Economists have recently argued that time inconsistency may play a central role in explaining inter-temporal behavior, particularly among poor households. However, time-preference parameters are typically not identified in standard dynamic choice models and little is known about the fraction of inconsistent agents in the population. We formulate a dynamic discrete choice model in an unobservedly heterogeneous population of possibly time-inconsistent agents motivated by specifically collected information combined with a field intervention in rural India. We identify and estimate all time-preference parameters as well as the population fractions of time-consistent and "naive" and "sophisticated" time-inconsistent agents. We estimate that time-inconsistent agents account for more than half of the population and that "sophisticated" inconsistent agents are considerably more present-biased than their "naive" counterparts. We also examine whether there are other differences across types (e.g. in risk and cost preferences) and find that these differences are small relative to the differences in time preferences.