Methodology

A. Empirical specification

Following the example set by Blomberg, Hess, and Orphanides (2004), we specified three estimation equations:

Equations (1)–(3) examine the determinants of the income per capita growth rate, the investment share, and the government spending share (G/GDP), respectively, where i = 1, …, N represents the country and t = 1, …, T indexes the time period. The independent variable terror is a measure of terrorist attacks, which is either transnational terrorist incidents or all incidents (both transnational and domestic), internal denotes internal conflicts, and external measures external conflicts. βs, γs, and ϕs are regression coefficients, while the remaining Greek letters indicate the disturbances. In Equations (1)–(3), each disturbance consists of three components: unobservable (time-invariant) country effect, subscripted with i; unobservable time effect, subscripted with t; and the classical random error. For example, α_i is the unobservable country-specific effect, λ_t is the unobservable time-specific effect, and ν_it is the stochastic error term. In Equation (1), political violence variables are added to the main determinants of economic growth. Equations (2)–(3) identify the potential channels through which the political violence variables slow down economic growth by either reducing investments or augmenting government spending through security expenditures.²

Our empirical approach was based on the behavior of the unobservable effects (see, e.g., Baltagi 2005). We performed tests to examine the presence of unobservable effects. If the effects were not present, we preferred ordinary least squares (OLS), which is consistent and efficient. If, however, unobservable effects were present, we applied the one-way fixedeffects estimator when there were only country effects (time effects), and the two-way fixed effects estimator when evidence suggested both time and country effects. In choosing the fixed effects method, we assumed that the unobservable effects were fixed parameters for estimation. Alternatively, we could regard the effects as random and apply generalized least squares (GLS), which implied that the unobservable effects were part of the disturbance and therefore independent of the observable explanatory variables. We implemented the Hausman test to investigate the correlation between the effects and the regressors. If the Hausman test supported independence between the observable regressors and the unobservable effects, we performed the random-effects estimator, in addition to the fixedeffects model, for sensitivity analysis.

B. Data

Our data were initially drawn from four sources: Penn World Table Version 6.2 (Heston, Summers, and Aten 2006), International Terrorism: Attributes of Terrorist Events (ITERATE) (Mickolus et al. 2006), Global Terrorism Database (GTD), and the UCDP/PRIO Armed Conflict Dataset, Version 4–2007 (Gleditsch et al. 2002). We constructed an unbalanced dataset for 42 Asian countries for 1970–2004.³ The sample countries included Asian countries for which we can get both macroeconomic and political violence data and, as such, include the main developing and developed countries within the region. We concluded the sample period at the end of 2004 since this is the last year of terrorism data for GTD, while we began at 1970 to increase the number of countries with macroeconomic data. Moreover, terrorism data for 1968–1969 is rather spotty because terrorism datasets were only started in 1968. By 1970, these datasets were better able to track incidents.

Macroeconomic variables—real GDP per capita in constant dollars, economic openness, investment share of real GDP, population, and government expenditure share of real GDP— were obtained from the Penn World Table Version 6.2. Based on data on real GDP per capita, we computed the growth of real GDP per capita as the difference in the log (ln) of GDP per capita of subsequent years. We also calculated the log of the index of country i’s openness at time t, which we denote by ln (open_it)

ITERATE and GTD are used to construct two alternative measures of terrorism. The number of transnational terrorist incidents per million persons (terr iter) indicates the level of transnational terrorist incidents normalized by the venue country's population. Five incidents in a year in a country with a population of 300 million should, ceteris paribus, have less of an economic influence than the same number of incidents in a country with a tenth of the population. We generally favored a terrorism measure where the number of events was used rather than a dummy that merely signals one or more events in a given year, since the latter does not indicate the prevalence of terrorism. Similarly, we assigned a measure (terr gtd) for all terrorist events—domestic and transnational—per million persons based on GTD data. The latter does not distinguish between domestic and transnational events. Moreover, we could not properly isolate domestic terrorist events by differencing ITERATE and GTD observations in a given country and time period, because these datasets rely on different sources and judgment calls. An event in ITERATE may or may not be in GTD. Finally, we had two indicator variables for conflicts. Based on UCDP/PRIO Armed Conflict Dataset, external was 1 if the country experienced an international conflict (interstate or extrastate) in a given year and 0 otherwise; similarly internal was 1 if the country experienced an internal conflict (internal or internationalized internal) in a given year and 0 otherwise. Table 1 [ PDF 12.6KB | 1 page ] summarizes information on data and sources.

Summary statistics are displayed for our variables in Table 2 [ PDF 24.1KB | 1 page ]. For 1970–2004, income per capita for a sample country grew on average by 2.3%. Investment share was about 15.5% of GDP, while government spending share was about 22.5% of GDP. On average, a sample country experienced 0.055 transnational terrorist incidents per million persons, while it experienced 0.369 terrorist incidents (of all kinds) per million persons. In any given year, external conflict was present in about 8% of the sample countries, while internal conflict was present in about 22% of the sample countries. This relative incidence of conflicts agrees with the impression gained from Figure 1.

ITERATE records essential information about transnational terrorist events such as its date, country location (start and end location), incident type, and so on (Enders and Sandler, 2006a). GTD does the same for all terrorist events. Both datasets rely on media accounts. ITERATE had a large reliance until 1996 on the Foreign Broadcast Information Service (FBIS) Daily Reports, which survey a couple hundred of the world's newspapers. ITERATE has continued to draw information from major newspapers, wire services, and other media outlets since 1996. ITERATE excludes not only attacks directed at combatants or occupying armies, but also attacks associated with declared wars or guerilla warfare.

We performed some cleanup of ITERATE data because it lists both a start and an end country for an incident. Typically, the start and end country locations were the same, but they differed for a small percentage of incidents. Our concern was when the start or end country lies outside of Asia—our region of interest. For 90 events, the incident started in an Asian country but ended outside of Asia. After reading the description of these 90 events, we determined that 16 of these incidents really took place in the Asian country of origin (e.g., a plane hijacked at an Asian airport). The other 74 events really took place outside of Asia and were dropped. There were 16 terrorist events that ended within Asia but started outside of Asia (e.g., letter bombs mailed from Europe to an Asian country). Eight of these observations were kept after further investigation. Finally, 47 incidents started in one Asian location (e.g., India) and concluded in another Asian location (e.g., Pakistan). After consulting the incidents' descriptions, 43 of them were assigned to the start country and the remaining four were assigned to the end location. Also, we did not include “terrorist events” coded as arms smuggling (incident type 22), since the use of these arms in a specific terrorist incident was not indicated. Moreover, the arms may be intended for purposes other than terrorism—e.g., war or crime.

GTD consists of event data on domestic and transnational terrorist incidents for 1970–2004. Owing to its inclusion of domestic events, GTD includes many more incidents than ITERATE. Like ITERATE, GTD requires that incident perpetrators seek a political, economic, religious, or social goal to qualify as a terrorist event. GTD also excludes actions associated with internal or external wars. To construct GTD, the National Consortium for the Study of Terrorism and Responses to Terrorism (START) Center at the University of Maryland first obtained the data from the Pinkerton Global Intelligence Services, which recorded observations on terrorist events based on wire services, government reports, and major international newspapers. These data were cleaned and updated by START. Apparently, the data for 1993 were lost (fell off a truck) and, so, are not currently in GTD. The GTD data are divided into two datasets: GTD1 covers 1970–1997, while GTD2 covers 1998–2004.⁴ GTD2 has not yet been “cleaned”; hence, we cleaned GTD2 of duplicate events and non-terrorist events for our study. In so doing, we combined GTD1 and GTD2 to provide a continuous dataset for 1970–2004, excluding 1993.

Figure 3 [ PDF 15.2KB | 1 page ] displays the annual number of terrorist incidents for 1970–2004, where the solid line represents all terrorist events from GTD and the broken line represents only transnational events from ITERATE. The left-hand scale is for GTD, while the right-hand scale is for ITERATE. The break in the GTD plot at 1993 corresponds to the missing year of data. There are some noteworthy observations. First, GTD typically recorded over 10 times as many events each year as ITERATE so that domestic events swamp transnational events in number. Second, the shapes of the two time series were surprisingly similar, which means that ITERATE may capture the rises and falls in terrorism even though transnational terrorism is only a small fraction of all terrorist events. Third, both time series suggested cycles in terrorist events (see Enders and Sandler 2006a), which is somewhat clearer for the ITERATE time series. Fourth, there are more transnational terrorist incidents and a greater variability for 1986–1994, which corresponds to the rise of Islamic fundamentalist terrorism as the dominant influence of transnational attacks. Fifth, in the two years following 9/11, there was a big increase in transnational terrorist attacks in Asia, which corresponds to geographical transference, identified by Enders and Sandler (2006b). Figure 4 [ PDF 14KB | 1 page ] depicts transnational and all terrorist events in terms of the annual number of events per million persons. With this normalization, the two series appear even more in sync.

Although not shown in the figures, the geographic distribution of terrorism is of interest. In terms of transnational terrorist incidents for 1970–2004, ITERATE ranks the top fifteen venues in descending order as follows: the Philippines; Pakistan; India; Cambodia; Afghanistan; Republic of Korea; Indonesia; Thailand; Japan; Australia; Tajikistan; Malaysia; Sri Lanka; Taipei,China; and People's Republic of China (see Appendix). In terms of transnational terrorism events per million persons, the top fifteen hotspots are: the Solomon Islands; Tajikistan; Fiji Islands; the Philippines; Singapore; Afghanistan; Cambodia; Georgia; Lao People's Democratic Republic; Australia; Malaysia; Hong Kong, China; Sri Lanka; Pakistan; and Republic of Korea. Some sparsely populated countries on the second list do not appear on the first list, while some populous countries (e.g., India, Indonesia, and People's Republic of China) on the first list do not appear on the second list. Nine countries show up on both lists. For GTD data, eleven of the top fifteen countries experiencing both forms of terrorism were also among the top fifteen venues for transnational terrorism (see Appendix). This suggests that domestic and transnational terrorism are correlated for many countries.

Download this Discussion Paper [ PDF 178.3KB| 32 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.