One flag, two rallies: Mechanisms of public opinion formation in Israel during the 2014 Gaza war

Transcription

1 One flag, two rallies: Mechanisms of public opinion formation in Israel during the 2014 Gaza war In the summer of 2014, the Israel Defense Forces (IDF) launched an extensive operation against Hamas in the Gaza Strip. During the seven-week Protective Edge operation (July 8-August 26), Israeli Prime Minister Benjamin Netanyahu experienced a massive boost in popularity his public approval ratings climbed from less than 50 percent to nearly 80 percent 1 and support for the military operation peaked at over 90 percent among the Jewish majority of Israeli citizen residents. This type of "rally-round-the-flag" (henceforth RRTF) reaction to a major military operation or security crisis is not unique to Israel. For example, public opinion scholars have identified rally periods that occurred in the United States during certain international conflicts (see Feinstein 2016a for a review) as well as in Britain during the Falklands War and the Gulf War (Lai and Reiter 2005; Lanoue and Headrick 1998; Norpoth 1987). Conceptually, scholars have usually portrayed the RRTF phenomenon as a unitary (rather than multifaceted) process. In his studies of the phenomenon in the United States, John Mueller (1970:22) originally defined a rally as a let s-getbehind-the-president effect, and most subsequent studies have adopted this definition, focusing on the popularity ratings of sitting presidents. A few studies, however, have expanded the concept to include increases in the approval ratings for presidential policies in addition to a boost in presidential popularity. Baker and Oneal (2001:661), for example, defined RRTF as "a surge of patriotism and public approval for [the president's] administration and its policies during an international crisis" (although the authors operationalization measured only presidential public 1 Based on data reported in Haaretz daily newspaper on April 11, 2014 and August 6,

2 approval ratings). 2 Following Baker and Oneal, this study defines RRTF as extraordinary increases in both support for the leader and approval of foreign policy. However, in contrast to previous research (including Baker and Oneal [2001]), the present study does not assume that RRTF is a unitary process, but rather considers the possibility that support for the leader and approval of policy are two separate outcomes that emerge simultaneously in rally periods via two distinct sets of mechanisms. THEORETICAL AND EMPIRICAL BACKGROUND Need for a novel theoretical model The RRTF phenomenon cannot be adequately conceptualized via extant theoretical models of attitude formation. First, attitude changes during rally periods deviate from the relatively stable and predictable patterns of public opinion (Page and Shapiro 1992:173) during these periods public attitudes do not reflect the slowchanging general ideological and political divides in a society. Second, conventional arguments about attitudes toward military engagement, which suggest that the public supports successful and low-cost wars (i.e., those that entail few casualties on our side), cannot explain rally effects because these effects emerge before the public has information about the outcome and cost of the military engagement (Berinsky 2007). Given these limitations of prior models of opinion formation, understanding what motivates individuals to rally around the flag requires the development of a novel theoretical model centered on more dynamic mechanisms of opinion 2 A few scholars have proposed other extensions to the definition and operationalization of rallyround-the-flag, adding, for example, increases in the public s trust in governmental institutions and the popularity of the Senate and the House (Hetherington and Nelson 2003; Parker 1995). 2

3 formation, which are activated during rally periods. Initial theoretical models of the RRTF phenomenon have highlighted the correlations between presidential popularity ratings and characteristics of the military operation/security crisis, including the official goal of military actions, the cues sent to the public by the political elite, and various aspects of the media coverage of the war. This body of research has shown that rally periods are likely to occur during defensive wars and unlikely to occur during interventionist wars (Jentleson 1992), and that support from the political opposition leads to greater support for military action (Berinsky 2007; Brody 1991; Zaller 1992). In addition, Groeling and Baum (2008) found two key factors that influence the emergence of a rally: 1) journalists decisions to cover or ignore particular speakers and messages and 2) whether the content of messages from certain speakers contrasts with politics as usual specifically, whether representatives of the opposition party explicitly endorse the president s policy. (See Groeling and Baum [2008] for a full review of media-related characteristics correlated with rally periods.) I expand previous theory by exploring the role of individual-level mechanisms of attitude change in rally periods. To develop a dynamic individual-level model of public opinion formation during rally periods in wartime, in the following theoretical discussion I weave together elements from communication studies, public opinion studies, and social and political psychology. I discuss the distinct extant theoretical arguments about public opinion formation, and explore how these arguments can be combined to jointly form a theoretical model of RRTF during wartime. I then test this model using panel data collected in Israel during and following the 2014 Gaza war. I also test competing or supplemental arguments. The results provide empirical evidence for a novel theoretical model of rally periods and contribute to the development of 3

4 pragmatic conflict resolution processes by offering a more nuanced understanding of the attitudinal dispositions embedded in national identities among individuals living in societies that are engaged in a prolonged violent conflict. Contextualizing a rally: Communication of national threat The electronic and print media play a key role in the emergence of rally effects during a war or security crisis. Before developing into a rally point, a security event first becomes a media event (Dayan and Katz 1992): nearly all news channels interrupt their routine programming and begin live-streaming reports about the event, which they season with dramatic special announcements as well as opinions and explanations from special experts such as retired military generals. This intense broadcasting transforms an event that may be geographically bounded into a national event that monopolizes the public conversation. In line with this assertion, my data (data collection methods are described in detail below) reveal intense news consumption in Israel during the rally period that followed the outbreak of the 2014 war with Hamas in Gaza: 97 percent of survey respondents reported watching the news more than once per day, and about three quarters watched the news at least once an hour. In contrast, after the war ended, more than a third of respondents (36 percent) reported watching news once a day or less frequently, and only one fifth reported watching once an hour or more often (see Table 1). Importantly, during war or security crises the media is the platform used to communicate the perceived threat to national security which then serves as the generator of rally effects both via the intensity of its coverage and the celebratory nature of the content of that coverage, which embraces an official rhetoric that charges the event with symbolic meaning as a test of the solidarity, bravery, and 4

5 determination of the nation. As I show elsewhere (Feinstein 2016a), in the United States, at least since World War II (earlier periods were not included in the study), one of the necessary conditions for the emergence of rally periods has been a presidential rhetoric that (via the media) propagated national threats and stressed the need to stand united and firm against the enemies who posed the threats. In the Israeli case, this type of national threat construction as well as a call for unity, bravery, and determination took center stage during the 2014 Gaza war. The most prominent agent of this message was Prime Minister Netanyahu. For example, when he first announced operation Protective Edge in a nationally televised address, he emphasized that, For the terrorist organizations, all parts of Israel are one front, and all citizens of Israel are one target. Therefore, we must stand together, as one, united, certain about our justness. 3 This message was immediately picked up and embraced by the Israeli media, which in addition to providing frequent updates about both the rockets falling in the southern region of Israel and the progress of the military operation in Gaza, also documented residents of other regions in Israel offering expressions of solidarity with their southern compatriots. 4 The concerns for national security that are boosted during rally periods are, therefore, sociotropic concerns (Huddy, Feldman, Taber, and Lahav 2005) about threats that are mainly symbolic. 5 These concerns center on the imagined national 3 (retrieved on January 3, 2016) 4 For example, Azari, Amram. July 8, Tel Aviv identifies: Here, we also feel like those in the south. News2 Online:Mako: (retrieved on February 8, 2016) 5 More generally, studies show that political opinions and behavior are not driven by egocentric concerns, but by sociotropic concerns (Kinder and Kiewiet 1979; Joslyn and Haider-Markel 2007) and the symbolic predispositions they evoke (Sears 2001). 5

6 group: the adversarial behavior of an enemy aimed at the nation s symbols or sovereignty, or at communities that belong to the nation, is viewed as an attack against the nation as a whole, and thus leads even individuals whose personal safety is not threatened to join the rally. In the case of Israel, rally outcomes emerged because even Israelis who resided outside the range of the rockets launched from Gaza perceived the rockets as a threat to their national security. Specifying the exact relationship between threat perception and news consumption is beyond the scope of this study, which focuses on the consequences of an increased sense of threat to national security. However, it is worth noting that during wartime, media exposure and threat perception may not have a simple cause and effect relationship at the individual level, but rather may interact in the form of a "reinforcing spiral" (Slater 2007): the media propagates a sense of threat, and then worried individuals turn to the media to obtain more information about the crisis. EXTENDING THE EXTANT THEORETICAL MODEL The construction of a severe threat to national security is the starting point for the theoretical argument developed in this study (i.e., I consider media coverage exogenous to the current model). The analysis outlines the mechanisms that convert this threat into a rally around the flag. 6 Specifically, based on studies in social and political psychology, which highlight the cognitive biases and emotions that emerge under perceived threats to the ingroup, the main task of this study is to specify the mechanisms that mediate the effects of threat perception on rally outcomes. I argue that support for the head of the state (the prime minister in Israel) increases as a 6 Perceived threats to national security can also have consequences other than RRTF, for example, individuals blaming the government for failing to protect its citizens. 6

7 result of amplified identification with the nation while increased approval for military action emerges due to the activation of two dispositions associated with status maintenance vis-à-vis the external rivals in a conflict: anger toward the enemy and nation-state chauvinism. In the following sections, I elaborate on these arguments and then use empirical data to test the newly developed theoretical model. Rallying behind the leader I argue that increased support for the national leader is a direct consequence of the activation of national identification, which results from perceived threats to the national group. National identification boosts support for the leader because it temporarily shifts membership in the nation to the center of the sense of self (Brubaker and Cooper 2000) at the expense of the memberships in sub-national or supra-national groups (e.g., political party, political ideology, ethnicity, or religion) that drive public opinion during non-rally periods. A rally behind the leader occurs when many citizens who are not part of the president s regular constituency (in contemporary Israel, this sub-set comprises individuals who identify with the political left ) shift from opposition or indifference to support due to increased identification with the national group. Figure 1 summarizes this proposed mechanism. << Figure 1 about here >> Jacobson s (2007) study of public opinion in the United States after the September 11 attacks presented compelling evidence for this process: the increase in Bush s popularity after the attacks occurred almost entirely among Democrats and independent voters. Therefore, in contrast to the claims of some of the earliest research on RRTF (see Mueller 1970:21 for a review), citizens rally behind the 7

8 leader not because of his executive role as head of the state, but rather because of his symbolic role as head of the nation. Importantly, if the leader s role as head of the state had been responsible for the RRTF in Israel, increased support would have been observed across the entire society and not only among the Jewish majority group (data about the attitudes of Arab-Israelis during the war are scarce, but reveal very low levels of support for both the government and the military operation 7 ). Rallying behind military action In this section, I argue that in the context of militarized conflict, boosts in support for military actions emerge via different mechanisms than increases in public support for the leader. The first of the two mechanisms outlined in this discussion, an increase in nation-state chauvinism, is triggered by the national identification process discussed above. The second mechanism, anger toward the enemy, is a direct effect of the perception of national threat. I focus on these mechanisms for two reasons: First, previous research has shown that both nation-state chauvinism and anger are pivotal to the formation of militant attitudes during international conflicts and security crises. Second, these two mechanisms do not require the assumption that the public receives sufficient information about a military engagement to assess its outcomes and costs, but rather hinge on the fact that chauvinism and anger are the products of the military confrontation itself, once it is embodied in the form of a media event. Nation-state chauvinism. Following Kosterman and Feshbach s (1989) seminal research, many studies in the United States have established that beliefs 7 See the Israel Democracy Institute s report, Eng.pdf (retrieved on January 4, 2016). 8

9 about the superiority of the nation-state are consistently positively correlated with militant views (e.g., Crowson 2009; de Figueiredo and Elkins 2003; Worchel and Coutant 1996). However, scholars have paid much less attention to the following concern expressed by Kosterman and Feshbach in the conclusion of their article, which points to the dynamic and context-dependent aspect of nation-state chauvinism: One cannot help but be concerned by the periodic waves of nationalism [i.e., nation-state chauvinism] that seem to sweep nations and the accompanying receptivity to belligerent actions. (Kosterman and Feshbach 1989:273; emphasis added) My explanation for the massive support for military action during rally periods emphasizes the increase in nation-state chauvinism due to national identification in response to perceived collective threats. This argument is grounded in a fundamental principle of how individuals behave as members of groups: as shown in numerous social psychology experiments, under conditions of external threat or competition, individuals who experience high levels of identification with the ingroup tend to display perceptual biases aiming to maintain or enhance their group s status (for a review, see Ellemers, Spears, and Doosje 2002) (see evidence for the operation of this mechanism at the national level in Feinstein 2016b). The literature suggests two mechanisms that may connect nation-state chauvinism to support for military action. First, political psychologists often assert that national chauvinism is accompanied by a desire to dominate other nations (e.g., de Figueiredo and Elkins 2003; Huddy and Khatib 2007; Kosterman and Fesbach 1989). As Herrmann and coauthors (2009:729) explained, national chauvinists expect that international relations should be governed by a norm of authority ranking with their nation on top, which during international conflicts means that 9

10 they expect opponents to succumb to the demands of "our" government, even if this requires the use of military force against defiant enemies. There is a second reason that nation-state chauvinism leads to support for military action: During war, ingroup bias perceiving one s own nation-state as a shining house on the hill precludes individuals from placing responsibility for the conflict on their own nation-state and rather leads them to place the blame entirely on the enemy. 8 In the context of the Israeli-Palestinian conflict, claims about the alleged moral superiority of Israel (e.g., in international and domestic speeches Prime Minister Netanyahu has referred to Israel as the only real democracy in the Middle East ) are an integral part of the discourse in Israel. Further, as Roccas, Klar, and Liviatan (2006) have shown, the tendency to justify the national group s actions and deny any criticism of the group is inherent to this type of national glorification. This tendency took center stage during the 2014 Gaza War: both official speakers and the public conversation described the IDF operation as a just war against a cruel and corrupt regime that not only attacked Israel, but also victimized its own people by launching rockets from areas with dense civilian populations that served as human shields. This type of blame attribution was expressed, for example, by Netanyahu in his televised address on the thirteenth day of the war: Hamas is responsible for the consequences that stem from its aggression. We make every effort not to hurt the residents of Gaza, while Hamas makes every effort so that the residents of Gaza would be hurt. We feel sorry for any harm of 8 Sociologists and social psychologists who study racial inequality in the United States have made a similar argument. Their studies show that attributing moral or normative inferiority to a minority group rationalizes its oppression by placing the blame for that group s suffering on its own wrongdoing (Bobo, Kluegel, and Smith 1997; Bonila-Silva 2006; Sears 1988). 10

11 innocent people, but whenever they are hurt the blame is on Hamas, and on Hamas only! 9 Anger. Increased support for the use of military force also has an emotional drive. Several studies in the field of political psychology have shown that individuals who feel anger toward the enemy tend to support the use of military power because being angry drives individuals toward action and risk taking (Huddy, Feldman, and Cassese 2007; Lerner et al. 2003; Skitka et al. 2006). Thus, feeling anger toward the enemy in a conflict drives a rally behind military action. The anger that motivates individuals to support the use of military power is linked to the same perceived collective threat that drives other mechanisms of attitude formation during rally periods. This link has been explored in the social psychological literature on group-level emotions, which attributes anger to both the perception that the entire group is under attack and identification with fellow group members who suffer because of the enemy s actions (Smith, Seger, and Mackie 2007). During crisis periods, anger emerges as a mass phenomenon due to the prevalent sense that the action of the nation s perceived enemies poses a threat to the nation and compatriots. This (righteous) anger then leads individuals to seek retaliation in order to maintain or restore the group s status (see Huddy and coauthors [2007] discussion of the effects of anger in public opinion in the United States after the September 11 attacks). Rallying behind military action: Theoretical summary. In sum, during rally periods, the perceived threat to the nation boosts support for military action through two complementary mechanisms: (1) Individuals experience increased nation-state chauvinism, and as a result are more likely to expect other countries to

12 submit to the authority of the ingroup government and, at the same time, are less concerned about the costs inflicted on the other side, which is held responsible for its own suffering. (2) Increased anger leads to a desire for retaliation and willingness to take risks. Figure 2 summarizes these two mechanisms. << Figure 2 about here >> DATA AND ANALYTICAL APPROACH Research design Ideally, the emergence of rally effects would be modeled via a comparison of panel data collected shortly before and during a rally period, but because the events that lead to rally periods are unforeseeable, this type of analysis is usually not feasible. An alternative option is to collect data during and after a rally period, which allows the researcher to check which attitudes, sentiments, and affective reactions distinguish individuals who supported the government and its military policy only during the rally period from individuals whose levels of support or opposition remained relatively stable. In this study, I collected survey data from a representative sample of the Jewish population in Israel (at the end of 2014, Jews composed 74.9 percent of Israel's population). 10 Data collection was conducted in two waves: the first survey (henceforth "wave 1") was conducted from July 24 through July 27, 2014 (during the ground assault of the IDF in Gaza). During that period, the Jewish majority in Israel was rallying behind the leadership of Prime Minister Netanyahu and the military operation in Gaza (the Arab minority was overwhelmingly against the war and in 10 Information retrieved on March 9, 2015 from Israel's Central Bureau of Statistics: 12

13 opposition to Netanyahu): an estimated 65 percent of Jews in Israel were at least somewhat satisfied with the way Netanyahu was handling his job, and less than 10 percent agreed (to at least a certain extent) that conducting a military operation in Gaza was a mistake. The second wave of the survey ("wave 2") was conducted four months after the war ended: from December 16 through December 29, By that time, Jewish Israelis rates of satisfaction with Netanyahu and support for the IDF operation had shifted markedly: only about 32 percent were at least somewhat satisfied with Netanyahu, and about 34 percent agreed to at least a certain extent that conducting a military operation in Gaza was a mistake. The survey was designed to test the theoretical arguments discussed above. The survey included questions about attitudes and feelings, as well as a set of demographic questions. 11 Because it was unclear how long the IDF operation in Gaza would last, I was compelled to conduct wave 1 as rapidly as possible, and therefore I chose an Internet-based survey design. Recent research has shown that given appropriate sampling methods, Internet surveys produce estimates similar to those produced by more traditional survey methods (Ansolabehere and Schaffner 2014). I hired Midgam, a leading polling firm that specializes in political attitudes, to conduct the study. Participants were recruited via ipanel, the largest database for Internet polling in Israel, with over 100,000 enrolled individuals who embody the full range of social, economic, and geographical facets of Jewish society in Israel. 12 The sample of 1,007 respondents was created by matching the distributions of core demographic 11 Throughout the data collection, the order of questions about attitudes and feelings was randomized to prevent potential ordering effects. 12 ipanel maintains the highest standards in the field of Internet surveys, and has both international accreditation from ESOMAR and local accreditation from the department of statistics at the Hebrew University in Jerusalem. 13

14 variables gender, age, religiosity, and district to their distributions in the entire Jewish population of Israel (see Appendix 1 for the distributions of demographic variables in the sample). Four months after the end of the war, all participants were invited to complete the survey again. No less than 75 percent of first-round respondents also participated in the second round (N=758). Because the attrition of participants from wave 1 to wave 2 might not be random, I tested for possible demographic differences between the respondents who participated in both waves and those who only participated in the first wave (the distributions of the demographic variables in the full and reduced samples are presented in Appendix 1). For most demographic variables age, education, religiosity, ethnicity, immigration history, marital status, political ideology, and income the difference between the two samples was minor and not statistically significant. I found a statistically significant difference for only two variables: gender and district, which I included as control variables in the analysis, thus accounting for the potential bias caused by differential attrition rates. Outcome variables The investigation focuses on two indicators of RRTF. First, public satisfaction with the head of state (in this case Prime Minister Netanyahu) was measured on a six-point Likert scale ranging from "very satisfied" to "very unsatisfied." Second, public support for the use of military power was assessed by asking respondents how strongly they agreed or disagreed that the decision to implement the IDF operation in Gaza was a mistake (again, responses were measured on a six-point Likert scale). I chose this item based on the assumption that it would produce a higher level of reliability in a longitudinal study than other 14

15 questions about support for the military operation, for which (at least when asked in Hebrew) the interpretation depends on whether the war is ongoing or has already ended. I reversed-coded respondents answers, so higher scores represent greater support for the military operation. Explanatory variables Each aspect of the proposed mechanisms of RRTF changes in ethnonational identification, changes in nation-state chauvinism, and changes in anger is represented by a variable. To reduce measurement error, sentiments and feelings are estimated by composite measures based on several survey questions (see Appendix 2 for a list of composite variables, their related survey questions, and factor loadings). << Table 1 about here >> FINDINGS AND DISCUSSION The main goal of the empirical analysis is to fit a model that integrates the variables associated with the mechanisms outlined above. To establish the superiority of an integrated model over explanations that focus on distinct mechanisms, I first conduct regressions that test direct effects, and then estimate structural equation models that test an integrated model including both direct and indirect effects. Step 1: Regression analysis of direct effects on rally outcomes As shown in Table 1, in wave 1 (administered during the war in Gaza) about 65 percent of the Jewish population in Israel was satisfied, quite satisfied, or very satisfied with the way Prime Minister Netanyahu was handling his job (only about 14 percent were unsatisfied or not satisfied at all), whereas five months later only 15

16 about 32 percent expressed at least some level of satisfaction, and about 36 percent expressed a moderate or strong level of dissatisfaction. At the same time, there was a similar, albeit somewhat less dramatic change in support for the IDF operation in Gaza: in wave 1, no less than 81 percent disagreed or strongly disagreed with the mistake statement, but four months after the war ended only about 64 expressed similar levels of objection. Table 1 shows significant changes in each of the focal variables. More respondents expressed high levels of agreement with the statement "Hamas poses a security threat to Israel" in wave 1 than in wave 2, and the level of anger toward Hamas was higher in wave 1 than in wave 2. In addition, the data reveal declines in both national identification and nation-state chauvinism. 13 To identify which of the changes in explanatory variables were directly associated with changes in the outcome variables, I conducted multivariate regression analyses via STATA. Because the goal is to test arguments about the mechanisms that generated changes in individuals' satisfaction with Prime Minister Netanyahu, both the outcome variable and some explanatory variables measure change. 14 The regression equation is as follows: Differences in means reported in this section are statistically significant at the.05 level or a higher level. 14 However an individual s score on an outcome variable may also change over time due to a change in the strength of the variable s association with explanatory variables (e.g., through priming or depriming; Tesler 2014). To check this possibility, I estimated the correlations each of the two outcomes had with the potential predictor. The results reveal coefficient consistency between the two waves of the survey. These findings suggest that the dramatic shifts in attitudes toward Prime Minister Netanyahu and the war were not due changes in the effects of explanatory variables, but rather to changes in individuals scores on these variables. 15 t 1 =wave 1, t 2 =wave2, and Δ=t 2 -t 1 ; The Xs are not followed by numerical subscripts because they may represent one or several explanatory variables. 16

17 ΔŶ = Yt 1 + b 0 + b 1 X + b 2 ΔX The equation includes four types of variables. On the left, ΔŶ is the change in the outcome variable, measured by subtracting each respondent s wave 1 score from their wave 2 score. On the right, Yt 1 measures the individual s wave 1 outcome score, which is included because the potential for change may depend on this score (including the so-called autoregressor is standard practice in models of transition). The list of predictors also includes the intercept (b 0 ) and two types of variables: (1) wave 1 explanatory variables whose starting point scores may affect the potential for change in the outcome variable (b 1 X), and (2) variables that measure the change in explanatory variables by subtracting wave 1 scores from wave 2 scores (b 2 ΔX). The discussion focuses on the coefficients of the change variables (the latter type). 16 I used ordered logistic regression in Stata to model changes in political attitudes. To account for the non-independence of information collected from the same respondent in two time periods, robust standard errors were clustered on the respondent's identification number. 17 Table 2 presents results of the analysis of two individual-level outcomes: change in satisfaction with Netanyahu and change in 16 I used survey weights in all analyses, but also ran the analyses without the weights and found nearly identical results. The weights were calculated based on 2013 National Census data, using the variables gender, age, religiosity, and district. 17 Because ordered logistic regression makes the proportional odds (or "parallel lines") assumption, I re-estimated the models using generalized ordered logistic regression with a sub-command that tests whether each explanatory variable violates the proportional odds assumption. The direction and statistical significance of the main effects did not depend on whether the simple or generalized form of ordered logistic regression was used. Therefore, I present the results of the simpler version for ease of interpretation. 17

18 support for or opposition to the decision to embark on a military operation in Gaza in the summer of The descriptive statistics in Table 1 show that between the two waves of the survey, the majority of respondents experienced either a decline of 1-2 levels on the satisfaction scale (55.4 percent) or no change at all (26.4 percent). More dramatic declines of 3-5 levels occurred among only 9.9 percent of the sample, and thus these higher-level categories were combined to avoid a small cell count problem during data analysis. Only 8.3 percent of respondents reported an increasing level of satisfaction and most of these respondents reported a 1-level increase (only four respondents increased their satisfaction with Netanyahu two or more levels). Therefore, all positive changes (becoming more satisfied or less unsatisfied) were combined into a single category. For similar reasons, for the variable that measures approval of the military operation in Gaza, I collapsed all declines of at least two levels into a single category, and all increases in approval into another category. << Table 2 about here >> Table 2 reports the coefficients of variables that measure change in an individual s scores from wave 1 to wave 2 (for the coefficients of other variables in the analysis, see Table S1 in the online supplement). The first two models test the direct effects of change in the assessment of the level of threat posed to Israel by Hamas. This variable had no statistically significant association with either changes in satisfaction with Prime Minister Netanyahu or changes in support for or opposition to the military operation in Gaza. Models 3 and 4 estimate the direct effects of change in identification with the Jewish ethno-national group and nationstate chauvinism, (to account for potential spuriousness, the model also controls for 18

19 patriotism, which is positively correlated with both national identification and nation-state chauvinism). In line with the theoretical predictions, change in satisfaction with the prime minister had a statistically significant positive association with change in ethnonational identification but no statistically significant association with either nationstate chauvinism or patriotism, while change in approval of the military operation had a statistically significant positive association with change in nation-state chauvinism but no statistically significant association with the other two sentiments. The results of Model 5 show statistically significant evidence of the hypothesized effect of change in anger felt toward Hamas on change in approval of the military operation in Gaza. Models 6 and 7 serve as an initial test of the theoretical proposition that the two rally outcomes emerged via separate mechanisms. Each outcome maintained associations with its unique predictors, and had no statistically significant associations with the predictors of the other outcome. To further check the robustness of these findings, I added individual-level demographic control variables such as gender, age, education, marital status, and area of residence to the analysis (due to space limitations I do not present these models). The coefficients of all explanatory variables maintained their size and statistical significance. To assess the substantive significance of the main coefficients of interest, I used the results of Models 6 and 7 to estimate how a decline of one standard deviation in an explanatory variable affected the predicted probability of each category of the outcome variable, while holding all other continuous variables in the model at their means and categorical variables at their medians. The results suggest that the effects were substantial: a one standard deviation decline in ethno-national 19

20 identification increased the probability of experiencing a decline of at least two levels in satisfaction with Prime Minister Netanyahu by 34 percent. At the same time, declines of one standard deviation in nation-state chauvinism and anger increased the likelihood of experiencing a decline of at least two levels in approval of the IDF operation in Gaza by 20 and 23 percent, respectively. In the theoretical discussion, I hypothesized that the effects of change in ethno-national identification on satisfaction with the prime minister would occur primarily among the opposition, because in normal times the individuals in this category oppose the prime minister (who does not represent their party or ideological preferences). Therefore, Model 8 tests an interaction between change in ethnonational identification and respondent s self-categorization as a member of the political right, center, or left. 18 The findings support the hypothesis: the effect of change in ethno-national identification on satisfaction with Prime Minister Netanyahu was much stronger among those in the left category, and the difference was statistically significant. Model 9 tested whether there was a similar moderating effect in the analysis of change in approval of the military operation in Gaza. The results show no statistically significant differences in the effects of change in either nation-state chauvinism or anger on change in support for military intervention between categories of political identification. This finding shows initial support for the claim that the two rally outcomes emerged through simultaneous but distinct processes. 18 In Israel, the right-left division refers to attitudes about the Israeli-Palestinian conflict: In general, the left title refers to supporters of the two-state solution (a small radical left advocates for a binational or non-national state), and right refers to supporters of the principle that the entire land of Israel from the Jordan river to the Mediterranean should be the nation-state of the Jewish People. 20

21 To summarize, the decline in satisfaction with Netanyahu was associated with a decline in ethno-national identification that occurred primarily among those on the left. In contrast, the decline in approval of the military operation in Gaza was associated with declines in nation-state chauvinism and anger, and was not moderated by political identification (or the moderating effect was too weak to produce statistically significant findings). Next, I use structural equation models to test my arguments about mediation effects. In addition, I test alternative explanations of changes in attitudes toward the prime minister and the military operation in the aftermath of the war. Step 2: Using structural equations to test mediation Mediating variables from the focal theoretical framework. To empirically test models that include both direct and indirect effects, I estimated structural equation models (SEM) via STATA. Because some outcome variables are not continuous, I used the generalized version of SEM (GSEM), matching the type of each regression in the model to the level of measurement of the outcome variable (ordinary list squares regression for continuous variables, and ordered logistic regression for ordinal variables). Structural equations allowed me to integrate the analysis of the two rally outcomes in a single model rather than testing them in separate models (as in the previous section). The first model (results shown in Figure 3) included all pathways and mechanisms to rally outcomes described in the theoretical discussion and summarized in Figures 1 and 2. The figure presents the estimated coefficients for variables that measure change in the predictors. In addition, each regression equation included the outcome variables and predictors as measured in wave 1 of the survey, 21

22 but these were omitted from the graphical presentation for ease of interpretation (for the full list of coefficients, see Table S2 in the online supplement). << Figure 3 about here >> The findings support all expectations about mediated effects. To check the statistical significance of indirect effects, I used the bootstrapping procedure recommended by Hayes (2009) with 5,000 repetitions, which produced a biascorrected confidence interval for each mediation effect. 19 All mediation paths shown in Figure 3 successfully passed this strict test. As an additional test of robustness, I estimated a "saturated" model (not shown), which adds a direct path between every pair of variables that are connected by an indirect path. The findings are nearly identical to those of the restricted model (shown in Figure 3), although the effect of perceived collective threat on ethno-national identification, which was only marginally statistically significant in the restricted model, is statistically significant in the full model. In summary, the SEM results suggest that change in the level of perceived national threat affected change in satisfaction with Prime Minister Netanyahu and change in support for the IDF operation in Gaza via two distinct paths of mediation. Jewish Israelis rallied behind Netanyahu's leadership due to increased identification with the ethno-national group, which was at least partly driven by concerns for national security. At the same time, they rallied behind the IDF operation in Gaza, 19 This procedure is superior to other tests of mediation because it does not require two problematic assumptions: (1) that in order to establish that a predictor has an indirect effect on an outcome it must first be established that it has a direct effect (the popular Baron and Kenny procedure makes this assumption), and (2) that the indirect effect has a normal distribution (as in the commonly used Sobel test). 22

23 both because the military confrontation with Hamas generated anger toward Hamas, and because ethno-national identification translated into nation-state chauvinism. Mediating variables from competing/supplemental arguments. In the next model, I add variables associated with competing or supplemental arguments about support for leaders and military action during wars. Figure 4 presents the results of model estimation (for the full list of coefficients, see Table S2 in the online supplement). << Figure 4 about here >> Four supplemental or competing arguments were tested. First, I test the possibility that rather than the two rally outcomes being the result of distinct processes, one outcome caused the other. Specifically, increased support for military action may lead to increased satisfaction with the leader who ordered the action, or alternatively, individuals may lose confidence in the leader following a costly and not very successful military operation, because they blame him for making the wrong decision about going to war or failing to achieve victory. To test this possibility, the model in Figure 4 estimates a direct path from change in approval of the military operation in Gaza to change in satisfaction with Prime Minister Netanyahu. The small and statistically not significant association between the two rally outcomes provides no support for this first competing argument. 20 Importantly, the coefficients associated with the focal arguments about change in satisfaction with the leader maintained their size and statistical significance in this analysis, which controls for changes in military success assessment. 20 To make the presentation of the results more straightforward, the model in Figure 4 treats change in support for military action as a continuous variable when it serves as a predictor. Running the analysis with the variable in its original categorical format yielded similar results. 23

24 Second, an influential argument about public opinion (based primarily on studies in the United States), suggests that the public tends to support military actions assessed as successful or likely to succeed, and object to military actions assessed as unsuccessful or likely to fail (Eichenberg 2005; Gelpi, Feaver, and Reifler 2006; Jentleson 1992; Larson and Savych 2005). The model in Figure 4 tests this argument by including variables that measure change in a respondent s assessment of military success. The findings provide partial support for this argument: a decline in satisfaction with Netanyahu was more likely for respondents whose assessment of military success declined at least two levels than for respondents whose success assessment either declined only one level, remained the same, or increased. At the same time, a decline in approval of the military operation was more likely for respondents whose success assessment declined by one level than for respondents whose assessment declined more sharply. Once again, the coefficients associated with the focal arguments maintained their sizes and statistical significance even when the model included variables that may represent an alternative or supplemental explanation for the outcomes. Third, I tested the possibility that fear caused by a perceived threat to personal security or the security of loved ones is an additional mechanism driving attitudes toward military actions. Political psychologists have shown that high levels of fear lead individuals to be risk aversive and thus less likely to support military action (Huddy et al. 2007; Lerner et al. 2003; Skitka et al. 2006). This prior finding suggests that changes in the level of perceived personal threat will be positively associated with changes in fear that will, in turn, be negatively associated with changes in support for military action. This proposition is supported by the data, but the effect is much smaller than the effects of the focal variables (in standard 24

25 deviation units, the semi-standardized coefficients are as follows: 10.5 for nationsate chauvinism, 3.5 for anger, and only 1.2 for fear). 21 At the same time, the coefficients associated with the core mechanisms once again remain intact. Fourth, a high level of public attention to media coverage during war could affect rally outcomes through mechanisms not included in the discussion or analysis (see Groeling and Baum 2008; Iyengar and Simon 1994; Scheufele, Nisbet, and Ostman 2005). Therefore, not including news consumption as a predictor might generate an omitted variable problem and result in biased findings. For this reason, the model in Figure 4 adds direct pathways from change in news consumption to change in each of the rally outcomes, as well as pathways from change in news consumption to change in assessment of personal and collective threats. 22 The results show a positive association between change in news consumption and change in assessment of both types of threats. In addition, there is a positive association between news consumption and change in support for military action (though it is statistically significant only at the 0.1 level), and a statistically significant negative association between media exposure and change in satisfaction with the prime minister. Explaining these findings in detail is beyond the empirical scope of this 21 However, the findings increase the generalizability of the argument that anger and fear have contrasting effects, which has been put forward in studies of public opinion in the United States (Huddy et al. 2007; Lerner et al. 2003; Skitka et al. 2006). 22 Although the original format of the variable measuring change in exposure to the news was ordinal, a dichotomous version (i.e., distinguishing major declines from a smaller decline, no change, or an increase) better expressed the effects of this variable on other variables in the analysis. Therefore, Figure 4 includes the dichotomous version of the variable (substantively, this version captures the difference between the intensive consumption of news during the war and ordinary levels of exposure to news). 25

26 research. Importantly, all coefficients reported in Figure 3 maintained their sizes and statistical significance in the Figure 4 model, which controls for direct media effects. CONCLUSION This study sought to improve the scholarly understanding of public opinion formation during rally-round-the-flag periods by making five contributions. First, this study extends the investigation of the RRTF phenomenon in wartime beyond the popularity of the head of the state (the focus of most previous studies) by also examining levels of support for the use of military power. The data suggest that each of the two rally effects had its own set of generating mechanisms. The second contribution of the study relates to its empirical scope. Some of the most fundamental arguments about political behavior in general, and the rallyround-the-flag phenomenon in particular, are based on incidents in the United States. Therefore, by taking on a portion of the burden of accumulating evidence from various political fields, the current study contributes to the development of a more general understanding of public opinion formation. Third, the study was designed to test arguments about mechanisms of public opinion formation during rally periods. Because rally periods represent a dynamic and context-dependent aspect of public opinion formation, the study applied a panel design that allowed the measurement of the degree to which each respondent s attitudes, sentiments, and feelings changed over time. Ideally, data would have been collected shortly before and during a rally period, but the unpredictability of rallygenerating events makes such a design unfeasible. Thus, for the current study, data were collected during and after a rally period, which is the optimal alternative. Still, it is possible that the findings represent not rally effects, but rather declines in the leader s popularity and approval of the military operation that emerged due to 26