Because their departures are difficultly observed, little is known about the performance of immigrants who leave a region and move to another. This paper shows conditions under which the (conditional) outmigration probability, work probability and the expected earnings of outmigrants are nonparametrically identified using data on immigrant sample attrition. We present an econometric model which extracts the information on outmigration behavior from sample attrition and allows to incorporate unobserved heterogeneity in the choice process in a very natural way, a feature not previously possible. The method presented in the paper is general enough to estimate models of internal and external migration. We apply our framework to estimate a panel data model where immigrants can choose whether to stay or leave a country. The model is estimated using the German Socio-Economic Panel (GSOEP). Outmigrants are found to have significantly lower labor market earnings and work propensities than permanent migrants.