The untold story of missing data in disaster research: a systematic review of the empirical literature utilising the Emergency Events Database (EM-DAT).

Global disaster databases are prone to missing data. Neglect or inappropriate handling of missing data can bias statistical analyses. Consequently, this risks the reliability of study results and the wider evidence base underlying climate and disaster policies. In this paper, a comprehensive systematic literature review was conducted to determine how missing data have been acknowledged and handled in disaster research. We sought empirical, quantitative studies that utilised the Emergency Events Database (EM-DAT) as a primary or secondary data source to capture an extensive sample of the disaster literature. Data on the acknowledgement and handling of missing data were extracted from all eligible studies. Descriptive statistics and univariate correlation analysis were used to identify trends in the consideration of missing data given specific study characteristics. Of the 433 eligible studies, 44.6% acknowledged missing data, albeit briefly, and 33.5% attempted to handle missing data. Studies having a higher page count were significantly (p < 0.01) less prone to acknowledge or handle missing data, whereas the research field of the publication journal distinguished between papers that simply acknowledged missing data, with those that both acknowledged and handled missing data (p < 0.1). A variety of methods to handle missing data (n=24) were identified. However, these were commonly ad-hoc with little statistical basis. The broad method used to handle missing data: imputation, augmentation or deletion was significantly (p < 0.001) correlated with the geographical scope of the study. This systematic review reveals large failings of the disaster literature to adequately acknowledge and handle missing data. Given these findings, more insight is required to guide a standard practice of handling missing data in disaster research.