The power of a promise: whom do governments’ security justifications convince to accept surveillance?

ABSTRACT To justify surveillance measures and gain them public support, governments use the promise of security. It is usually claimed that individuals are more willing to have freedom and privacy restricted than waiving a promise of increased security. However, empirical evidence to support this claim has been scarce—especially from a comparative perspective. Focusing on surveillance measures, this paper shows that people do indeed express greater acceptance of restrictions when these are justified by promises of security, being one of the first to examine this across 29 countries on all continents. Based on data from the ISSP, it investigates to which degree the effect of a security-based justification is moderated on the micro and macro level, with surprising results: While the effect does not differ between different levels of government support and political orientation, it differs significantly depending on how liberal-democratic the country is. The effect of the security-justification is very pronounced in liberal democracies, while it is even reversed in rather autocratic countries, meaning that individuals seem to be rather suspicious towards security justifications in non-democratic countries.


Introduction
In the last two decades, research on public opinion surrounding security, freedom, and privacy has become abundant. When dealing with attitudes, much of the literature relies on the assumption that security preferences are shaped by a security-liberty trade-off, assuming that individuals choose their preferred policy by advantaging one value (i.e. security) at the expense of the other (i.e. freedom or privacy) (e.g. Davis and Silver 2004;Garcia and Geva 2016;Merolla and Zechmeister 2009), and is drawing on social psychological ideas such as Maslow's (1943) hierarchy of human needs. The need for security is believed to be so important to humans that it comes second only to the need to have physiological requirements met, such as hunger and thirst, while liberty is rather categorized as luxury item. A growing number of researchers, however, casts doubt on the trade-off framework and the assumption that security and liberty cannot be achieved simultaneously, when in fact they are not necessarily antagonistic but may be compatible (Degli Esposti, Pavone, and Santiago-Gómez 2017, 72). Furthermore, it has been shown that individuals, too, do not necessarily consider security and liberty to be antagonistic or even to be intrinsically linked (e.g. van den Broek et al. 2017, 15). However, individuals are exposed to the juxtaposition of these two concepts in daily life, especially when it comes to privacy intrusions. Governments justify surveillance by emphasizing individual benefit and collective welfare and cautioning citizens about the dangers posed by crime and terrorism. In the public discourse, the promise of security is the prevailing legitimizing principle for surveillance (Bajc 2007(Bajc , 1579. Governments declare to their citizens that their proposed measure will enhance security-and thus makes potential restrictions of civil liberties inevitable. Rendering an issue a security problem underscores its urgency and allows for greater latitude in addressing that issue and to enforce more far-reaching measures (Bajc 2007(Bajc , 1579. 1 In this context, security can be understood in terms of citizens' safety-elsewhere defined as 'freedom from fear' or 'human security' (Pavone and Degli Esposti 2012, 558). Thus, even though individuals seem to know that privacy and security are not necessarily antagonists, they are constantly told that they must accept restrictions to the former to enjoy the latter.
Interestingly, the focus of the literature seems to be so much on the (non-)meaningfulness of trade-off, balance, or equilibrium images that the role of how security promises might affect privacy attitudes has not been thoroughly addressed yet. This is surprising since promissory politics in general and the promise of security in specific are assumed to significantly influence individuals' preferences (Aradau 2014). People might take security promises into account when forming their opinions on policies since they (a) permanently want to diminish the threat of possible future damages (e.g. Beck 1992) and (b) are exposed to the state whose legitimacy is based on promising security to its citizens (Campbell 1992, 56; see also Hobbes 2 ). It is believed that governments are increasingly taking advantage of this and justifying more and more policies with the need to safeguard security (Königs 2022).
Even though some authors have noted that the interplay between attitudes towards privacy and security might depend on cultural, social, and geographic contexts (e.g. Budak, Rajh, and Recher 2017;Menichelli 2017), the degree of variation and factors involved have not been thoroughly investigated. The purpose of this paper is not to pitch privacy against security and to draw causal conclusions but merely to find out what role the promise of security plays in the evaluation of certain privacy interventions by the government. Our central research questions are: (1) Do privacy attitudes differ when surveillance is employed with the promise of security? (2) How can we explain differences between attitudes towards security-justified and non-justified surveillance?
Based on data from the International Social Survey Programme (ISSP) and additional data at the country level, we conduct a cross-national multilevel analysis in 29 countries with overall more than 22,000 respondents. Literature assumes that when governments justify surveillance measures to enhance security, people are more willing to give up freedoms. There is, however, no cross-national empirical evidence of this hypothesis so far. Our findings suggest that a security frame does indeed lead to significantly higher acceptance of surveillance measures. The effect, however, varies significantly across individuals' predisposition on the micro level as well as their context on the macro level, namely the country they are living in. The regression results show that the 'framing effect' correlates strongly positively with the level of liberal democracy in the respondent's country. Support for government and political orientation, however, do not correlate significantly with the difference in acceptance. These results are particularly interesting against the background of the discrepancy between the presumed increasing demand for security on the one hand and the simultaneously growing skepticism about surveillance measures on the other hand. This implicates that the justification of restricting privacy for security reasons leads to greater resistance in societies that are less liberal-democratic.

The promise of security
Although the idea to watch over people might be as old as civilization itself (Laidler 2008, 17), it was considered only 'little more than a prop for thrillers or science fiction movies' (Petersen 2012, 7) without causing any palpable change in the attention paid to privacy questions for a long time. Today however, the phenomenon of surveillance pervades almost all aspects of social life in Western societies (Lyon 2009, 449). Since the middle of the last century not only the types and number of surveillance technologies have gradually increased but also the number of people and places that are being monitored. Governments, corporations, and individuals gather, store, process, assess, and use data on individuals or groups involving asymmetric power relations, potentially leading to control, manipulation, and domination (Fuchs 2013). In the public discourse, one often differentiates between government (or state) surveillance and corporate surveillance. Historically, surveillance is closely connected to the evolution of the nation state and modern bureaucracy (Giddens 1987, 181). 'Government surveillance' is used to define measures initiated and carried out by the executive branch of the state. It remains a popular term even though it is hard to define who stands for 'the government' or 'the state' since lines between public and private sector become increasingly blurred (Lyon 2015, 145).
Governments have an intrinsic interest in ensuring that the public evaluates their policies positively or, at least, does not reject them-on the one hand, so that the policies can be implemented more easily, and on the other, because it wants to be re-elected. Research on public opinion has repeatedly shown that 'patterns of selection, emphasis and exclusion that furnish a coherent interpretation and evaluation of events' (Norris, Kern, and Just 2004, 4)-a process coined 'framing' (Goffman 1974)-can influence citizens' thoughts and support of issues. Put simply, a frame, like framing something as relevant for national security or promising that a measure will enhance citizens' security, is a rhetoric element used to set and influence the agenda-for instance by politicians who are seeking public support for their policies. In most cases, frames subconsciously emphasize certain values and/or core beliefs (Chong 1996).
Even though academia is interested in many facets of the privacy-surveillance nexus, the views of citizens on (government) surveillance have not yet been thoroughly investigated (Reddick, Chatfield, and Jaramillo 2015, 129). Only a few studies have focused on personal attitudes concerning government surveillance so far and systematic understanding of citizens' views on and behavioural responses to government surveillance is lacking (ibid.). Due to definitory differences, authors adapt diverse theoretical models to explain people's attitudes (Watson, Finn, and Barnard-Wills 2017). Contrasting security to liberty is one of the most popular approaches in surveillance and privacy research (see contributions in Friedewald et al. 2017). This approach is often applied in the form of discrete-choice models (e.g. Patil et al. 2014) or as a conflict of values. Based on the assumption that competing values and demands in a society can only be realized at the expense of each other, privacy and security are conceptualized as a virtual zerosum game. As outlined above, the meaningfulness of this approach is debated and many scholars are very skeptical of the binary trade-off model, especially given that despite political framing as such, individuals do rather not seem to contrast privacy and security (e.g. van den Broek et al. 2017, 15). Generally, several studies have shown that public response to surveillance varies between individuals and across countries (e.g. Dencik and Cable 2017, 764). The question, however, to what extent individuals react differently to security promises, i.e. frames, and how these differences can be explained has been dismissed in the literature so far.
Surveillance measures can be seen as government policies (Nakhaie and de Lint 2013). It is assumed that citizens have a positive attitude towards policies if they (are likely to) benefit from the corresponding content (Trüdinger 2020, 319). Security is considered a human need; we thus argue-in line with the framing and securitization literature-that: H1: Compared to no justification given, a security frame enhances attitudes towards government surveillance. 3

Whom do justifications convince?
Framing effects are not universal but shaped through individual characteristics by serving as 'critical intervening variable[s] between opinion and information' (Haider-Markel and Joslyn 2001, 522). We thus hypothesize that there are variables that moderate the postulated relationship, namely influencing its direction and/or strength. While the securitization literature has not yet extensively addressed the question of who is (more likely to be) securitized 4 , there has been growing interest in how framing might lead to moderation and mediation of attitudes (Borah 2011, 252). From the basic assumption that the human capacity to process information is limited (Fiske and Taylor 1984) scholars draw different conclusions regarding the effect of framing. On the one hand, there are those who attribute a mitigating effect to frames: when a frame is added to the equation of attitude formation, other relevant considerations are trumped by the new thoughts stimulated through the frame (Lee, McLeod, and Shah 2008, 703). Lee, McLeod, and Shah (2008) call this stimulation of thoughts 'hydraulic patterns' since the new considerations abruptly become the crucial element suppressing everything else. In their study on attitudes on framing of counterterrorism policies, Brinson and Stohl (2012) find empirical evidence for this mitigating effect of frames. On the other hand, scholars find that framing might lead to an intensification of preexisting attitudes. In his theory of 'motivated reasoning', Kunda (1990) argues that in their attitude formation, individuals draw on existing 'preconceptions' and thus evaluate new policy-related information consistently with their prevailing opinion. Motivated reasoning thus leads to belief polarization (Cook and Lewandowsky 2016). Individuals with different preconceptions may become even more distant from each other by the same new information. Studies show that frames have a strong polarization effect on politically contentious issues (e.g. Dharshing, Hille, and Wüstenhagen 2017).
In view of our second research question, we are therefore interested in how respondents' attitudes towards government surveillance differ when a security justification is added. From the literature, we generate two possible predictors on the micro level (individual's predispositions or attitudes)-government support and political orientation. Cross-national research has repeatedly shown that context can influence individuals' attitudes. We thus derive an additional macro predictor to account for differences between countries, namely a country's level of liberal democracy. Individuals are nested within countries, meaning that the respondents in one country are equally exposed to the macro variables (e.g. a country's economic or political situation).
There is much literature on effects of government support on individuals' attitudes toward (proposed) policies. One of the findings is that the more trustworthy citizens perceive the government to be, the more likely they are to consent to its demands and regulations (Levi and Stoker 2000). This also seems to hold true with regard to the acceptance of surveillance measures (Davis and Silver 2004;Friedewald et al. 2015, 19;Nakhaie and de Lint 2013, 160;van den Broek et al. 2017, 27): Government surveillance is approved when the state is trusted to handle the assessed data with care. At the same time, one can assume that individuals do also trust the state to not misuse the urgency of a security need. In the securitization literature, securitization moves are assumed to only be socially effective if securitizing agents (i.e. the government) can emphasize the existentiality of a threat to justify an emergency action (Buzan, Waever, and Wilde 1997, 6). If the agent does not enjoy the necessary inter-subjective relationship to convince the audience of the exceptionality of the action, the act of securitization is not successful (see e.g. Vultee 2010, 88). We thus hypothesize that highlighting national security as a justification for surveillance measures should be particularly persuasive to those who tend to be supportive of the government. H2: Government support is positively correlated with the susceptibility to the security frame. 5 Policy researchers have shown that the strength of framing effects particularly depends on political orientation (Chong and Druckman 2013;Dharshing, Hille, and Wüstenhagen 2017). Left-right ideology is seen as a 'product of motivated social cognition' (Jost and Hunyady 2005, 264). Supporters of the two camps usually disagree on a whole range of issues and often do not share a common underlying understanding, e.g. regarding taxes, migration, or sexuality. It is assumed that the quotidian distinction between the political left and the political right tracks 'pervasive ideological differences in orientation toward uncertainty and threat' (Jost and Napier 2011, 91). Jost and Napier (ibid.) provide some evidence that uncertainty avoidance and more attention to threats are associated with more conservatism and less liberalism.
Empirical findings almost cohesively conclude that people who report a rightist political orientation will be more supportive of surveillance measures (e.g. Friedewald et al. 2015, 21). van den Broek et al. (2017, 29) explain their findings by the fact that in general, left-wing oriented respondents are more critical of public actors' activities while 'law and order' have become core issues of rightist policies. Based on Jost and Napier's (2011) findings, we assume that rightist attitudes are closely connected to uncertainty avoidance and more attention to threats, and thus to being more susceptible to security frames while leftist individuals might be less likely to 'buy' governments' emphases of national security. H3: Rightist political attitudes are positively correlated with the susceptibility to the security frame.
According to Brettschneider (2010, 77), a country's commitment to democratic government implies a commitment to protecting individual privacy. Collective self-determination, for example in the form of political participation opportunities, presupposes a minimum of individual self-determination. In democracies, citizens must be granted the right to resolve their private affairs for themselves. In public discourse, intensive government surveillance is connected to rather autocratic regimes. There is historical evidence of autocratic surveillance states (e.g. Nazi Germany or the Soviet Union) as well as current (e.g. China); however, there is not necessarily a correlation between the level of democracy and the density of surveillance. Empirical data show that there are countries that rank high in democracy indices and still apply high levels of surveillance measures, for example Denmark (Pavone, Degli Esposti, and Santiago-Gómez 2015, 70). Generally, it is argued that government's goal behind employing surveillance is safeguarding order and control (Weller 2012, 58). It is often broadly justified to fight 'enemies of the state' (Bigo 2012, 280). The difference between democracies and autocracies lies in the question who is defining the 'enemy of the state', i.e. the source of insecurity. While in democracies, there must be a minimal consensus on what qualifies as source of insecurity, this is not necessarily the case in non-democratic countries. Since non-democratic governments can be held less accountable for their surveillance measures than democratic ones (Dandeker 2005, 40-42), security is also more often used as a reason for surveillance that serves other purposes (Wirth, Maier, andLaumer 2019, 1340). Security justifications should thus be more accepted when made by governments in liberal countries then in autocratic ones.
H4: A country's level of liberal democracy is positively correlated with the susceptibility to the security frame.

Data and methodology
In contrast to classical framing experiments, data are not derived by comparing framing effects of different experimental groups with a reference group but by comparing answers to two questions of the same respondents. Data stem from ISSP's 2016s Role of Government module (ISSP Research Group 2018). Question (a) reads: 'Do you think that your government should or should not have the right to monitor e-mails and any other information exchanged on the Internet?' and question (b) is: 'Some people think that governments should have the right to take certain measures in the name of national security. Others disagree. Do you think that your government should or should not have the right to collect information about anyone living in your country without their knowledge?' with a four-point response scale each. We make use of the assumption that the questions are 'alternative descriptions of what is essentially the same decision problem' (LeBoeuf and Shafir 2003, 77). While (b) emphasizes that the measures are taken 'in the name of national security', (a) does not. 6 We will thus in the following speak of question (a) as unjustified and (b) as security-justified. In our analysis, the unjustified statement provides 'the baseline preference against which to judge the impact of framed conditions' (Chong and Druckman 2007, 106). While this design does not enable us to draw causal relationship, it is well suited for analyzing the relationship between response patterns due to the size of the data set.
It is obvious that the questions do not only differ in the security frame but also regarding different forms of surveillance. Question (a) is more specific than question (b). 'Monitoring e-mails and other information exchanged on the Internet' is more easily imaginable than 'collecting information'. It is assumed that respondents have stronger opinions on specific question than on general questions (Kaplan, Luchman, and Mock 2013). The effect of the security-justification is thus underestimated (if at all) rather than overestimated. In addition, we assume that individuals link the second sentence of question (b) ('Do you think your government should … ') with the first about the provision of security.
Here, too, we can assume that the effect of the security-justification is rather underestimated. In a similar vein, one could argue that any justification (as outlined above, this could also be monetary or convenience justifications) would enhance the acceptance of a measure. It remains opaque whether respondents see a salient difference between 'monitoring' and 'collecting information'-and their equivalents in the translated versions of the questionnaire. While we are aware of the differences between the two questions, the combination of the above assumptions suggests that the effect of security-justification tends to be underestimated and that our assumptions are conservative rather than too liberal. Since items and modules are not randomized in the ISSP, there is the potential of anchoring effects. However, since (a) is always posed before (b) at least the security frame cannot influence the attitude towards the unjustified item.
To identify differences in the response behaviour of single individuals, our analyses are based on stacked data. We duplicated the data for each respondent and their response behaviour; a dummy variable allows us to distinguish the originals (0) from the copies (1). In the next step, we created a new variable 'attitude towards surveillance' and assigned the answer to the unjustified item (a) to the originals (0) and the answer to the security-justified question (b) to the copies (1). In the following, a three-level model (level 1: differences within respondents, level 2: differences between respondents, level 3: differences between countries) was calculated. The first hypothesis is tested based on the association between 'attitudes towards surveillance' (dv) and dummy (iv), while hypotheses 2-4 are tested based on the association between 'attitudes towards surveillance' (dv) and the interaction of the corresponding variable with dummy (iv).
Although it is a recognized explanatory variable in empirical social research, this ISSP module does not ask about trust in government. To determine respondents' government support, we decided to use their voting behaviour and whether they voted for the incumbent government. We share Anderson and Tverdova's (2003:, 103) assumption that '[by] determining who wins and who loses, elections create a lens through which people evaluate their political context' (see also Houston et al.'s (2016) empirical validation). Equating incumbent government vote with trust in government is certainly not optimal but seems to be the best of the available options for including an explanatory variable that is so central to the model given the existing literature. Government party data stems from the Database of Political Institutions (Cruz, Keefi, and Scartascini 2016). Using the voting variable has two disadvantages: Firstly, the question is not asked in some countries (see Fn. 11). Secondly, in some countries of our sample (Australia, Belgium, Turkey) voting is mandatory which might influence government support (as proposed by Lundell 2012). In our sample, there is neither a significant relationship between mandatory voting and government support, nor is participation significantly higher in these countries than in others in the ISSP data set. We thus believe that whether or not voting is mandatory should not influence the results.
Political orientation is also determined via respondents' voting behaviour. The underlying assumption is that there is ideological congruence between voters' preferences and their political representatives, a linkage that is found in many empirical studies (e.g. Golder and Stramski 2010;Klingemann et al. 2006). Accordingly, we linked the ISSP individual data with data from the Manifesto Project (Volkens et al. 2020) via the respective elected parties. For all respondents who indicated their voting decision, the Manifesto Project's left-right variable is used to reflect the political orientation. It is based on the work of Laver and Budge (1992) and measures political orientation on a scale from far left (−100) to far right (100).
For a measure of liberal democracy, we used the Liberal Democracy Index from the 'Varieties of Democracy Project' (V-Dem) (Coppedge et al. 2021). In V-Dem's understanding, liberal democracy 'embodies the intrinsic value of protecting individual and minority rights against a potential 'tyranny of the majority' and state repression more generally' (Coppedge et al. 2016, 582). The index is based on expert evaluations of a country's electoral democracy plus three additional components: the rule of law ensuring respect for civil liberties, judicial constraints on the executive branch, and legislative checks and oversight of the executive (ibid). 7 The literature suggests that not only does the status quo have an impact on respondents' attitudes but that the experience of authoritarian regimes continues to have cultural effects on attitudes toward surveillance today. Research on surveillance in post-communist societies is particularly extensive (see e.g. Los 2003Los , 2005Svenonius and Björklund 2018). While it is beyond the scope of this article to deal with this in detail, an additional model was calculated to control for the post-communist legacy. 8 We control for age, gender, education (ISSP data) and the country's terror legacy 9 (data from University of Maryland's Global Terrorist Database (2019)). In addition, we calculated an additional model including a 'migration background' 10 as further control variable.
The full data set comprises more than 48,000 respondents from 35 countries (including countries from all regions of the world), surveyed between 2015 and 2018 (data set version 2.0.0). In the analysis, the sample size is much smaller (22,443). 11

Analysis and discussion of results
Since we assume variation on the micro and the macro level, a multilevel approach is necessary. As proposed by Hox (2010:, 55-59), this assumption is tested by first calculating an intercept only model (where the intercept is modelled as random, and no explanatory variables are included) (see Online Appendix 1 for all models). In this null model, we see that about 45% of the variability in respondents' attitudes towards surveillance are due to differences at the second and third level. Even though there is no agreed threshold, everything above five per cent in between-group variation would certainly justify a multilevel approach.
The dependent variable has four scale points. Since it leads to the same results as an ordered logit model (see Online Appendix 2), we calculate a three-level linear model. 12 In following Gelman's (2008) approach, all numeric variables are standardized by dividing them by two standard deviations. As proposed by Heisig and Schaeffer (2019), we include the justification variable as random slope since it is situated on the lower level in all cross-level interactions.
The significantly positive correlation of the security justification dummy variable with attitudes towards surveillance (see Table 1) indicates that acceptance of security-justified surveillance is significantly higher than of unjustified surveillance. This is verified by a Wilcoxon matched-pairs signed-rank test. We thus cannot reject H1 and assume that compared to no frame, a security-frame enhances attitudes towards government surveillance.
However, the difference of attitudes does not seem to be true for all respondents: while 33% of the respondents are more supportive of surveillance measures when confronted with a security frame, 47% of the respondents do not change their attitudes and 20% of the respondents indicate more negative attitudes (see Figure 1). In the next step, we seek to explain the differences between the unjustified and the security-justified statements.
As indicated above, the hypotheses are tested based on the coefficient for the respective interactions between the justification variable and the explanatory variables. We see that while the interaction coefficient for liberal democracy is highly significant, this is not the case for trust in government and for political orientation. To better interpret what the coefficients mean, we will look at the distributions of the margins: Stacked data set; Robust standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Figure 2 indicates for both individual-level variables (government support and political orientation) that the security-justified measure has a higher average acceptance than the unjustified measure among respondents at each level. In particular, the two parallel graphs for political orientation show that the acceptance gap between the two measures is roughly equal at all levels of the x-axis. This is also true for government support, although it is somewhat harder to see graphically. In line with the non-significant coefficients, the difference from the unjustified item does not-as hypothesized-increase with higher government support or a righter political orientation respectively.
Contrary to what was assumed, government support does thus not significantly positively correlate with the susceptibility to the security frame, neither can we indicate a significantly negative relationship which would also have been conceivable based on the  assumption that the exceptional nature of securitization might be particularly persuasive to otherwise government-critical individuals (see Fn. 5). Our results suggest that the relationship between the security agent (i.e. the government) and the audience (i.e. the population) does not play a role for the success of a securitization move. Support for governments' security justification thus does not seem to depend on whether individuals personally support the government or not. Public opinion research suggests that individuals differentiate between different state actors and prefer some actors' surveillance over others (Hallinan, Friedewald, and McCarthy 2012, 267). The ISSP questions leave open to the respondents who is meant with 'the government'-elected politicians, law enforcement, intelligence, or others. It therefore remains unclear whether there really is no relationship between government support and the susceptibility to the security frame or whether different understandings of who is 'the government' create 'noise'. Notwithstanding, the present results add to the finding that the role of the audience and its relationship with the securitizing actor(s) seems to be undertheorized (Côté 2016) and emphasize the need for further research in this field. For political orientation, no significant correlation with the security justification was found either. This finding suggests that on average, leftists and rightists are equally influenced by a security justification. Hypotheses 2 and 3 must thus be rejected. For the level of liberal democracy, however, things look different: the greyish graph has a significantly larger slope than the black graph. It follows that the more liberal a country is, the greater the differences resulting from the security justification. Hypothesis 4 can thus be accepted. The peculiarity here is that in less liberal countries, the security frame even seems to have a negative effect. While we would have expected that the security justification would not lead to any significant attitudinal differences compared to non-justified surveillance in illiberal states, people seem to support surveillance less when it is framed as security-related than when it is not justified. In Hungary, India, and Turkey-three countries that V-Dem classifies as rather illiberal-we can for example see that on average, respondents accept non-justified surveillance measures more strongly than those that are security-justified. In rather liberal Sweden and Norway, in contrast, respondents express far more positive attitudes for the securityjustified measures. One possible explanation for this finding draws on Bigo's (2012) idea of what is defined as 'enemy of the sate' and the supposedly different connotation of the concept of national security: it is believed that in democracies, the term rather emphasizes preserving the territorial integrity of the state from external invasion, while it focuses 'more intently on internal threats to the political integrity of the nation' (Peceny, Beer, and Sanchez-Terry 2002, 19) in autocracies. Compared to no justification given, the security-justification might thus trigger different connotations depending on the countries' level of liberal democracy. In any case, the results invite closer research; especially because the correlation between (standardized) coefficients is comparably strong. The inclusion of control variables does not alter the effects. While older individuals and people living in countries with a terror legacy exhibit higher levels of acceptance, higher educated tend to have lower levels of acceptance. The difference between genders is not significant. Having a migration background and living in a country with a communist past neither have a significant relationship with acceptance of surveillance measures in general nor are their relationships with susceptibility to the security frame significant.
In the literature on acceptance of surveillance measures, it has been noted that cultural, social, and geographic contexts matter for the acceptance of surveillance measures (e.g. Budak, Rajh, and Recher 2017;Menichelli 2017). Our results do not only add to the literature but do also suggest that the macro level may be more important than previously thought since the country-level variable is the only variable of interest that has a significant relationship with the security justification. To better understand the role of the context, further cross-national research, ideally not restricted to Europe, has to be conducted. In addition to more detailed analysis of a country's legacy (terrorism, past regimes, or colonialism (see McCulloch and Pickering 2009, 636)), candidates for additional macro-level explanatory variables could be a country's level of surveillance (Budak, Anić, and Rajh 2013, 105), more nuanced measures of liberal democracy or related concepts such as press freedom (Bakir 2015) as well as cultural indicators (Gellman and Dixon 2011) such as Hofstede's cultural values (1980) which have been commonly used to predict privacy attitudes (see e.g. Budak, Rajh, and Recher 2017;Dinev et al. 2006).
Regarding the two strands of literature postulating the effects of framing in relation to predisposition, we also investigated whether the attitudinal differences between the respondents regarding the unjustified claim on government surveillance can also account for individuals' attitudinal differences. As for the polarization assumption, we test whether an individuals' attitude towards unjustified surveillance is correlated with what we call the susceptibility towards the security frame, e.g. the degree to which the attitudes between unjustified and justified surveillance differ. The regression results indicate that there is a highly significant and negative relationship between attitudes towards unjustified surveillance measures and the attitudinal difference, suggesting that the lower the baseline attitude towards surveillance, the higher the susceptibility for securityframing. This contradicts polarization assumptions about framing effects whose proponents would have postulated a positive relationship.
The hydraulic patterns assumption postulates that the strength of predictors decreases when individuals are exposed to a frame. When comparing predictors for attitudes towards unjustified and security-justified surveillance (see Online Appendix 3), we find that that the direction of coefficients does not change between the models (except for gender). Regarding the strength of the relationships, we see an increase in the magnitude of most correlation coefficients, e.g. of government support, political orientation, and liberal democracy. However, there are decreases for age, gender, and education. The hydraulic patterns hypothesis can thus not univocally be supported: while it holds true for the sociodemographic variables, we find that framing leads to an increase of certain predictors' influence.

Conclusion
The aim of this paper was twofold: we firstly sought to test whether one can identify a framing effect concerning security within the ISSP dataset and secondly, we explored factors that explain the postulated effect. While the results regarding the first question -surveillance measures are indeed assessed more positively when framed as being conducted in the name of security-are not surprising but join the large body of literature on securitization, the findings regarding the second shed light on a to-date scarce research field. The multi-level regression analysis showed that a country's level of liberal democracy significantly correlates with the susceptibility to the security frame: individuals in liberal democracies are indeed more likely to react to a security frame than those in less liberal-democratic societies. Interestingly, in the latter, the item without justification even leads to higher acceptance of surveillance measures than the security-justified one. Government support as well as political orientation do not show the hypothesized effects. While for the former, the interaction with the frame does not correlate significantly with acceptance of surveillance measures, the latter shows the reversed relationship: acceptance is always higher when framed as security-relevant, i.e. on the whole left-right spectrum. The fact that government support is not positively correlated with the susceptibility adds to the scarce literature on the relationship that surveillance agents and audience have. Our findings suggest that their relationship is to be neglected when forecasting the success of security justifications. Due to the unique ISSP data, this paper is one of the first to provide empirical evidence that is not limited to a state or a region but draws conclusions from attitudes of people from all continents. Nevertheless, future research might further disentangle the relationship between audience and surveillance agents-especially with regard to who is thought of when speaking about 'the government'.
Certainly, the present study suffers from some limitations: Watson, Finn, and Barnard-Wills (2017) argue that surveys that are not designed to specifically analyze people's attitudes towards surveillance-or privacy-related issues are useless for research on that matter. We do not generally agree with this point of view, as we are convinced that more general surveys, like the ISSP, can also add value to the understanding of surveillance. Nevertheless, the authors are right in that more general surveys do not consider certain factors that could at least conceivably influence attitudes to privacy and surveillance. For this study, we would have particularly wished for additional information on why people have certain attitudes towards surveillance, questions about individuals' threat perception as well as more questions related to liberal values on an individual level. In addition, the study of attitudes towards surveillance is potentially always limited due to a biased non-response rate: since surveys are themselves essentially a form of surveillance, some individuals might interpret them as a privacy invasion (Haggerty and Gazso 2005;Marx 2007). However, no significant correlation is found between the countries' (proxy measures of the) national level of surveillance 13 and the respective non-response or refusal rates.
A point for criticism of the present study might certainly be the study design. In contrast to experimental designs, we can only make limited statements about the treatment effect and causalities. Although the attitudes towards unjustified and justified statements about surveillance are statistically independent, it cannot be ruled out that respondents were primed by the initially posed questions and thus influenced in their responses to subsequent questions. In addition, the fact that the questions did not only vary with regard to the (non-)presence of the justification may lead to a bias of the effects. A further limitation derives from limited knowledge about the individuals: According to Baele, Coan, and Sterck (2018, 465), it is problematic to estimate framing or securitization effects for highly contentious issues-with government surveillance certainly being one. If issues have a long history on the political agenda in real-life, securitization might already have taken place and researchers are no longer able to formulate neutral statements.
There is thus the risk of underestimating framing or securitization effects. Since statistically significant differences can still be found, this point can be neglected.
Despite these limitations, our findings point to important implications for the understanding of how frames influence individuals' attitudes towards surveillance thereby adding an international perspective to the existing literature. This paper is one of the first to provide empirical context about which individuals are most likely to be persuaded by the promise of security. Especially the strong and significant correlation for the level of liberal democracy builds an interesting bridge to the research of individual policy attitudes in autocracies and democracies. Further research will be necessary to explain individual attitudes towards government surveillance and people's willingness to trade privacy against security. In particular, the finding that the effect of the frame differs significantly depending on the democracy level of the respondent's country invites future research.

Notes
1. The evidence of the positive (and even causal) relationship between surveillance and security, however, is inconclusive and fairly contradictory. Systematic reviews of related research for example find that CCTV reduces crime only to a small degree. CCTV is most effective in reducing vehicle crime in car parks or parking lots; and it has little to no effect on crime in public transport and city centre settings (Welsh and Farrington 2002). Other research indicates that, instead of deterring or preventing crime, surveillance merely displaces crime in various forms (Clarke 1994) and several studies conclude that surveillance has no effect at all (Ditton and Short 1999;Gill et al. 2006;Phillips 1999). An exception is Park et al.'s study (2012): In Gwang Myeong City (South Korea), the number of robberies and thefts in the areas with CCTV installed decreased by 47.4%. For other forms of surveillance like Internet surveillance, it is difficult or even impossible to quantify effectiveness of surveillance. Cayford and Pieters's (2018) study supposes that there is no evidence of information technologies' surveillance being effective in improving security. 2. According to Hobbes, people needed to surrender their right to do as they please and submit to the will of any sovereign to create a better life and protect themselves from a state of war. 3. This is not to say that endorsing reasoned surveillance measures cannot be a rational decision. In his essay on the security-liberty debate, Solove (2008) suggests that an individual's or executive's preference for one or the other should depend heavily on external factors, such as how likely a threat to liberty is or how effective a countermeasure would be. While the latter will not be varied across respondents, we seek to control for the former, namely the country's terror attack legacy. In addition, security is not the only justification that makes individuals possibly giving up their privacy. Market research literature has shown that individuals are willing to give up privacy if they receive other barter objects like monetary discounts (e.g., Acquisti, John, and Loewenstein 2013, 253) or non-monetary equivalents such as convenience (e.g., Viscusi and Zeckhauser 2003, 116) in exchange. 4. See e.g., Côté's (2016) criticism of the omission and the lack of common understanding of the role of the audience from the research agenda. 5. On the other hand, it could also be argued that security justifications have little impact on the acceptance from those who already support the government but rather, by shifting the issue into the non-normal, primarily influence those who (in non-exceptional times) would rather oppose the government. Due to the lack of empirical evidence of this direction of relationship, we test H2 in its above form. 6. The addition 'in the name of national security' thereby implicates two things: Looking at the speech act itself, the utterance of security can be understood as a promise implicating that acceptance of the audience is necessary to make the securitization successful (Waever 1995, 55). Secondly, it can be interpreted as a justification understood as the government explaining itself to the public (Waldron 1987, 149). According to Bonilla (2022, 15), promises as justifications have become increasingly critical to political strategy. Since it is our aim to investigate whether the promise is successful, we will continue to speak of the addition 'in the name of national security' as justification. 7. While some authors generally question validity of the use of democracy indices, Boese (2019) shows that compared to other indices (Polity IV and Freedom House Index), V-Dem stands out with respect to the underlying definition and measurement scale, as well as the theoretical justification of the aggregation procedure. 8. Theoretical and empirical research suggests that people in former communist states are more positive about Internet and digital surveillance (Friedewald et al. 2015, 55) and there is a structural lack of understanding of human rights protection and privacy (Nagy 2017, 447-448). Surveillance is framed as a measure to protect national culture and national political independence (ibid.). A society's experiences with surveillance is not limited to post-communist states, Tsapogas (2017, 226) finds that the rule of the Greek military and its surveillance still influences citizens' attitudes towards state surveillance. Other studies find an effect of Germans and Italians still bearing a legacy from their dictatorships in the 1930s and 40s (Fonio 2011;Gill 1995;Luebcke and Milton 1994). However, due to operationalization and space restraints, this will not be addressed in this article. 9. As surveillance is often framed as an 'anti-terror measure', actual terrorism is likely to affect people's acceptance of these measures. We use the University of Maryland's Global Terrorist Database's country-specific score between zero and ten for the relative impact of incidents. 10. Especially in studies focusing on surveillance in the US, immigration background or ethnic group membership are often-tested variables. Researchers include these categories in order to explain behavioural variance in people on a group statistical level, differentiating between an autochthonous population and culturally or racially others. Several authors find that domestic surveillance is closely correlated with racial surveillance (Ryan 2018, 43) leading to a 'hyper-surveillance' of certain ethnic groups (Goffman 2009;Remster and Kramer 2018). When tested, studies almost unanimously find that ethnic group membership has an effect on the attitude towards surveillance. Preponderantly, researchers chose to use the dichotomy White/non-White and find a negative effect on the attitude towards surveillance for the latter (Bali 2009, 247;Davis and Silver 2004, 31;Spriggs et al. 2005, 42;Viscusi and Zeckhauser 2003, 107). As proposed by the Conference of European Statisticians' 'Recommendations for the 2020 Censuses of Population and Housing' (UNECE 2015, 136), we define persons with a 'migration background' as persons whose both parents were born abroad. 11. This is mainly due to missing data in the ISSP's vote choice variable (Great Britain, the Philippines, Suriname, and Thailand) as well as Taiwan and Venezuela not being included in the Manifesto Project dataset. Additional analyses (see Online Appendix 4) show that the exclusion of any variable related to individuals' vote choice (and thus significant increase of N) does not significantly alter the results. 12. Based on graphical analyses, we find that level 1-residuals are normally distributed. However, heteroscedasticity cannot be ruled out. Thus, we decided to apply robust standards errors. There must be a sufficient number of upper-level cases to properly determine estimates with standard maximum likelihood estimate-especially if there are explanatory variables at the upper level. Although our 29 countries might be enough, we calculated an additional model based on restricted maximum likelihood estimates, as proposed by Elff et al. (2021). 13. Stoycheff, Burgess, and Martucci (2018, 479) propose to measure the degree of government surveillance with Freedom House's Freedom on the Net Index' 'violations of user rights' dimension that assesses government monitoring of individuals' Internet activity as well as negative 'repercussions for online speech and activities, such as imprisonment, extralegal harassment, or cyberattacks' on a 0-40 scale, with higher scores indicative of greater surveillance (Freedom House 2016). 14. Graphs were created with plotplain package by Bischof (2017).