Unobserved Inputs in Household Production

With few exceptions, empirical household production studies ignore unobserved inputs. We demonstrate that without additional assumptions, the estimable impacts of the observed inputs cannot provide informative estimates of their marginal products due to contaminating variation in unobserved inputs; not even the sign of marginal impacts can be ascertained. Instrumental variables cannot solve this problem since every candidate for an instrument affecting an observed input, including experimental assignments, would also affect unobserved inputs choices through the budget constraint, invalidating this variable as an instrument. We show that under certain additional assumptions an appropriately specified empirical model can provide bounds for true marginal products. Our main point is that unless one is willing to make assumptions of this nature, estimated effects would have no useful interpretation. Almost all existing empirical studies of health, child development, and job-training programs fail to account for this issue, rendering their conclusions incomplete and possibly misleading