Risky Coping

Abstract

This paper examines whether, in an effort to cope with adverse shocks, poor people with limited coping capacity take large risks (risky coping). About two years after a tropical cyclone in Fiji, many people decided to apply for dangerous international jobs involving casualty risk through a recruitment agency that later turned out to be a fraudster. The analysis reveals that victims with damaged housing are more likely to undertake this risky investment strategy than non-victims. I show evidence that disaster victims use this strategy for risk coping, but not because they have become less risk averse.

Acknowledgments

I wish to thank my field team – Jonati Torocake, Viliame Manavure, Viliame Lomaloma, and 16 enumerators – for their advice, enthusiasm, and exceptional efforts on behalf of this project. Special thanks are owed to the Fijians of the region who so willingly participated in the survey. The Cakaudrove Provincial Office in Fiji offered valuable institutional support. This paper has benefited significantly from the comments and suggestions of Marcel Fafchamps. This research has been made possible through support provided by the Sumitomo Foundation, the Japan Society for the Promotion of Science, and the Ministry of Education, Culture, Sports, Science and Technology in Japan. Any errors of interpretation are solely the author’s responsibility. The data and code used in the paper are available from the author upon request.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. Fraud is a pitfall in international labour migration, especially from developing countries. Many labour migrants rely on recruitment agencies having information about and market power in foreign job markets, and their potential for fraud and abuse has been noted (World Bank, 2006).

2. Fraud victim data are potentially underreported and/or misreported out of embarrassment (for example, Ennis, 1967; Walsh & Schram, 1980). My data on job applications were collected before people became aware of the fraudulence, and thus these data should be free from such fraud-induced measurement errors.

3. Some country-level analyses show evidence for migration driven by weather shocks and environmental degradation (for example, Marchiori, Maystadt, & Schumacher, 2012; Reuveny & Moore, 2009).

4. Fijians’ fraud victimisation through their coping responses in the reconstruction phase intimates that natural disasters can be adversely associated with crime in a broader and more persistent way than current criminology research suggests (for example, Harper & Frailing, 2010; White, 2012).

5. The province, consisting of part of Vanua Levu, all of Taveuni, and other small islands, significantly lags behind the main island, Viti Levu, where the state capital, two international airports, and most tourism businesses are situated. Between 1879 and 1916, the British colonial government brought over 60,000 indentured labourers from India for sugar plantations. Fiji is divided almost evenly between native Fijians and Indo-Fijians. Distinct from the two other provinces (Bua and Macuata) on Vanua Levu Island, Cakaudrove is not in a sugar cane area; as a result, the Indian population is relatively small (16%) and their presence is limited to towns. My study focuses on native Fijians in rural areas.

6. Marriage across different kin groups in the village or different villages is common, and this paper focuses on the kin groups to which households currently belong.

7. Although the agency’s recruitment drive did not cover all villages in the region, all eligible individuals were effectively attempted victims, because people in villages not covered by the recruitment drive were privately informed of the job opportunity.

8. At the time of interviews, 80 per cent of applicants were still waiting for the result, 10 per cent were approved, 8 per cent were rejected, and 2 per cent had withdrawn their application, though none of the applicants with an approved application got the job.

9. Based on the survey conducted in 2003 among nine villages in the same study area (including seven villages covered in the 2005 survey), household crop income seven to ten months after the cyclone was already neutral to the magnitude of crop damage; that is, households intensified crop rehabilitation in proportion to the magnitude of crop damage (Takasaki, 2011c). Crop damage (about F$174 on average) was also not significantly correlated with the incidence of housing damage.

10. According to the 2003 survey, for six months after the cyclone, households received a huge amount of emergency food aid, the value of which is about three times the average crop damage (Takasaki, 2011c). Thus, crop damage was well compensated for in the relief phase. This also suggests that crop damage is unlikely to affect job application in the reconstruction phase.

11. Determinants of migration often include demographic factors, education, asset/income, home location, and migrant network, as well as shocks (for example, Borjas, 1989; Stark, 1991). Sociologists categorise determinants of crime victimisation into exposure, attractiveness of potential targets, guardianship, and proximity to potential offenders (Cohen, Kluegel, & Land, 1981). The first three are often captured by demographic factors, asset/income, and police access, respectively (for example, Barslund, Rand, Tarp, & Chiconela, 2007). Home location, police access, and proximity to potential offenders (for example, their targeting) are fully controlled for by village/mataqali fixed effects in my models. Migrant networks in the Middle East are very limited – acquaintance with someone who already had gotten the same job is very uncommon in the sample.

12. Household land holdings (under the customary tenure) when job-application decisions were made should be almost the same as those at the time of interviews used here. Household non-land asset holdings are not controlled for, because non-land assets at the time of interviews can be affected by job application and the present data lack information about non-land assets at the time of application decisions.

13. This is because in anticipation of labour migration and remittances, the household may adjust its earning efforts, and any unobserved factors that determine income, such as skills, may also influence application decisions (even income measured before the job application, which the present data lack, would be endogenous). Still, household permanent income is controlled for by demographic factors and assets.

14. When three dummies for secondary-school completion or above (18% of eligible individuals), secondary-school incompletion (37%), and primary-school completion (28%) are used (with primary-school incompletion or below as a base), none of the three estimated coefficients are significantly different from each other.

15. Although the present data cannot capture individual military experience, anecdotal evidence suggests that individuals with such experience are not very common in the sample; indeed, households with a member currently working in the military are rare.

16. Standard errors clustered by household in Ordinary Least Squares/2SLS with mataqali dummies are very similar in most cases (the estimated coefficients are the same as those reported here).

17. As a majority of households make only one application, if any, the damage effect on individual application is smaller than that on household application.

18. The cluster-specific FE estimates of Equation (1) and the corresponding FE2SLS estimates with village as a cluster either without or with controls (not shown) are very similar to those reported in Table 2. This suggests that unobserved housing location within villages is not a main source of bias in the FE estimates.

19. The job-application rates among 787 and 452 eligible individuals belonging to households with and without housing damage are .224 and .133, respectively.

20. Flood shocks experienced by neighbours in the same mataqali (village subgroup), for example, are not significantly correlated with own housing damage in the first-stage Equation (3) with mataqali fixed effects, which fully control for mataqali-level covariate flood shocks; this is also true in the model with village fixed effects.

21. Experiencing a flood shock of any magnitude increases the probability that housing is damaged by .29 (F value for this excluded IV is 50 and 41 at the household and individual levels, respectively), that the household applies for the job by .12, and that the individual applies for the job by .07.

22. Distinct from damage incidence, measurement errors in damage value can be significant, causing attenuation bias that also can be controlled for in the FE2SLS estimation (given the instrument orthogonality condition).

23. Flood shock is a strong instrument (F value for this excluded IV is 75 and 60 at the household and individual levels, respectively). A very large flood increases damage value by over 300 per cent. The corresponding reduced-form results are consistent with the FE2SLS results.

24. The FE2SLS estimates of aid effects – for either aid dummy or amount – are negative and considerable in magnitude with statistical significance, though they are much smaller in magnitude than the corresponding damage effects; interpreting these results with potential bias requires caution, however.

25. Disaster victims may also strengthen their beliefs about disaster occurrence in the future and then seek more insurance against future disasters. Then, even if disaster victims become more risk averse, they may take the risk now for future insurance. My data do not allow me to distinguish such an ex-ante insurance motive from a hypothesised ex-post coping motive.

26. These estimation results are strongly consistent with the descriptive findings reported in panels C and D of Table 1. Estimation results for all controls are very similar between cluster-specific FE and FE2SLS.

27. For example, 10 per cent maximal size means that a 5 per cent hypothesis test rejects no more than 10 per cent of the time. These critical values are for Cragg-Donald F statistic (Cragg & Donald, 1993). Across specifications, most of the F values for the excluded IVs in the first stage are above 30.

28. The AR statistic can reject either because the endogenous regressors are nonzero or because the instrument orthogonality conditions fail.

29. The combination of the zero labour-quantity effect on household application and the positive labour-quantity effect on individual application among disaster non-victims implies that even though households that make more than one application are not common, non-victims still tend to do so when they have more labourers.

30. The estimated damage effects depend on labour endowment as follows. First, the positive damage effects on household application are not strongly differentiated by labour quantity (although the estimated coefficient for housing damage itself is not statistically significant, the estimated damage effects for possible numbers of male working-age adults are statistically significant; column 1). With more labourers, the damage effects on individual application significantly decrease (column 2); the damage effects are significant only for households with one or two male working-age adults (.43 and .24, respectively, with a 1% significance level; .06 with three adults). This is because a majority of households select one applicant, if any. Second, the damage effects are statistically significant only among heads and the educated (the estimated damage effects are .40 and .28, respectively, with a 1% significance level; columns 3 and 4). This result for heads is caused partly by the household decision to select one applicant, if any, because headship is strongly negatively correlated with the number of male working-age adults in the household (correlation: -.45); in particular, when households have only male working-age adult, he is likely to be a household head. In contrast, individual education is uncorrelated with the number of male working-age adults.

31. Whereas younger adults are more likely to be an applicant than older adults, individual application is neutral to agricultural land (Table 2). Adding an interaction term of housing damage and age/land to Equation (2) reveals that these relationships hold regardless of housing damage and that the damage effects are not differentiated by age/land (results not shown).

32. The estimated positive labour-quantity effect on household application among disaster victims (.079) is statistically significant at an 11 per cent significance level (column 1).

33. The estimated labour-quantity effect on household application is statistically stronger than that in the model with the housing damage dummy – the estimated effect with mean housing damage value (.05) is statistically significant at a 1 per cent significance level (column 1). The estimated positive labour-quantity effect on individual application among disaster non-victims (.058) is also considerable, though the result is statistically weak (column 2).

Additional information

Funding

This work was supported by the Japan Society for the Promotion of Science [16402012]; Ministry of Education, Culture, Sports, Science and Technology in Japan [17653026]; Sumitomo Foundation [033253].