Suchergebnis

Titel des Experiments
Human vs. Algorithmic Auditors: The Impact of Entity Type and Ambiguity on Cheating Behavior

Autoren
Marius,Protte; Behnud,Mir,Djawadi

Kurzbeschreibung des Experiments
The field experiment will be conducted at a public event hosted by our university. The classic die rolling task by Fischbacher & Föllmi-Heusi (2013) is adapted to a card draw, in which participants have the opportunity to increase their own financial gain by overstating their drawn number. Subjects draw a random number between 1 and 6 from an urn containing 100 cards with an unknown distribution. They then confidentially reported their drawn number via a computer interface. After submitting their report, participants complete a questionnaire.

Every two hours, among all participants from that timeframe, one is randomly selected to be paid their reported number multiplied by €15 (payoff range: €15 to €90), pending a “Verification Part”.
The Verification Part comprises up to two lotteries:
• In Lottery 1, a number between 1 and 10 is randomly drawn. If this number is greater than the participant’s reported number, no verification is necessary, and the full payoff (reported number × €15) is paid. If the number is less than or equal to the reported number, the participant’s actual drawn card is checked.
• If the reported and actual numbers match, the full payoff is paid.
• If they mismatch, Lottery 2 is triggered: An urn containing numbers from 1 up to the reported number is used to randomly draw one number. If the drawn number is less than or equal to the actual number, the full payoff is still granted. Otherwise, the payoff is reduced to the actual number × €7.50 (payoff range: €7.50 to €37.50).

Theoretically, for a risk-neutral decision-maker, the payoff-maximizing strategy is to always report a 6, regardless of the actual number drawn.

We implement four experimental treatments: Human (H), Machine (M), Human Blackbox (HB), and Machine Blackbox (MB). In the Human treatment, the verification process is conducted by a human agent (i.e., the experimenter) using numbered balls drawn from urns, whereas in the Machine treatment, it is executed by a computerized, rule-based algorithm (visualized through animations on a computer screen). Procedures in the Blackbox versions (HB and MB) mirror their respective non-blackbox treatments, except that the verification rules are not disclosed to participants.
The underlying verification rules and their associated probabilities are held constant across all treatments; only the auditor’s entity (human vs. machine) varies.

Downloads

Institutional Review Board Certificate (English)