Long-Term Change in Conflict Attitudes: A Dynamic Perspective

Abstract A large literature examines how citizens in violent conflicts react to the conflict's events, particularly violent escalations. Nevertheless, the temporal nature of these attitudinal changes remains under-studied. We suggest that popular reactions to greater violence are typically immediate but brief, indicating short-term emotional responses to physical threats. Over the longer term, however, public opinion is more commonly shaped by non-violent events signaling the adversary's perceived intentions, reflecting slower but deeper belief-updating processes. We support this argument using dynamic analyses of comprehensive monthly data from Israel spanning two full decades (2001–20). Rather than violence levels, we find that long-term changes in Jewish attitudes on the Israeli–Palestinian conflict follow non-violent events implying Palestinian preferences, particularly failed negotiations and out-group leadership changes. Our findings underscore the importance of public opinion's temporal dynamics and show that non-violent events, which are often overlooked in the literature, play a prominent role in shaping long-term attitudes in conflictual contexts.

Nevertheless, in some cases, violent events can instead increase popular support for peaceful agreements and moderate politicians (Arian, Shamir, and Ventura 1992;Gould and Klor 2010;Tellez 2019). Such conciliatory attitudes are more likely when the violence activates anxiety and weariness, concerns about future hostilities, and a greater desire to end the bloodshed (Beber, Roessler, and Scacco 2014;Hazlett 2020;Huddy et al. 2005).
While the influence of violence has been researched extensively, several behavioral studies find that public opinion also reacts to non-violent events conveying new information about the adversary's goals and the conflict. Such signals can include direct or indirect statements of out-group intentions (Hall et al. 2018;Halperin et al. 2011;Leshem and Halperin 2020), bilateral negotiation summits (Rosler, Cohen-Chen, and Halperin 2017), or positions taken by international actors (Shelef and Zeira 2017). They are typically accepted more easily when reaffirming negative preconceptions of the conflict and out-group (Halperin and Bar-Tal 2011;Nyhan and Zeitzoff 2018;Sheafer and Dvir-Gvirsman 2010), though positive signals from the adversary can sometimes induce hope (Leshem 2019).
Nevertheless, most studies, whether focused on violent events or non-violent signals, share similar vagueness about their dynamic dimension. This ambiguity leaves several important questions under-studied: How long does the influence of different events endure in the public's mind?; How quickly do they take effect and at what pace do they erode?; and To what extent do they shape the long-term trajectory of public opinion? Many studies-particularly tests for causal mechanisms in controlled or quasi-controlled environments-examine attitudinal change immediately after specific events or stimuli. Others analyze their correlations with attitudes at a later but frozen point in time, be it days, weeks, or years after the event took place. These static lags between cause and effect are often selected by arbitrary data-collection limitations, rather than for theoretical or empirical reasons. Finally, even when the data include multiple periods, most time series tend to be short and sparse, and lack in-depth analyses of long-term temporal structures. 1 Immediate Threats and Long-Term Belief Updating How, then, do violent and non-violent events influence public attitudes over time? All events can trigger emotional and cognitive reactions, yet we contend that violent and non-violent occurrences activate these mechanisms differently. Consequently, we expect that their temporal influence on public opinion-their immediacy, duration, and erosion rate-would vary too.
Events involving violence against the in-group pose an instant physical threat. Such palpable threats tend to trigger strong negative emotions and attitudes due to heightened mortality salience (Pyszczynski, Solomon, and Greenberg 2003) and concern for the in-group (Huddy et al. 2002;Wohl, Branscombe, and Reysen 2010). Since such emotional reactions are linked to survival instincts, we expect them to appear immediately and strongly when the violence occurs but then wane quickly as the physical threat grows distant. Violent events also convey information to the targeted population. Political violence is used to gain attention and signal the out-group's determination, capability, and willingness to impose painful costs to achieve its goals (Kydd and Walter 2006;Nussio, Böhmelt, and Bove 2021). Such signals, however, are often unsuccessful (Abrahms 2006;Gould and Klor 2010). Violence is likelier to communicate new information in nascent or dormant conflicts with high uncertainty about the adversary's goals and behavior. Yet, once recurring, even in large intervals, violent actions become expected and familiar. Targeted group members, accordingly, become cognitively desensitized and less likely to change their attitudes (Liebes and Kampf 2007;Nussio 2020).
Cognitive desensitization, nevertheless, does not undermine short-term emotional reactions to violent threats (Nussio 2020). Hence, in contexts where the adversary's goals and violence are familiar, we expect that violent events cause only an immediate but fleeting influence on attitudes. As discussed earlier, the literature has mixed expectations about the direction of this influence. Accordingly, we propose two competing hypotheses: Hypothesis 1a (H1a): Greater violence will increase hawkish attitudes about the conflict immediately but only for a brief period.
Hypothesis 1b (H1b): Greater violence will decrease hawkish attitudes about the conflict immediately but only for a brief period.
The second type of events signal new information about the adversary and the conflict without using violence. This broad category can take various idiosyncratic forms, including elite actions, international incidents, or domestic political events. Nevertheless, several common features are required, as such events must: portray new information about the adversary's goals or the conflict's future; be publicly and broadly visible; and lack a palpable physical threat. 2 Consider, for example, public negotiation meetings, which send visible signals about the adversary's demands and willingness to compromise (or lack thereof). A similar cue appears when the out-group selects new leaders with more moderate or extreme positions about the conflict. Actions by the international community can also send salient indications about the conflict's future.
Contrary to violence, the lack of physical threat in non-violent events should weaken immediate emotional reactions and increase the role of cognition (Gordon and Arian 2001). Moreover, such signals are typically more complex. Whereas violence sends a simple message due to its embedded danger-the adversary physically threatens the in-group-non-violent events carry greater uncertainty. Deciphering such new and complex information requires additional topdown cues and public narratives that take longer to process (Berinsky 2009;Zaller 1992). Hence, non-violent events signaling new information about the adversary should influence popular attitudes slower. However, once they do, this cognitive channel can update beliefs more fundamentally and for lengthier periods. Therefore, we suggest the following hypothesis: Hypothesis 2 (H2): Non-violent events carrying salient negative (positive) new information about the adversary will increase (decrease) hawkish attitudes about the conflict more slowly but for longer periods than violence.
One caveat is in order. In extreme cases, the emotional influence of violence can create lasting traumas that sustain long after the danger passes. This often occurs in individuals that experience violence directly and develop post-traumatic stress disorder (Bonanno and Jost 2006;Hirsch-Hoefler et al. 2014), yet they are usually too few to shift aggregate public opinion. 2 Some non-violent events, especially large-scale protests against the in-group, are perceived as physically threatening due to their mass nature and out-group stereotypes (Manekin and Mitts 2022). Since public opinion reflects subjective threat perceptions, we exclude such occurrences from this category. However, outstanding atrocities, such as indiscriminate mass violence or forced population transfers, can form lasting collective victimhood narratives that are passed down through generations (Balcells 2012;Lupu and Peisakhin 2017;Rozenas, Schutte, and Zhukov 2017). Our argument disregards such extreme cases given their rarity, but researchers studying violence of this scale should adjust their analyses accordingly.
In what follows, we test our hypotheses empirically using dynamic analyses of hundreds of monthly public opinion surveys conducted in Israel regularly over two decades. The data's high frequency and longevity allow us to analyze how aggregate attitudes on the conflict respond to various violent and non-violent events, how quickly they react, and how long the change endures. Before elaborating on our methodological approach, we first discuss our case study and data.

The Israeli Case
The Israeli-Palestinian conflict has long been a central case study for political behavior in conflictual contexts. The conflict has been a constitutive issue in Israeli society and politics (Arian and Shamir 2008;Shamir and Arian 1999;Yakter and Tessler 2022), yet despite its enduring presence, it has been quite dynamic. In the past decades, the Israeli public faced differing degrees of violence, including two large-scale violent campaigns (intifadas), ebbs and flows of small-scale terrorist attacks and rocket shelling, and extended calmer periods. 3 The nature of violence itself varies too, shifting periodically between local stone throwing, lone-wolf attacks, large-scale suicide and car bombings, and rocket shelling. The conflict has also provided multiple non-violent opportunities to assess the other side's intentions. The past decades included several rounds of peace negotiations, some prolonged with meaningful advancements, particularly in the 1990s and briefly in the mid-2000s, and others shorter and futile. Both sides also made leadership changes over the years, signaling alternating popular support for more combative or moderate visions of the conflict.
The temporal variation in violent and non-violent events deem Israel particularly fitting to test our argument. They also mark the boundaries of our analysis, a point discussed further in the article's conclusion. Most significantly, the conflict's rich temporal dynamics are testable due to its protracted nature. As such, our broader inferences should be considered carefully where hostilities are experienced as a single shock with long-term traumatic effects and/or the adversary is relatively unfamiliar. Nevertheless, as Peffley, Hutchison, and Shamir (2015, 819) observe, Israel's experience with violence is "distinctive but not unique" and comparable with dozens of other countries. As more societies endure recurrent political violence, so do Israeli patterns become more relevant and generalizable.
Public Attitudes about the Conflict: The Peace Index Data Our primary outcome of interest is Israeli public attitudes toward the conflict. To examine these attitudes over time, we pooled hundreds of public opinion surveys from the Peace Index, a survey project established in the 1990s at Tel Aviv University. The Peace Index has regularly conducted monthly representative surveys with a small set of recurring questions about the peace process, alongside interchanging questions on current events. These data, therefore, offer a unique opportunity to examine national attitudes on the conflict using identical questions in short and regular intervals over two full decades.
We analyze two dependent variables based on a pair of recurrent questions asked regularly starting in July 2001. The first gauges ideological support for negotiations as follows: "What is your position on conducting peace negotiations between Israel and the Palestinian Authority?" The second measures pragmatic hope for an agreement as such: "Do you believe or not believe that negotiations between Israel and the Palestinian Authority will lead in the coming years to peace between Israel and the Palestinians?" Both offer a four-point answer scale from "Strongly in favor/believe" to "Strongly opposed/do not believe at all." Both questions tap into hawkish/dovish attitudes about the conflict, its future, and the adversary. Accordingly, we expect both indicators to react similarly to the conflict's events. Nevertheless, we analyze them separately because their core sentiments-ideological preference and pragmatic expectation-are interrelated but not analytically identical (Leshem and Halperin 2020). While their similarities and differences are outside the scope of this article, identifying potential variations and parallels can add further depth to our analysis and directions for future research.
Whereas each monthly survey draws a new respondent sample, their representative design produces a reliable time series of aggregate public attitudes. Accordingly, our two dependent variables measure the monthly aggregate net support for each statement, that is, the total share of negative answers subtracted from the share of positive answers every month. Since Arab citizens have been included inconsistently throughout the series, we aggregate only answers from Israel's Jewish majority. Moreover, we expect Arab citizens, many of whom identify as ethnic Palestinians, to display different attitudinal patterns regarding the conflict. Hence, the reactions of minority ethnic groups in ethnonational conflicts merit a separate theoretical and empirical framework outside the scope of this article.
Both variables comprise 227 months from July 2001 to May 2020. Of these months, twenty-five have missing observations. Section 2 in the Online Supplementary Material (OSM) demonstrates that the missing observations are spaced far apart and uncorrelated with the conflict's violence and non-violent events. Additionally, both indicators display strong serial autocorrelation. Hence, we impute the missing observations using a simple linear interpolation. Section 2 in the OSM offers additional discussion and robustness tests for this choice. Figure 1 plots both time series. The data show that aggregate net support for negotiations has remained positive throughout the past twenty years, notwithstanding occasional ebbs and flows. Aggregate net hope of their prospects is consistently lower than, though noticeably correlated with, the former. We examine the dynamic nature of these trends using two complementary methods. First, we estimate a general error-correction model (GECM) with prespecified explanatory variables to find average reactive patterns throughout the entire period. Secondly, we employ a structural breakpoint analysis to inductively identify critical junctures of long-term attitudinal change and then qualitatively investigate nearby events and implied mechanisms. Together, the two methods establish a robust dual test for our hypotheses. We discuss each set of findings in turn.

Independent Variables
We begin by estimating attitudinal changes by several independent variables measuring monthly violence levels and non-violent signals. Two variables gauge the monthly levels of Palestinian violence experienced by Israelis. The first counts the monthly number of Israeli casualties by Palestinian actions. 4 Given the country's mandatory military service and high sensitivity to combat deaths (Levy 2012), we count both civilian and security forces fatalities. The second variable measures the monthly number of rockets shot at Israel from the Gaza strip. 5 Palestinian factions began using rockets in the early 2000s and quickly increased their volume and range, now reaching the most populous areas in central Israel. Whereas these rockets are less deadly than other forms of violence, they are aimed indiscriminately at civilians and cause meaningful psychological, social, and political effects (Besser and Neria 2009;Getmansky and Zeitzoff 2014;Zeitzoff 2014). Since casualties and rockets have unevenly high peaks and decreasing marginal influence, we take the natural logarithm of both. 6 Next, we construct two variables to measure non-violent events signaling Palestinian intentions. Coding such events ex ante is not trivial. Since they can take multiple idiosyncratic forms, an open coding scheme may introduce coder bias in favor of high-impact events, while overlooking comparable incidents with low influence. Thus, we focus conservatively on two event types that can be coded systematically. 7 The first variable is a dummy indicating months with public Israeli-Palestinian negotiation meetings. Public negotiation summits send salient signals about the outgroup's demands, willingness to compromise, and distance from the in-group's positions. Our data include five negotiation meetings between 2001 and 2020 that were publicly known in real time. Such meetings can send either a negative or a positive signal, depending on their conclusions. However, due to their futility throughout our sample, we code all these meetings as negative signals.
The second variable gauges Palestinian hawkish leadership selection using a categorical score that indicates months signaling either a militant (1) or moderate (−1) leadership change. The popular choice of hawkish or moderate leaders sends a visible signal about greater public support for their visions of the conflict. Our sample includes four such moments. In January 2005, a positive signal was sent when the relatively moderate Mahmoud Abbas was elected president after Yassir Arafat's death. 8 The three negative events signaled popular support for Hamas, a militant Islamist movement that has continuously engaged in violence against Israel: in March 2006, 4 Casualty data are from B'Tselem (see: https://www.btselem.org/statistics). 5 We code rocket data using monthly reports by the Meir Amit Intelligence and Terrorism Information Center (https:// www.terrorism-info.org.il/en/). 6 To deal with zeros, we take the natural log of each month's count plus 1. Since high peaks in casualties and rockets often indicate periods with direct combat, we do not model military operations separately. 7 Naturally, the media is a primary platform for non-violent signals. Yet, for our purposes, the media is a mediating channel, rather than an initiator of external events. Moreover, media coverage in Israel displays a stable tendency to frame the conflict's events negatively (Sheafer and Dvir-Gvirsman 2010;Wolfsfeld 2004). Hence, we leave questions about mediaframing effects outside our analysis and treat it as an invariable constant. 8 Most Jewish Israelis perceived Arafat to be a violent extremist. In the October 2004 Peace Index survey, weeks before his death, 74.9 per cent of Jewish respondents stated that Arafat controlled street-level Palestinian violence and 78.7 per cent described him as a terrorist, rather than legitimate statesman. Conversely, in January 2005, after Abbas' election, 57.3 per cent of Jewish respondents agreed that he is making sincere efforts to stop Palestinian violence and only 31.7 per cent disagreed.
Hamas formed the Palestinian government for the first time after winning a majority of seats in the legislative election; in March 2007, it headed a new unity government with the more moderate Fatah party; and in July 2007, after the unity agreement collapsed, it forcibly took control over the Gaza Strip and formed a parallel government to Fatah's West Bank administration. 9 We also include three control variables. First, we calculate the share of Israeli cabinet ministers from right-wing parties to control for similar dampening signals about the preferences of fellow Israelis. 10 Secondly, we use the real average monthly wage of Israeli hired workers (constant 2011 prices) to control for the state of the economy, which may influence the public mood regardless of the conflict. 11 Third, we include a monthly time trend (that is, a monthly counter) to account for a possible monotonic decline in support or hope for peace irrespective of specific events. Section 1 in the OSM presents descriptive plots of all variables.

Model Setup
To examine the dynamic relationship between the two attitudinal variables and our independent variables, we estimate a GECM using an ordinary least squares (OLS) regression (De Boef and Keele 2008). The GECM regresses changes in the dependent variable on its own lagged value and on both the first differences and the lagged values of the independent variables. Formally, the GECM is specified as follows: where Δy t is the change in the dependent variable, α 1 estimates the influence of its levels one period earlier (also known as the error-correction term), b ′ 0 estimates the immediate effects of a one-unit change in independent variables X t , and b ′ 1 estimates the lagged effect of the latter's levels in the previous period. The inclusion of both first differences and lagged values separates each independent variable's immediate and delayed effects on the outcome. The model's errorcorrection term (α 1 ) reflects the monthly rate at which the outcome adjusts back to its former levels after the explanatory variables changed. Hence, it tells us how quickly their influences erode.
To avoid bias, the GECM requires that all variables have the same order of integration (Grant and Lebo 2016;Keele, Linn, and Webb 2016). Multiple unit root tests, specified in Section 4 in the OSM, indicate that all our variables are stationary except the share of Israeli right-wing cabinet seats. This variable, therefore, is modeled only at first difference.
We select the GECM's proper number of lags using a general-to-specific approach, detailed in Section 5 in the OSM. In short, we iterated our models with different combinations of lag lengths for each variable, picking the number of lags that optimized model fit and coefficient t-tests scores. The procedure favored a single monthly lag for logged rockets, logged casualties, and average wage, and two-monthly lags for negotiation meetings and Palestinian leadership changes, foreshadowing a temporally intricate effect, as we shall see. 12 9 A potential concern may suggest that non-violent events can increase violence, establishing a combined chain effect. To reject this possibility, we regressed violence levels on negotiations and leadership changes, and found null results. Section 3 in the OSM presents more details. 10 Our argument implies that in-group elite cues are a possible mechanism for information processing following salient non-violent events. To verify that changes in the Israeli cabinet do not mediate the influence of non-violent signals, we re-estimated our models without this covariate. The results remain unchanged. 11 Israel's consumer satisfaction survey, which measures economic perceptions more directly, is only available starting in 2011. The average wage, nevertheless, is strongly correlated (r = 0.75) with this index. 12 Algebraically, once the GECM includes an independent variable's first and second lag, its first difference (ΔX t ) is replaced with its second difference (Δ 2 X t ).
Finally, we tested for serial autocorrelation in our models using Ljung-Box and Breusch-Godfrey tests. In both series, dynamic completeness is achieved after adding the first difference of the lagged dependent variable (Δy t−1 ). Additional Ljung-Box tests reject concerns of seasonality in the data. Section 6 in the OSM details these tests. Table 1 presents the GECM estimations. For ease of interpretation, both net attitudes are measured on a scale of 1-100, rather than percentages. The first-difference coefficients indicate that greater violence leads to an instant decline in both net support and net hope for peace. On average, a one-standard-deviation increase in the logged number of rockets immediately lowers the aggregate net support for negotiations by 1.57 points and net hope in their prospects by 1.15 points. Similarly, a one-standard-deviation increase in logged casualties instantly decreases net support for negotiations by an average of 1.92 points and net hope by 1.72 points. These results corroborate the hypothesis that greater violence immediately dampens public attitudes regarding resolution (H1a) and rejects a positive relationship (H1b). Conversely, we do not see sudden attitudinal changes following negotiation meetings or Palestinian leadership changes, consistent with their expected lagged effect (H2).

GECM Findings
We examine the model's longer temporal dynamics in two steps. First, we calculate each explanatory variable's long-run multiplier (LRM), presented at the bottom of Table 1. Calculated as LRM x = −(β 1 /α 1 ), these scores reflect the total cumulative influence of each independent variable across all future months. We estimate the LRMs' confidence levels using the bounds test proposed by Webb, Linn, and Lebo (2020). The test has three possible outcomes: no long-term relationship (a t-statistic below the lower bound); a statistically indeterminate long-term relationship (a t-statistic in between the bounds); and a statistically significant long-term relationship (a t-statistic above the upper bound). Most LRMs in our model fit the middle category, indicating ambiguous statistical confidence. Rockets, however, have a strictly insignificant long-run effect. 13 Secondly, we compute the temporal distribution of these cumulative effects over time. This calculation is based on each variable's immediate effect at time t (β 0 ), follow-up effect at t + 1 (β 1 ) and t + 2 where relevant, and monthly erosion rate thereafter (α 1 ). Figure 2 plots the monthly distribution of each variable's long-term influence after a one-off one-standard-deviation increase. Negotiations and hawkish leadership change, which are binary/categorical, are assumed to increase by one unit.
The resulting patterns tell a temporally intricate story. In both models, the negative influence of rocket shelling lapses after a single month. An increase in casualties has a similarly fleeting effect on the aggregate hope for peace, though its influence on the willingness to negotiate is slightly less forgiving: a one-standard-deviation increase creates an immediate negative decline of 1.92 points, but its cumulative effect reaches 6.23 points, on average, over the next four to five months. More casualties, therefore, do not deepen pessimism about the prospects of peace beyond a single month, but they do have a slightly larger and longer dampening influence on support for diplomacy. While mostly similar, this nuance is intriguing enough to justify additional research. Nevertheless, the broader pattern supports H1a: greater violence levels undermine support and hope for compromise immediately but also rather briefly.
Non-violent political events exhibit the opposite dynamic. Both negotiation meetings and hawkish leadership changes cause a lagged but longer-term demoralization in public attitudes, 13 According to Getmansky and Zeitzoff (2014), the expanding rocket range gradually shifted more Israeli voters rightward. We run two tests to rule out similar temporal heterogeneity in our models. First, we re-estimated our analysis while interacting the logged rocket count-its first difference, its first lag, and then both simultaneously-with a dummy indicating the period after rockets were first shot at the Tel Aviv metropolitan area (in November 2012). Secondly, to gauge a more gradual increase in threat, we also interacted the logged rockets variables with a simple yearly count. All interactions produced null results.
barely moving at first but then plummeting two months later and eroding slowly thereafter. Curiously, hawkish leadership changes create a positive immediate effect, perhaps reflecting an initial hope for pragmatic moderation under the pressures of leadership. Nevertheless, this effect is statistically insignificant and turns negative one month later. While the LRM bounds tests are statistically indeterminate regarding the full extent of these long-run influences, the GECM indicates that the largest dip, taking place at t + 2, is statistically significant for both negotiations and leadership changes. These results, therefore, support H2, indicating that informational signals have a lagged but prolonged influence on aggregate public attitudes regarding resolution.

An Inductive Analysis of Influential Events
The GECM estimation is instructive but has several limitations. First, it models average attitudinal changes after events that can be coded systematically, yet such incidents can take various idiosyncratic forms that are hard to operationalize ex ante. This is particularly true for singular moments, such as deeply traumatic violent acts or other unique events reshaping prior conceptions. Secondly, the statistical indetermination of the LRM bounds tests muddies our full confidence in the long-term influence of non-violent signals. Finally, the average patterns provide little information about the real-world dynamics accompanying moments of prolonged attitudinal change.
To address these issues, we employ an inductive structural breakpoint analysis of the two series. Rather than prespecified events, structural breakpoint analyses detect moments of significant change based on the underlying structure of the data. Once such points are identified statistically, we can qualitatively examine real-world events that occurred at these times and assess their fit with our hypotheses (Caporale and Grier 2005;Wawro and Katznelson 2014). Theoretically, we expect that long-term breaks in the two series will follow salient non-violent events signaling Palestinian intentions and the conflict's future, rather than violent escalations. Moreover, if our hypotheses are too narrow, this approach can discover other influential event types that we may have missed. We estimate structural breakpoints in our data using Bai and Perron's method, which allows for multiple breaks in each series (Bai and Perron 1998;Bai and Perron 2003). A Zivot-Andrews unit root test confirms that both attitudinal series are breakpoint stationary. We use a standard trimming parameter value of 0.15 and determine the number and position of the estimated breakpoints with a sequential l + 1 breaks versus l test. 14 Since both series show signs of trend stationarity, we include a temporal trend term in the estimation.  Table 2 summarizes the descriptive characteristics of these breakpoints. The first two columns indicate whether each structural breakpoint displays changes in absolute levels and/or a long-term trend and their direction. 15 These shifts show both the similarities and differences between the two series. Both ideological support for negotiations and hope about their prospects grew similarly during the early 2000s and experienced a parallel sharp drop in early 2006 (Points A and D) that lasted several years. 16 The aggregate net support for negotiations, though not net hope, Figure 2. Lag distribution after a one-standard-deviation increase in each independent variable (dummy/categorical variables increase by 1). 14 A robustness check identifies the same breakpoints with Bayesian Information Criterion (BIC) and Liu-Wu-Zidek (LWZ) tests. 15 The direction and significance of each change are corroborated econometrically by interrupted time-series models detailed in Section 7.1 in the OSM. 16 Whereas the GECM estimates average effects across all pre-coded events, the breakpoint analysis identifies only the most temporally influential occurrences. The latter, accordingly, underscores larger effects than the former. temporarily bounced back in early 2009 (Point B) before gradually eroding again. Practical hope for peace, conversely, never rebounded after 2006. The last breakpoints in both series-September 2013 (Point E) and October 2016 (Point C)-exhibit smaller changes in levels without shifting the long-term trend.
Which factors explain these breaks? The third column in Table 2 summarizes the relative rocket and casualty levels at each point compared to their sample averages. Corroborating our hypotheses and earlier findings, none of the long-term breaks occurred in particularly violent moments. By contrast, as the fourth column indicates, all moments can be linked with non-violent events that sent salient signals about Palestinian preferences and/or the conflict's future. Moreover, most involve leadership changes or failed negotiations. To gain better insight into their influence, we turn to examine each moment in greater detail. We pay particular attention to the first three breakpoints, which exhibit the largest structural changes in both attitudinal levels and trends.

January-April 2006: Hamas's Electoral Victory
The first notable breakpoint in both series occurred in January-April 2006 (Points A and D). The early months of 2006 featured two political developments: the Palestinian legislative elections and subsequent government formation (January and April); and the Israeli general election (March). Of the two, the Israeli election seems less consequential. The incumbent party, Kadima, won the election handily. According to the post-election Peace Index survey, most respondents supported the winning coalition and split along expected partisan and ideological lines. The Israeli election, therefore, reinforced the political status quo.
The Palestinian election, by contrast, seems like a watershed moment. The Palestinian Legislative Council election was held for the first time in a decade after the Fatah party's long single-party reign. Yet, defying earlier expectations, the more extremist Hamas won most seats. In reality, Hamas's success reflected public frustration with Fatah, infighting among their opponents, and advantageous electoral rules, rather than popular support for its militant agenda (Shamir and Shikaki 2010). However, to Israeli eyes, the victory signaled Palestinian endorsement of violent extremism. In the January 2006 Peace Index survey, 60 per cent of Jewish respondents stated that Hamas's victory posed an existential threat to Israel and 74 per cent predicted little to no chance that Hamas will eventually recognize Israel's right to exist. Moreover, 55 per cent of Jewish respondents opposed direct negotiations with a Hamas-led government and 87 per cent estimated that there is little to no chance of reaching a peace agreement with it.
The negative signal about Hamas's victory developed further in the following weeks with additional elite cues, fitting the lagged structural break in aggregate hope in April 2006. Following the election, the Middle East Quartet-the United States, the European Union, the United Nations, and Russia-demanded publicly that any Hamas-led government recognize Israel, accept previous bilateral agreements, and commit to non-violence. Hamas, however, blatantly rejected these conditions, triggering severe international and Israeli economic sanctions once the new government was formed. The Israeli public took notice: the Peace Index survey conducted in late March 2006, two months after the election, showed an increase in the share of Jewish respondents doubting that Hamas will moderate its violence (79 per cent compared to 50 per cent in January) and objecting to direct negotiations with the movement (57 per cent compared to 54 per cent in January). Hence, Hamas's public refusal to disavow violence and recognize Israel after its victory, combined with aggressive international and Israeli delegitimization campaigns, explicated the notion of Palestinian preference for violent extremism over compromise. The result was a prolonged drop in aggregate Jewish-Israeli support and hope for compromise.  mind. In the March 2009 Peace Index survey, 62 per cent of Jewish respondents estimated that the Israeli government would strive to maintain a good relationship with the United States regarding peace negotiations and would face severe pressure from Obama if not. Hence, the flurry of preparations in the spring of 2009 sent a visible signal that the US administration was determined to revive the peace process and that agreeing to negotiations was in Israel's best interest. The lack of a similar rise in hope for peace may indicate that many waited for substantial signs that the Palestinians were similarly committed. Subsequent developments over the following months help explain the slow but renewed decline in support for negotiations. The momentum continued during the summer, with Obama's dovish Cairo Speech and Netanyahu's acceptance of the two-state solution in his Bar-Ilan University address. Furthermore, in November 2009, Israel announced a ten-month settlement construction freeze following intense US pressure. By its conclusion in September 2010, Israeli and Palestinian representatives were set to meet for direct peace talks.
Yet, the first cracks appeared early and gradually expanded. Despite Netanyahu's endorsement of the two-state solution, he developed a visibly strained relationship with Obama and repeatedly demanded strict preconditions for an agreement, particularly formal Palestinian recognition of Israel as the state of the Jews. During the September 2010 talks, Israel's leadership continued to raise this demand publicly and refused to extend the settlement freeze without it. Most Israelis internalized this cue: in the Peace Index survey from October 2010, 75 per cent of Jewish respondents justified Netanyahu's demand for Palestinian recognition and 81 per cent agreed that the Palestinians do not accept Israel's existence and would destroy it if they could. The negotiations quickly imploded, leading the frustrated US administration to shift focus to other foreign policy areas. Both the initial momentum in early 2009 and its subsequent failure are mirrored well in the rise and fall of aggregate support for negotiations, corroborating our earlier findings of the lagged but long-term influence of failed diplomacy. 17

September 2013: More Futile Negotiations
The last two breakpoints exhibit smaller but noticeable breaks in attitudinal levels without changing their previous temporal trends. Nevertheless, they too align with discernible non-violent signals about the adversary and conflict's future. The first (Point E) shows a decline in practical hope for peace in September 2013, two months after another round of failed negotiations initiated by US Secretary of State John Kerry. Unlike the positive momentum in 2009, Kerry's initiative never gained steam and ended with a whimper. While the Jewish-Israeli public was initially skeptical, it nevertheless grew more doubtful after their failure. In the Peace Index survey from June 2013, before the negotiations started, 71.3 per cent estimated that Kerry's initiative had low or very low success odds. After the meetings, this share grew to 79.4 per cent in July and 81.2 per cent in September. Hence, even a swift, low-key diplomatic failure caused an additional long-term decline in aggregate hope for peace. Fitting our GECM findings, this change became more pronounced a month or two after the direct meetings, as the extent of the failure was properly processed.

October 2016: Trump's Election
The last structural breakpoint (Point C) identifies rising support for negotiations after October 2016. The most notable event at this moment was US President Donald Trump's electoral victory in early November. 18 In the Peace Index survey conducted right after the election, 48.5 per cent of Jewish 17 Of these intricate developments, our GECM codes only the 2010 summit as the relevant public signal of the Palestinians' true intentions. Hence, the negative pattern in our estimation fits this chronology. 18 Although the model identifies October as the breakpoint, November is within the margin of error. Figure 3 verifies that the bounce occurs in the following months. Moreover, we could not identify other notable events in October.
Israelis estimated that Trump favors Israel over the Palestinians and 61.8 per cent assumed he would not oppose and would even support settlement construction. By contrast, only 22.2 per cent stated that Obama was friendly to Israel throughout his term. Hence, Trump's victory signaled unconditional US support for Israeli demands, leading to a better bargaining position than the Palestinians. This cue helps explain the structural bounce in willingness to negotiate even as practical hopes for peace remained low. This signal does not involve Palestinian intentions, yet it too illustrates the importance of new non-violent information about the conflict's balance of power.

Conclusion
Public opinion is a key aspect of violent conflicts, establishing bottom-up pressures that can escalate or moderate the conflict's future path. Accordingly, a large literature explores various factors influencing public attitudes in conflictual contexts, with a particular focus on violent incidents. Nevertheless, past findings mostly highlight short-term or static attitudinal reactions, providing incomplete accounts of their full temporal dynamics and other types of influences.
Using two decades of monthly surveys from Israel, we argue that popular reactions to realworld events have a meaningful, yet largely under-studied, temporal dimension. Our findings show that Jewish-Israeli support and hope for compromise decreases immediately after violent escalations, but this influence lapses quickly and leaves little to no mark on the long-term trajectory of public opinion. Conversely, non-violent events carrying visible informational signals about the adversary and the conflict-particularly failed negotiations and Palestinian leadership changes-have a lagged but larger and longer effect on public attitudes. This conclusion is supported by both average patterns found with a GECM and an inductive structural breakpoint analysis matched with real-world events.
These divergent temporal patterns are consistent with our suggested theoretical logic. The sudden but fleeting effect of violence implies not only an instinctive emotional reaction to a palpable danger, rising and falling with its physical threat, but also cognitive desensitization. Meanwhile, the lagged but longer-lasting reactions to non-violent signals suggest deeper belief updating following gradual cognitive processing of new information and top-down cues. While we do not test these mechanisms directly, our aggregate findings open new avenues for micro-behavioral research on their operation, different temporal implications, and interrelations. Moreover, given the literature's focus on violence, more theoretical and empirical work is needed on different types of informational signals in conflictual environments. Our findings, for example, imply that unexpected information (for example, Hamas's surprising electoral victory) may have a longer attitudinal influence than expected events. This point is particularly important for real-world conflict resolution efforts. Our findings show that public attempts to resolve conflicts by international and domestic actors can be a double-edged sword if unsuccessful. In Israel, futile negotiation initiatives, especially when pushed by outside actors, not only failed to advance peace, but also left an enduring negative mark on Jewish-Israeli attitudes. Similarly, international and domestic framings of the outgroup's political choices, such as the negative campaign against Hamas's looming government, can have long-lasting adverse implications for public attitudes about resolution. Even indirect cues, such as unconditional one-sided support by international actors, can move public opinion in meaningful ways (see Shelef and Zeira 2017). Insofar as popular attitudes matter for peace, actions and cues by global and local elites must be taken carefully and with proper consideration of their potential long-term repercussions.
As noted earlier, while our focus on the Israeli-Palestinian conflict offers important insights, it also raises questions about generalizability. Three issues stand out, marking both the promise and limits of our conclusions, and paths for further comparative research. First, the longevity, intensity, and embeddedness of the Israeli-Palestinian conflict raise the concern that Israelis have become uniquely desensitized to violence (see Nussio 2020). This worry, however, is inconsistent with a large body of evidence showing that the conflict's violence constantly triggers emotional, attitudinal, and behavioral reactions in Israel (for example, Berrebi and Klor 2008;Besser and Neria 2009;Getmansky and Zeitzoff 2014;Gould and Klor 2010;Hirsch-Hoefler et al. 2014). Moreover, multiple findings from other Western countries show that sparser terrorist and civilian attacks have likewise caused only short-lived attitudinal shifts regarding security policies, social trust, and outgroup resentment (Arvanitidis, Economou, and Kollias 2016;Breton and Eady 2022;Castanho Silva 2018;Economou and Kollias 2019;Finseraas and Listhaug 2013;Geys and Qari 2017;Sharkey and Shen 2021). Thus, we expect similar temporal patterns elsewhere, including in shorter and less salient violent contexts, though more comparative dynamic analyses are needed.
Secondly, the period that we analyze had few positive diplomatic breakthroughs. Our data, accordingly, do not test how aggregate public attitudes react to positive negotiation advancements, which often also involve temporary setbacks and violence. Descriptive data from the 1990s, the heyday of the Israeli-Palestinian peace process, fail to show notable spikes in Israeli attitudes during constructive diplomatic developments (Hermann and Yuchtman-Yaar 2002). This pattern fits the expectation that negative signals are more influential than positive ones. Nevertheless, recent experimental research finds that optimistic information about the adversary can increase support for compromise under certain conditions Leshem and Halperin 2020). Hence, we need additional dynamic analyses of attitudinal changes in periods and regions experiencing real progress toward resolution.
Finally, our analysis of Jewish-Israeli popular reactions focuses on the stronger group in an asymmetrical conflict. However, these processes may manifest differently among minority or weaker groups in uneven settings. Jaeger et al. (2012) find not only similar short-lived Palestinian reactions to targeted Israeli violence, but also a greater sensitivity to collateral violence, which is endured more regularly in the territories. Accordingly, more direct comparisons of dynamic reactions across asymmetrical power hierarchies are required. Our findings, we hope, help facilitate such future endeavors.