This paper uses a unique panel data set and data envelopment analysis (DEA) to obtain estimates of technical efficiency for 492 traditional rice plots in Côte d'Ivoire. The objective of this paper is to explore the importance of explicitly controlling for exogenous shocks to production in technical efficiency estimation. We show how omission of such variables in highly stochastic production environments can lead to serious inferential errors, with potentially significant policy implications. Conventional DEA estimation of a production frontier, followed by second-stage Tobit estimation of the correlates of plot- level technical efficiency, suggest widespread and substantial inefficiency related to crop fragmentation and seed varieties. However, when one controls for unobserved groupwise cross-sectional and intertemporal heterogeneity and introduces measurable exogenous shocks into the second-stage estimation, managerial characteristics become jointly insignificant and state-conditional technical efficiency becomes nearly universal. The implication is that conventional technical efficiency estimates that refute the classic Schultzian "poor but efficient" hypothesis may be incorrect because they ignore farmers' vulnerability to adverse states of nature against which they cannot insure.