Data source and design In this study, we used data from the ‘Bangladesh Disaster-related Statistics-2015’, the largest and most recent disaster-related survey in the country conducted by the Bangladesh Bureau of Statistics (BBS) (BBS, 2016). A disaster was defined as a major adverse event resulting from natural processes of the earth, e.g. cyclones, drought, floods, erosion, volcanic eruptions, earthquakes, tsunamis, and other geologic processes. The survey was designed to measure the socio-economic characteristics of households and population, damage and loss of land and residence, health and sanitation conditions, different disasters faced, and perception and knowledge about climate change in the disasterprone areas. The respondents were asked for the disaster-related information between 2009 and 2014. All 64 districts of Bangladesh were considered as a sampling frame consisting of mauzas/mahallas (lowest administrative unit). Respondents were questionnaire was used to collect data in two phases. In both phases, the field operation was conducted for 45 days. About 1800 data collectors and employees of the Bangladesh Bureau of Statistics (BBS) were employed in the data collection after four days of comprehensive training,
We estimated the impact of natural disasters on the selected outcome variables. The key outcome variables were injury, disability, and death due to different disasters in the last six years. The ‘Bangladesh Disaster-related Statistics-2015’ survey collected self-reported injury and disability information associated with natural disasters. In this study, the injury was defined as any accidental force applied to the body during the natural disasters that caused harm (Schuh-Renner et al., 2019). The difficulty with mobility, basic activities (e.g. dressing and bathing) of daily living after affected by the natural disasters were considered as disability (Manini et al., 2017). The participant was asked, ‘Member of your household, who got sick, either injured or got disabled due to natural disaster during 2009–14’ to capture the injury and disability information. The prevalence of injury, disability, and death were estimated per 100,000 populations across different socioeconomic characteristics (e.g. age group, sex, marital status, education, occupation, physically challenged condition, disaster warning, disaster preparedness, number of disasters faced, place of residence, region of residence, and asset quintile). Physical or psychological situation that hinders the daily movement, sensations, and activities of a man or woman was defined as the physically challenged condition (BBS, 2016). The dummy variable ‘got disaster warning’ denoted 1 if a household received any advanced notice or forecast on disaster and 0 otherwise. Preparedness was defined as measures that are designed to ensure that communities will have the knowledge and understanding of their risk environment to enable them to better cope with disaster associated causalities (BBS, 2016). Number of disasters faced was defined by how many times the person was affected by a disaster. We analysed the effects of drought, flood, water logging, cyclone, tornado, storm/ tidal surge, thunderstorm, river/coastal erosion, landslides, hailstorm, and other disasters in this study. Households’ wealth was categorized into five quintiles ordered from the poorest to the richest based on the available assets of the household, including housing material, sanitation facilities, access to utility services, and access to drinking water.
Statistical analyses Principal component analysis (PCA) was applied to survey responses on ownership of a set of key assets and the values of the index were based on the first principal component (Vyas & Kumaranayake, 2006). Household size was adjusted while estimating PCA score. PCA is a commonly used technique when computing asset indices; although traditionally applied to continuous variables (Filmer & Pritchett, 2001). Higher scores of the index indicated more affluent households. The chi-square test was performed to assess the association between categorical variables. We applied three different logistic regression models to identify the associations between individual/household characteristics and disaster-related injury, disability, and death. In first two models, disaster-related injury and disability were treated as dependent variables and other socio-economic characteristics as independent variables (e.g. age group, sex, marital status, education, occupation, physically challenged condition, disaster warning, disaster preparedness, number of disasters faced, place of residence, region of residence, and asset quintile). However, in the third model, disaster-related death was analyzed for all of these characteristics except marital status, education, occupation, and physical challenges due to the unavailability of the information. All statistical analyses were performed using STATA version 13 (StataCorp., 2013).