Mental Models of Incentives

2026-05-13

2026-07-31

- Respondents' effort beliefs in the MidCorp incentives vignette.
- Four prespecified binary mental-model components derived from those effort beliefs.
- Behavioral index summarizing the four mental-model components.

Respondents indicate their beliefs about how much effort an agent will provide in a contextualized principal-agent setting.
Respondents evaluate four contracts:
- Contract 1: USD 8,000 fixed pay and no bonus.
- Contract 2: USD 12,000 fixed pay and no bonus.
- Contract 3: USD 8,000 fixed pay and a 0.5 percent sales bonus.
- Contract 4: USD 8,000 fixed pay and a 10 percent sales bonus.

For Contract 1, respondents choose whether the salesperson would exert no extra effort, some extra effort, or high extra effort. For Contracts 2 to 4, respondents report whether effort is higher under Contract 1, the same, or higher under the current contract. These raw effort beliefs are denoted E_ij.
The four prespecified binary mental-model components are:
- B_i1: positive effort without performance incentives.
- B_i2: a positive fixed-wage response.
- B_i3: a small bonus does not increase effort (even when additional effort would be feasible from a cost-benefit perspective).
- B_i4: a regular bonus does not increase effort or may reduce effort.

The four B_ij components are the prespecified primary outcome family for mental-model analyses. A summary outcome is the behavioral index, defined as the average of the four B_ij components. If one or more components have little empirical heterogeneity, the preferred index-level outcome may be replaced with a weighted index or an index that excludes some components. Open-ended text is a supporting qualitative and mechanism measure: it is coded as behavioral, non-behavioral, or other relative to the benchmark principal-agent model, and behavioral responses are then assigned to mechanism categories.

See the separate pre-analysis plan and full survey instrument for exact wording and screenshots.

- Open-ended reasoning about one contract response, used as a supporting qualitative and mechanism measure.
- Contract choice in an incentivized experiment based on Falk and Kosfeld (2006, Hidden Costs of Control).
- Exploratory contract rankings from the vignette.

Open-ended reasoning data is coded in two steps. First, responses are coded as behavioral, non-behavioral, or other relative to the benchmark principal-agent model. Second, behavioral responses are assigned to individual mechanism categories motivated by the literature. The main coding will use a researcher-defined codebook and LLM classification; research assistants will code a subsample as a robustness check.

After the mental-model vignette and background questions, respondents in sessions with the adoption task participate in an incentivized contract-choice experiment based on Falk and Kosfeld (2006). Respondents choose whether to impose control and separately choose how much to pay a worker. Wages range from 0 to 120 points in increments of 10 points.
Respondents are matched to employees who signed up on Prolific, and one decision is randomly chosen for employee payment. The respondent with the highest profit in each class receives a bottle of Champagne, which is handed out in class.
The primary adoption outcomes are an indicator for imposing control, an indicator for paying a positive wage, and the wage amount as a continuous variable.
The vignette also asks respondents to rank the contracts on a seven-point scale. These ranking measures are secondary and exploratory.

See the separate pre-analysis plan and full survey instrument.

The study population is MBA students at one of the leading global business schools who participate in one of two course in May 2026 and consent to use of their data for research.

The survey is distributed as a mandatory homework assignment that students complete as a teaching exercise

Mental models of incentives are measured by eliciting beliefs in the context of a principal-agent vignette. A between-subject priming treatment studies the influence of salient intrinsic motivation on mental models.
In addition, respondents complete a contract adoption task and provide additional background information through a survey. The adoption task implements a stylized contract-choice experiment based on Falk and Kosfeld. Respondents choose whether to impose a minimum-effort rule and choose an unconditional wage.

Not available

Treatment assignment is computerized in the oTree survey software. The allocation is 30% treatment and 70% control. The survey software sorts participants by start time of the survey and assigns treatment in blocks of 10, with 7 control assignments and 3 treatment assignments randomly shuffled within each block.

Separately, the contract selected for the open-ended reasoning question is randomized with weights that oversample theoretically informative responses. This randomization affects only which explanation is elicited.

Individual student-level

No

Not applicable, since no clustered treatment assignment

Expected n = 100 to 150 MBA students. The realized sample depends on class attendance and consent.

30% treatment and 70% control. With an expected total sample size of n = 100 to 150, this implies approximately 30 to 45 respondents in treatment and 70 to 105 respondents in control.