1This is a remarkably small drop in arrivals compared to the lower revenues from the tourism sector after the disaster. Households in the lowest and highest quantiles appear to be smaller in size (column 6) compared to households in the middle quantiles. Given the exposure to the tsunami disaster, our treatment group is more represented by the high-income households compared to the households in the lowest quantiles (column 8).
This clearly shows that the recovery after the first crash is biased towards the wealthy. A similar pattern is observed for the post-disaster years 2009 and 2012, except for the lowest quantile and households in the 99% of the income distribution (in columns 10 and 12). The distribution of recovery benefits by year after disaster again shows a decline in 2009; these results are shown in Figure 4.
In the post-disaster 2009, the households in the lowest and the highest quantile's recovery (apart from the negative recovery in the 6th quantile) are only statistically significant (column 10). Recovery in the higher quantiles (8th and beyond) in 2012 returns to the level of the first post-disaster year or even further extended after a reduction in the year 2009. The Gini coefficient of household income for the entire sample (figure 1 ) shows an increase of inequality in 2002 compared to the year 1995, then a reduction in the first post-disaster year 2006.
These results clearly show an increase in inequality in the first post-disaster year and then a reduction in inequality over the post-disaster years.
Robustness analysis
The coefficient of variation for consumption also shows a decrease during post-disaster years and remains lower in recent years than in 1995. Compare the income of tsunami-affected and non-affected households separately in Figure 2 (a and b), the affected region ( Figure 2 a) looks more equal than the unaffected regions. Comparing the two groups in the pre- and post-disaster years, the affected region's income inequality increased from 1995 to 2002, but decreased in the post-disaster period and ended up lower than the first survey year.
The results indicate a reduction of inequality in affected areas compared to unaffected regions during the post-tsunami period. In contrast, the Gini coefficients (in Figure 2 a) show that the affected regions are more unequal in consumption than the non-affected regions in the first survey year. Both groups' inequality is increased until 2002, but the inequality among the unaffected regions is increased compared to affected regions in the first post-disaster year 2006.
It then decreases for both groups in the years after the accident; more importantly, the Gini coefficient of the affected region remains lower than the non-affected region in the last year after the 2012 disaster. The coefficient of variation follows an almost similar pattern to the non-affected regions, but the affected regions show a decrease in the Gini coefficient in the years after the disaster (Figure 2 b). Alternatively, we estimated quantile treatment effects using the change-in-change (CIC) method of Melly and Santangelo (2015).
Our estimates in the quasiquantile regression appear reliable, as a similar pattern of income recovery revealed in the change method. 3 Despite the change in change approach relaxing the assumption of parallel trend, the assumption of conservation of rank order is still not eliminated. Further, conditional and unconditional quantile estimates are still not reliable if the assumption of rank conservation does not hold.
Violation of this assumption is possible in this case if the treatment correlates with other, time-varying, unobserved factors and, as a result, if the households are placed in a different quantile (higher or lower) in the first year after the disaster in the later years after the disaster. . Given the increased income due to the tsunami (as revealed by De Alwis and Noy, 2016), low-income households are likely to move into high income quantiles in the post-disaster period. If the ranking is violated, the observed higher income/consumption gains towards the high-income quantiles in the post-disaster years in this study could be due to a shift in the ranking of lower-income households upwards in the distribution in the later post-disaster years than treatment effect on households in high income quantiles.
Conclusions
Further, the normalized incomes for the selected quantities (Quantiles 2, 5, 7 and 9) in Figures 7, 8, 9 and 10 show a much closer residual income trend between the two groups during the pre-disaster period. However, contrary evidence emerges when the recovery is estimated as a percentage of median income for the observed income groups. Estimates of the recovery as a percentage of average consumption show a U-shaped recovery4 across all groups—poor to rich—in the short and medium term.
All groups recovered in the short term, the recovery of households in the middle of the distribution fades in the medium term, and the recovery is sustained only among the wealthy groups in the long term. Overall, this observation points to a long-term increase in consumption inequality in the affected regions. 5 In this case, this is likely a shift in the position of low-income households upwards in the distribution in the later post-disaster years, as the poor achieved significantly more recovery shortly after the disaster.
Consequently, the observed recovery of higher quantiles in later years after the disaster could be misleading. It is possible that the rich-favored recovery revealed in the last years after the disaster overshadows the sustainability of the recovery achieved by poor households in the first year after the disaster. Another limitation to note is that the results are only valid if there are no spillover effects between affected and unaffected regions, which the analysis has not yet ruled out.
Household covariates include gender, age, years of education, ethnicity of household head and household size. Household income (interest variable) includes only paid, agricultural and non-agricultural incomes, remittances, transfers, dividends, property rent and cash income and does not include loans, asset sales, withdrawal of savings, insurance compensation and other adhoc gains (see Appendix 1).
The Impact of Natural Hazards and Disasters on Agriculture, Food Security and Nutrition: A Call to Action for Resilient Livelihoods. The impact of natural disasters on human development and poverty at the municipal level in Mexico. The Impact of Natural Disasters on Income Distribution: An Analysis of Hurricane Katrina, Journal of Regional Analysis and Policy.
Long-term perspectives on the response to the 2004 Indian Ocean tsunami: A joint follow-up assessment of the linkages between relief, rehabilitation and development. Predicting long-term business recovery from disaster: A comparison of the Loma Prieta earthquake and Hurricane Andrew. The impact of natural disasters on income inequality: Analysis using panel data over the period 1970-2004.
Appendices
