On the estimation of treatment effects with endogenous misreporting