There are at least four methodological challenges that confront any economic (or social science) work on international migration and remittances. These problems include: simultaneity, reverse causality, selection bias and omitted variables. This introduction reviews each of these challenges, and suggests possible solutions (for more, see McKenzie and Sasin, 2007).
First, many of the decisions on international migration are made at the same time as other household decisions. For example, a household may decide to send its oldest male to work abroad at the same time that it decides to send its youngest daughter to school. As a result, variables that “cause” international migration may also “cause” household patterns of consumption and education.
The second problem is reverse causality. For instance, while international remittances may help reduce poverty in the developing world, the level of poverty may also influence the amount of remittances received by a particular country. Thus, any attempt to analyze the impact of remittances on poverty that fails to consider the reverse causality between these two variables might lead to erroneous conclusions.
The third problem is selection bias, which refers to the “selectivity” of people who tend to migrate and to receive remittances. If, for example, households with more education or income are more likely to produce migrants, then it is impossible to identify the effects of migration by simply comparing the characteristics of migrant and non-migrant households.
Fourth, when households produce migrants or receive remittances on the basis of unobservable characteristics – characteristics like the risk averseness of the household head – then the problem of omitted variable bias arises. For example, it is possible that households with more risk averse heads will be less likely to produce migrants, but it is very difficult to collect data on this issue.
To meet these various methodological challenges, at least five possible solutions have been proposed in the literature. Most of the studies included in this anthology employ one or more of these solutions.
The first, and perhaps best, solution is to use a randomized, “natural” experiment whereby individuals desiring to pursue international migration are denied the right to migrate (by a lottery system, for example), thereby creating a “control group” of would-be-migrants to compare with a group of actual migrants (see e.g. McKenzie, Gibson and Stillman, 2006). Comparing the characteristics of would-be-migrants to those of actual migrants would then yield accurate information on the causal motives for migration. Unfortunately, however, it is very difficult to conduct such randomized, “natural” experiments in the developing world to such an extent that the only real example at this time of such a natural experiment is MeKenzie, Giibson and Stillman (2006).
A second, and slightly less difficult, solution is to use panel data. Panel data, which includes repeated observations on the same household over two or more time periods, is a good solution because by taking “first differences” between various variables it becomes possible to eliminate many of the methodological problems discussed above. Unfortunately, however, panel data sets on international migration and remittances in the developing world are relatively rare.
A third solution is to construct a “counterfactual” situation, that is, to artificially construct what the status of a migrant household would have been had that household not produced a migrant. For example, if the topic is remittances and income, then it would be necessary to estimate the income of a migrant household by imputing the value of that migrant had he stayed and worked at home (see e.g. Barham and Boucher, 1998).
A fourth solution to use econometric procedures to regress the outcome of interest (for example, poverty) on a set of independent variables, and then supplement this approach with a sample selection procedure, like the two-stage Heckman model (see e.g. Acosta et al., 2007). Here the selection model is used to estimate the size and direction of the selection bias. However, the difficulty comes in specifying an exogenous variable that “causes” migration or the receipt of remittances in the first-stage equation, but has no direct impact on the dependent variable in the second-stage equation.
A fifth, and quite common, solution is to use instrumental variables. A good instrumental variable, one that is correlated with the explanatory variable but uncorrelated with the outcome variable, can eliminate many of the biases that arise from endogeneity, selection bias and omitted variables. In practice, however, selecting a good instrumental variable in migration and remittances work can be difficult. For example, assume that migration is the explanatory variable and poverty is the outcome variable of interest. The challenge is then to find an instrumental variable (like distance, for example) that is correlated with migration but exogenous to the outcome variable, poverty.
As noted above, many studies employ one or more of these solutions to the problems of simultaneity, reverse causation and selectivity. It is not uncommon, for example, to find instrumental variables used in conjunction with panel data. Other studies estimate counterfactual situations with the use of instrumental variables (see Acosta et al, 2007).