The Softest of Targets: A Study on Terrorist Target Selection

Abstract

What factors may lead a terrorist organization to attack undefended and unaware civilians—or as we style them, “soft targets”? We examine two distinct processes in choosing to attack soft targets: the one-time decision to begin attacking soft targets and the continued use of violence against such targets. Our analysis points to ideology—specifically religion—as the decisive factor in turning to target civilians, while organizational factors such as group size and network centrality are related to the number of incidents perpetrated.

KEYWORDS:

• Terrorism
• targeting
• social networks
• ideology
• quantitative modeling

Notes

∗p <.05;

∗∗p <.01.

^†All Vuong statistics have p-values between 0.1522 and 0.1297.

1. MIPT generally does not distribute its dataset—see http://www.tkb.org/Faqs.jsp#Incident_Dataset. However, we have obtained permission to extract and use data from the online pages.

2. There is a growing body of literature, particularly in the area of participation, which underlines the importance of social networks for social movements. See, for instance, Dixon and Roscigno (2003); Khagram, Riker, and Sikkink (2002); and Klandermans and Oegema (1987).

3. Portions of the data description have also appeared in Asal and Rethemeyer, 2008.

4. Incidents may be extracted by group using this URL: http://www.tkb.org/IncidentGroupModule.jsp. Note: the figures obtained at a later date may not match these, as MIPT is continuing updating this database.

5. The authors have no affiliation with MIPT and played no role in the original collection of data included in the MIPT dataset.

6. We are indebted to Brian Lai of the University of Iowa for this insight.

7. Impressionistically, it appears that many of the missing groups are new organizations that have sprung up in Iraq since the U.S. military action began.

8. For a more complete description of how ideology is coded in the MIPT database, see http://www.tkb.org/DFI.jsp?page=method

9. Throughout we refer to “the decision not to kill” because ZINB models the zeros in the data—those organizations that choose not to kill.

10. Stata 9.0 includes two “canned” approaches to exposure control in all count models: one where exposure is modeled as the natural log of units exposure with the coefficient constrained to be 1.0 (the “exposure” option); the other being when exposure is modeled as a linear variable, also with the coefficient constrained to be 1.0 (the “offset” option). Our tests found that the assumed constraints on the coefficient were not supported by the actual data. Thus we opted to control for exposure through a standard independent variable.

11. Stata includes several optimization algorithms, including their modified Newton-Raphson (NR) algorithm, the Berndt-Hall-Hall-Hausman (BHHH) algorithm, the Davidon-Fletcher-Powell (DFP) algorithm, and the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm. Stata also has a maximization option called “difficult” for likelihood functions with non-concave areas. None of the algorithms or options were able to find a solution. This was also true when organizational age was entered as a quadratic rather than a linear variable. Thus our conclusion that POLITY2 was difficult to model because it has relatively little relationship to the dependent variable.

12. The easiest explanation—that older organizations have more chances to attack soft targets—should be precluded by the exposure control we implemented.