The hypothesis presented in Todaro (1 969), that the likelihood of finding work in town influences the rate of rural-urban migration, now enjoys the status of a received doctrine. Assuming potential migrants indeed respond to this employment probability, the model of Harris and Todaro (I970) demonstrates that, in certain parametric ranges, urban job creation may actually exacerbate unemployment and even reduce national product. This result has had considerable influence on policy formulation in LDCs, by emphasising that, in the urban sector, the social opportunity cost of labour may not be insignificant despite burgeoning unemployment. Yet neither the Todaro hypothesis nor prevalence of the Harris-Todaro parametric range has been adequately, empirically explored. Many estimates of macro migration equations do exist, normally relating the proportion of population migrating to average wages in differing locations and occasionally to average population characteristics. But in Lucas (I 975), I show that the popular nonlinear specification of such macro functions may well display serious specification error bias; a nonlinear function of the aggregate variables is not simply the average of underlying micro migration decisions related to the disaggregated variables. Thus, although a few estimates of macro migration equations have also incorporated average unemployment rates, usually in developed country contexts and with mixed results, these analyses are at best very circumscribed tests of the Todaro and Harris-Todaro theories. On the other hand, surprisingly few estimates of micro migration response functions have appeared. In part, this probably reflects the difficulty of dealing with unobserved variables. For example, suppose whether or not each person moves to town is to be related to his or her wage at home compared to the relevant wage in town. In a typical survey, for those individuals who remain in the village the wage they would command in town is not observed; for those who have moved to town, the prior rural wage is usually not reported. DaVanzo (I976) deals with this problem of unobserved variables, in a US context, by introducing a multivariate wage equation to predict potential wages in alternative locations for each person using their observed personal characteristics. However, this broad approach suggested by DaVanzo has not previously been extended to incorporate unobserved employment probabilities in alternative locations. IThus, in surveying the literature in i980, DaVanzo concludes: