Terrorist choice: a stochastic dominance and prospect theory analysis

Abstract

The paper explores terrorist choice by applying two well-known theoretical frameworks: stochastic dominance and prospect theory (PT). We analyse each pair of attack methods that can be formed from the RAND-MIPT database and the Global Terrorism Database. Instances of stochastic dominance are identified. PT orderings are computed. Attention is accorded to the identification of ‘trigger points’ and the circumstances that may lead to an increased likelihood that a terrorist will select an attack method associated with a higher expected number of fatalities, i.e. a potentially more damaging attack method.

Keywords:

• Terrorist choice
• Prospect theory
• Stochastic dominance
• Attack methods
• Trigger points
• Fatalities

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ When is something more risky than before? And when is something more risky than something else? There is not a straightforward answer to these questions and, unfortunately, no definitive ‘proof’ that one definition of ‘riskier’ is better than another definition (Rothschild and Stiglitz 1970, 226–227). According to Rothschild and Stiglitz (1970), possible answers to the question include (1) Y is riskier than X when Y is equal to X plus some noise; (2) Y is riskier than X if every person who is risk-averse prefers X to Y; (3) Y is riskier than X if Y has more weight in the tails of its probability distribution; (4) Y is riskier than X if Y has a greater variance or standard deviation than X.

² An example is an attack on an embassy.

³ Assassination is an exception. However, the outcomes (fatalities) per ‘assassination’ attack are normally distributed. That is, the normality assumption cannot be rejected for ‘assassination’. The Shapiro–Wilkes statistic is 0.245.

⁴ The authors thank the referee for this suggestion. These are cases where fatalities are zero, not because the operation failed but because the operation was pre-empted. Such cases, if included in the data-set, would reduce the mean or expected number of fatalities for planned attacks. The potential effects on the preference orderings are unclear. If each type of attack method is subject to a similar proportion of pre-emptions, the relative number of expected fatalities may not be significantly impacted.

⁵ See Phillips and Pohl (2014).

⁶ Its cumulative distribution lies to the right of its alternative past the point at which the two cumulative distributions intersect.

⁷ It is implicitly assumed, of course, that the parameter values determined by Tversky and Kahneman (1992) apply in this context. Clarifying this point is a matter for future research.