Abstract
Surveys typically use hypothetical questions to measure subjective and unverifiable concepts like happiness and quality of life. We test whether this is problematic using a large survey experiment on health and subjective well-being. We use Prelec’s Bayesian truth serum to incentivize the experiment and defaults to introduce biases in responses. Without defaults, the data quality was good and incentives had no impact. With defaults, incentives reduced default biases in the subjective well-being questions by inducing participants to spend more effort. Incentives had no impact on the health questions regardless of whether defaults were used.