Bias correction for within-group estimation of panel data models with fixed effects and sample selection