We present three LDV panel estimators with individual effects, that allow for endogenenous time-invariant regressors. To our knowledge, besides joint MLE, no alternative consistent parametric estimators exist. We also address whether linear models may provide better small sample estimates of marginal effects. The small sample behavior of LDV models and linear models are described in a Monte Carlo study of a Probit model. The Three LDV panel estimators behave differently in designs with varying degrees of endogeneity and instrument quality, but are not preferred to cross-sectional estimators. Linear estimators do not improve estimates even with misspecified distribution of individual effects.