Testing Network Theory through an Analysis Migration from Mexico to the United States

Spittel, Michael
Working paper no. 1999-01


Very recently several writers in the study of international migration have recognized the role of social networks, or ‘migrant networks’, as an important force in explaining the perpetuity of international migration (Massey, 1987; Boyd, 1989; Massey et al., 1993, 1994; Portes, 1995). However this network effect has never been directly empirically examined. Traditional analysis using logit/logistic models makes unrealistic assumptions of the unmeasured common causes and, more often, does not control for relevant measured characteristics which may be correlated to the individuals in the network (Massey and Palloni, 1992).

In order to avoid these estimation problems a multi-state event history model, developed by Alberto Palloni and Douglas Massey, is employed using household level data from the Mexican Migration Project and the parameters of a Weibull function are estimated. This study specifically looks at the relationship between father and eldest son. For simplicity, the states are specified as unidirectional. State one is defined as both father and son reside in Mexico and have never migrated; state two is defined as when only the father has migrated to the United States; state three is defined as when only the eldest son has migrated to the United States; and state four is defined as when both father and son have migrated to the United States. A set of controls are added to the analysis, namely education, occupation and age, period effects, common conditions, as well as controls for unmeasured heterogeneity-which control for unmeasured ‘common causes’ between the pairs under examination. Predicted probabilities are then compared and contrasted and life tables are created.

When examining the migration experience between father-son pairs, the effects of social networks maintain as the most significant force in affecting the risk of migrating after controlling for selection and common causes, both measured and unmeasured. In particular, the probability of migrating is higher for those fathers whose son had migrated first (transition 3->4) compared to households where the father migrated first (transition 1->2). The hazard is also higher for those sons whose father had migrated first (transition 2->4) compared to those where the son migrated first (transition 1->3), net of individual effects and community characteristics.