Abstract
We assess Dugan and Chenoweth’s Rational Choice-based argument regarding moderating effects of indiscriminate conciliatory state actions on levels of terrorist attacks in Israel-Palestine, utilizing data drawn primarily from declassified security records on Israeli state actions during the First Intifada (1987–1992). This type of data source, we argue, contains a more accurate ratio of conciliatory to repressive actions than Dugan and Chenoweth’s media-based data, given state authorities’ attempts at concealing repressive actions while publicizing conciliatory actions during times of intense conflict. We discuss differences in results—including, most centrally, no support for the hypothesized effect of conciliatory state actions—highlighting the theoretical payoffs of examining the Political Process–related factor of varying levels of state control over the political environment.
In an innovative article, Dugan and Chenoweth (2012) argue that states seeking to design effective counterterrorism strategies should consider moving beyond the sole reliance on repression as a means of reducing the expected utility of engaging in terrorism, to offer conciliatory actions as a means of raising the expected utility of abstaining from terrorism. They argue further that offering indiscriminate conciliatory actions as rewards produces a larger decrease in terrorism than more discriminating rewarding measures. Moreover, indiscriminate conciliatory actions produce a larger decrease in terrorism than either discriminate or indiscriminate repressive measures. The rational choice logic underlying these arguments is as follows: While not necessarily offsetting grievances, offering conciliatory actions as rewards indiscriminately would improve upon the status quo of terrorists’ constituents. This in turn would help state authorities win the legitimacy battle, undermine terrorist organizations’ resources, and hence their capacity to inflict harm (Dugan and Chenoweth, 2012: 602–603).
Dugan and Chenoweth base their arguments on a cross-national dataset on Government Actions in Terrorist Environments (GATE) which, combined with data from the Global Terrorism Database (GTD), is applied to the Israeli-Palestinian environment between June 1987 and December 2004. Statistical analyses of the relationship between Israeli state actions and Palestinian attacks throughout the entire period, along with an analysis of three sub-periods, the authors name tactical regimes—the First Intifada (1987–1993), the Oslo Lull (1993–2000), and the Second Intifada (2000–2004)—yields revealing results. Specifically, repressive actions are either unrelated to the frequency of terror attacks or are related to subsequent increases in terror; conciliatory actions—most meaningfully indiscriminate ones—produce decreases in terrorism, depending on the tactical regime.
Impressed by Dugan and Chenoweth’s study, we are nonetheless concerned about their sole reliance on international media sources. We share the authors’ point that the chances of the Israeli-Palestinian conflict being unrepresented in international news coverage are slim given its centrality. We also agree that regardless of the open sources one uses, it is inevitable that some events will miss public scrutiny, hence “no reason to believe this issue would be systematic” (Dugan and Chenoweth, 2012: 607). We have a good reason, however, to believe that a systematic bias may exist in the ratio of reported conciliatory actions to repressive actions. Given states’ interest in tightly regulating, if not controlling the flow of information during times of conflict, and the international scrutiny involved in high-profile cases, it is plausible to expect state authorities to invest a great deal of effort in concealing certain actions while publicizing others. Consequently, we expect media coverage to reflect this bias given the tight restrictions and pressures journalists and editors face, resulting in a misguided image of the conflict.
While supplementing data from local media, either Israeli or Palestinian can reduce this bias, access to state archives is more promising in this respect. The newly developed First Intifada Dataset (FID)—based primarily on field security reports of the Israeli Defense Force (IDF)—contains a strikingly different ratio of conciliatory to repressive actions compared to GATE’s (approximately 1:16 and 1:3, respectively), and therefore offers a unique opportunity to assess the robustness of Dugan and Chenoweth’s findings, as applied to one of their tactical regimes: The First Intifada. We report on results from a partial replication of Dugan and Chenoweth’s study using our dataset (i.e. limited to the First Intifada tactical regime and the independent variables) and discuss the theoretical implications of differences we find in results. Analytically invested in a Political Process Theory, we discuss how attention to varying levels of state’s control over the political environment, particularly the capacity, and ability to implement repressive and conciliatory actions, enhances our understanding of which type of state actions has an effect on insurgent attacks and the nature of that effect. An implication for policymakers seeking to design an effective strategy for reducing terrorism is also discussed. Before doing so, we briefly describe the First Intifada Dataset and the replication process.
The First Intifada Dataset
The FID represents a sizable compilation of contentious events, related explicitly and directly to the First Intifada. It contains daily observations of an exceptionally wide-ranging catalog of types, modes, and forms of contentious events (n = 23,266) involving the multitudes of parties and actors in the cycle of contention between January 1987 and January 1993. All events were gathered from non-media-based sources, inclusive of Israeli state records and IDF field reports, non-state Israeli, Palestinian, and foreign sources. A more detailed discussion of the rationale, structure, and particularities of FID is found elsewhere (Alimi et al., 2019).
Replicating Moving beyond Deterrence
The replication process involved the reconstruction of Dugan and Chenoweth’s data, study design, and analyses. Beginning with the dependent variable—Palestinian terrorist attacks—we followed the authors in relying on GTD, 2010 version. A careful reading of the authors’ materials regarding attacks’ inclusion criteria (i.e. an attack perpetrated by Palestinians, involving at least one Israeli target in Israel or the Palestinian territories), assisted in filtering GTD cases and reaching their sample of attacks. Specifically, we reached 491 attacks between June 1987 (GATE’s starting point) and December 1992 and, given the lagged nature of the dependent variable and inclusion of control variables (see Table 1), 471 attacks for the 61 months covering December 1987 to December 1992, on which their analysis was conducted.
Comparison of Model Specification.
GTD: global terrorism database.
Moving to the independent variables, the replication required recoding existing FID variables as well as coding new ones. Here, too, we first consulted GATE’s definition of government actions and the rules for coding these actions as repressive, conciliatory, and neutral or ambiguous. Specifically, Dugan and Chenoweth define Israeli government action as any type of action, broadly defined to include verbal and physical, actual and intended actions, toward Palestinian terrorists and their constituents in Israel and the Palestinian territories. To determine if Israeli government actions are conciliatory or repressive (or neutral and ambiguous), from the Palestinian perspective, the authors established a seven-point Conciliation—Repression Scale, modeled after Goldstein’s (1992) scale, that “features distinctions in the intensity of the action as well as its relative placement of the action on a conciliation-repression spectrum” (Dugan and Chenoweth, 2012: 608).
Coding Israeli government actions in FID according to the authors’ Scale involved three adjustments to our original codes. Following Dugan and Chenoweth’s attempt to single out specific actions, the first adjustment involved identifying FID events that included two or more actions that took place in the same place and the same time, originally coded as a “combination,” and unpacking them according to their constitutive actions. Second, because Dugan and Chenoweth’s definition of government actions included intentions regarding future actions, a type of action originally excluded from FID, all materials were searched a second time to identify and include those actions. Third, to match GATE’s inclusion of political statements, procedures, and declarations as repressive or conciliatory actions, events originally coded in FID as institutional were re-coded. The result was 4826 repressive actions, 244 conciliatory actions, and 59 neutral/ambiguous actions (not included in the analysis).
The last adjustment regarded new codes, reflecting the target type of the action. Except for actions coded as neutral/ambiguous, all other repressive state actions were coded according to whether they attempted to single out those “suspected” parties (i.e. discriminate) or directly affected those people not suspected of involvement in terrorist activity (i.e. indiscriminate). Respectively, conciliatory actions were coded according to whether they rewarded only those parties relevant to the lack of terrorist activity (i.e. discriminate) or the entire population (i.e. indiscriminate). 1 All the data were organized as count data and aggregated at the monthly level.
Finally, because our replication is based on the First Intifada tactical regime, the issue of comparing across different tactical regimes due to the different mix of conciliatory and repressive tactics during each, as observed by Dugan and Chenoweth, was excluded as a vector from all statistical models (see Table 1). We also excluded the issue of controlling for differences in data collection techniques in GTD before and after 1998 (i.e. GTD2). This exclusion is due to the different organizations responsible for data collection (i.e. Pinkerton Global Intelligence Service before 1998), coupled with the fact that since 1998 data were collected retrospectively, inevitably undercounting attacks that were documented only in sources that are no longer available (Dugan and Chenoweth, 2012: 607). All other vectors, inclusive of four attacks-based control variables, were included.
Comparing Yields
Dugan and Chenoweth used Negative Binomial Regression (NBR) and Generalized Additive Models (GAM) to test five hypotheses regarding the effect of Israeli state actions on Palestinian terrorist attacks. These hypotheses were organized according to three statistical models, each measuring a particular dimension of state actions. Specifically, the first model tests the hypothesis that any Israeli action will have a moderating effect on Palestinian attacks; the second disaggregates Israeli actions, testing the hypotheses that repressive or conciliatory actions unto themselves will moderate terror attacks; and, the third disaggregates further, testing the hypotheses that indiscriminate repressive actions will lead to increases in attacks, while indiscriminate conciliatory actions will have the greatest moderating effect. We followed suit and ran the same statistical tests and specifications, 2 using FID data for the various state action variables, and R (version: 3.3.1) (see Supplemental Appendix).
Beginning with descriptive statistics, the values presented in Table 2 highlight not only the quantitative differences between GATE and FID but also the differences between FID’s ratio of conciliatory to repressive actions and GATE’s. Whereas the overall number of conciliatory actions and the monthly average of such actions are quite similar, the same cannot be said about repressive actions, with FID showing five times higher average of repressive actions compared to GATE. In addition, unlike GATE, FID data suggest that Israeli repressive actions were almost three times more likely to be discriminate.
Means, Standard Deviations, and Proportion of Zeros: First Intifada (n = 61).
SD: standard deviation.
Turning to inferential statistics, we see that the differences in quantity have qualitative implications. The comparisons of both NBR and GAM results, presented in Table 3 and Figure 1 respectively, are revealing. Regarding model 1, testing the hypothesis that any Israeli action would lead to fewer terrorist attacks by Palestinians, we find no support and note the positive coefficient—similar to Dugan and Chenoweth. Some evidence of the positive yet statistically null relationship between actions and attacks is offered by the GAM results (Figure 1). However, in sharp contrast to GATE-based results, FID-based partial predictions suggest hardly any effect of past month actions on current month attacks, regardless of the level of state actions.
Negative Binomial Coefficients and (SE), First Intifada, n = 61.
SE: standard error.
p ⩽ 0.05; ** p ⩽ 0.01 (two-tailed test).
Partial Predictions Summary of State Actions from Past Month on Terrorist Attacks in the Current Month.
Differences between Dugan and Chenoweth’s findings and ours become even more marked in models 2 and 3. It seems that the key difference between GATE-based results and FID-based results relates to the hypothesized effect of conciliatory actions. Dugan and Chenoweth hypothesize that conciliatory actions lead to fewer attacks (H2, model 2) and that this decrease will be larger when actions are applied indiscriminately (H5, model 3). These hypotheses receive partial support: it is discriminate not indiscriminate conciliatory actions that cumulatively lead to a decrease in subsequent attacks. In contrast, our results indicate no support whatsoever. Specifically, while Dugan and Chenoweth’s expected pattern exists, the relationship between conciliatory actions and attacks is statistically null regardless of whether state actions are discriminate or indiscriminate. The partial predictions in Figure 1 offer important insight: while conciliatory actions do have an effect, it is neither pronounced nor unambiguous as indicated by GATE-based results.
A final note relates to the Rational Choice-based expectation that more repression leads to less terrorism (H3) unless it is applied indiscriminately (H4). Although in both sets of data the results are not statistically significant, with noticeably low coefficients, it is nevertheless worth noting the opposite dynamics when comparing FID-based partial predictions to GATE-based ones. In contrast to GATE-based results, there is some evidence of an increase in the number of attacks in the fairly low levels of state repressive actions. When unpacked according to discriminate and indiscriminate actions, it seems that in the higher level of repressive discriminate actions—indicated by a fairly narrow point-wide confidence interval bands and representing a fair amount of months—we find some evidence of decreased terrorist attacks. Regarding indiscriminate repressive actions, it seems that in the mid-level state actions bring about some increase in attacks (i.e. backlash effect). We relate to the lack of cumulative effect of attacks from previous month below.
Discussion
How can we account for the lack of a meaningful and unambiguous moderating effect of conciliatory state actions—particularly indiscriminate ones—on levels of terrorist attacks in Israel-Palestine, as hypothesized and substantiated by Dugan and Chenoweth? As terrorists, just like any dissident actor, calculate costs in relation to benefits and, consistent with the logic that guided Dugan and Chenoweth, rewards become meaningful in relation to penalties, the different ratio of conciliatory to repressive actions in FID compared with GATE is expectedly consequential. Thus, it is likely that the expected utility of abstaining from terrorism for Palestinian terrorists’ constituents will be low when facing such a massive amount of repressive actions for any single conciliatory action, in which case Israeli authorities will be on the losing end in the legitimacy battle. It is also possible that a decrease in Palestinian terrorist attacks was the result of the considerable amount of Israeli repressive actions in and of themselves (i.e. raising the costs of engaging in terrorism), regardless of conciliatory actions of any type. Yet, Israeli repressive actions do not emerge as a meaningful predictor of levels of Palestinian terrorism either.
Two related answers we put forth as directions for further exploration are aggregation problem and inappropriate time interval. Regarding aggregation problem, it is true that higher temporal levels of aggregation help reveal an increase or decrease in the number of certain sub-categories of events, which constitute only a fraction at lower temporal levels, facilitating the detection of macro-level patterns. Yet, it is also true that higher temporal levels of aggregation are associated with the loss of information regarding patterns and dynamics unfolding at lower temporal levels (Hannan, 1985). It is not unlikely that aggregating FID’s data at its original day level would facilitate the detection of meaningful patterns which are missed once data is aggregated at the monthly level similar to GATE’s. Attempts at re-aggregation when possible, as is the case with GTD (used for constructing the dependent variable), may involve flattening out potentially meaningful variations in cases where the granularity of the data is considerably low and lack of data at a given time unit is problematically treated either as representing no activity or as a missing value. It is plausible to suggest that given the high level of control Israel managed to exert on the work of domestic and foreign journalists and, more broadly, the flow of information from and into the arena of contention during the first Intifada (Falloon, 1986; Wolfsfeld, 1997), GTD likely captures only a fraction of information on the level and type Palestinian terrorist attacks.
Regarding inappropriate time interval, while choosing the appropriate interval at which actors respond to one another is never easy, and while such a decision is typically theory-driven, the decision should also be based on the features of the phenomenon under investigation and be empirically informed (Shellman, 2004). The possibility that we may be losing information regarding patterns and dynamics unfolding at lower temporal levels becomes even more plausible given the characteristics of cycles of contention which are highly intensive, developing their own momentum and logic which include—in addition to rational calculations—factors such as moral outrage, opportunities, and vengeance (Brym and Araj 2006; Della Porta, 2011; Tarrow, 2011). These features are particularly salient in, and applicable to ethnonational cycles of contention, such as the First Intifada (Hiltermann, 1991; Kuttab, 1988; Schiff and Ya’ari, 1990) and other popular struggles for national liberation (e.g. Cyprus, Spain, and Northern Ireland). These are typically mass-based, broad in scope, and remarkably intensive, such that actions of both the state and insurgents are rapidly surpassed, at times in a matter of days or even hours. Moreover, these cycles of conflict and contention, labeled by some as irregular wars (Kalyvas, 2006) or national insurgencies (Wood, 2003)—tend to go beyond matters of the legitimacy of authorities and state policy to issues of self-determination and control over land and population.
The theoretical implications of all this regard the importance of taking into consideration the changing larger political contexts that shape rational calculations and influence the capacity not only of insurgent terrorist groups, as Dugan and Chenoweth do, but also of state authorities. Specifically, we should consider levels of state control over the political environment, which is a function of the set of opportunities and constraints state authorities face internally and externally when coping with domestic challenges (McAdam, 1999; Tarrow, 2011) as well as the implementational capacity to carry out repressive and concessive measures (Gurr, 1970; Rucht, 1996). Translated into the ability to initiate and control events, to regulate the flow of information and gather intelligence, or to mobilize elite support and resources, material and nonmaterial alike, these aspects play an important role in shaping contentious interactions. Relatedly, and by extension, they offer an important elaboration of the relationship between state actions and dissident terrorist attacks.
To illustrate in concluding, it is possible that the reason for the overall moderating effect of discriminate (not indiscriminate) repressive actions on terrorist attacks found in the GAM analysis relates to Israel’s intimate knowledge of key Palestinian activists involved in terrorism, and the ability to expose and neutralize them. Following the same logic, the lack of a pronounced moderating effect of Israeli indiscriminate conciliatory actions on Palestinian terrorism may suggest something about Israel’s lack of propensity to engage in such actions and, equally important, ability to implement them. In addition to Israel’s traditional intransigence regarding Palestinian demands, expressed for example in the systematic refusal to recognize the existence of a Palestinian people or to allow the meeting between Israelis and PLO members, it is important to note the meaningful challenges faced by Israeli authorities when attempting to promote and carry out more conciliatory measures vis-à-vis the Palestinians. Internally, the primary and most determined challenge involved ultra-radical Jewish settler groups that increasingly took the law into their own hands in the face of what they considered insufficient safety measures provided by the IDF. Along with growing instances of settler militias sabotaging the implementation of collective concessions (e.g. erecting checkpoints and roadblocks after being lifted by the IDF), any sign of willingness by Israel to offer concessions to the Palestinian population saw a surge of settler violence directed against Palestinian property and residents, at times also against IDF forces who came to the scene to control the situation. A case in point is the dramatic increase in settler violence, following the publication in early April 1989 of Israeli Prime Minister Shamir’s plan of free elections in the Occupied Territories, which lasted for weeks and resulted in the death of several Palestinians as well as the injury of an Israeli soldier (Zertal and Eldar, 2004). Interestingly, the wave of Jewish settler violence consolidated despite the immediate and unequivocal rejection of Shamir’s plan by PLO organizations and institutions based in Tunisia (Jerusalem Media and Communication Center (JMCC), 1991), who constituted a second source of challenge to Israel’s ability to implement conciliatory measures. Notwithstanding tensions and discords between the Territories-based and Tunisian-based Palestinian leadership, the latter’s status and power were robust enough to prevent any peace initiative from materializing without its explicit endorsement. Thus, even when Israel was willing to engage in conciliatory initiatives, as was the case with new peace initiatives proposed by East Jerusalem Palestinian public figures or with Secretary of State James Baker’s peace plan, following “Desert Storm,” there was little chance for any such initiative to move forward without the approval of the Tunisian-based PLO leadership.
Regarding policy implications, finally, it is possible to offer an important corrective to the Rational-Choice-based counterterrorism strategy offered by Dugan and Chenoweth. Granted an interest of state authorities to raise the expected utility of nonviolent behavior, the implication is that in cases of high levels of state control, conciliatory measures should be discriminate, directed at those key-players and central network cliques in terms of popularity and legitimacy among the insurgent population. Under these circumstances, indiscriminate conciliatory actions are more likely to be perceived by the insurgent organizations and their constituents as attempts to “go over their heads.” Whereas the former strategy can potentially prompt turf-fighting between insurgent organizations, leading to occasional deflection against the state, the latter strategy is likely to have unintended consequences of further reinforcing support for, and legitimacy of the insurgent organizations, and, consequently, their capacity to inflict harm. These and other leads are worth further explorations based on a more comprehensive and representative set of data regarding both state actions and insurgent attacks.
Supplemental Material
sj-docx-1-psw-10.1177_14789299211014011 – Supplemental material for Rethinking Moving beyond Deterrence: A Partial Replication Study
Supplemental material, sj-docx-1-psw-10.1177_14789299211014011 for Rethinking Moving beyond Deterrence: A Partial Replication Study by Eitan Alimi and Gregory Maney in Political Studies Review
Footnotes
Acknowledgements
The idea for a replication study came up in late 2016, just before my close friend and collaborator Greg Maney suffered an incurable violent recurrence of brain cancer. I thank Nachman Ben-Yehuda, Yagil Levy, Charlotte Ryan, and Sidney Tarrow for offering constructive criticism on a previous version. Special thanks to Micha Mandel for offering invaluable assistance and guidance with the statistical analysis. I also thank Laura Dugan for assisting me with the replication of her and Erica Chenoweth’s study.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: The research is based on funding from U.S. National Science Foundation (0958743), Israel Science Foundation (1232/14).
Supplementary information
Additional supplementary information may be found with the online version of this article.
Notes
Author Biography
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
