Hierarchies and decision-making in groups: experimental evidence

The study addresses how different forms of hierarchy affect group decision-making outcomes in economic tasks. While much research analyzes individual decisions, many real-world decisions occur in groups embedded in hierarchies. The paper examines whether hierarchy type influences risk-taking in lotteries and performance on intellective tasks. Hierarchy is implemented as formal authority of a leader over group choices after discussion. Five group decision structures are compared: majority voting (no hierarchy) and hierarchies where leadership is assigned by random draw, election, age (seniority), or merit (financial literacy). Using a student sample, incentivized tasks, and Bayesian and frequentist analyses, the study tests whether hierarchy type changes risk choices and the likelihood of correct answers in intellective tasks. The hypotheses: H1: Groups with leaders assigned by irrelevant characteristics (age or random) perform worse on intellective tasks than groups with leaders assigned by merit. H2: Groups with hierarchy make more safe choices in the lottery task than groups without hierarchy.

Research across management, sociology, and psychology shows mixed effects of hierarchies: they can reduce conflict and enhance coordination, but may deter motivation and innovation. Political economy and psychology compare decision rules (unanimity, majority), but typically do not study lotteries with hierarchical authority. Experimental economics on groups shows heterogeneous results: groups sometimes exhibit stronger or weaker biases than individuals; groups may align more with game-theoretic predictions in strategic settings; risk-taking on behalf of others depends on moderators (framing, recipient identity). Prior economic experiments on group non-strategic risky choice (e.g., Baillon et al., 2016; Baker et al., 2008; Masclet et al., 2009; Rockenbach et al., 2007; Shupp & Williams, 2008) document that groups can choose safer lotteries or accumulate higher expected value, with results depending on task features and decision rules. Studies on violations of expected utility show groups can be as consistent as individuals or less prone to errors. Organizational psychology distinguishes intellective tasks (correct answers) from judgmental tasks (no single correct answer) and posits conditions when hierarchies help: procedural interdependence, perceived legitimacy, and alignment of competence with authority (Halevy et al., 2011). There is little direct evidence on how different hierarchy assignment mechanisms influence performance in economic experiments; Baillon et al. (2016) compare decision rules but not hierarchies. Hence, testing hierarchy types in lottery and intellective tasks is novel. The paper derives H1 (merit-based leaders outperform random/age-based) and H2 (hierarchy increases safe choices).

Design: Within-subject laboratory-style experiment (paper-and-pencil) with three stages and six treatments.

• Stage 1 (Treatment 1): Individual decisions (baseline) — each participant completed a Holt-Laury (2002) lottery task (10 paired choices A vs. B) and three intellective tasks (applications of framing effects, Ellsberg paradox, and basic probability).
• Stage 2 (Treatment 2): Group without hierarchy — groups of three decided by majority vote on the same types of tasks.
• Stage 3 (Treatments 3–6): Group with hierarchy — groups of three decided after 10 minutes of discussion, but the leader had formal authority over the final group decision. Four hierarchy assignment mechanisms were implemented within-subject: random (die roll), election by group, age-based (oldest), and merit-based (highest score on a financial literacy test).

Tasks and incentives:

• Lottery task: 10 choices between safer Lottery A (low variance) and riskier Lottery B (high variance). Payoffs in KRW with probabilities increasing by 10 percentage points per row. Under risk neutrality, switching from A to B is optimal when the high payoff probability exceeds 40%. The task measures risk attitude; multiple switches indicate inconsistency. One row per lottery task was randomly selected for payment by a 10-sided die.
• Intellective tasks: Different items each round, modeled after Curșeu et al. (2013) and Huang & Wang (2010), covering framing, Ellsberg-type ambiguity, and basic probability.

Procedures and sample:

• Location: Yonsei University, Mirae Campus, Wonju, Republic of Korea; no dedicated lab (paper-and-pencil sessions). Instructions in Korean (with back-translation) or English as appropriate; read aloud.
• Recruitment: Undergraduate students, unrestricted by major, age, or gender; first economic experiment at the site; no restrictions on prior participation.
• Sessions: Pretest plus three sessions; total N=99 participants; average session length 103 minutes; average earnings KRW 19,200 plus 5,000 KRW show-up fee; minimum wage context provided. After data cleaning for missing values: N=96 participants forming 32 groups. Each group experienced five group treatments (voting + four hierarchies), yielding 160 group-level observations. A total of 128 participants served as leaders in one or more hierarchy treatments.
• Group size: Three (to ensure majority voting in no-hierarchy condition and comparability with prior studies).
• Payment: Identical within-group for group treatments based on the group/leader’s choice on randomly selected rows; all choices incentivized.

Analysis:

• Bayesian analyses conducted in JASP with default priors (per Rouder et al., van Doorn et al.) and Bayes factors interpreted via Jeffreys/Andraszewicz categories.
• Frequentist analyses conducted in Stata/SE 17: ordered probit/probit regressions; additional non-parametric tests (Friedman, Cochran’s Q; Wilcoxon; McNemar) as applicable. Predicted probabilities and marginal effects computed with SPost/Long & Freese.

Ethics: Approved by Yonsei University IRB (June 26, 2015); informed consent obtained; procedures aligned with the Declaration of Helsinki.

Descriptive (individuals, N=96):

• Mean number of safe choices: 4.8333 (of 10).
• Inconsistent lottery choices: 6.25%.
• Correct intellective answers: 32.29%.

Descriptive (groups, 32 groups × 5 treatments = 160 observations):

• Number of safe choices (mean): voting 5.0000; random 5.0000; vote 5.0938; age 5.1563; merit 4.9688.
• Inconsistent lottery choices (%): voting 3.125; random 3.125; vote 12.5; age 3.125; merit 6.25.
• Correct intellective answers (%): voting 16.125; random 28.125; vote 21.875; age 21.875; merit 53.125 (highest).

Bayesian repeated-measures ANOVA (groups):

• Number of safe choices: BF01=47.719 (very strong evidence for no differences across hierarchy types).
• Inconsistent lottery choice: BF01=4.860 (moderate evidence for no differences across hierarchy types).
• Correct intellective answer: BF01=0.105 (strong evidence for differences). Post-hoc vs. voting: merit vs. voting BF01=0.019 (very strong evidence that merit-based groups perform better); other comparisons showed anecdotal to moderate evidence for no differences.

Regression (group level, baseline=voting):

• Ordered probit (safe choices): No hierarchy type significant.
• Probit (inconsistent choice): No hierarchy type significant.
• Probit (correct intellective answer): merit hierarchy coefficient 1.0884 (SE 0.3478), significant at 1% (positive effect vs. voting). Predicted probability increase vs. voting: +0.389 (p=0.000).

Leaders: within-subject Bayesian comparisons (individual vs. as leader; n per comparison=32):

• Across hierarchy types, leaders showed more safe choices and fewer/equal inconsistencies descriptively, but Bayes factors BF01 ranged 1.354–5.295, indicating anecdotal to moderate evidence for no change in leaders’ behavior relative to their individual baseline.

Determinants of leaders’ choices (n=128 leader-observations; baseline=merit leader):

• Number of safe choices (ordered probit/OLS): hierarchy type not significant; individual risk attitude predicts leader risk choices (each additional safe choice individually → +0.6757 safe choices as leader, p<0.01). Female leaders made fewer safe choices than males (−0.5442, p<0.05 in OLS ME discussion).
• Inconsistent lottery choice (probit): age-based leadership reduced inconsistency by 5.3 percentage points vs. merit (p<0.05 in marginal effects). Korean leaders less likely to be inconsistent (−6.7 pp, p<0.01).
• Correct intellective answer (probit): relative to merit leaders, random (−0.164, p<0.05), elected (−0.186, p<0.01), and age (−0.219, p<0.01) leaders had significantly lower probabilities of correct answers (average marginal effects). Korean leaders had a 26.3 pp lower probability of correct answers than non-Koreans (p<0.01).

Ideal-type predicted probabilities (holding controls at sample values):

• Across Korean/non-Korean women/men, moving from merit to age-based leadership produced the largest decreases in probability of correct intellective answers (about 22.7–33.5 pp decreases; typically statistically significant). No significant changes for inconsistent lottery choices.

Hypotheses:

• H1 confirmed: merit-based leadership yields higher performance on intellective tasks than age- or random-based (and also vs. elected) leadership.
• H2 rejected: no evidence that hierarchies increase the number of safe choices relative to no hierarchy.

The findings indicate that hierarchy type matters for intellective tasks but not for risk-taking in lotteries. Groups with leaders appointed based on demonstrated competence (merit) achieved substantially higher rates of correct answers, consistent with theories that legitimacy and alignment of competence with authority improve group performance. Conversely, hierarchies based on age, random assignment, or election underperformed relative to merit in intellective tasks. Risk preferences and consistency in lottery choices did not vary across hierarchy types, suggesting that hierarchical structure does not shift group risk attitudes in this context. Leader-level analyses show that leaders’ individual risk preferences carry over to group decisions in lotteries, while the mechanism of leader selection strongly predicts performance on intellective tasks. The absence of significant within-person changes when individuals become leaders suggests that the hierarchy effect on intellective performance is not due to leaders changing their behavior per se, but to the selection mechanism aligning leadership with competence. These results align with organizational theories (e.g., Halevy et al.) emphasizing the importance of perceived legitimacy and competence-based status in enhancing coordination and decision quality, and complement mixed prior evidence on group decision-making by isolating hierarchy assignment as a key moderator. Managerially, they suggest that competence-based leadership selection can improve decision quality on intellective problems without materially affecting groups’ risk posture in lottery-like choices.

This study contributes experimental evidence that the mechanism by which leaders are appointed in small groups crucially affects performance on intellective tasks: merit-based leaders yield markedly better outcomes than leaders appointed by age, random assignment, or election, while hierarchy type does not affect groups’ risk-taking or consistency in lottery decisions. The within-subject design, combining Bayesian and frequentist analyses, isolates these effects across multiple hierarchy types. Future research should examine additional decision tasks and outcomes, explore other hierarchy mechanisms and contexts, investigate the observed gender and cultural heterogeneities, and broaden subject pools beyond students to enhance external validity (e.g., managers; varied age ranges; different group sizes).

• External validity: student sample from a single university; culturally specific context (South Korea); narrow age range (19–30; mean ~21).
• Group size fixed at three; real-world teams vary in size and dynamics.
• Simplified, formal hierarchies in the lab may not capture complexities of organizational hierarchies (informal power/status, longer-term interactions).
• Paper-and-pencil setting without a dedicated lab; potential language/translation nuances (Korean/English).
• Intellective tasks limited in scope; results may depend on task domain.