Pay Every Subject or Pay Only Some

One technique employed by budget-conscious researchers is to pay only some subjects for their choices in an experiment, or pay smaller stakes to all subjects. We test the effect of paying some subjects versus paying all subjects in the context of risk preferences, controlling for the difference in stakes induced by paying only some subjects. Over two experiments, we demonstrate that paying some subjects yields lower levels of risk aversion than paying all subjects, but more risk aversion than paying all subjects lower stakes. Paying some subjects also impacts the ordering of subjects by elicited risk aversion. Neither probability weighting nor standard experimental demographics were correlated with subjects' differences between these conditions. We exploit our multiple measurements of risk aversion to estimate a simple structural model of latent risk aversion and derive a correction factor to approximate the results as if all subjects were paid high stakes. Our findings imply that paying some subjects high stakes meaningfully impacts the elicited level of risk aversion, although it better approximates the experimental ideal of paying all subjects high stakes compared to paying all subjects lower stakes.