Empirical studies of tax and benefit incidence routinely ignore behavioral responses and measurement errors. This paper offers an econometric method of estimating the mean benefit withdrawal rate (marginal tax rate) allowing for incentive effects, measurement errors, and correlated latent heterogeneity in incidence. Under the method's identifying assumptions, a feasible instrumental variables estimator corrects for incentive effects and measurement errors, and provides a bound for the true value when there is correlated incidence heterogeneity. A case study for a large cash transfer program in China indicates that past methods of assessing benefit incidence using either nominal official rates or raw tabulations from survey data are deceptive. The program entails a nominal 100 percent benefit withdrawal rate -- a poverty trap. However, the paper finds that the actual rate is much lower, and clearly too low in the light of the literature on optimal income taxation. The paper discusses likely reasons based on the qualitative observations.