Does Terrorism Increase after a Natural Disaster? An Analysis based upon Property Damage

Abstract

Does an emergency such as a natural disaster lead to a surge of terrorism? This paper contributes to the emerging literature on this issue. We consider the experience of 129 countries during the period 1998–2012 to determine the effect of a natural disaster on both domestic as well as transnational terrorism. We also control for endogeneity using expenditure on health care and land area in a country as instruments. In contrast to the existing literature, we measure the extent of terrorism by the value of property damage. The results indicate that after natural disasters, (a) transnational terrorism increases with a lag, and (b) a statistically significant impact on domestic terrorism is not observed.

Keywords:

• Natural disasters
• Domestic terrorism
• Transnational terrorism
• Empirical research

Notes

¹ The GTD is an open-source database including information on terrorist events around the world from 1970 to 2014 (with additional annual updates planned for the future). Unlike many other event databases, the GTD includes systematic data on domestic as well as transnational and international terrorist incidents that have occurred during this time period and now includes more than 140,000 cases. For each GTD incident, information is available on the date and location of the incident, the weapons used and nature of the target, the number of casualties, and – when identifiable – the group or individual responsible. Statistical information contained in the GTD is based on reports from a variety of open media sources. Information is not added to the GTD unless and until they have determined the sources are credible.

² Property damage is weighted between 0 and 3 following inputs by the Global Peace Index Expert Panel in particular with the advice of Expert Panel member and terrorism expert Dr. Ekaterina Stepanova, Head of the Peace and Conflict Studies Unit at the Institute of World Economy & International Relations.

Detailed information on the GPI panel is as follows:

•	Professor Kevin P. Clements, Chairperson Foundation Chair of Peace and Conflict Studies and Director, National Centre for Peace and Conflict Studies, University of Otago, Dunedin, New Zealand.
•	Dr. Sabina Alkire, Director, Oxford Poverty & Human Development Initiative, University of Oxford, United Kingdom.
•	Dr. Ian Anthony, Research co-ordinator and leader of the Arms Control and Non-proliferation Programme, Stockholm International Peace Research Institute, Sweden.
•	Mr. Vasu Gounden, Founder and Executive Director, African Centre for the Constructive Resolution of Disputes, Durban.
•	Mr. Nick Grono, CEO, Walk Free, formerly deputy president, International Crisis Group, Brussels, Belgium.
•	Dr. Manuela Mesa, Director, Centre for Education and Peace Research and president, Spanish Association for Peace Research, Madrid, Spain.
•	Dr. Ekaterina Stepanova, Head, Unit on Peace and Conflict Studies, Institute of the World Economy and International Relations (IMEMO), Russian Academy of Sciences, Moscow, Russia.

³ Our analysis indicated that the correlation between number of incidents and damage due to domestic terrorism was 0.19 and that between number of incidents and damage due to transnational terrorism was 0.04.

⁴ The database is compiled from various sources including the United Nations, governmental and non-governmental agencies, insurance companies, research institutes, and press agencies (see Table II at http://www.emdat.be/guidelines). As there can be conflicting information and figures, Center for Research on Epidemiology of Disasters has established a method of ranking these sources according to their ability to provide trustworthy and complete data. In the majority of cases, a disaster will only be entered into EM-DAT if at least two sources report the disaster’s occurrence in terms of deaths and/or affected persons. The final figures in EM-DAT usually originate from the priority source, but they can also be completed by a secondary source. In certain cases, a secondary source can become a primary one. This can be the case, for example, when final figures are made available long after the disaster has occurred. Also, some sources are used for specific disasters (i.e. USGS for earthquakes, WHO for epidemics).

⁵ We use data from the World Bank. Comparable numbers are also available at the KILM database of the International Labour Organization.

⁶ All the variable data sources are listed in Table III.

⁷ In the Appendix 1, we present additional regressions that control for population density. There is no substantial change in the results.

⁸ Although we report results with two-year lagged damages in this paper, in our research, we have also considered specifications with a one-year lag and with no lag and found them to be statistically insignificant.

For transnational terrorism, the coefficient of natural disaster damage had a p-value of 0.752 when we used one-year lagged damages and had a p-value of 0.166 when we used contemporaneous damages. For domestic terrorism, the corresponding p-values were 0.993 and 0.759, respectively. We also considered a three-year lag going beyond what earlier literature has focused on when evaluating the impact of natural disasters on transnational and domestic terrorism damage and found the coefficients to be statistically insignificant. Specifically, the p-values were 0.717 and 0.462, respectively, for transnational and domestic terrorism.

⁹ The justification for using lagged values of the IVs and contemporaneous values of the other independent variables is as follows: the exogenous variables such as political rights, civil liberties exhibit autocorrelation. Further, since they do not have any influence on damage due to natural disasters, their inclusion in a lagged fashion as instruments would be problematic as it will introduce multicollinearity. To avoid these issues, we adopt the framework described in Equation (3(3) ).

¹⁰ A question we were asked in seminars is the following: Do we observe changes in land area during the period of study? The answer is that yes, we do observe changes for 37 countries (such as Syria, Ukraine, China, Iraq, Norway, the U.S., North Korea) and therefore it is appropriate for selection as an instrument in a FEs panel model.

¹¹ The first-stage F-test yielded a p-value < 0.0001. The over-identification test indicated a p-value of 0.5047. In the Stock–Yogo test, Cragg–Donald Wald F statistic is 43.09. These indicate we have strong and valid IVs.

¹² The first-stage F-test yielded a p-value < 0.0001). The over-identification test indicates a p-value of 0.0990. In the Stock–Yogo test, Cragg–Donald Wald F statistic is 43.19. These indicate we have strong and valid IVs.

¹³ The results of the first-stage estimation are reported in Table A1.

¹⁴ There could be some concerns regarding the possible existence of multicollinearity between measures such as HDI and GDP per capita. In order to detect multicollinearity, we calculated the variance inflation factor (VIF = 5.57) and tolerance (0.179). These statistics show that multicollinearity is not a problem for this study.

¹⁵ As a robustness check and to show how the results change when we consider property damage as the outcome measure, we evaluated the impact of natural disasters on deaths due to transnational terrorism. Our results support the findings of Berrebi and Ostwald when the focus is on fatalities. Specifically, we find that transnational attacks against the government increase with a shorter lag (one year) following the disaster. Since the independent variable in this case is a count variable, we use FEs negative binomial regression models as is commonly done in the literature. We have included the results table in the Appendix 1 (Table A7).

¹⁶ As an additional robustness check, we evaluated the impact of the damage distribution on the qualitative nature of our results. Specifically, we developed quantile regression models (see Appendix 1, Table A9) to analyze the relationship between log-transformed transnational terrorism damage and lagged natural disaster damage. Our findings indicate that our original results of a statistically significant positive relationship between transnational terrorism damage and lagged natural disaster damage hold at the 0.75 quantile.

¹⁷ The results of the first-stage estimation are reported in Table A2.

¹⁸ As a robustness check and to show how the results change when we consider property damage as the outcome measure, we evaluated the impact of natural disasters on deaths due to domestic terrorism. Our results support the findings of Berrebi and Ostwald when the focus is on fatalities. Specifically, we find that domestic attacks against the government increase after a two-year lag following the disaster. Since the independent variable in this case is a count variable, we use FEs negative binomial regression models as is commonly done in the literature. We have included the results table in the Appendix 1 (Table A8).

¹⁹ As an additional robustness check, we evaluate the impact of the damage distribution on the qualitative nature of our results. Specifically, we developed quantile regression models (see Appendix 1, Table A10) to analyze the relationship between log-transformed domestic terrorism damage and lagged natural disaster damage. Our findings indicate that a statistically significant positive relationship holds between domestic terrorism damage and lagged natural disaster damage at the 0.75 quantile. This finding while contradictory to our original results is understandable when one considers the economic damage that is associated with major or catastrophic disasters (when one considers 0.75 quantile) and the significant influence these rare yet economically profound events can have on ensuing terrorism attacks and resulting property damage.