The 21st Century climate challenge

under the age of five countries that had a DHSwith a geographic positioning system (GPS)module two to three years following a disasterwere selected. The selection of countrieswith GPS modules was necessary, especiallyfor countries where some administrative districtswere more affected than others. For adultwomen selection was limited to major disastersthat had occurred during the 1970s and 1980s;with the requirement that the disaster in questionoccurred at least 15 years prior to the firstDHS. See table for country coverage and samplecharacteristics.MethodologyThis approach borrows from impact evaluationtechniques widely used in the social sciences.For children under the age of five, the outcomeindicators used were: stunting (low height forage), wasting (low weight for height) and malnourishment(low weight for age). For adultwomen 15–30, the outcome indicator was educationaloutcome. In the absence of longitudinaldata, a set of synthetic before and after cohortswere constructed and their outcomes comparedusing logit regressions with a difference-in-differenceapproach, controlling for individual,household and community characteristics.To construct the cohorts, children andadult women in DHS were identified and theirbirth dates tracked. The subject’s birth date andbirth location were then crosschecked againstthe occurrence of a natural disaster as indicatedin EM-DAT. The following groups wereidentified:• Subjects born before a disaster in an areathat was subsequently affected (born before,affected—group 1, affected).• Subjects born before a disaster in an areathat was not subsequently affected (born before,not affected—group 1, not affected).• Subjects born during a disaster in an areathat was affected (born during, affected—group 2, affected).• Subjects born during a disaster in an areathat was not affected (born during, not affected—group2, not affected).Using these different groups, the followingmodel was estimated:nˆφ = — 1 N Σ[(y ai=1– i2 ya i1)– (yna – i2 yna)]where y i2is the outcomeiin question for the i th person. 2At each step, a set of control variables wereused to identify the effects of specific characteristicson children’s nutritional outcomes. Theseincluded individual variables (the sex of thechild, birth intervals and such maternal characteristicsas mother’s age and education) andcommunity-level variables (e.g., urban/rurallocation). A regression analysis was then conductedto isolate the specific risks associatedwith being affected by a disaster.For adults, if it is assumed that disasters area deterministic process, then virtually every indicatorincluding household socio-economiccharacteristics is determined by early exposureto a disaster, and is therefore endogenous.As a result, only variables that can reasonablybe assumed exogenous, such as religion, wereincluded.Most of the results are shown and discussedin chapter 2 and in Fuentes and Seck 2007.Notes1 Guha-Sapir et al. 20042 Cameron and Trivedi 2005TableCountry coverage and sample characteristicsCountry Year of survey Sample size Stunted (%) Malnourished (%) Wasted (%)ChildrenEthiopia 2005 9,861 43.4 37.8 11.1Kenya 2003 5,949 32.5 20.2 6.7Niger 1992 6,899 38.2 38.9 14.5AdultsYear of surveySamplesizeNo education(%)At least primaryeducation (%)At least secondaryeducation (%)India 1998 90,303 35.3 50.5 33.6HUMAN DEVELOPMENT REPORT 2007/2008 363