"THE DARK FIGURE” OF TERRORISM: UNDERSTANDING THE REPORTING AND NON-REPORTING OF TERRORIST EVENTS

This study aims to conduct an inquiry into the extent and implications of potential biases in widely used open source (OS) terrorism databases. The underreporting or misreporting of terrorist events is examined through the perspective of rational choice theory, in which media reporting is explained by the expected benefits and estimated costs to the agents in the media. Comparison of OS accounts with ostensibly more reliable official accounts has provided evidences to gain insights into the rate of reporting, and extent and nature of discrepancies between OS and official accounts. Due to the limited number of cases and no available official terrorism data in the US, indirect evidence is gathered in the first stage by comparing OS accounts with the official Supplemental Homicide Report (SHR) on gang-homicides from the US in 2011, a phenomenon that resembles terrorism in some essential ways, and for which open-source coverage rates can be estimated. In the second stage, direct evidence is sought through the comparison of the Global Terrorism Database (GTD) with the Turkish National Police (TNP) terrorism accounts (1996-2012).Descriptive and multivariate statistics (logit and multinomial logit) are used to assess the coverage of incidents. For robustness and sensitivity checks, findings are obtained under alternative terrorism definitions and stratification methods. In addition to testing the statistical significance of variables, cross-validation analyses are used to estimate the out-of-sample predictive performance of the models and contributions of each variable in predicting OS coverage. The analyses suggest four substantial conclusions for the two settings studied: (i) OS accounts significantly under-report terrorism and gang homicides relative to official accounts; (ii) the under-reporting in OS accounts is not random; incident characteristics — victim/target, offender, and incident types, temporal and spatial factors — and rational factors — especially newsworthiness — significantly influence the likelihood of OS reporting; (iii) severity — especially the existence and number of fatalities — is the strongest predictor of the OS coverage; and (iv) the other factors that predict underreporting in OS accounts differ substantially across the two settings. Relying on these findings, this study concludes with the assessment of these reporting discrepancies and their potential influence on the accuracy of existing terrorism research, and suggests some practical implications for the readers, the users of the OS data, and the data collectors.