We consider a rank-order tournament model involving a tournament administrator (employer or principal) and two agents competing for the prize (the promotion), and test experimentally the principal's inferences and the agents' responses. Specifically, one agent has an exogenous advantage, which is known and observable to the principal. The principal's incentives are designed such that it's in their best interest to design a promotion rule that levels the playing field ex post. We will test whether i) the principals promote optimally and if not, ii) agents anticipate this distortion and adapt effort accordingly. This pre-registration extends pre-registration #14254.
Intervention
The interventions in the HUMAN condition, in which principals are played by a subject and winning rule determined ex post (no commitment): 1) In Baseline (SYM), the agents are fully symmetric. 2) In Treatment (ASYM), the agents are asymmetric.
The interventions in the COMPUTER conditio, in which the winning rule is fixed ex ante and known to agents (commitment). Note: SYM: Symmetric Agents, ASYM: Asymmetric agents. 1) EO SYM: Symmetric agents, winning rule is set to 0, meaning agent with higher work output wins. Equal opportunities. 2) EO ASYM : Asymmetric agents, winning rule is set to 16, meaning agent with higher performance, but not necessarily work output wins. Equal opportunities, affirmative action implemented. 2) UO ASYM : Asymmetric agents, winning rule is set to 0, meaning agent with higher work output wins. Unequal opportunities since agents are asymmetric, so no affirmative action implemented.
Experimental design
See below.
Design details
Randomization method
First, participants randomly draw a card that assigns them to their computer / cubicle. This determines whether they are a principal or agent (in COMPUTER: whether they are agent 1 or 2); and in which of the conditions they are, also in which matching group of 9 participants they are (cluster). HUMAN: In each round, a group of 1 principal and 2 agents will be formed randomly by the computer (repeated random re-matching). COMPUTER : In each round, a group of 2 agents will be formed randomly by the computer (repeated random re-matching). Finally, the agent's error term (noise) is randomly drawn by the computer in each round, too.
Randomization unit
Technically, randomization happens at the unit level (drawing a card at the individual level, and treatments are within-session randomized). We will cluster error terms on matching group level, since there might be correlated error terms within matching group (in case within-session randomization is not feasible due to participant numbers).
Clusters
HUMAN:
Matching groups of 9, which means 216/9=24 clusters (matching groups) on which the SE's will be clustered on.
COMPUTER:
48 subjects per treatment (matching groups of 6, so 8 clusters).
Observations
HUMAN: 216 participants in total, 72 agent type 1, 72 agent type 2, 72 principals.
COMPUTER: 144 participants in total, 48 per treatment, (72 agent type 1, 72 agent type 2, no principals)
Arms
HUMAN: We will collect 216 participants. Given that we have sessions of the size of 18
participants, this will be 216 participants in total. This yields 24 matching groups à 9
participants; and 72 prinicipals, 72 agents of type 1, and 72 agents of type 2.
There will be 2 treatments. Thus, 216/2=108 participants per treatment arm. Of those, in
each treatment: 36 principals, 36 agents type 1, 36 agents type 2.
COMPUTER: 144 participants in total, 48 per treatment, (72 agent type 1, 72 agent type 2, no principals). Matching groups of 6, so 8 clusters per treatment.
Primary outcomes
HUMAN:
1) Principal's choice of the promotion rule, in which they decide how to assign the high
prize as a function of the difference of the output of agents (scale: -100 to 100).
2) Beliefs of agents regarding principal's choice of the allocation rule (scale: -100 to
100)
3) Agent's effort provision (effort choice 0 to 100).
COMPUTER:
1) Agent's effort provision (effort choice 0 to 100).
Secondary outcomes
The outcomes are directly observable and will not be constructed.
Cited by
No citation data available for this registration yet.