Working Paper: NBER ID: w23470
Authors: Jonathan De Quidt; Johannes Haushofer; Christopher Roth
Abstract: We propose a technique for assessing robustness of behavioral measures and treatment effects to experimenter demand effects. The premise is that by deliberately inducing demand in a structured way we can measure its influence and construct plausible bounds on demand-free behavior. We provide formal restrictions on choice that validate our method, and a Bayesian model that microfounds them. Seven pre-registered experiments with eleven canonical laboratory games and around 19,000 participants demonstrate the technique. We also illustrate how demand sensitivity varies by task, participant pool, gender, real versus hypothetical incentives, and participant attentiveness, and provide both reduced-form and structural analyses of demand effects.
Keywords: experimenter demand; behavioral measures; treatment effects; Bayesian model; demand sensitivity
JEL Codes: B41; C91; C92
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| positive demand treatment (J23) | increased actions (G34) |
| negative demand treatment (D12) | decreased actions (I12) |
| incentives (M52) | increased effort (D29) |
| experimenter demand (C99) | participant behavior (C92) |
| demand treatment sensitivity (C22) | task and demographic variations (J21) |
| gender (J16) | demand treatment response (C22) |