05 | .30 | | 尽责性 (Big5) / Conscientiousness | 0.00 | -0.01 | 0.01 | .86 | | 政治技能得分 / Political Skill Inventory (PSI) | 0.00 | -0.05 | 0.05 | .30 | | 犹豫不决指数 / Indecisiveness Index (II) | 0.00 | 0.01 | -0.01 | .82 | | 风险偏好 / Risk appetite | 0.00 | 0.03 | -0.03 | .49 | | **总样本量 / Count** | **555** | **287** | **268** | — | *注：客观技能和性格特质得分已进行了标准化（均值为 0，标准差为 1）处理。* 完成本研究的参与者平均能获得 35 英镑的报酬。我们对经理和普通员工实施了差异化的激励结构：普通员工享有固定的 33 英镑出场费，在协作任务中的产出完全不与其最终所得关联；而经理人除 25 英镑的基本报酬外，还在团队最终得分中抽取高达 £4–£12 的团队绩效奖金（排名前 40% 的高绩效经理获得 12 英镑，后 40% 的仅拿 4 英镑）。之所以如此设计，主要是为了模拟现实中企业往往对管理层提供高弹性的绩效绑定、而对基层员工提供基于固定薪酬的市场常态。 Participants who completed the study were paid £35 on average, with a minimum payment of £29 and a maximum of £41. The individual tasks were incentivized with a bonus of £0–£4. Managers received a flat payment of £25 at the end of the group experiment, plus a team bonus of £4–£12 that depended on team performance in one randomly selected round. Workers did not have performance incentives for the group tasks and were paid a fixed rate of £33 for the group session. Workers received the same average payment as managers. --- ### II.F. 谁想成为经理？ / II.F. Who Wants to Be a Manager? 在正赛开始前，我们在两组中都采用一致的指导语询问受试者成为经理的意向。从图 III 的意向直方图可以看出，总体意愿分布大致均匀，但在“自我推荐组”中，最终当选为经理的人其自评领导意愿高度向右偏，几乎有一半的人直接打了 10 分。 We elicited preferences for being a manager in both arms of the experiment. Figure III presents the distribution of manager preferences among managers for both arms of the experiment. The left panel shows the distribution in the lottery arm. The distribution is fairly uniform, suggesting that neither role was dominantly desirable (mean response = 6, standard deviation = 3). The right panel shows the distribution of preferences among managers in the self-promotion arm. By design, managers in this arm strongly prefer to be in charge, with almost half the managers responding with a 10 on the 1–10 scale. ![Figure III / 图 III](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/qje/141/2/10.1093_qje_qjag004/1/m_qjag004fig3.jpeg?Expires=1788379800&Signature=zg5aOpMXbxVqkVfVf4-Qv366lsuSdVcy57ObS4jECrbwOUXD3ndNwGctJIKl1N~sp-sJCLRbL2ihRxw9vLppK2migILx9~w6R~IDldMEn3mE89GaEzCCCI3~LBPa0YRThOG3bUHaFZkEJ3dTYX1pPsDSDNFaUGRijxz2wC8-RS4WFYqqWq9tmzd-DQei7Zc12Ipe1o~S8vE2NfpuwQHYVXAGw8KwVmEUiXmRCG7zcJpPt8JpL3tYzkFPaxTAbHFSy8lUmSjboF2AZv~LyVVR3ZCdoiOcmlJET4Y57soCXI7tp5pqMYh-eBePary08KCgBucq5Qs6lYQzNnEvyjJPQQ__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA) **图 III / Figure III: Histogram of Preference to be a Manager, by Treatment Arm / 两组当选经理后的管理意愿直方图对比** 表 II 报告了什么样的人具有更强烈的管理偏好和领导意愿。结果表明：那些自评外倾度极高、风险偏好较高、以及男性受试者，在统计学上呈现出较强的管理意向。特别的是，女性并不会因为业务技能强而更想领导，男性业务能力越强则管理偏好越显著。 Table II explores the associations between various individual characteristics and wanting to be a manager. Column (1) shows coefficients from a model where willingness to manage (on a scale from 1 to 10) is regressed on a full set of individual characteristics. The three variables that are most strongly correlated with wanting to be in charge are extraversion, risk appetite, and being male. Columns (2) and (3) present regression results separately for men and women. The relationship between high extraversion and wanting to be a manager is driven largely by men. #### TABLE II / 表 II: Associations with Willingness to Manage / 受试者管理意向的关联因素回归 | 自变量类别与名字 / Predictor Variables | 全样本 / Full sample (1) | 牛人 / Men (2) | 织女 / Women (3) | | :--- | :---: | :---: | :---: | | **人口学基础特征 / Demographics** | | | | | 女性 / Female | -0.46* | — | — | | 年龄 / Age | 0.00 | 0.01 | -0.04 | | 研究生身份 / Graduate student | 0.26 | 0.61 | -0.02 | | 工作经验年限 / Work experience | 0.02 | 0.08 | -0.04 | | **客观技能表现 / Skills** | | | | | 业务解题技能 / Production skills | 0.27* | 0.43** | 0.03 | | 经济决策技能 (AG) / Economic decision-making | 0.21 | 0.60*** | -0.18 | | 流体智力 (Ravens) / Fluid IQ | 0.23 | 0.04 | 0.42* | | 社交情绪感知 (RMET) / Emotional perceptiveness | -0.13 | -0.02 | -0.17 | | **自评性格特质 / Personality & Styles** | | | | | 外倾性 / Extraversion (Big5) | 0.49*** | 0.78*** | 0.22 | | 开放性 / Openness (Big5) | 0.23 | 0.46* | 0.15 | | 宜人性 / Agreeableness (Big5) | -0.24 | -0.50** | -0.01 | | 神经质 (反向评分) / Emotional stability (Big5) | 0.33** | 0.26 | 0.34 | | 尽责性 / Conscientiousness (Big5) | -0.01 | 0.06 | 0.05 | | 政治技能自评 / Political Skill Inventory | 0.09 | 0.06 | 0.20 | | 风险偏好得分 / Risk appetite | 0.42*** | 0.43** | 0.39* | | **总样本数 / N** | **509** | **265** | **244** | *注：回归系数显著性：\* p < 0.1, \*\* p < 0.05, \*\*\* p < 0.01。标准误计算在个体层面进行。* --- ## III. 实验室证据 / III. Experimental Evidence ### III.A. 经理重要吗？ / III.A. Do Managers Matter? 表 III 报告了在排除组员客观技能后，一个标准差的管理变动对最终团队产出所带来的平均净贡献。列 (1) 是我们的基准回归：好经理（质量高一个标准差的经理）能显著且因果地使团队整体绩效提升 **0.221** 个标准差（且极其显著，通过随机化推断所得的 p 值为 0.03）。为了便于横向对比，如果我们像列 (2) 那样在回归时不控制小组成员的基准业务能力，由于有些经理随机分配到了聪明的组员，好经理效应的标准差扩大到 0.284 个标准差。在同一水平上，将组下两名普通员工的业务解题能力提高一个标准差，对团队产出贡献为 0.265 个标准差。这意味着，一个优秀经理对团队生产效率的改善价值，几乎等同于将整队普通员工的总生产素质提升一个标准差。 We estimate large and stable manager effects using our repeated random assignment method. The top row of Table III presents estimates of manager effects ($\hat{\sigma}_\alpha$), accompanied by p-values from randomization inference. The second row of Table III presents analogous estimates for workers. The middle panel of the table presents coefficients on the measures of productivity used to condition group scores in model (1). Column (1) presents our preregistered results. The typical manager effect (i.e., the impact of receiving a manager who is one standard deviation above average) is 0.22 standard deviations ($p = .03$). Comparing the results in columns (1) and (2) allows us to contextualize the importance of overall manager quality with the value of having more productive workers. We find that the overall impact of a one standard deviation better manager is 0.28 standard deviations (column (2)), while increasing the productive skills of the workers by one standard deviation increases team production by 0.27 standard deviations (column (1)), suggesting that manager quality matters about as much to team production as the overall productive capacity of workers. #### TABLE III / 表 III: Magnitude of Manager and Worker Effects / 经理与普通员工对团队净绩效的影响度量 | 回归检验模型 / Models | (1) 基准 | (2) 纯效应 | (3) 方差 | (4) 细颗粒 | (5) 非线性 | (6) 熟人 | (7) 风险 | (8) 抽签组 | (9) 抽签组 | | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | | **经理效应 / Manager effect ($\hat{\sigma}_\alpha$)** | **0.221** | **0.284** | **0.222** | **0.218** | **0.219** | **0.214** | **0.215** | **0.179** | **0.253** | | 随机化推断 p值 / RI p-value | .03 | <.01 | .04 | .05 | .04 | .05 | .04 | .194 | .05 | | **员工效应 / Worker effect ($\hat{\sigma}_\Omega$)** | **0.036** | **0.205** | **0** | **0** | **0** | **0.074** | **0.057** | **0** | **0** | | 随机化推断 p值 / RI p-value | .49 | .04 | — | — | — | .48 | .48 | — | — | | **控制变量 / Controls** | | | | | | | | | | | 经理自身业务技能控制 | 控制 | — | 控制 | 细颗粒 | 非线性 | 控制 | 控制 | 控制 | — | | 普通员工总业务技能控制 | 控制 | — | 控制 | 细颗粒 | 非线性 | 控制 | 控制 | 控制 | — | | **其他细分控制项** | | | | | | | | | | | 团队技能方差控制 | — | — | 是 | — | — | — | — | — | — | | 经理是否认识组员控制 | — | — | — | — | — | 是 | — | — | — | | 经理风险偏好控制 | — | — | — | — | — | — | 是 | — | — | | **样本类别 / Manager sample** | **全部** | **全部** | **全部** | **全部** | **全部** | **全部** | **全部** | **抽签** | **抽签** | | **总小组样本数 / Groups** | **728** | **728** | **700** | **700** | **700** | **700** | **700** | **360** | **360** | 列 (3) 到 (9) 展示了稳健性检验。在控制了小组成员技能在不同模块上的细颗粒分布差异、加入非线性非参数拟合、控制实验中偶遇熟人的影响、控制风险偏好后，经理固定效应依然显著和稳健。我们还执行了“留一法交叉验证”（Leave-one-out cross-validation，即每次排除一个经理某一轮次的数据，仅凭其余三轮的成绩来拟合该经理的固定效应，以此预测该经理人在未参与拟合的那个轮次中的净成效），结果在统计学上呈显著预测，表明实验室中经理固定效应存在系统性关联。 Columns (3)–(9) present robustness checks, demonstrating that our results are unlikely to be driven by specifics of the conditioning model. Estimated manager effects are robust to all these controls. Following our pre-analysis plan, we also test robustness using a leave-one-out procedure. We find that manager contributions from three rounds of data predict team performance in the holdout round ($p < .01$). Overall, our leave-one-out analysis suggests that the manager effects we are estimating robustly predict performance. --- ### III.B. 自我推荐的经理表现差于随机抽签的经理 / III.B. Self-Promoters Perform Worse than Lottery Managers 接下来，我们检验那些表现出较强管理意向、并在“自我推荐组”中当选的管理者的表现。自我推荐上位的经理，其带队的平均得分显著不如那些随机抽签任命的经理。如表 IV 列 (1) 所示，自我推荐经理带队的总产出甚至低了 **0.128** 个标准差（在统计上显著）。即便在列 (2)-(9) 中增加各种组员流体智力、全员技能乃至经理自身特征的控制，这一负面影响也未消除，稳定在 -0.10 个标准差左右。 Next we examine the performance of participants who express a strong desire to be a manager. Table IV contrasts the performance of groups managed by a lottery manager with those of a self-promoted manager. We regress group performance on a binary indicator being in the self-promotion arm. Column (1) shows that teams led by self-promoted managers do worse on average than teams led by lottery managers. The estimate is noisy (0.13 standard deviations, $p = .09$), but we can reject that self-promoters do substantially better. Columns (2)–(8) add a series of controls for group characteristics. Throughout these specifications, the magnitude of the difference between self-promoted and lottery managers is around −0.10 standard deviation. This magnitude is meaningful: on average, groups with self-promoted managers perform about as poorly as groups with mean fluid intelligence almost one standard deviation below average. #### TABLE IV / 表 IV: Difference in Performance (Lottery vs Self-Promoted Managers) / 抽签产生经理与自我推荐当选经理带队表现差异回归 | 解释变量 / Predictor Variables | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | | **自我推荐组 (相对于抽签组) / Self-promoted** | **-0.128*** | **-0.127*** | **-0.119*** | **-0.115*** | **-0.105** | **-0.100** | **-0.096** | **-0.098** | **-0.113** | | (回归系数标准误 / Std. Error) | (0.074) | (0.070) | (0.069) | (0.069) | (0.069) | (0.070) | (0.070) | (0.071) | (0.074) | | 小组总解题能力 / Production skills | — | 0.189 | 0.146 | 0.142 | 0.138 | 0.138 | 0.138 | 0.140 | 0.134 | | 成员平均流体智力 (Ravens) | — | — | 0.104 | 0.098 | 0.094 | 0.104 | 0.104 | 0.103 | 0.101 | | 成员平均经济决策技能 (AG) | — | — | — | 0.019 | 0.011 | 0.010 | 0.010 | 0.009 | 0.008 | | 成员平均情绪感知 (RMET) | — | — | — | — | 0.028 | 0.016 | 0.019 | 0.023 | 0.018 | | 经理风险偏好 / Manager Risk | — | — | — | — | — | -0.003 | -0.001 | 0.001 | 0.021 | | 组内熟人熟知度控制 | — | — | — | — | — | — | 0.076 | 0.073 | 0.090 | | 教育与经验均值控制 | — | — | — | — | — | — | — | 控制 | 控制 | | 性格与政治敏感度自评控制 | — | — | — | — | — | — | — | — | 控制 | | **观测样本数 / Obs** | **728** | **728** | **728** | **728** | **728** | **700** | **700** | **700** | **700** | *注：显著性：\* p < 0.1, \*\* p < 0.05, \*\*\* p < 0.01。* 为什么强烈想当经理的人表现得更差？探索性分析指出：过度自信可能扮演了关键角色。我们发现：那些渴望成为经理的人，其在赛后对本团队整体产出的自评表现高估了 0.41 个标准差，高出抽签经理，且呈现出一定的性别差异（男性展现出更明显的过度自信）。更深层的原因在于，这些主动自我推荐的经理在“人际交际技能”上倾向于自我膨胀：自评情商出众的受试者，在客观的情绪辨识测试（RMET）中实际得分不高，二者呈 -0.37 的负相关。过度自信阻碍了他们客观评估其组员的实际工作进度和情感状态，导致协调效率下降。 Why might people who prefer to be managers fail to outperform people who were randomly chosen, some of whom did not want the job? We hypothesize that self-promoted managers are overconfident, especially about their social skills. People who want to be in charge were much more overconfident than people who do not have strong preferences for being a manager ($d = 0.41$ standard deviations, $p < .01$). These results align with field studies on the overconfidence of managers and executives (e.g., Malmendier and Tate 2015; Huffman, Raymond, and Shvets 2022). Second, we find that self-promoted managers are particularly overconfident about their social skills. Among self-promoters we find a strong negative correlation between self-reported people skills and performance on the RMET (correlation = −0.37, $p < .001$). This relationship was not significant for lottery managers. Overall, we find suggestive evidence that self-promoted managers are overconfident about their performance in general and about their social skills in particular. --- ### III.C. 实验室中，什么特质能预测一个好经理？ / III.C. What Predicts Being a Good Manager in the Lab? 为了厘清能够用于指导经理选拔的底层技能特质，我们着重分析“抽签组”这一不受自我选择偏差污染的样本。表 V 列 (1) 表明唯有两项客观特质能稳健预测其管理绩效：**流体智力（Ravens 测验，相关系数 0.22）**和**经济决策技能（分配博弈，相关系数 0.23）**。相比之下，包括年龄、性别、工作年限、Big 5 性格、自评政治技巧及风险偏好在内的所有软性特征，其与管理贡献的关联度在统计学上均未表现出显著性。而在自我推荐经理的样本（列 2）中，自命不凡的“社牛专家”（政治特长自评，PSI）其客观管理表现甚至呈现出显著的负相关（相关系数 -0.24）。这提示，选拔中层管理者不应过度依赖自吹自擂。 Table V explores the characteristics associated with being a good manager in the experiment. Table V separately reports predictors of management performance for lottery managers (column (1)) and self-promoted managers (column (2)). We focus primarily on the lottery managers, as the relationships between manager performance and manager characteristics in the self-promoted treatment arm are moderated by the filter of self-selection. Only two measures clearly predict manager performance in the lottery arm: fluid intelligence (Ravens) and economic decision-making (assignment game). No other characteristics strongly predict managerial performance. Among self-promoted managers we find negative correlations between management performance and both self-reported extraversion ($\rho = -0.23$, $p < .05$) and self-reported political skill ($\rho = -0.24$, $p < .05$). In other words, when managers are selected by self-promotion, the managers who think they are a “people person” are less successful in the job. #### TABLE V / 表 V: Correlates of Management Performance / 个人基准特质与客观管理绩效的相关系数 | 个人特征分类与名字 / Characteristics | 抽签组经理 / Lottery Managers (1) | 自我推荐组经理 / Self-promoted (2) | 全样本经理 / Full sample (3) | | :--- | :---: | :---: | :---: | | **客观核心技能 / Skills** | | | | | 流体智力 (Ravens) / Fluid IQ | 0.22 (.04) | 0.29 (<.01) | 0.25 (<.01) | | 经济决策技能 (AG) / Economic decision | 0.23 (.03) | 0.10 (.34) | 0.16 (.03) | | 情绪辨识 (RMET) / Emotional perceptiveness | -0.02 (.87) | 0.19 (.07) | 0.09 (.21) | | **性格特质自评 / Personality & Styles** | | | | | 外倾性 / Extraversion (Big5) | -0.07 (.51) | -0.23 (.03) | -0.14 (.06) | | 开放性 / Openness (Big5) | -0.19 (.08) | -0.04 (.73) | -0.12 (.13) | | 宜人性 / Agreeableness (Big5) | -0.18 (.10) | 0.04 (.70) | -0.05 (.52) | | 神经质 (易波动) / Neuroticism (Big5) | 0.10 (.37) | 0.03 (.76) | 0.05 (.50) | | 尽责性 / Conscientiousness (Big5) | 0.00 (.99) | -0.02 (.88) | -0.02 (.84) | | 政治交际技能 (PSI) / Political savvy | -0.04 (.72) | -0.24 (.02) | -0.14 (.06) | | 风险偏好 / Risk appetite | 0.07 (.53) | -0.12 (.26) | -0.03 (.69) | | **人口学基础特征 / Demographics** | | | | | 年龄 / Age | 0.00 (.99) | 0.15 (.16) | 0.07 (.35) | | 女性 / Female | -0.10 (.38) | -0.10 (.33) | -0.09 (.25) | | 工作年限 / Years of experience | -0.07 (.54) | 0.07 (.54) | 0.00 (.99) | *注：括号内为相关系数的双尾显著性 p 值。* --- ### III.D. 量化实验室中不同经理选拔机制的影响 / III.D. Quantifying the Impact of Manager Selection Mechanisms in the Lab 我们通过模拟进一步对比了在不同选拔情景下最终当选经理在实验室中的客观管理成效。如图 IV 所示，从自我推荐机制（Self-promotion）转型为**基于经济决策技能（AG）**选任经理，最终上线的经理人在客观管理水平上将有 **0.7 个标准差** 的显著提升。其带队绩效显著优于纯随机任命的“摇号机制”（Lottery），也优于单纯提拔业务骨干的“彼得原理”机制（Peter principle，即提拔个人业务技能最高的人）。 We quantify the benefits of skill-based promotions by comparing the effect of different selection mechanisms on average manager contributions. In addition to comparing self-promoted and lottery managers, we use data from the lottery arm to simulate counterfactuals in which managers are selected on specific skills. Figure IV compares the results of six selection mechanisms. We examine self-promotion, lottery, and choosing managers based on four skills: fluid IQ (Ravens), economic decision-making (AG), emotional perceptiveness (RMET), and individual production skills (i.e., the Peter principle). Different selection mechanisms have large effects on manager quality and group performance. Compared with a regime of self-promotion, selecting managers based on economic decision-making skills yields managers who are 0.7 standard deviations better in terms of their manager effects. ![Figure IV / 图 IV](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/qje/141/2/10.1093_qje_qjag004/1/m_qjag004fig4.jpeg?Expires=1788379801&Signature=Ep6hhs-NvfIQLW1N-pTLmBqwCWfefBRRU~NstAzTgvsb88PKrgyswChmurkgek-p8lruGapBOe1f2Swe8DBbv3l9UDh7PPTRodLTR7uGeCHdnuNw-JgB~YV5rf7b3HWdJ-0o2erCt5ML35PEL83jVfzWxBsV~~crAcsi2PR~qn-rE7T-WB3CopcFKen1eRTZYlSjLhg6EavvNyzzBgVYHHKDwLZ5h0i6tmsio6TPvqAL3phCAUQCFr8a9u-kWnbVnW1gT9s4CO-vN9IWT-459fJUWEU3gK7lhdGhj1n5WC0g~7IYkoMXxZ5bi4QXRdii6nHAhcq8VJqpH52pYJb7hg__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA) **图 IV / Figure IV: Comparing Methods of Appointing Managers / 实验室中六种不同经理选任机制下的最终平均管理绩效对比** --- ## IV. 实地验证研究 / IV. Field Evidence ### IV.A. 实验参与者的劳动力市场表现 / IV.A. Labor Market Outcomes of Experimental Participants 我们首先在现实劳动力市场中追踪了实验室参与者。通过领英（LinkedIn）数据，我们获得了 73 名在实验室担任过经理的受试者的公开职业履历。在控制了学历、流体智商和性格特征等传统控制项后，我们发现：在实验室协作中测得的因果管理表现 $\alpha_i$ 每提高一个标准差，受试者在现实世界中的平均晋升频率显著增加 **0.16** 次/年。高管理技能受试者平均每 2.3 年获得一次晋升，而其余普通表现者平均需要 5.3 年。 First we focus on the labor market outcomes of our experimental participants. We hypothesize that managers who succeed in the lab will be more likely to succeed in the labor market. From our original sample of 184 managers, we find publicly available labor market data on LinkedIn for $n = 73$ participants. This LinkedIn sample represents people in the early stages of their career: mean age is 27.5 and participants have an average of 4.9 years of work experience. A one standard deviation increase in management performance $\alpha_i$ is associated with 0.16 more promotions per year ($n = 73$ managers, $p = .001$). Good managers in our experiment (+ 1 standard deviation) are promoted every 2.3 years on average, compared with 5.3 years for other participants ($n = 73$, $p = .03$). --- ### IV.B. 大型零售企业中经理的技能 / IV.B. Managerial Skill in a Large Retail Firm 我们随后在南美一家大型连锁零售巨头开展了实地研究，该企业在单一国家拥有 500 家超市门店。这一研究场景具有典型的三层层级治理架构（总部、中层店长、前线理货员/收银员）。我们招募了其中的 225 名在职中层店长，让他们参加了与实验室中完全相同的客观核心能力测评，并追踪了其在长达 28 个月内的门店月度销售面板数据。 We turn to our field evaluation, which studies managers in a multibillion-dollar retail firm in South America. Our data set comes from one country, in which the firm owns 500 grocery stores. The firm has a three-layered organizational structure: central headquarters, middle managers, and front-line workers. We recruited a sample of 225 managers for whom we could observe work history, including any transitions between stores. Managers completed a set of assessments analogous to the broad measures of individual skill used in the lab experiment. The manager data are linked to a panel of store-level data covering the period February 2020 to May 2022. For each store we observe monthly sales and employment along with complete records about the identity of managers and their movements. #### 1. 双向固定效应估计方法 / 1. Empirical Strategy 为了从由于门店区位优势或当地人流带来的环境干扰中精准剔除出“店长个人管理能力”对月度销售额的净贡献弹性，我们遵循了著名的 **AKM 模型（Abowd-Kramarz-Margolis 模型，这是由阿博德、克拉马兹和马戈利斯于 1999 年提出的一种开创性的计量经济学分析框架，主要应用于劳动经济学中，通过追踪雇员在不同企业（或经理在不同门店）之间的流动，将难以观测的雇员固定效应与雇主固定效应解耦并进行独立估计）**。我们对该连锁店中进行的店长跨店轮换流动轨迹进行了数学关联，建立如下双向固定效应估计模型： To estimate the value added by managers in terms of log monthly sales, we leverage the mobility of managers across stores to estimate manager and store fixed effects, with the following model: $$\log(Sales_{smt}) = \alpha + \theta_m + \psi_{s(m,t)} + \delta_t + \gamma demo_{st} + \epsilon_{smt}$$ 其中，因变量 $\log(Sales_{smt})$ 是门店 $s$ 在店长 $m$ 管理下第 $t$ 个月的总销售额对数值；$\theta_m$ 代表该店长的个人客观管理水平（店长固定效应）；$\psi_s$ 代表门店自身的不可观测永久性质量（门店固定效应，如地段优劣、周边竞争）；$\delta_t$ 是月份时间固定效应（控制普遍的宏观和季节波动）；$demo_{st}$ 则是该门店当期所拥有的前线员工队伍的核心平均基本特征（控制员工变动）。该模型通过追踪同一个店长去往不同门店前后、或同一个门店换了店长前后的业绩净跳变，实现了店长固定效应 $\theta_m$ 与门店固定效应 $\psi_s$ 的解耦。 Where the dependent variable is $\log(Sales_{smt})$, defined as the log of total sales in store $s$ reporting to manager $m$, in month $t$; $\theta_m$ is a fixed effect for the manager, $\psi_s$ refers to a store fixed effect, and $\delta_t$ is a date (monthly) fixed effect that accounts for the seasonality of sales. 我们对店长在不同门店之间流动的潜在选择性偏差进行了对称性与水平性残差检测（即检验上迁与下迁门店时业绩损益是否对称、平移前是否存在趋势变异），较好地支持了该连锁店店长轮换的条件外生独立性（Conditional Mean Independence），控制了选择性匹配所引发的估计偏误。 We performed several tests analyzing manager mobility and its potential correlation with store performance dynamics. We confirm that the moves of managers across stores appear to be conditionally mean independent of the match-specific component, demonstrating that the gains and losses from moving managers across stores are roughly symmetric and origin store trends are flat. #### 2. 实地估计与技能预测结果 / 2. Results 中层管理者在真实的超市场景中到底能带来多大影响？我们对换店长这一节点开展了事件研究（Event Study）。结果显示，在一位高于门店平均历史水平的优秀店长派驻到位后，门店销售额呈现出显著且平稳的跳升，4个月后销售额净增加达 13%。整体来看，**店长管理质量提高一个标准差，对应的门店销售额增加 25%（年单店平均增收达 520 万美元）**。 What is the economic impact of improving manager quality in this field context? To answer this question, we conduct an event study in which the events reflect the arrival of a manager who raises the managerial quality of the store. We find that the arrival of a higher-quality manager results in a large, immediate, and significant increase in sales. After four months, we estimate that an above average manager raises sales by 13%. We estimate that the arrival of a good manager (defined as a manager who is one standard deviation above average) increases sales by 25% which equates to an increase in annual turnover of US$5.2 million. 最引人注目的是：我们在实验室中探明的那套客观决策测验，在实地店长评估中同样起到了决定性预测作用。如表 VI 所示，**经济决策技能（分配博弈得分）每高出一个标准差，店长在实地回归中测得的管理业绩固定效应（$\theta_m$）显著提高 0.192 个标准差（$p < 0.05$）**。而且在加入所有人口和工作背景变量（列 5）后，这一技能指标依然保持显著。 We ask whether the predictors of managerial performance in the lab also predict performance in the field. This provides an important test of external validity, in that we hypothesize similar underlying skills should drive success in both contexts. We find that this is indeed the case. We regress manager fixed effects on the three broad measures of managerial skill we used in the experimental study. The results are presented in Table VI. The best predictor of performance is the measure of economic decision-making: a one standard deviation increase in economic decision-making skill is associated with an increase in manager fixed effects of 0.19 standard deviations. #### TABLE VI / 表 VI: Predictors of Manager Performance in the Field / 哪些技能能预测店长的实地真实管理绩效？ | 解释变量 / Predictor Variables | (1) | (2) | (3) | (4) | (5) | | :--- | :---: | :---: | :---: | :---: | :---: | | **经济决策技能 / Economic decision-making** | **0.192** | — | — | **0.242** | **0.273** | | (标准误 / Standard Error) | (0.069) | — | — | (0.097) | (0.101) | | 流体智力 (Ravens) / Ravens score | — | -0.033 | — | -0.053 | -0.089 | | 情绪感知 (RMET) / Emotional perceptiveness | — | — | -0.152 | -0.184* | -0.180* | | 女性店长 / Female | — | — | — | — | 0.161* | | 年龄 / Age | — | — | — | — | -0.151 | | 工作年限 / Experience | — | — | — | — | 0.047 | | **观测样本店长数 / Obs** | **94** | **94** | **94** | **94** | **94** | | **店长特质可解释度 / $R^2$** | **0.177** | **0.145** | **0.164** | **0.210** | **0.238** | *注：回归在店长层面进行。自变量均标准化为标准差形式。标准误在店长流动的‘联通集’（Connected Set）层面进行聚类调整。\* p < 0.1, \*\* p < 0.05, \*\*\* p < 0.01。* --- ### IV.C. 量化实地选拔机制的影响 / IV.C. Quantifying the Impact of Managerial Selection in the Field 我们随后对连锁零售企业实施技能选拔机制的经济成效进行了情境模拟。如果我们结合实验室测得的精确无偏差提拔绩效弹性和超市实地销售增长数据，可以模拟出：相比目前企业现行的“主观提拔”或“业务骨干提拔”政策，**如果企业在选拔店长时，仅考虑那些在经济决策测验（AG，分配博弈）中得分排在前三分之一的候选人，各门店的平均销售额由于管理效率改善能够提升 16%**。 Last, we quantify the economic impact of skill-based hiring in the field. This is the field analogue of the experimental analysis comparing the effect of different selection mechanisms on typical managerial performance. We combine cleanly identified estimates from our lab experiment with estimates of the value of good managers in terms of store-level sales. Shifting to an approach that only considered candidates from the top third of scorers on the assignment game would increase average monthly sales by 16%. --- ## V. 机制分析 / V. Mechanisms 优秀的中层管理者是通过哪些高超的行为手段，促成其团队在各种干扰和高压任务下取得卓越成功的？我们在实验室和实地零售店两个维度，对监督、分工与激励这三大机制进行了拆解。 ### V.A. 监督机制 / V.A. Monitoring 第一大机制是**监督（Monitoring）**。在三人协作中，由于团队得分取决于最低模块得分。这就导致，在协作结束时，如果有员工仍在解答一个得分已经高出其他板块许多分（高出 10分以上）的模块，他的工作时间就会被浪费。我们将其定义为“监督失误”。在实验室任务中，13%的小组在结束时存在这种明显的无用功。然而，我们发现，**优秀经理的监督失误率（7.1%）仅为普通经理（13.3%）的一半**。管理才能越高的经理，其监督失误越少。 In the group task, monitoring helps avoid situations where workers are wasting their time on tasks that don’t contribute to group performance. We define a monitoring failure by the manager as having any group member working on a module at the end of the collaborative production task that is substantially greater than the minimum module score. Over the course of the experiment, 13% of groups finished the task with at least one person working on a module that was not contributing to group success. Failures in monitoring were strongly related to overall manager contributions (correlation = -0.34, $p < .001$). A manager one standard deviation above average reduced the error rate from 13.3% to 7.1%. In other words, good managers had roughly half the rate of monitoring errors. ### V.B. 根据比较优势分配任务 / V.B. Allocating According to Comparative Advantage 第二大机制是**分工（Delegation）**。根据组员在第一阶段测试中的业务特长，在开场时实现全员处于其自身最具“比较优势”工种上的初始分工，对后续表现至关重要。我们发现，**能够做出合适初始分工的经理，其带队的最终总分比不曾分对的小组高出 0.50 个标准差**。 Next we examine the quality of managers’ allocation decisions. We use information on participants’ individual module test scores to assess whether managers found the optimal assignment. The probability that a manager finds the optimal starting assignment is positively associated with overall manager performance ($\rho = 0.18$, $p = .02$). Groups whose managers always start with the best assignment scored 0.50 standard deviations higher than groups with managers who never start with the best allocation ($p < .01$). ### V.C. 激发员工付出努力 / V.C. Motivating Workers to Exert Effort 第三大机制是**激励与士气掌控（Motivation）**。由于员工没有任何绩效奖金，我们通过拟合员工在某个经理治下与单兵作答时解题效率的系统性因果提升，来分离出该经理在言语激励、团队气氛营造上的净影响力。回归模型设定如下： We interpret the rate at which workers solve puzzles as an indirect measure of motivation. We fit a simple model to help understand variation in each worker’s productivity across managers and modules: $$Y_{ik[m]} = \nu_m + \delta_k + \beta X_{ik} + e_{ik}$$ 其中，$\nu_m$ 代表该经理对团队士气和工作热情的净激励效应，而在自变量中我们精细控制了该员工单兵作答的固有业务能力 $X_{ik}$。由于我们彻底排除了由于员工自己很聪明或经理辅导员工答题（我们检验了经理个人能力对组员解题率没有任何溢出帮助）带来的影响，$\nu_m$ 是相对干净的士气感染力指标。结果显示，**这一士气感染力与该经理人在带队表现中的净因果贡献高度共振（相关系数达 0.39）**。 where $\nu_m$ represents manager fixed effects, $\delta_k$ are domain fixed effects, and $X_{ik}$ represents worker $i$’s estimated solo ability. Overall, we find that motivational fixed effects ($\nu_m$) are an important driver of managerial performance ($\hat{\rho} = 0.39$, $p < .001$). --- ### V.D. 实验室中哪些机制最重要？ / V.D. Which Mechanisms Matter Most in the Experiment? 为了对比这三大机制哪一个更重要，表 VII 将经理客观固定效应同时对这三大机制进行标准化回归。结果表明：**员工士气与工作热情维护（回归系数 0.383）**和**防范员工做无用功的监督机制（回归系数 -0.348）**的作用较为明显，其后是初始任务分配（回归系数 0.136）。 To understand the relative importance of the three channels, Table VII regresses estimated manager effects ($\alpha_i$) against three manager-level parameters. We find that all three mechanisms independently contribute to managerial contributions (column (5)) and that the motivational and monitoring channels are the strongest predictors of overall manager performance. #### TABLE VII / 表 VII: Which Mechanisms Predict Managerial Performance? / 哪一种底层机制对经理带队成绩的贡献度最大？ | 机制自变量（均已标准化）/ Explanatory Variables | (1) | (2) | (3) | (4) | (5) | | :--- | :---: | :---: | :---: | :---: | :---: | | **初始工种分配 / Initial allocation** | **0.204*** | — | — | **0.178*** | **0.136** | | **监督失误率 / Monitoring errors** | — | **-0.335*** | — | **-0.321*** | **-0.348** | | **员工士气激励 / Worker motivation** | — | — | **0.391*** | — | **0.383** | | **回归模型总可解释度 / $R^2$** | **0.042** | **0.113** | **0.154** | **0.145** | **0.292** | *注：标准误与显著性：\* p < 0.1, \*\* p < 0.05, \*\*\* p < 0.01。* --- ### V.E. 实地研究中的机制分析 / V.E. Mechanisms in the Field 在真实的超市经营中，我们通过分析**“缺货率”（Stockout Proportion）**来对上述监督机制在实地零售表现中的作用进行横向验证。缺货属于由于店面监督和库存盘点不及时带来的运营瓶颈。事件研究（图 V）展示：在派驻了一位优秀的超市店长后，**门店因店面漏盘、漏报导致的货架缺货率立刻呈现出显著且持续的缩减，缩减了 0.25 个百分点，相当于在基准缺货率基础上降低了 3.7%**。而且，在分配博弈测试中表现出色、展现出极高决策才能的店长，其带店的货架缺货情况有着显著降低。 We find suggestive evidence in support of the hypothesis that monitoring is an important channel for managerial performance. Specifically, we examine stockouts, a situation in which managers fail to monitor stock levels, leading to products becoming unavailable. We find that good managers reduce the rate of stockouts. We estimate that the arrival of an above average manager reduces the proportion of products out of stock by roughly 0.25 percentage points, or 3.7% of the mean. Finally, we find a strong marginal correlation between assignment-game score and the rate of stockouts in stores, a one standard deviation increase in managers’ assignment-game score being associated with a 0.35 percentage point reduction in stockouts. ![Figure V / 图 V](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/qje/141/2/10.1093_qje_qjag004/1/m_qjag004fig5.jpeg?Expires=1788379801&Signature=IRSwUKyhT6ozxzmYZfPiN1ODhSYRKme3Oydbotd~nN0vZc4f351ZWj4nX5l4yqDLPN3y0MR8vlAJIaZZ8n6CnoJQAnHIT5FY2nbSDPWsuyNLbvbm0ivcx1no8OxLRDJjlnR~CYMZ9uCm4xBXMOG~1AGBiuJwGUTQGMSu~gQ5OLjsukCteDquh5Y1xvk8au5P0Nuuy-nCicxniyvImPoLGpVpSwj9RyPHtOHTuQjlRlveuj7omrvR-B~eWDHYCad8knuamRXYe5Lkx6uJ-I8qjNK6-zUu8I9WOB~9oK6ajRSEQ480FH23Vf7mKyQdxkrcq-HA7yUP~IThV0RBfNWr0Q__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA) **图 V / Figure V: The Impact of Above-Average Managers on Stockouts / 超级店长走走走上任对超市内货架商品缺货率带来的改善事件研究图** --- ## VI. 结论 / VI. Conclusion 本研究基于严密的实验室协作和大型超市的真实长周期面板数据，双向证实：选任高效的中层管理者，对提升组织绩效具有重要意义。并且我们指出，在传统上受到重视的自评性格特质（如 Big 5 外倾性）、极度强烈的权力欲望（自我推荐）不仅不能保证客观管理成效，反而容易使管理层滑入过度自信和漏盘、无视团队瓶颈的深渊。通过引入经济决策测验（AG，分配博弈）等前瞻性客观技能指标来筛选和选拔新经理，能够为企业和组织带来显著且持续的绩效增长。 This article develops a novel, skills-based approach to prospectively identifying good managers. Shifting to an approach that screens candidates based on economic decision-making and other task-relevant skills could substantially improve managerial quality and team performance. Overall, our findings emphasize the importance of good management and suggest that skill measures are better predictors of manager performance than personality traits or preferences to be in a leadership position. Firms that engage the widest possible set of candidates and screen for economic decision-making and other skills could see substantial improvements in managerial quality and team performance.