Previous Work

How does living under the threat of terrorism affect voting? This question is related to three strands of extant literature: (1) the effect of violence on political behavior, (2) the strategic nature of terrorist - government interactions, and (3) retrospective voting and issue ownership.

Existing studies explore various effects of violence and terrorism. While Pape (Reference Pape2003) suggests that terrorism, and in particular suicide bombings, are effective against democratic voters who pressure their governments to provide concessions to terrorists, Abrahms (Reference Abrahms2006, Reference Abrahms2012) questions the overall effectiveness of terrorism as a political strategy. Many papers assert that terrorism and counterterrorism lead to cycles of violence (Haushofer, Biletzki, and Kanwisher Reference Haushofer, Biletzki and Kanwisher2010; Jaeger and Paserman Reference Jaeger and Paserman2008). However, others provide evidence that some forms of counterterrorism might be effective in reducing further insurgent violence (Lyall Reference Lyall2009), or in curbing future terrorist attacks (Benmelech, Berrebi, and Klor Reference Benmelech, Berrebi and Klor2010a ).

Looking at the Israeli-Palestinian context, Berrebi and Klor (Reference Berrebi and Klor2008) use the variation in Palestinian suicide terrorism across time and space to argue that attacks shift Israeli voters toward right-wing parties that are less supportive of granting political concessions to Palestinians. They estimate that one suicide attack within three months before an election increases right-wing parties’ vote-share in the targeted locality by 1.35 percentage points. Moreover, suicide attacks increase the support for the right-wing bloc beyond those localities in which they take place. Gould and Klor (Reference Gould and Klor2010) use a similar identification strategy to make a more nuanced argument: while terrorism increases the vote-share of right-wing parties, it also shifts the entire political spectrum to the left, thereby making voters and parties more supportive of accommodating Palestinian political demands. Yet both Berrebi and Klor (Reference Berrebi and Klor2008) and Gould and Klor (Reference Gould and Klor2010) only examine fatalities from suicide terrorism. However, fatalities from suicide terrorism could influence electoral behavior through one of two channels. (1) Casualties caused by terrorism or war can have a strong effect on voting. Previous research shows that support for war, and the incumbents in office, are sensitive to fatalities (Kibris Reference Kibris2011) and the context in which they occur (Gelpi, Feaver, and Reifler Reference Gelpi, Feaver and Reifler2006). Additionally, the political psychology literature suggests that exposure to violence influences intergroup attitudes (Canetti-Nisim et al. Reference Canetti-Nisim, Halperin, Sharvit and Hobfoll2009) and political conservatism (Bonanno and Jost Reference Bonanno and Jost2006). (2) Terrorism can also strike fear into a targeted populace (Huddy, Feldman, and Cassese Reference Huddy, Feldman, Cassese, Crigler, MacKuen, Marcus and Neuman2007), and affect voting through the fear of future attacks rather than past casualties. By only looking at suicide terrorism, it is unclear whether shifts in the electorate are due to casualty aversion, or the fear that terrorism creates. Using rocket attacks from Gaza, which caused relatively few casualties but provoked widespread fear (IICC 2009), we are able to isolate the effect of fear over terrorism on voting behavior.

Outside the Israeli-Palestinian conflict, Kibris (Reference Kibris2011) finds that Turkish nationalistic parties’ vote-shares increase following fatal attacks by Kurdish insurgents against police targets. However, in the case of Spain, Bali (Reference Bali2007) and Montalvo (Reference Montalvo2011) show that the 2004 Madrid bombings mobilized opposition voters and shifted the electoral map to the left against an unpopular foreign policy.Footnote ² Several studies suggest that political responses to terrorism might be mediated by psychological variables. For instance, Canetti-Nisim et al. (Reference Canetti-Nisim, Halperin, Sharvit and Hobfoll2009) find that exposure to terrorism leads individuals to adopt more exclusionary attitudes towards minorities. Especially relevant to the current study, Zeitzoff (Reference Zeitzoff2013) shows that residents living in localities with high exposure to rockets from Gaza show more altruism to their ingroup than those in low exposure localities. Other studies are less sanguine, finding that rocket exposure is associated with increased adolescent violence and aggression (Henrich and Golan Reference Henrich and Golan2013). Habituation to terror can also shape voters’ behavior. Becker and Rubinstein (Reference Becker and Rubinstein2011) show that reduction of services and goods purchased in Israel during the Second Intifada occurred only among occasional consumers, since regular users of such services are able to adjust their behavior. Numerous other studies draw a connection between war-time, nationalism, and patriotism (Baker and Oneal Reference Baker and Oneal2001; Skitka Reference Skitka2005). Huddy, Feldman, and Cassese (Reference Huddy, Feldman, Cassese, Crigler, MacKuen, Marcus and Neuman2007) show that perceptions of threat post-9/11 were more likely to favor aggressive military actions. Finally, Abramson et al. (Reference Abramson, Aldrich, Rickershauser and Rohde2007) connect George W. Bush’s 2004 presidential election win to his ability to appear to be a strong leader, an especially important factor considering the import US voters placed on combatting terrorism. This is also confirmed by Oates, Kaid, and Berry (Reference Oates, Kaid and Berry2009), who find that fear was an important component of Bush’s campaign in 2004 that projected Bush as a safe and secure leader for troubled times.

Review of these studies suggests that exposure to terrorism might increase, decrease, or have no effect on electoral support for right-wing parties. What complicates this question even further is the strategic nature of terrorist-government interactions. Powell (Reference Powell2007) shows that both terrorists and governments choose actions that maximize their utility, given their expectations regarding the other side’s actions. Terrorists choose the targets and the timing of their attacks strategically. For example they may try to derail peace talks (Kydd and Walter Reference Kydd and Walter2002), trigger government response that would facilitate terrorist recruitment (Bueno de Mesquita and Dickson Reference Bueno de Mesquita and Dickson2007), or maximize government concessions (Lake Reference Lake2002; Pape Reference Pape2003). Thus, they target those populations that are more likely to respond in the desired manner, either by voting for right-wing parties (if terrorists’ goals are to “spoil” talks or facilitate recruitment), or by voting for left-wing parties (if terrorists’ goals are to extract concessions). Governments, who are motivated to protect against attacks as well as to increase their vote-shares, are also strategic in their response to terrorism, and protect targets to minimize the overall expected impact of attacks (Powell Reference Powell2007).Footnote ³ Thus, in equilibrium, terrorist attacks might have no effect on voting, or they might strengthen public support for the incumbent.

Finally, previous research on electoral behavior can illuminate how terrorism affects voting. Theories of retrospective voting suggests that voters evaluate candidates based on their past performance and achievements in areas such as economic policy (Fiorina Reference Fiorina1978; Kiewiet and Rivers Reference Kiewiet and Rivers1984). Alternatively, Petrocik (Reference Petrocik1996) and Wright (Reference Wright2012) argue that some parties have an advantage in certain areas, and they tend to benefit when their issues become salient. In the Israeli context, Schofield and Sened (Reference Schofield and Sened2005) and Arian and Shamir (Reference Arian and Shamir2008) argue that right-wing political parties enjoy a valence advantage on security issues. In the context of conflict and terrorism, there is evidence that incumbents lose electoral support following attacks and casualties, as expected if the retrospective voting theory is correct (Gelpi, Feaver, and Reifler Reference Gelpi, Feaver and Reifler2006; Karol and Miguel Reference Karol and Miguel2007; Gassebner, Jong-A-Pin, and Mierau Reference Gassebner, Jong-A-Pin and Mierau2008; Bali Reference Bali2007; Montalvo Reference Montalvo2011). However, Berrebi and Klor (Reference Berrebi and Klor2008) and Koch and Tkach (Reference Koch and Tkach2012) find that in Israel incumbents are not punished for suicide attacks, and the right-wing bloc increases vote-shares regardless of the incumbent’s political affiliation, in accordance with the valence theory.

Using Rocket Range to Understand Terrorism

In this article we provide several contributions to understanding the connection between terrorism and voting. First, the nature of rockets and the circumstances of their use allow us to circumvent the problem of strategic targeting. Palestinians who fire rockets are confined to Gaza, and the impact of their attacks depends on rocket technology that determines the range and accuracy of their rockets.Footnote ⁴ Thus, at any given point in time, those within the rockets’ range are threatened not because they are perceived to be attractive targets, but because this is how far rockets can reach. Within the range, the targets of rocket attacks are strategically chosen. However, due to limited accuracy and varying launching skills, everyone within the range has some likelihood of being hit. Crucial to our purpose is that selection into the range is not driven by strategic considerations, but by availability of technology and the constraint to launch rockets from within Gaza. In fact, we show that those residing within the range do not differ substantially from those living outside the range, and we attribute the difference in voting behavior to being threatened by rockets. This is an improvement over existing studies, both in the Israeli context as well as beyond, that so far focus on the impact of those terrorist attacks whose targets are not determined by a random choice.

Second, by examining the range rather than the actual targeting, we argue that experiencing an attack is not necessary to influence voting, and that voting could be also affected by the mere threat of being a target. Focusing on threat (living in the range of fire) is also closer to the view of terrorism as a strategy that seeks to instill fear in a target populace, rather than to destroy the opponent militarily (Abramson et al. Reference Abramson, Aldrich, Rickershauser and Rohde2007).

Range of Rockets

Our main explanatory variable is Range _{i, t} that indicates whether a locality is within the range of rockets one day before elections.Footnote ¹⁵ Being in the range depends on two factors: the maximum distance that a rocket can travel from Gaza, and locality’s distance from Gaza.

The information on rockets’ range comes from the Israel Defense Forces and Home Front Command websites. To determine whether a locality is within the range of rockets, we used ArcGIS to calculate the shortest distance between a locality and the Gaza perimeter. A locality is considered to be within the range if one day before election that distance is less than or equal to the rockets’ range. Some localities, mostly regional councils, are composed of discontinuous units. We consider such localities to be within the rockets’ range if the minimum distance between all units composing a locality and Gaza perimeter is less than or equal to the rockets’ range.Footnote ¹⁶ In the Robustness Checks section, we examine alternative measures of distance.

Voting

Voting data are available from the Israel Social Sciences Data Center (ISDC) at The Hebrew University of Jerusalem.Footnote ¹⁷ We classify Israeli parties into left-wing, centrist, and right-wing blocs using the classification proposed by Arian and Shamir (Reference Arian and Shamir2008), and similar to Schofield and Sened (Reference Schofield and Sened2005) (see Table 2). These blocs differ with respect to their positions on the Israeli-Palestinian issue: right-wing parties are less willing to provide concessions to Palestinians than centrist and left-wing parties.

TABLE 2.

Parties in Political Blocs (Vote-Shares in Parentheses)

Our main dependent variable is RightShare _{i, t} that measures the right-wing bloc’s vote-share in locality i in election t (1999, 2003, 2006, and 2009).Footnote ¹⁸ In addition, we examine vote-shares of nationalistic, religious ultra-orthodox parties, Russian immigrants’ parties, as well as vote-shares of Likud, Labor, and Shas parties. We calculate vote-shares by dividing the number of votes a bloc, a subset of a bloc, or a party received by the number of valid votes in a locality in a given election. Vote-share variables are continuous, and range from 0 to 1.

We focus on vote-shares as our dependent variable for two reasons. First, they reflect voting behavior, and are therefore directly relevant to understanding how terrorism threat affects voters. Second, Israel has a highly proportional representation system, with vote-shares closely corresponding to seat-shares in the parliament. Thus, they not only represent changes in voting behavior, but also have direct implications for government composition and policies.

Control Variables

We control for a broad set of variables that can affect voting. First, we include locality-level time-varying demographic attributes available from the Central Bureau of Statistics: population size (logged), median age, ratio of males to females, and the share of Jewish population in a locality. We use a one-year lagged value of these variables. In addition, we control for the share of Jewish population with family origin in Asia, Africa, and in the former Soviet Union because previous research shows that these groups of voters are more likely to support right-wing parties (Berrebi and Klor Reference Berrebi and Klor2008). Locality-level data on residents with origin in Asia and Africa are available from the 1995 and 2008 censuses, whereas similar data on the share of residents with origin in the former Soviet Union was recorded only in the 1995 census. We therefore use the 1995 values of these three variables in specifications without locality fixed effects.Footnote ¹⁹

To control for local economy, we include standardized locality mean wage adjusted for inflation using 2006 as the base year.Footnote ²⁰ Education can also affect voting, and we control for the share of high school graduates among residents who are 17–25 years old in each locality-year.Footnote ²¹

Migration could pose a problem if population composition within the range changes over time. We control for population movement in and out of localities by including the net migration divided by locality’s population.Footnote ²²

Previous studies demonstrate that other forms of terrorism, and in particular suicide attacks, affect voting (Berrebi and Klor Reference Berrebi and Klor2008; Gould and Klor Reference Gould and Klor2010). We, therefore, control for the logged number of local fatalities due to suicide attacks three months prior to an election. We obtain information on local fatalities by combining data from Benmelech, Berrebi, and Klor (Reference Benmelech, Berrebi and Klor2010b ) and Getmansky (Reference Getmansky2011).

In addition to the broad set of locality-level time-varying controls, we include locality fixed effects that account for observable and unobservable time-invariant locality characteristics, such as distance to border, distance to Gaza, or to the West Bank, distance to the center, topography, and whether a locality is a regional capital. In specifications without fixed effects, we explicitly control for some of these attributes.

Finally we include year dummies to control for secular trends that affect all localities in a given year, such as the national economy, quality of electoral campaigns, leaders’ characteristics, and events related to the Israeli-Palestinian conflict that influence Israel as a whole.

Econometric Model

Our goal in this article is to identify the causal effect of being in the range of rockets on voting. We estimate the following linear model:

(1) \begin{equation} Right Share_{i,t} \!=\!\alpha {\times} Range_{i,t} {\,+\,}\bm{\beta {\times} (X_{i,t-1})} {+} \gamma _t\,{+}\,\mu _i\,{+}\,\varepsilon _{i,t}, \end{equation}

where RightShare _{i, t} is the vote-share of right-wing parties in locality i in election t, Range _{i, t} is a binary indicator of whether locality i is within rockets’ range one day before election t, and X _{i, t − 1} is a vector of lagged time-varying locality-level controls, as explained above. γ _t is an election year dummy that captures secular trends, μ _i is the locality fixed effects, and ε _{i, t} is the locality-clustered robust error that captures unobserved determinants of voting in year t.Footnote ²³

The coefficient α identifies the effect of being in the range on right-wing parties’ vote-share. We expect α to be positive and statistically significant if localities within the range have a higher right-wing vote-share than localities outside the range, conditional on X _{i, t − 1} , μ _i , and γ _t . A negative and statistically significant α would imply that localities within the range of rockets are less supportive of right-wing bloc, and more supporting of centrist or left-wing parties. An α coefficient that is statistically not significant would imply that being in the range of rockets does not affect voting.

Is Rockets’ Range Exogenous to Voting?

We use the variation over time in the range of rockets to explore the effect of being in the range on voting. Given that time invariant locality-level characteristics, such as distance to the Gaza Strip remain constant, and controlling for time-varying locality attributes, we study how voting is affected by being in the range of rockets. Our key identification assumption is that the range of rockets is exogenous to voting. Essentially, we assume that localities come into the range of rockets for reasons that are unrelated to their anticipated voting behavior. To investigate this assumption, we examine whether observable variables that are known to be correlated with voting also determine whether certain localities are within the range of rockets. If localities within the range are similar to those out of the range, then any differences in voting can be attributed to living in the range of rockets. Assessing similarity between within- and out-of-range localities is also important to minimize extrapolation bias (King and Zeng Reference King and Zeng2006).

Table 3 presents the summary statistics of the various variables we use to compare localities within and outside the range.Footnote ²⁴

TABLE 3.

Summary Statistics of Pretreatment Variables

Overall, the distributions of these variables is very similar across within- and out-of-range localities. For many variables (gender ratio, share of immigrants from the former Soviet Union, the mean wage, presence of a regional capital, and turnout in 1999) localities within the range have similar localities that are out of the range. Some variables, however, have nonoverlapping areas. Localities within the range are less populous than those outside the range, and their median age is higher, implying fewer children than in localities outside the range. Localities within the range have higher shares of residents with family origin in Asia and in Africa, and a lower share of high school graduates. Furthermore, localities within the range have higher right-wing vote-shares in the 1999 election and they are on average closer to the border than localities outside the range, but not by much.

It is important to note that even those variable that do not completely overlap have a broad range of common support, implying that for many localities within the range there are similar localities outside the range. Notwithstanding, in the robustness checks section, we drop observations with nonoverlapping values (outside of common support) to make sure our results are not driven by outliers.

Summary

In this article, we are asking how the threat of being targeted by terrorists affects voting behavior. We answer this question by focusing on the effect of being in the range of rockets fired from the Gaza Strip on voting in 2003–2009 Israeli parliamentary elections. We provide evidence that localities in the range of rockets are very similar to localities that are not in the range, and therefore the difference in voting is due to being in the range.

Our findings suggest that entering the rockets’ range substantially affects voting and increases support for the right-wing bloc, and especially for nationalistic parties. We find that right-wing vote share is 2–6 percentage points higher in localities that are within the range. We attribute this finding to the valence advantage that right-wing parties in Israel have with respect to security threats. Consistent with the valence explanation, we find that right-wing parties are not punished by voters, and in some cases even gain additional votes in localities that become part of the range during right-wing parties’ term in office. We demonstrate that these results are driven by being in the range, and do not depend on actual targeting, and on changes in the composition of the population and the electorate. Furthermore, we show that our results are robust to various model specifications, and different methods of defining and measuring the range.

Implications

Our finding that the threat of terrorism leads voters to support right-wing parties provides a plausible mechanism for why some conflicts endure: voters who are threatened with violence elect candidates who are less willing to make concessions. This further reduces the bargaining range between conflict participants, and decreases the chances of reaching a negotiated solution. Our results also provide empirical support for theories that posit that hard-line groups may use violence and the reaction it engenders to foment conflict, and discredit more dovish actors (Kydd and Walter Reference Kydd and Walter2002; Bueno de Mesquita and Dickson Reference Bueno de Mesquita and Dickson2007).Footnote ⁴⁰

Our project is also related to recent studies on the efficacy of terrorism from the perspective of the terrorists (Abrahms Reference Abrahms2006 Reference Abrahms2012). If the terrorists’ goal is to scare a population into accepting the terrorists’ demands, our article suggests that this might be counterproductive. However, if the goal is to ensure conflict between the two sides, and marginalize doves on both sides, then it may be effective.Footnote ⁴¹ By using the randomized threat of rockets from Gaza, our study avoids issues of endogeneity—nonrandom selection of targets and strategic interaction between terrorists and targeted government—that limit the interpretations of previous empirical studies.

In addition to finding that voters within the range harden their attitudes towards the perpetrating group, we also detect an interesting incumbency effect. Rather than punishing incumbents who hold office during increases in the threat of rockets, our findings show that voters turn to the right-wing Likud party regardless of incumbency (and they do not punish right-wing parties even if the range expanded during their term in office). This finding parallels recent work that suggests voters choose parties who they feel “own” issues (such as the case of right-wing parties and security in Israel) (Wright Reference Wright2012). This has serious implications for bargaining and negotiations—violence leads to hard-line parties coming to power, making settlement less likely and future violence more likely.

We argue, perhaps counterintuitively, that the nature of the threat from rockets’ range provides a generalizable mechanism for how violence influences political behavior in other conflicts. Previous studies find that exposure to violence affects voting and political preferences (Berrebi and Klor Reference Berrebi and Klor2008; Gould and Klor Reference Gould and Klor2010), but that this effect dissipates if violence increases beyond a certain threshold, as the targeted population accepts high levels of violence as part of life (Jaeger et al. Reference Jaeger, Klor, Miaari and Paserman2012). If violence has diminishing returns in its impact on political preferences, then most of the substantial effects will manifest with violence that does not inflict a large number of casualties, like in the case of suicide attacks. Despite the number of rockets that have been launched into southern Israel, there have been relatively few casualties. Furthermore, the localities that are within the range in 2001–2009 have not been major targets of suicide bombings or other forms of terrorist attacks. We argue that the focus on threat and low levels of lethality is more relevant to understanding the broader impact of terrorism (e.g., in the US and Europe)—where most people are threatened by, rather than directly exposed to, the violence. This focus on the effect of rockets’ range differentiates our studies from purely retrospective studies of violence (Bellows and Miguel Reference Bellows and Miguel2009; Kibris Reference Kibris2011), as voters are weighing their past exposure and fear from future exposure (Huddy et al. Reference Huddy, Feldman, Taber and Lahav2005). Given the stated goal of terrorism to instill fear in a target populace (Abramson et al. Reference Abramson, Aldrich, Rickershauser and Rohde2007), this is a cleaner measure of the effect of terrorism.

Our research also has serious policy implications. Previous research on terrorism and political behavior has mostly focused on suicide terrorism. This research is unable to adjudicate on whether the main effect of terrorism on political behavior operates through casualty aversion or through living under threat. Using data on the range of rockets from Gaza, we show that merely being in range of terrorism, and not necessarily experiencing a high number of casualties, is sufficient to have a strong influence on voting. This implies that governments face a significant trade-off. Do they take every effort to warn residents and emphasize the wider level of threat, thereby fanning the effects of terrorism, or do they seek to reduce perceptions about threat and provide mechanisms for coping? This question may depend on the political affiliation of the leaders, with right-wing politicians showing a preference to increase the warnings over terrorism and heighten the threat level, rather than dampen it down (and left-wing politicians the opposite).Footnote ⁴²

Finally, our findings also add nuance to other studies of indiscriminate violence, such as Lyall (Reference Lyall2009) who finds that randomized artillery shelling in Chechnya reduces insurgent attacks through intimidating the targeted population. Conversely, we find that randomized threat of violence increases hardline attitudes. Taken together, these findings suggest that the effect of violence (indiscriminate or otherwise) may be context dependent. We conjecture that the targeted actors’ military strength moderates this relationship. Israel is a stronger military actor compared to Hamas. Conversely, the Chechen rebels were weaker compared to the Russian army. Exposure to violence thus may harden the populace of a targeted stronger military actor, leading them to be more willing to fight. Yet, it may lead the populace of the weaker actor to favor withdrawal, and concessions to the perpetrating actor. Given the role of civilian victimization in several recent conflicts—Syrian Civil War, US drone strikes in Pakistan, and post-Mubarak violence in Egypt—this hypothesis deserves further scrutiny.