Eliciting the marginal propensity to consume (MPC) in surveys has become an important source of data for both researchers and policymakers. This talk will report on survey experiments designed to study two puzzles that have emerged in this literature. The between-study puzzle is that average MPCs vary substantially across surveys. We show in a randomized survey experiment that elicited MPCs are quite sensitive to question wording. The within-study puzzle is that that elicited MPCs are poorly correlated with measures of liquidity (and other observables). We conducted text analysis of open-format follow-up questions to a standard MPC elicitation question. The results suggest that the within-study puzzle is primarily driven by nonstandard behaviors (such as mental accounting) rather than respondent misunderstanding of questions or other measurement problems.