Migration responses to adverse household shocks: do family relationships and types of shocks matter?

Abstract

This paper estimates how family relationships between migrants and their households affect migration and remittance responses to three different household shocks: an earthquake, the death of a family member, and livestock death. Using agricultural household panel data from El Salvador, we find that migrant responses to negative shocks change across family relationships and the type of shock. Sisters of the household head return home after an earthquake, while sons and brothers migrate away after livestock deaths, and household heads migrate away after the death of a family member. We also find that remittances received by the household significantly fall in response to earthquakes.

Keywords:

• household panel data
• Central America
• El Salvador
• international migration
• remittances
• family relationships

Notes

1. On January 13, 2001, at least 944 people were killed, 5565 injured, 108,261 houses destroyed, 169,692 houses damaged, and more than 150,000 buildings were damaged. On February 13, 2001, 315 people were killed, 3399 were injured, 44,750 houses were destroyed, and 16,752 houses were damaged. (‘2001 El Salvador earthquakes’, n.d.).

2. Another example of this research is Ratha (2003).

3. One exception that we are aware of is Gubert (2002), who uses a cross-section analysis to look at remittance response to income shocks and does consider family relationships. Here, only migrants who were brothers or sons of the household head were considered (along with the head themself); brothers were found to be a positive determinant of remittances.

4. In Spanish: Fundación Salvadoreña para el Desarrollo Económico y Social.

5. We use non-resident Latino workers to capture the most accurate wage rate faced by Salvadoran immigrants to the US.

6. We split livestock into small stock (chickens, goats, ducks, etc.) and large stock (cows, horses, oxen) and examined the deaths of each separately. For small stock we only indicated that a household experienced small livestock death shock if they lost more than 10 small stock in a year. The results for large livestock deaths were consistently insignificant and thus dropped from analysis.

7. We also considered using data on number of dependents in the household, but these results were always insignificant across all regressions and thus dropped from the analysis.

8. Note that the specifications shown below examine changes in number of migrants at the household level, aggregating all migrants. All variables are measures at the household level i.

9. At the same time there is no reason to believe that a household’s distance to an epicenter of a random, natural event would be correlated with unobservable household characteristics that have changed over time.

10. All IV regressions also report test statistics for over-identification and under-identification. First, the Hanson j-statistic provides a test of over-identification, where the joint null hypothesis is that the instruments are valid instruments, meaning they are uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. Since the p-value is 0.08, one can reject the valid instruments only at the 8% level. Also reported is the Kleibergen_Paap LM test statistic for under-identification of whether the equation is identified, meaning the excluded instruments are correlated with ‘earthquake impact.’ The statistic is significant at the 0.01% level, thus the equation is identified. Finally, we report the Wald F statistic for weak identification.

11. Again, all IV regressions also report test statistics including: Hanson j-statistic for over-identification, Kleibergen_Paap LM test statistic for under-identification, and the Wald F statistic. Here, the results show that all instruments are valid, except for column 7 where the p-value for the Hanson j-statistic is 0.05.

12. Small livestock refers to the death of 10 or more small livestock, such as chickens, turkeys, goats, etc. Large livestock were also considered but results were insignificant.

13. Again, all IV regressions also report test statistics including: Hanson j-statistic for over-identification, Kleibergen_Paap LM test statistic for under-identification, and the Wald F statistic. Here, the results show all instruments are valid.