Abstract This study addresses the issue of how accurately parameters characterizing a stochastic, discrete behavioral process can be estimated using data from a discrete-time panel study. Specifically, the study examines the likelihood and magnitude of error in estimating the parameters characterizing the behavioral process based on observations of the process obtained at discrete time points. The results of the study offer strong indications that the likelihood and magnitude of error can be both very large. Furthermore, reducing the interval between survey waves improves estimation accuracy but only very slowly. The study results point to the need to obtain continuous data as accurately as possible from surveys made at discrete time points, using carefully designed and administered recall questions.