Econometric models estimating parameters for agricultural policy analysis increasingly rely on unbalanced panels of farm‐level data. Since such models have often been estimated through simplified approaches, in this paper we show that adopting more sophisticated panel data techniques may be very important for obtaining more reliable estimates of policy parameters. We also extend the two‐stage procedure proposed by Shonkwiler and Yen (1999) and Tauchmann (2005) for the analysis of censored data to account for heteroskedasticity and correlation of the error terms of the first‐stage probit models.