With democratic backsliding becoming serious concerns in many countries around the world, scholarly interest in this dangerous phenomenon has spiked in recent years1,2. A particular feature of democratic backsliding, is the delegitimization of critical voices from the opposing side. Specifically, dissent—the expression of disagreement with the (dominant) group’s norms, actions, or decisions3,4—appears to be extremely relevant and essential to the functioning of democratic societies, as drawing attention to alternative views and courses of action can facilitate reform and social change5,6,7. Accordingly, efforts to (re)legitimize critical voices in this context may be integral to attempts to protect democratic principles, but the issue has yet to be addressed from an intervention perspective within the vast literature on democratic erosion. Meanwhile, the delegitimization and exclusion of critical voices discourages constructive discourse, leading societies to become more oppressive and thus less democratic8. To safeguard democracy, it is therefore crucial that societies safeguard the freedom of expression and the right to criticize.
In different countries around the world, those who seek to defend democratic institutions and values have been delegitimized in public discourse9,10,11. Namely, civil actors and movements that strive to protect democracy and human rights have found themselves under attack for criticizing regimes and policies in countries like Brazil12, Poland13, Russia14, and Israel15. Even in Western democracies like the United States, critical voices are not always accepted. Planned Parenthood is a case in point: Despite offering several services that are broadly accepted by the public (e.g., the testing and treatment of sexually-transmitted infections), the organization suffers from widespread delegitimization over a subset of its activities, namely the highly-polarized topic of abortion16.
In this research, we examine whether and how the perceived legitimacy of critical voices can be increased in the Israeli context, where civil and political actors have been struggling to uphold democratic values in the face of widening polarization17,18 and an anti-democratic political agenda19. The New Israel Fund (NIF) is one of the most prominent civil actors to have taken on the role of defending Israeli democracy, and it has faced delegitimization as a result. This organization represents many critical voices, especially from the shrinking political left, which have in recent decades faced long and hard delegitimization campaigns from both the political leadership and NGOs and media outlets associated or aligned with it.
As a representative case of both the local and international protection of liberal democracy, we employ the case of the NIF, a high-profile and widely delegitimized umbrella NGO. We conducted this research in collaboration with the aChord Center, an applied research center in Israel, and the NIF. The latter organization funds a multitude of other local organizations working to promote equality, minority rights, fair social services and other fundaments of liberal democracy, as well as organizations promoting an end to Israel’s occupation of Palestine—a position more associated with the Israeli Left. Through these activities, the NGO raises and amplifies voices that are critical of the Israeli government, its policies, and the activities of some of the country’s sacred institutions (e.g., the Israeli army).
Meanwhile, public intolerance of internal criticism has increased alongside the violent escalation in the Israel-Palestinian conflict20, and efforts meant to weaken and delegitimize peace-advocating and left-wing civil society actors in Israel have become more prominent and successful in recent years21,22. Delegitimization campaigns directed at the NIF, labeling its employees as “traitors,” “foreign agents,” and “enemies of the state”22, have focused on the NGO’s support of organizations that publicly document, criticize, and strive to end Israel’s occupation of the West Bank and Gaza Strip. The delegitimization the organization has endured is prototypical for how other critical civil society organization have been treated in Israel and how critical voices are generally delegitimized and often silenced in the context of democratic erosion. Given the crucial role they play in upholding democracy, it is essential to ask: what intergroup psychological interventions might contribute to the public (re)legitimization of critical voices, allowing them to be heard across polarized divides?
To answer our research question, we start by considering the delegitimization process and its psychological underpinnings, rooted in social identification and categorization. The delegitimization process is essentially a process of social categorization, which can take many forms. Specifically, it is the “categorization of groups into extreme negative social categories which are excluded from human groups that are considered as acting within the limits of acceptable norms and/or values”23 (p. 170), and is particularly salient in highly-polarized (e.g., the U.S24.) and conflict-ridden (e.g., Israel25) societies. It inherently follows an “us vs. them” mentality26, and may lead to harmful outcomes, such as conspiracy theories27, self-censorship7, endorsement of non-compromising attitudes, and participation in political violence28.
Notably, whereas delegitimization shares some of the same characteristics with dehumanization29,30, it is a separate phenomenon in that does not necessarily involve likening the group in question to animals, or seeing them as less-than-human. A normative concept in principle, delegitimization is reinforced by societal norms of what and who is appropriate in which context23. Political leaders and the media often encourage public endorsement of delegitimization by depicting critical voices as threatening31. Threat perceptions, in turn, amplify negative attitudes towards the group identified with the threat32,33, and a common reaction to an internal threat is to expel or exclude its source from the group in order to ensure cohesion26.
According to Social Identity Theory (SIT), individuals tend to categorize themselves and others into groups, identify with their own group as part of their self-concept and compare and prefer their group over others34. In this sense, delegitimization of critical voices follows a psychological pattern similar to other forms of political animosity and prejudice: the mere categorization of people into political ingroups and outgroups provides an incentive to discriminate34,35,36. Additionally, any case of ingroup criticism can threaten cohesion and the validity of ingroup norms and induce social identity threat3,25, especially in the presence of external threat37.
By portraying critical voices as separate from and threatening to the group’s status, safety, or even survival, anti-democratic regimes and their supporting structures (e.g., media, government-allied NGOs) create a hostile environment for critical voices, distancing them from the common ingroup identity and presenting them as malicious. In practice, the delegitimization of actors from the ingroup often comprises accusations of disloyalty and serving foreign interests (i.e., by receiving support from international organizations or other countries9,38,39). It is the “othering” of a group, its portrayal as foreign and unfamiliar, that enables its delegitimization.
But can this process be reversed? Paraphrasing Kelman, the process of legitimization refers to the recategorization of a target from illegitimate to legitimate. The ultimate legitimization outcome is therefore the acceptance and perception of a political actor (a system, group or person) as morally acceptable40. Per the relevant literature, we considered two broad approaches to the relegitimization of critical voices.
To address this, interventions based on a social (re)categorization approach, involving highlighting commonalities between the critical voices and the general public (i.e., the mainstream), can either change the perception of groups within an existing common boundary41,42,43 or attempt to recategorize entirely by emphasizing the role of a common superordinate identity, as suggested by the Common Ingroup Identity Model44,45. Following this logic, we developed three categorization-based interventions that we hypothesized would increase the perceived legitimacy of the delegitimized group: (1) highlighting mainstream activities (i.e., activities broadly-accepted within the public consensus) by the delegitimized target, thus highlighting common grounds; (2) constructing a common identity based on a value-based recategorization; and (3) presenting within-group disagreement as strengthening the common ingroup.
A second approach, based on the moralizing and value-oriented nature of legitimization processes, focuses on highlighting the inconsistencies between the delegitimizing attitudes individuals hold and their core values. At its core, the delegitimization of a group might be inconsistent with individuals’ held values and beliefs. This approach assumes that the natural desire for consistency46,47 will lead individuals—who often unknowingly deviate in their attitudes from their core values—to reconsider conflicting attitudes when the discrepancies are highlighted. Interventions revealing such inconsistencies have previously been found effective in reducing support for undemocratic practices48 and prejudice41,49.
This approach can be implemented using several techniques41. First, paradoxical thinking interventions confront individuals with an exaggerated representation of their held attitudes, which—when successful—leads to a sense of identity threat stemming from the inconsistency between this representation and one’s self-image50. Analogy-based interventions, on the other hand, can reveal inconsistencies between attitudes and core values through the presentation of similar cases from different contexts51,52. Finally, interventions applying proximal temporal framing of a potential future problem as more urgent53—for example, by presenting the group’s delegitimization as a slippery slope which is likely to harm democracy and conflict with one own’s preferences and desires—is another way in which an inconsistency between delegitimizing attitudes and core values can be brought to light. Consequently, we hypothesized that interventions employing (1) paradoxical thinking, (2) analogy, and (3) a presentation of the imminent threat of delegitimization to democracy could contribute to a group’s relegitimization.
Our main hypothesis, in accordance with common practice when conducting intervention tournaments49, is that each of the six interventions could increase the NGO’s Perceived Legitimacy relative to a control condition. Due to the polarized context and the NGO’s identification with the political left, we consider a potential interaction with ideology. Israeli leftists are more likely than other ideological groups to support the organization and view it as legitimate. In contrast to leftists, rightists in Israel are ideologically resistant to the NGO’s more controversial activities and are therefore more likely to see it as illegitimate and actively delegitimize it. Finally, centrists, a growing group in Israel with considerable political power, may be slightly more sympathetic to the organization than rightists, but are nonetheless affected in their attitudes by the mainstream delegitimized view of the NGO.
With this classification in mind, we treat legitimization as a holistic process. Unlike existing known delegitimization scales in the literature25,28,54, which mostly aimed to capture Jewish Israelis’ delegitimization of Palestinians, in this research, in the context of democratic backsliding, it was important for us to capture the legitimization of civil society and other critical voices, who share a common ingroup with the delegitimizing actor(s). Such societal legitimization might comprise different desired outcomes for different audiences: The first, which is the focus of this paper, addresses the need to increase legitimizing attitudes and behaviors towards the delegitimized group—i.e., the group’s perceived legitimacy—among those who view it more hostilely (i.e., mainly rightists and centrists). The second, outside the scope of this paper, is concerned with enhancing the active engagement (i.e., sense of identification and wish to act together) of those closer and more similar to the delegitimized group (in this case, political leftists).
Method
The study consisted of two waves of online surveys, with the second measurement taking place about 2 weeks after the first. The study was approved by the Hebrew University Ethics Committee. All analyses were conducted in R Version 4.3.0. All data was collected online and saved through the Qualtrics survey platform. T1 data was collected between December 10, 2020 and February 16, 2021. T2 data was collected from February 21, 2021-March 21, 2021 following pre-registrations on AsPredicted on February 19 and February 21, 2021 (see & We obtained informed consent in the beginning of each survey, and participants were compensated for their participation by the survey company. All statistical tests were two-sided unless otherwise noted. Data distribution was assumed to be normal but this was not formally tested. Although the study was pre-registered, several methodological adjustments were made during its execution. Below, we outline the key deviations from our original pre-registration.
Deviations from pre-registration
Several deviations from our pre-registered plan occurred due to practical and methodological considerations. First, the study was initially pre-registered as two separate studies (differed primarily by the specific interventions included in each) due to an unplanned expansion of intervention conditions. To ensure a robust baseline, we later merged the datasets and combined the two (identically-operationalized) control groups after confirming no significant baseline differences. Second, due to an oversight, our pre-registration stated that no data had been collected at pre-registration. Although hypotheses were formulated solely based on expected intervention effects at T2, which was collected post-registration, it would have been more accurate to say that some data (i.e., the baseline measurement, T1) has already been collected.
In terms of our hypotheses and DVs, our original hypotheses emphasized total of nine intervention effects on specific ideological groups, according to its relation to the outgroup in question and estimated ideological fit of the different interventions included in the project. However, as we decided to focus the paper on perceived legitimacy (rather than the other engagement factor, see more below), the main text zooms in on a subset of the six interventions that are relevant to the theoretical focus, with results pertaining to the other interventions included in Supplementary Note 2 and Supplementary Table 7. Strong main effects across the full sample led us to focus on overall effects in the main text, with group-specific analyses vis-à-vis the pre-registered hypotheses also reported in Supplementary Note 2. Additionally, we pre-registered a three-component measure of legitimization but conducted an Exploratory Factor Analysis (EFA) to optimize item compatibility with our study’s unique legitimacy measure. The resulting two-factor structure (Perceived Legitimacy and Engagement) was adopted, while results using the pre-registered approach are reported in Supplementary Note 3.
The following deviations were made from the registered analysis: we used a linear mixed-effects model (lmer in R) instead of a traditional mixed within-between ANOVA to better account for repeated measures (T1 and T2) and the three-way interaction of time, condition, and ideology. Mixed-effects modeling provided greater statistical power and robustness to unbalanced data. In addition to our pre-registered exclusion criteria (failing attention checks), we excluded duplicate responses and participants who did not complete all DV items at both time points as including either would interfere with our intention to properly analyze the data. Three exploratory variables—Willingness to Act on Facebook, Perceived Threat and Support for Violence—were included in the main text due to their relevance to group legitimacy. Finally, we pre-registered an expected 20% dropout rate, estimating a final sample of ~1947 participants. However, the actual dropout rate was 30.5%, resulting in 1,691 responses at T2. Analyses comparing dropouts versus participants who completed both waves revealed no significant differences in baseline legitimization levels (t(1478.2) = −1.28, p = 0.200, 95% CI [−0.19, 0.04]) or ideological group identification (χ²(2) = 4.20, p = 0.122), suggesting attrition was not systematically related to key variables.
Participants and procedure
We reached out to participants online via an Israeli survey company (iPanel). As per our pre-registration, we removed participants who failed attention checks (n = 137 in T1). In addition, we excluded from analysis duplicate responses (n = 36 in T1) and participants who did not complete all DV items, either because they were screened in the beginning of the survey based on demographic information or dropped out (n = 891 in T1). In Wave 1, we recruited 2433 Jewish Israelis (Mage = 45.6, SDage = 15.8, range: 18–86). Of them, 52% self-identified as men (n = 1266), 48% as women (n = 1167), and none chose the option “prefer not to say”. In terms of political ideology, 30% of participants self-identified as rightists (n = 742), 31% as centrists (n = 749), and 39% leftists (n = 942). We intentionally recruited large-enough sub-samples of participants who identify with Israel’s three main ideological groups to allow for the proposed moderation analysis. They were invited to take part in a study on social and political issues. They completed informed consent followed by a short demographics questionnaire followed by number of measures regarding different political groups and organizations (for the full list of measures, see Supplementary Methods; measures relevant to our study are reported below).
We invited all participants from the first wave (T1) to participate in the intervention tournament in the second wave (T2), reaching a total of 1691 participants (the 1220 exposed to the relevant interventions were included in the analysis presented in the paper). As per our pre-registration, we removed participants who failed attention and reading checks (n = 251 in T2). In addition, we excluded from analysis duplicate responses (n = 85 in T2) and participants who did not complete all DV items, i.e., dropped out early (n = 138 in T2). Participants were invited to participate in another study on social and political issues. The sample was also representative in terms of age and gender (Mage = 44.9, SDage = 15.3, range: 18–86). Of them, 52% self-identified as men (n = 873), 48% as women (n = 818), and none chose the option “prefer not to say”. In terms of political ideology, 30% of participants self-identified as rightists (n = 508), 30% as centrists (n = 506), and 40% leftists (n = 677) (see Supplementary Table 1 for a summary of demographic characteristics, and Supplementary Table 2 for demographics per condition).
After completing informed consent and an attention check, participants were assigned to conditions through block randomization (based on their political ideology—right, center, or left—and baseline perceived legitimacy level, resulting in a total of 12 blocks) to one of the conditions and responded to the same outcome measure as well as other exploratory items (see Supplementary Methods). They were informed that they would be presented with a recent post from the NGO’s Facebook page and then asked to respond to questions about it. Following the exposure to one of the conditions, they responded to a few questions about the post (e.g., “how would you react if you saw the post on Facebook?”), the main outcome variables and some exploratory measures.
A sensitivity analysis using R for a between-subjects design with 7 independent conditions, a sample size of 1220, an alpha level of 0.05, and a desired power of 0.80, indicated that the study would have 80% power to detect an effect size as small as Cohen’s f = 0.106, which corresponds to Cohen’s d = 0.213. Of individuals who entered the survey in T1, 3% did not agree to participate and another 8% dropped out while filling out the survey. Of individuals who entered the survey in T2, 1% did not agree to participate and another 7% dropped out while filling out the survey.
Construction of the perceived legitimacy measure
For the perceived legitimacy outcome, we opted to use a self-created measure, as existing scales did not fully capture our intended construct. Prior delegitimization measures25,28,54, which were constructed in the context of Jewish-Palestinian relations and the Ethos of Conflict54, did not adequately address the political dimension of our research interest. Similarly, existing measures of political (in)tolerance55,56, while applicable in political contexts, did not sufficiently capture the normative element of legitimacy, specifically when referring to an NGO.
We synthesized components from both types of measures. For instance, our item “In my opinion, people who [support the NGO] are traitors,” is in line with Hammack et al.’s28 delegitimization item “Most Palestinians support terrorism”, while our item “In my opinion, it is appropriate for artists and public figures to participate in events organized or supported by [NGO name]” resembles a reversed adaptation of Crawford’s56 political intolerance item “I think that this group should not be allowed to organize in order to influence public policy.” Maintaining the theoretical conceptualization of delegitimization as a process of social categorization, we included items considering ingroup membership (e.g., “In my opinion, it is appropriate for organizations to receive financial support from [NGO name]”). Additionally, we incorporated context-inspired items such “In my opinion, [NGO name] is a legitimate body” and “I would boycott an event organized by [NGO name]”.
To identify the underlying factors behind the outcome measure composed of legitimizing and delegitimizing perceptions and sentiments regarding the NGO, we conducted EFA. In T1, we included 16 self-report items tapping into perceived legitimacy, beliefs, attitudes, desired behavior, and identification regarding the NGO, all rated on a 1 (not at all)—6 (to a great extent) scale. We retained 13 items in T2, removing three that we deemed repetitive or less relevant (see more information in Supplementary Methods). The mean score of the 13-item scale in T1 was 3.00 (SD = 1.2), just below the midpoint (α = 0.95). First, we evaluated the correlations between the variables and confirmed that none were too highly correlated (r > 0.90) (Fig. 1). A Kaiser-Meyer-Olkin test (Overall MSA = 0.94) and Bartlett’s test for sphericity (χ² = 31532.99, df = 78, p < 0.001) confirmed that the items were sufficiently correlated to perform factor analysis57,58.
The heatmap displays Pearson correlation coefficients (r) between all study variables for the full T1 sample (n = 2433). Positive correlations are shown in blue colors, and negative correlations in reddish-orange colors. The intensity of the color indicates the strength of the correlation, with darker colors representing stronger relationships. White represents correlations near zero. Statistical significance is indicated by asterisks (*p < 0.05, **p < 0.01, ***p < 0.001). Correlation values are displayed in each cell with corresponding significance markers. Non-significant correlations are displayed in lighter shades. Variables are clustered hierarchically based on similarity of correlation patterns. This visualization uses a colorblind-friendly blue/reddish-orange diverging palette that maintains distinction for all types of color vision.
We performed both scree plot analysis and parallel analysis to determine the optimal number of factors. The scree plot indicated a clear 2-factor solution with a sharp drop after the first and second factors (eigenvalues: 8.07, 1.80, 0.73) (see Fig. 2), while parallel analysis suggested a possible 3-factor solution (see Fig. 3). We conducted EFA using principal axis factoring with quartimin rotation, as we anticipated the factors would be moderately correlated.
The scree plot displays eigenvalues (y-axis) for each factor (x-axis) extracted from the factor analysis. Factors are arranged in descending order of eigenvalue magnitude. The blue line with markers represents the eigenvalues, while the dashed red horizontal line indicates the Kaiser criterion (eigenvalue = 1) commonly used for factor retention decisions. Based on this criterion, two factors were retained (eigenvalues > 1). This visualization uses a colorblind-friendly blue and red color scheme to ensure accessibility for all types of color vision.
The parallel analysis scree plot compares observed eigenvalues from factor analysis (blue solid line with triangles) with eigenvalues from simulated data (black dotted line with circles) across factors. Factors where the observed eigenvalue exceeds the simulated eigenvalue are considered significant and retained for further analysis. Based on this analysis, three factors were retained as suggested by the parallel analysis results (parallel analysis suggested number of factors = 3). The visualization uses a colorblind-friendly color scheme with distinct line styles and shapes to ensure accessibility for all types of color vision.
We compared 2-factor and 3-factor solutions and selected the 2-factor model based on theoretical interpretability, parsimony, and the Kaiser criterion (eigenvalues > 1). This solution, supported by eigenvalues of 7.82 and 1.44, explained 64.6% of variance, with factor loadings ranging from 0.36 to 0.87 (Factor 1) and 0.56 to 0.87 (Factor 2), and item communalities from 0.44 to 0.87. The two factors, which distinguish legitimacy judgments from engagement intentions, were moderately correlated. While the 3-factor solution explained slightly more variance (67.8%), the third factor’s eigenvalue (0.46) fell well below the conventional Kaiser criterion threshold of 1.0, and it added minimal explanatory power relative to the increased model complexity. Thus, the 2-factor solution offered a clearer, more parsimonious structure. For items that cross-loaded on both factors, we assigned them to the scale where they had the higher loading. We calculated composite scores for each factor by averaging the respective items. See Supplementary Table 3 for full factor loadings.
Eventually, the first factor comprised items that refer more directly to the perception of the NGO as politically legitimate (e.g., “In my opinion, the [NGO name] is a legitimate body,” “I would boycott an event organized by the [NGO name]” [reversed]). The second factor represented closer engagement with the group (e.g., “I feel that the [NGO name] represents me through its activities and the organizations it supports,” “I would consider taking part in an activity organized by the [NGO name]”).
Cronbach’s alpha coefficients in T1 for Factor 1, hereby referred to as Perceived Legitimacy, and Factor 2, hereby referred to as Engagement, were both 0.93 in T1, indicating excellent internal consistency. As expected, the two factors were moderately to strongly correlated (r(2431) = 0.68, 95% CI [0.66, 0.70], p < .001), supporting the use of an oblique rotation. We calculated composite scores for each factor by averaging the respective items.
Materials
In T1 participants filled out a long questionnaire with most items being used for exploratory and applied purposes by the NGO. Participants who returned in T2 completed a number of measures examining their reactions to the post and their perceived legitimacy levels, as well as some exploratory measures (for all measures see Supplementary Methods).
Unless otherwise noted, participants responded to all items on a 1 (“not at all”)—6 (“strongly agree”) scale. The following measures were included:
Demographics (T1)
Participants reported demographic information including their age, gender, religiosity, education, income level, and area of residence.
Political ideology (T1)
Political ideology was determined based on an ideological identification question ranging from 1 (extreme right) to 7 (extreme left). Political ideology can carry different meanings depending on the context59. In the highly affectively polarized context of our study, the self-defined categories of “left,” “right,” and “center” relate to the delegitimized group in different ways. Specifically, this self-categorization signals a specific orientation towards Palestinians in Israel and the Occupied Palestinian Territories (with leftists more likely to support an end to the occupation and the peaceful resolution of the Israeli conflict and rightists more likely to view Palestinians as a security threat that needs to be controlled), which carries implications to their orientations towards groups that are critical of the state’s policies in this regard. We therefore treat political ideology as a categorical variable, coding participants who chose 1 (extreme right)—3 (moderate right) as rightists, those who chose 4 (center) as centrists, and those who chose 5 (moderate left) -7 (extreme left) as leftists.
Perceived legitimacy (T1–T2)
Following the results of the EFA, Perceived Legitimacy consisted of seven items (αT1 = 0.93; αT2 = 0.94), e.g.,: “In my opinion, the [NGO name] is a legitimate body”; “In my opinion, people who support the [NGO name] are traitors” (reversed item); “In my opinion, it is appropriate for artists and public figures to participate in events organized or supported by the [NGO name]”; “I would boycott an event organized by the [NGO name]” (reversed). We used T1 as a baseline measurement of the NGO’s Perceived Legitimacy. The full list of items appears in the Supplementary Methods.
Willingness to act on Facebook (T2)
The willingness to share the intervention post on one’s Facebook page was measured through one item: “to what extent would you be likely to share this post on your Facebook page?”
Perceived threat (T1–T2)
The perception of threat from the NGO was measured using two items that tap into threat on the symbolic (i.e., “The [NGO name] harms the values and identity of Israeli society”) and realistic levels (i.e., “The [NGO name] constitutes a real threat to the security of the state”). The items were strongly correlated at both T1 (r(2431) = 0.88, 95% CI [0.87, 0.89], p < 0.001) and T2 (r(1682) = 0.86, 95% CI [0.84, 0.87], p < 0.001). The T2 measure was used in a t-test as an exploratory outcome variable.
Support for violence (T1–T2)
Support for violence in its various forms was measures four items (i.e., “To what extent do you agree with the use of each of the following practices towards political opponents?”), including support of shaming, verbal abuse, vandalism and physical harm. The scale showed excellent internal consistency at both T1 (α = 0.94, 95% CI [0.93, 0.94]) and T2 (α = 0.90, 95% CI [0.89, 0.90]). The T2 measure was used in a t-test as an exploratory outcome variable.
link
