Explaining attitudes to secret surveillance in post-communist societies

ABSTRACT This article seeks to explain public attitudes to secret surveillance. Secret surveillance, for example wiretapping by intelligence agencies, is a controversial activity that affects fundamental civil liberties in any democratic system. Several large research projects have recently attempted to explain how people form opinions about surveillance in general. Thereby privacy concerns and institutional trust are often highlighted. In this article, we argue that earlier research uses a too narrow definition of attitudes to surveillance and that secret surveillance is particularly sensitive due to its opaque character. We introduce a two-dimensional concept that focuses on rationalistic and emotional responses to surveillance. Drawing on new data from three post-communist societies – Estonia, Poland, and Serbia – we show how institutional trust is mainly responsible for explaining acceptance of secret surveillance, but not how one feels about it. Instead, it is the level of ontological insecurity and privacy concerns that explains this second dimension. The results are theorised and implications for future research are discussed.


Introduction
Secret surveillance is a controversial activity that affects the fundamental civil liberties in any democratic system (Raab et al. 2012). Recent years have witnessed several large-scale surveillance scandals in Europe and North America, and the picture slowly emerges that intelligence agencies have been acting beyond their competencies in a number of instances (Berendt 2015;Greenwald 2014;The Economist 2013). In the wake of these events, questions about trust in intelligence agencies and attitudes to surveillance have gained a new momentum. Several large scale studies have been completed that shed new light on how people form opinions about surveillance (Denemark 2012;Friedewald et al. 2015;Patil et al. 2014;Strauss 2015). These projects demonstrate the key role of social and institutional trust in acceptance of surveillance. However, there are many questions left unanswered, especially in the context of post-communist societies with their history of extensive secret surveillance.
First, and paradoxically, the Central and East European countries where authoritarian surveillance practices were endemic less than 30 years ago constitute a gap in the research on public attitudes to surveillance. The specificities of the region, including widespread abuse of surveillance information, distrust in public institutions, and a culture of corruption suggests that other explanations may pertain to these cases than in the West (Semukhina and Reynolds 2014;Stepanović 2015). This article addresses the topic of secret surveillance in post-communist societies. Second, previous studies focused on explaining support for surveillance in terms of whether citizens accept or reject certain practices. However, other aspects of attitudes to surveillance have been studied to a much lesser degree. Whether one views surveillance as a necessary evil or not tells us little about how one feels about surveillance. One may, for example, believe that retention of internet logs is necessary to fight the spread of child pornography while still feeling uneasy about having one's communications monitored. In this article, we develop a two-dimensional theory of surveillance that captures both these dimensions. It reports results of survey research in Estonia, Poland, and Serbia during winter 2014/spring 2015 on attitudes to secret surveillance, trust, and security. It thereby expands the gaze of trust research to the field of secret surveillance.
The aim in the present study is to explain public attitudes to surveillance. The research question that guides the analysis is What makes people hold positive or sceptical opinions about secret surveillance? We answer this question by developing and testing surveillance theory through a series of hypotheses on a new data set. The post-communist context serves as the empirical locus of this study, which is suitable due to the recent history of excessive secret surveillance. The results point to an important role for trust, but that privacy concerns and peoples' general level of insecurity play an even more prominent role in the evaluation of secret surveillance. Thus, deeper sentiments of personal or, as we argue, cultural origin in some respects explain attitudes to surveillance better than e.g. institutional trust. While not generalisable to the whole post-communist population (e.g. Central and East Asia), we argue that these results are representative for European post-communist societies.
The article is structured as follows: Below we discuss earlier research and identify gaps in the literature. We subsequently outline a two-dimensional view on secret surveillance and formulate a series of hypotheses to be tested in the following analysis. After a methodological discussion, we present the results. The analysis lends support to the hypothesis that trust in institutions is associated with one dimension of attitudes to secret surveillance. However, high levels of existential concern, sometimes referred to as ontological insecurity, seems to be a far more salient factor than trust, in particular behind feelings about secret surveillance. The article ends with a theoretical discussion of these results and recommendations for future research.

State of the art
In this section, we first look at research on surveillance in post-communist societies, and then discuss how surveillance and trust are related from the perspective of earlier research. We show that trust has been put forward as the most prominent explanation for acceptance of surveillance, but that recent work increasingly looks at surveillance vulnerability in addition to acceptance. We identify gaps in the literature that concern perceptions of security and research on post-communist societies.

Surveillance and post-communist studies
In post-communist studies, the existential insecurity that characterised the period after the fall of the Iron Curtain has been a very important factor behind concepts like "cultural trauma" and "the post-communist condition" (Holmes 1997;Sztompka 2000). Accounts about whether the different communist regimes in Europe were effective or not differ, but there is agreement that communist surveillance and political control were detrimental to social trust (Horne 2014;Hosking 2013;Sztompka 1999). It is however striking that few research efforts have been made to study how post-communism is related to attitudes of surveillance.
Post-communist societies are characterised by the fact that they share recent experiences of communist regimes. Despite the vast differences in time and space between varieties of communism in the regionfrom the brutal Romanian regime to the comparatively more lenient Yugoslav one, most people can be expected to have a relatively clear picture of what secret surveillance entails and what risks it poses. This stands in opposition to the diffuse views of populations that have not experienced surveillance-intensive communist regimes (Horne 2014;Jackson et al. 2011;Kornai, Rothstein, and Rose-Ackerman 2004;Mieriņa and Cers 2014;Semukhina and Reynolds 2014). There is a specific literature devoted to public attitudes in post-communist societies, in which the concept of legacy plays an important role. Communist legacies are cultural effects on peoples' attitudes that remain from the communist times (Wittenberg 2015). Legacies stand in contrast to rationalist explanations holding that people form their opinions mainly based on perceptions of contemporary institutional efficiency and personal interests (Clark 2002). In the empirically oriented literature, the rationalist explanation has had the strongest position. Studies demonstrate that low levels of institutional trust and lack of civic engagement in Eastern European countries can be explained by poor performance and widespread corruption, rather than a post-communist culture of distrust (Clark 2002;Hooghe and Quintelier 2014;Mishler and Rose 2001). Still, it is an interesting fact that to date, very little research on surveillance has been carried out in post-communist states.
Of the few studies that have been carried out on surveillance in post-communist societies, two have identified particular privacy-sensitive and surveillance-sensitive strata in the populations. Székely, in his pioneering survey from 1990 in Hungary, describes that 16% of the sample population shared critical attitudes to surveillance as their only common denominator (Székely 1991, 43). Twenty-five years later, Budak et al. developed this view and could show that in the Balkans, a more tech-savvy and a privacy-sensitive stratum exist, as well as a more traditional and surveillance-sensitive one . However, these studies could not explain these results and did not take existential insecurity in consideration, which seems unfortunate because we know that insecurity has been an important factor in other aspects of political life in post-communist societies (Caparini and Marenin 2005). Finally, Friedewald et al. (2015, 55) noted an interesting difference in attitudes towards foreign state surveillance (item QB3) between Western and Eastern Europe in the Privacy and Security Mirrors (PRISMS) data. Respondents from post-communist societies were more prone than non-post-communist respondents to agree with the statement "I trust governments that monitor internet and digital communications, even if they are from another country". However, the researchers could only find weak correlations to institutional trust indicators, and did not follow up on this track in the subsequent analyses. This study intends to expand on this research by focusing on three very different cases within the population of post-communist states. We thereby hope to produce a more complete picture of how attitudes to surveillance are formed. This study further fills a gap by highlighting the role of existential, or ontological, insecurity in the formation of attitudes to surveillance. This is a particularly important topic in a time of resurgence of authoritarian attitudes in Central and Eastern Europe.

Trust and surveillance
In the context of surveillance, trust enjoys a peculiar position. In line with the phrase "trust is good, control is better" that is often attributed to Lenin, the a priori expectation is that surveillance and trust stand in sharp contrast (Möllering 2005, 284). Szrubka (2013) suggests that this is the case in his analysis of attitudes to CCTV in Poland. However, as Ellis, Harper, and Tucker (2013, 1) remark, this may be a fallacy, because "trust is a primary constituent of the relational dynamic of most surveillance systems". In other words, surveillance may deteriorate trust, but at the same time any surveillance system depends on a certain level of trust to be legitimate. Research has for example shown that where trust in the government is low, support for anti-terrorism policing is too (Denemark 2012;Nakhaie and de Lint 2013). Further, anxiety about government surveillance correlates with low support for antiterrorism policies (Best, Krueger, and Pearson-Merkowitz 2012). Trust in institutions has also been shown to be central to acceptance of intrusive surveillance technologies (Friedewald et al. 2016;Patil et al. 2014). Exploring this dynamic, several works on surveillance deal with the concept of trust (se e.g . Neyland 2006;Szrubka 2013), but often do not study it empirically. They therefore do not tell us very much about the relationship of trust with attitudes to surveillance, or other factors such as privacy or security.
For more empirical accounts of attitudes to surveillance and their relationship with trust we turn to quantitative studies. There exist only a rather small number of international surveys that focus on (secret) surveillance to date. Watson and Wright have written a comprehensive review of existing surveys (Watson and Wright 2013), in which the authors show that a small subset of those studies include questions of trust. These surveys indicate that people tend to trust public agencies rather than private companies, and that trust is not definite but contingent upon transparency and accountability of the organisations carrying out the surveillance (Watson and Wright 2013, 129f). This means that the level of trust seems to depend on either specific institutions or types of surveillance. 1 However, the authors state that previous research has not been able to establish a clear link between trust and public perceptions of privacy and surveillance, and that this is an area where more work is needed (Watson and Wright 2013, 131).
Watson and Wright also found lower levels of trust reported in post-communist societies, which is in line with more general studies on the regions as well (Dimitrova-Grajzl and Simon 2010;Horne 2014;Pop-Eleches and Tucker 2014). Rajh (2012, 2015) have studied the Balkan countries in the Extended Model of Online Privacy Concern project (PRICON), and found that concerns about privacy and surveillance is dependent on the level of education and trust in law enforcement institutions. Generally, attitudes to surveillance in this research field have hitherto been explained with reference to institutional trust and opinions about privacy and technology (Watson and Wright 2013). Together with education these factors largely account for attitudes to surveillance.
More recent research has substantially expanded the scope of the field, providing high quality input for further research. In a European context, the European Commission's 7th Framework Programme projects PRISMS, SurPRISE and PACT, which were recently completed, constitute milestones in this respect. 2 However, results have been somewhat contradictory, andimportant for this articlethe focus was mainly on "open" and not "secret" surveillance. The PRISMS and SurPRISE projects highlighted institutional trust and their findings downplayed the role of peer-to-peer, or social trust (Friedewald et al. 2016;Pavone, Degli-Esposti, and Santiago 2015). The PACT project focused on three surveillance contextstravel, health, and the internet. PACT highlighted the role of "general trust" as a determinant of all attitudes to surveillance practices. General trust is here an index of confidence in the voting system, the role of technology in society, trust in government and attitudes towards businesses in general (Patil et al. 2014, 82). Since it is a mixture of both institutional trust (in government) and abstract notions about society, these results tell us little about what actually explains attitudes to surveillance. While the PACT project accentuated the role of socio-economic factorsas a predictor of general trust, and in extension surveillance, SurPRISE showed that the perceived intrusiveness and effectiveness of security technology, in combination with institutional trust, could largely account for attitudes to surveillance (Patil et al. 2014, 66f;Pavone, Degli-Esposti, and Santiago 2015, 154). SurPRISE used so-called citizen summits as data sources, 3 which yielded less generalisable but more in-depth data. The results that explicitly focus on secret surveillance (smartphone location tracking by "security agencies") highlight transparency and accountability of public agencies for surveillance to be acceptable (Szénay 2014, 18f), and thus problematise their autonomy as a key factor. In all these studies, the view on what influences attitudes to surveillance and privacy shifts somewhat depending on the specificity of each project, but institutional trust and privacy concerns are typically two of the main factors.
We see the biggest gaps in the literature in the role of other aspects than institutional trust and privacy, such as security concerns and social trust, which are hitherto understudied. First, the conceptualisation of security typically refers to fear of crime, data security, or direct physical harm (Watson and Wright 2013, 132). However, as important as these may be to for example acceptance of CCTV and consumer data protection, in the context of secret surveillance where the subject is often unaware of the surveillance, these concepts are insufficient. PRISMS results did point not only to institutional trust, but also to the role of fear and anxiety (Friedewald et al. 2015). However, although it was noted that anxiety plays a role in explaining attitudes to surveillance, there was no substantial attempt made to explain this factor or develop this explanation. The PRISMS and PRICON surveys measured security concerns in terms of perceptions of vulnerability to surveillance Friedewald et al. 2015). This concept represents a major step on the way but the construct that it aims to capture is feelings of (in)adequacy and (in)competence for protection rather than the more general feeling of insecurity that, for example, Giddens has theorised in terms of ontological insecurity (Giddens 1991). Vulnerability is too diffuse because it does not separate worries about surveillance from the technical capacity to protect one's privacy. In some cases that may be preferable, but ontological insecurity arguably focuses on a different construct that we believe is more important as a predictor of attitudes to surveillance. This factor has yet to be conceptualised sufficiently.
Second, social trust is put forward as an important factor in for example the PRISMS project, where Friedewald et al. (2015, 97) note that: "The important role of trust, in people, in institutions as well as in the whole societal environment, is regularly confirmed in surveys." This is true for acceptance of security measures in the case of fear of crime, as Hummelsheim-Doss (2017) explains, but not necessarily when it comes to secret surveillance. In the PRISMS survey, which stands alone in actually measuring social trust, it was recognised as a predictor for institutional trust and hence indirectly affecting acceptance of surveillance (Friedewald et al. 2015, 93f). Social trust itself had only a small effect on acceptance of surveillance (Ibid. 2015, 132f). In the works discussed above this is the only attempt to assess the direct impact of social trust on attitudes to surveillance. Corroborating previous results is thus important to validate the theoretical assumption that surveillance is not related to trust in people.
The surveillance studies are, in our opinion, not entirely clear on how people form opinions about surveillance. The present study adds another piece to this puzzle by studying cases where surveillance and trust have very specific historical connotations, i.e. postcommunist societies. Below we discuss the particularities of the region and then continue to outline a two-dimensional view on surveillance.
We below continue to discuss conceptual issues and to outline a series of hypotheses to be tested in the analysis.

Modelling surveillance in post-communist societies
In this section, we discuss conceptual issues of surveillance and outline a model, from which we derive a series of hypotheses to be tested using the Like Fish in Water dataset. The section has two parts: the first argues that attitudes to surveillance should be studied in two dimensionsemotional positions and cognitive evaluations, i.e. as feelings and acceptance of particular surveillance practices. Looking at only the latter, which is the common approach, misses a central political aspect of surveillance. We end the section by outlining four hypotheses.

A two-dimensional view of surveillance
The words "surveillance" and "secret surveillance" in this study refers exclusively to covert techniques and practices of information gathering about people, occurring without the monitored subject's knowledge or approval. When we discuss attitudes to surveillance we thus refer to the way that people relate to the construct of secret surveillance. Such surveillance is typically carried out by law enforcement and intelligence agencies, although as the revelations by Edward Snowden showed, large corporations are increasingly being involved, especially in digital surveillance (Lyon 2015, 36f). Nonetheless, secret surveillance in the context discussed here is typically directed to, managed by, or designed for state intelligence agencies and/or law enforcement. Secret surveillance as discussed here is thus something completely different than, say, airport screenings or CCTV in public space. It is politically more sensitive and closely related to core issues of power and security (Caparini 2014).
While acceptance is the typical measure of choice when discussing approval of for example wire-tapping, we believe that this conceptualisation of attitudes to surveillance fails to grasp a highly political aspect of secret surveillancethe unease of being watched. As Denemark (2012) has shown, most people would cognitively accept secret surveillance practices by intelligence agencies as a necessity when confronted with terrorism, disinformation campaigns, or the risk of war. However, approving of surveillance towards these ends, where both the end and the surveillant institution is possible to comprehend, is one thing; how one feels about surveillance in general, as a more and more salient ingredient in peoples' ordinary life and secret surveillance expanding beyond public control, may be something completely different. Looking only at whether certain measures are perceived as legitimate or not fails to reveal these other sentiments. The basic question whether surveillance practices are acceptable is thus not sufficient, since it fails to shed light onto other, very real concerns that people may have.
Following this line of thought, we believe that surveillance can be more fruitfully conceived in more than one dimension. When researchers use acceptance of surveillance practices as the dependent variable they suggest that there is direct relationship between what people think about a potentially problematic practice and whether they think that it is justifiable. It could be, for example, a technology's intrusiveness (Friedewald et al. 2015;Strauss 2015). Drawing on the work of Fishbein and Ajzen (1975), the present study conceptualises attitudes to surveillance not only as the evaluation of a certain practice, but also the beliefs about it on an emotional level. In other words, a person may be principally highly concerned about surveillance in general yet still accept intrusive anti-terrorism policies because it is perceived as a necessity. Emotions point in the direction of cultural explanations since they represent something that are more or less stable, and learned whereas cognitive judgements are based on available object-specific information and thus depend on individual abilities to evaluate this information (SFB Affective Societies 2016). We argue that the emotional and cognitive dimensions are both equally central to any discussion on attitudes of surveillance, and in extension their legitimacy. To subsume one of these dimensions to the other prevents us from seeing the whole picture. The first part of our model is thus the difference between cognitive acceptance of and the more fundamental emotional beliefs, or feelings, about secret surveillance.

Hypothesising surveillance in the Like Fish in Water survey
While we cannot study the formation of beliefs and evaluations per se, we can study what factors can predict their occurrence. Drawing on our discussion on earlier research above, we believe that attitudes to surveillanceas acceptance and as feelingscan be accounted for by three factors: (a) trust in people, and/or the institution(s) responsible for surveillance; (b) the degree of privacy sensitivity of the individual respondent; and (c) the individual's existential concerns on a societal level. However, not all factors influence acceptance and feelings in the same way. Institutional trust is expected to influence acceptance to a higher degree than insecurity and social trust. We expect insecurity, in turn, to influence feelings about secret surveillance. Privacy concerns should theoretically influence both dimensions equally. Social trust, finally, is a wild card. It can be conceptualised as dependent on institutional trust or rather as a personality characteristic (Sønderskov and Dinesen 2016). During communism, social trust at times replaced trust in the state (e.g. in Poland in the early 1980s); while earlier, the security apparatus made trusting anonymous people impossible (Horne 2017;Kochanowicz 2004;Lühiste 2006). How much of this that is still present today is difficult to predict, but an a priori assumption that we present below is that social trust affects both dimensions of attitudes to secret surveillance. We now introduce a series of hypothesis that are subsequently tested in the analysis.
The first factor, institutional trust, should correspond with levels of acceptance of surveillance. If institutions are not considered to be trust-worthy due for example to corruption, which is common in many post-communist societies (Holmes 2006;Kostadinova 2012), we expect that respondents will disapprove of their competences. We do not expect that institutional trust is a major factor when explaining feelings about surveillance because our assumption is that this emotional category is regulated by other factors such as privacy and security. Echoing the debate in post-communist studies about cultural and rationalist explanations, we assume that institutional trust is a result in more or less rational evaluations of public institutions' performance (Rus and Iglič 2005).
As we discussed above, social trust has been described as an important factor in previous attitudinal studies on surveillance, but rarely tested or theorised. We are therefore interested in the influence of this type of trust as well. We propose that high social trust implies that one trust people's integrity to "do the right thing". People who report a high level of trust in anonymous others generally have a more positive outlook and are more engaged socially and politically (Sønderskov and Dinesen 2016;Stolle 2001). Whereas the relationship between institutional trust and acceptance of surveillance is performance-oriented and instrumental, social trust is expected to have a broader relationship with attitudes to secret surveillance. If this is true it would affect both dimensions such that high social trust predicts high acceptance of and low worries about secret surveillance. In line with this reasoning, we formulate the hypotheses that: H1a: Trust in institutions predicts acceptance of secret surveillance, but is unrelated to feelings about surveillance.
H1b: The level of social trust predicts acceptance of and worries about secret surveillance.
Privacy sensitivity is also derived from earlier research and follows naturally from the topic at hand. Surveillance does not necessarily exclude the possibility for privacy, but the type of surveillance discussed heresecret surveillanceentails substantial privacy infringement (Luca 2015). Privacy is a key value that for many respondents stands in conflict to surveillance (Watson and Wright 2013, 125f). As Székely showed in his landmark study in Hungary in 1991, privacy was an important value to people in communist societies (Székely 1991). Privacy concerns is also one of the factors that affect attitudes to all types of surveillance in the PRISMS data (Friedewald et al. 2015, 84). We expect that due to the political history and the many surveillance-related scandals in all post-communist countries, privacy may have a large explanatory value in this study as well. A respondent who is concerned for his or her privacy is more likely to hold sceptical attitudes to secret surveillance. Concern for privacy affects attitudes to surveillance, as acceptance and worries, equally, since privacy may activate both cognitive evaluations and emotional sentiments. Grenville 2010). The hypothesis reads: H2: Concerns about one's privacy predict acceptance of and worries about secret surveillance.
Notwithstanding the explanatory effect of key factors like trust and privacy, our main contribution to the literature lies in the area of security. Earlier research suggests that security threats are related to more positive attitudes to surveillance (Cohrs et al. 2005;Denemark 2012;Friedewald et al. 2015;Sparks 2003). Post-communist populations have lived with existential insecurity for a long time -first during communism (see e.g. Flam 1998) and later in its aftermath. Democratic and capitalist transformation brought a stark economic polarisation in all post-communist societies, and as weak welfare states, levels of social and economic insecurity rose dramatically in the 1990s (Sztompka 2008). Due to the socio-economic polarisation, an actual rise in crime rates, and in some cases political campaigns focusing on the threat of crime, fear of crime rose dramatically towards the end of the 1990's in most post-communist societies (Caparini and Marenin 2005;Holmes 2009;Łoś 2002;Waszkiewicz 2011). For domestic and external reasons, the economic crisis of 2008-09 also hit post-communist societies particularly hard (Kattel and Raudla 2013;Voinea 2013). Since we also know that subjective well-being in post-communist societies is particularly highly correlated with GDP development (Bartolini, Mikucka, and Sarracino 2017), it follows that these populations have been under a sustained economic and social stress, resulting in recurring periods of insecurity. According to what we know from previous research, this should lead us to expect that post-communist populations hold particularly positive attitudes to surveillance.
Giddens coined the term ontological insecurity to denote a general form of existential anxiety (Giddens 1991). Ontological insecurity reflects lack of confidence in the future and it is sensitive to rapid changes in society and transitions that makes this future look less transparent and the prospect of one's own life-chances less promising. Bauman and Lyon also contributed to this debate with their analysis of liquid fear and liquid surveillance (Bauman 2006;Bauman and Lyon 2013). In both Giddens and Bauman and Lyon one finds the expectation that existential anxiety paves the way for positive attitudes to surveillance. This would be true for both personal, "inner", control as well as social and bureaucratic control of various kinds (Giddens 1991). However, this assumption is complicated by the fact that our field of study concerns a post-communist context. There is reason to expect that secret surveillance in these societies is a highly controversial issue, particularly among those who were monitored by secret services during communism (Svenonius and Björklund 2016). Nevertheless, we expect the present to trump the past and propose that high levels of ontological insecurity are associated with low levels of worries about surveillance, as well as having a positive effect on acceptance of surveillance by secret services: H3: The level of insecurity predicts acceptance of and worries about secret surveillance.
These hypotheses are tested using the Like Fish in Water data set below. We show that ontological insecurity can indeed explain attitudes to surveillance in terms of worries, but in an unexpected way. Trust in institutions and privacy concerns are also important variables when explaining attitudes to surveillance. Institutional trust is most important to acceptance of secret surveillance whereas privacy and ontological insecurity better explain worries about secret surveillance. Below we continue by discussing methodological issues.

Survey design and case selection
The Like Fish in Water (LFiW) survey includes three countries that were selected based on two variables: a communist past and institutional trust. First, all societies included must share a communist past. We use post-communist societies as a test population because of the shared experience of surveillance that we ascribe to this region. To make this analysis possible it was essential that the societies included share certain political experiences, most notably that of communist rule. It is important to state that the regimes in Estonia, Poland, and Serbia were very different. However, the political repression through surveillance by the secret police and other institutions carried similar traits in most communist societies (Persak and Kamiński 2005;Maddrell 2014;Pop-Eleches and Tucker 2014;Wisser and Blanco-Rivera 2015). Especially surveillance methods can be considered rather similar, although the intensity with which they were applied or the notoriousness of the secret police were very different both within and between cases (Raab, et al. 2012, 23). The difference between the cases is accentuated by their association with the Soviet Union and the nature of the respective security apparatuses (Persak and Kamiński 2005;Tang 1983;Weiner and Rahi-Tamm 2012). The Estonian Republic was an occupied territory in the Soviet Union after the second world war, Poland was an autonomous communist state but closely tied to the Soviet sphere of influence, and Serbia, as part the former Socialist Federal Republic of Yugoslavia, represents a case of relative autonomy from the Soviet Union (Dulic 2009, 281f;Gerner 2009, 224f;Perchoc 2013, 244f). Second, the cases must differ in institutional trust (see Jackson et al. 2011). The selection reflects this requirement: Estonia is on par with the Scandinavian countries, Poland in the middle, and Serbia scores among the cases with lowest trust levels (together with Ukraine and Russia, which were discarded because of the general destabilisation in the region due to the Russian annexation of Crimea). The cases thus represent a diverse selection in Seawright's and Gerring's terms (2008). As our argument builds on communalities across these three states, the selection is a particularly tough test. If the hypotheses can be confirmed based in this selection they should be robust enough to hold for the region as a whole.
LFiW is an international survey commissioned by the Like Fish in Water: Surveillance in Post-Communist Societies project. Fieldwork was carried out by Ipsos during winter 2014/2015 in Poland and Serbia, and early spring 2015 in Estonia. The LFiW survey covered issues of trust, the political system, corruption, security and surveillance. Sample frame populations were the respective national populations, ages 18-80, from which a random probability sample of around 1000 respondents were interviewed faceto-face. With interview duration of about 20 min the survey was relatively short, which ensured a comparatively high response ratio, average or above in all countries. Out of the original data set of 3034 observations, 2744 and 2555, respectively, could be used in the two series of regressions in Tables 4 and 5 below. Post-stratification weights were used to compensate for representation biases in terms of gender, age, education and type of settlement (urban/rural). 4 Response rates for the LFiW data set are displayed in Table 1 below.
The high refusal and no-contact rates foremost in Poland stand out in this survey. The stated reasons for refusal were mainly lack of interest, but also to a lesser extent distrust and bad timing. Polish urban residents were particularly hard to reach. Interviewers were instructed to pay four visits to each selected household, but this measure does not seem to have had much effect among Polish non-responders. As shown in Table S1 in the supplementary materials, the response rates do not deviate from other surveys carried out using similar methodology. To compensate for missing data specifically on the income variable, values were imputed based on age, gender, education, employment and type of settlement (urban/rural) using STATA 15's multiple imputation commands. 5

Indicators
Dependent variable: attitudes to secret surveillance As we argue above, the construct "attitudes to secret surveillance by intelligence agencies" entails two different dimensions: on the one hand as feelings, here operationalised as degree of worries about surveillance by intelligence agencies in general, and on the other as acceptance of such practices performed by each country's most well-known intelligence agency. 6 The conceptual difference between worries and acceptance is vital: "Worries about surveillance by intelligence agencies" probes respondents' attitudes to secret surveillance in general, as an abstract phenomenon, and not contingent upon any particular type of surveillance technology or institution. The survey item used to measure worries about surveillance is a question on a quasi-continuous 11-degree scale: "How worried are you personally about the following? -Surveillance by intelligence agencies". 7 The item is part of a series of items concerning personal worries, which will also be discussed in the next section. It allows the respondents to report their degree of personal concerns about surveillance, while not linking them to any distinct surveillance practice. It is therefore particularly well suited to grasp the emotional dimension of the construct.
Acceptance of secret surveillance is operationalised as the degree of autonomy that respondents were willing to assign to intelligence agencies to carry out surveillance operations. This second measure probes further into the issue of legitimacy and trust in security agencies, and similar formulations have been used in previous research (Skinner, Cameron, and Friedewald 2014 item QB23). The acceptance variable in this study also serves as a benchmark for comparison with other surveys and was constructed in a similar way to an item in the PRISMS survey. If the results point in a similar direction as earlier research, this strengthens the validity of the data. The wording of the questions is: "Do you think that the secret services, such as [NAME OF AGENCY] should have an unrestricted mandate to use any surveillance methods they see fit?" Respondents could reply on a four-degree ordinal scale, using four levels of institutional restraints. 8 The summary statistics of the two dependent variables are displayed in Table 2. Increasing values represent more worries and less acceptance, respectively. Looking closer at the two measures together, the relationship between worries and acceptance is displayed in a spineplot (Figure 1). It shows the relative size of the different response options in two variables, here acceptance of (vertical) and worries about surveillance (horizontal). Spearman's rho is low, only −0.115 (p < 0.001; N = 2661). We view this slight correlation as normal due to the fact that they measure two aspects of the same construct. In the spineplot, we see that the ratio of "No, never" responses on the acceptance variable tends to increase with the value of the worries expressed by respondents between 0 and 10, which explains the small correlation between them.
The figure tells us that the share of respondents who do not trust intelligence agencies to make correct decisions autonomously, i.e. the "No, never" and "Yes, court order" responses, is higharound 75-80%. While this does not necessarily imply that respondents believe courts have higher integrity than intelligence agencies, the table does show that a vast majority believes these decisions are not to be made internally by the agencies themselves (we see that this share varies, but lies somewhere around 20-30%). There is thus a large share of the respondents who view surveillance by intelligence agencies as more or less unproblematic, but the majority does not think so. Instead there is a segment that is very concerned about surveillance and even rejects it all together. In the spineplot above we see that this group largely explain the slight correlation between the  worries and acceptance variables, with many respondents with extreme response options on both items.

Independent variables
The models analysed below include four main predictor variables and four control variables: institutional trust, social trust, privacy, and ontological security are the main explanatory variables. The controls are education, income, age, and country of residence. We describe these in turn.
To test the impact of trust on attitudes to secret surveillance we use two measurements. First, institutional trust is operationalised as a composite variable of trust in the police, in intelligence agencies, 9 in courts (legal system), in the tax agency, and in government. 10 The variable is constructed using the same institutions as the PRISMS survey, with the exception of businesses, which is included in their model, and intelligence and tax agencies, which are included here. The logic behind collating these institutions in one variable is to probe the trust that respondents hold in public sector institutions in general, and law enforcement in particular. We also aimed to avoid one particular institution affecting the general picture in the analysis. The most controversial of these is perhaps the tax agency, which has been identified in earlier research to be a rather common surveillance reference in post-communist contexts (Szrubka 2013). Presumably this is related to the realignment of tax revenue collection in post-communist societies; from a system built mainly on enterprise tax to one focused on personal income tax (Appel 2011, 36). In Kornai's (1992) words, many citizens suffered from "tax illusion" and were taken by surprise that the state was interested in collecting income tax, using public employment records. A composite mean was calculated from the five items and the indicator thus has the same range as the original items, 0-10. Second, social trust is measured by a standard item with an 11-degree scale. We include this factor in the models, as discussed above, but we do not treat it as one of the most important predictors. The wording of the question is a standard item: "Generally speaking, would you say that most people can be trusted, or that you can't be too careful in dealing with people?" Both trust questions were drawn from the European Social Survey (European Social Survey 2016). 11 The third predictor is privacy. Privacy concerns are difficult to measure across different semantic and cultural contexts (Watson, Finn, and Barnard-Wills 2017;Zureik and Harling Stalker 2010). In recent surveys, researchers have therefore made use of vignettes to facilitate a shared understanding of the construct. This was the case in the recent PRISMS survey, and also the Globalisation of Personal Data survey of 2006 (Friedewald et al. 2016;Zureik and Harling Stalker 2010). There are certainly good reasons to use scenarios, especially in cross-cultural survey research. In the LFiW survey, however, we choose to refer directly to the notion of privacy. 12 To measure privacy sensitivity, the LFiW survey included a series of items similar to the "worry about secret surveillance" described above, i.e. using the same 11-degree scale. The question probed for respondents' privacy concerns regarding a range of issues of variable severity, from sending emails to "video surveillance of public places by local police", and "private companies mapping your living habits". 13 In order to avoid measuring privacy concern by reference to internet surveillancea technology that a substantial share of the sample may be unacquainted withwe selected open-street video surveillance as a proxy measure for privacy sensitivity. 14 The item is not directed to secret surveillance per se, but is widely known as the main symbolic representation of surveillance and may therefore be a more tangible indicator of privacy sensitivity than items probing online environments. 15 Even though cameras, in a global perspective, are considered relatively uncontroversial (Doyle, Lippert, and Lyon 2012), the difference between observations is what matters here. We thus assume that video surveillance, as a marker of public surveillance, to some extent transcends the methodological difficulty of measuring privacy, at least across European cultural contexts. This is certainly true for our target countries, where this technology is increasingly ubiquitous (Björklund 2013;Mitrovic 2015;Sojka 2013).
The final hypothesis concerns ontological insecurity. It is measured using a composite variable consisting of respondents' reported personal concerns about terrorism, corruption, structural unemployment, social inequality, and war. Taken together these items reflect a general feeling of confidence about the direction in which "society" is heading. The question is: "How worried are you personally about the following?", to which respondents were asked to answer on an 11-degree scale, from "not at all worried" to "extremely worried". A composite variable was calculated in the same manner as with institutional trust. To rule out co-dependency between the dependent and independent variables (representing ontological insecurity and worries about secret surveillance), a confirmatory factor analysis was carried out, showing that worries about secret surveillance does not load onto the other five items. 16 Finally, we control for higher education, income, age, and country of residence. These demographics have been shown to be important in earlier research, although the way that they have been ascribed with meaning differs. Levels of education may affect attitudes to surveillance, usually people with higher education are more sceptical to surveillance than those without Grenville 2010). Higher education is a binary variable indicating whether the respondent has gone through higher education or not (by higher education we refer to post-secondary school education). Income has also been discussed in earlier research. Since the survey included different measures of income levels in different currencies, this variable has been standardised within each country. Due to item non-response, income data was imputed (see estimation section below). Age is included in the estimation as the age in years and age squared. An interesting suggestion from research on political participation in post-communist societies is "the generational gap" thesis, which assumes that older respondents are less reluctant to participate in political institutions and organisations (Friedewald et al. 2015, 86;Pop-Eleches and Tucker 2013). Translated to a surveillance context, older people may be more critical to secret surveillance, since they are the ones with the most accurate memory of the communist period. The reason to include a quadratic term is thus to control for possible non-linear effects. Country of origin, finally, is included as three binary variables. "Serbian" was used as the baseline in all regressions and is therefore omitted in these tables. We now continue to describe the model specification for the analysis below. Differences within the population of post-communist societies are potentially significant and since case selection strategy was to generate a diverse sample of cases we expect considerable differences between the countries. Summary statistics and survey question specifications for all variables are available in Table 3.

Estimation methods and model specification
The analysis below proceeds in two steps: bivariate correlation analysis as a first test of the hypotheses, and regression analyses to further analyse the interrelationship between the variables described above. For the correlation analysis, we use Spearman's rank coefficient (Spearman's rho) to estimate the association between the variables. 17 In the more detailed analysis of surveillance worries, linear regression is suitable to assess the direction and size of the impact, not just the association, of the independent variables. It is easily interpretable and frequently used in this type of analysis. However, it is ill suited for categorical variables such as the "acceptance of surveillance" measure. Here, ordered logistic regression was used instead. Results from this regression is reported in terms of odds ratios, which show the relative change in odds probability of a specific result. Because model estimation on imputed data does not follow the same maximum likelihood estimation principles as on ordinary data, goodness of fit statistics cannot be easily obtained. In the linear models below and in the supplementary materials, average R2 statistics were calculated using STATA's mibeta command, written by Yulia Marchenko (StataCorp). Average R2 is not exact but an indication on the actual R2. Since these values are almost identical to the statistics using non-imputed data without the income variable, it is a good approximation. STATA 15 have been used to estimate all models.
We estimate four models that refer to the set of hypotheses defined in the introduction, as well as a baseline and a combined model using all predictors. It is important to note that worries about secret surveillance indicate the increasing degree of worry; and that acceptance has as its maximum value the "No, never" response. It designates decreasing degree of acceptance, respectively, with each unit increase. The dependents are thus coded as to make it easier to interpret the results: the higher the value, the more sceptical the respondents.
The first model is an empty baseline including only the controls. Model 2 focuses on trust in institutions. High trust in public institutions should, according to H1a, be a good predictor for acceptance of surveillance, but unrelated to feelings about surveillance. Model 3 is social trust (H1b), where we expect a positive effect on both worries and acceptance: if people are trustworthy, one can also entrust them with carrying out surveillance. Model 4 tests privacy sensitivity (H2). We expect that respondents with a high degree of privacy sensitivity will be more worried about surveillance and less accepting of such practices by intelligence agencies than those who are unconcerned about privacy. Model 5 includes ontological insecurity as an explanatory factor for attitudes to surveillance (H3). Here we expect a negative relationship with worries because a high degree of general anxiety would, theoretically, coexist with positive attitudes to surveillance in terms of less worries. We also expect that ontological insecurity affects acceptance of intelligence agencies' activities positively. The last model includes all independents and thus tests their relative strength.

Unpacking attitudes to secret surveillance
The first step is to show the correlations between the dependent and independent variables. This provides us with an indication about whether there is a significant relationship between the variables or not. If not, we can reject the hypotheses. Figure 2 shows Spearman's rho for the two dependent variables and the seven independents. We see clearly that worries and acceptance show different effects. Most variables display small coefficients that are almost negligible. As expected, institutional trust is larger and negative when it comes to acceptance of surveillance. Somewhat surprisingly it also correlates with worries about surveillance. Social trust displays no significant relationship with either dependent variable. Interesting correlations are also ontological insecurity and privacy sensitivity, which are comparatively strong for worries but not for acceptance. Also, surprisingly, ontological insecurity has a positive coefficient and not negative one as expected. Among the controls, country of origin seems to be most relevant, but age also has a small significant correlation. The figure implies that the predictors, except for social trust, are associated to the dependent variables in ways that indicate that we may possibly reject the null hypotheses. The task now is to verify the significance, estimate the sizes and directions of the effects, and their internal dynamic.

Worries about surveillance by intelligence agencies
With the expectations from Figure 2 in mind, we now take a look at the effects in the regression models. Table 4 shows coefficients and standard errors for the linear regression analysis using worries about secret surveillance.
Based on the theoretical model, several of the results in Table 4 are surprising. Contrary to expectations according to the theoretical model, institutional trust is a significant predictor for the emotional dimension of surveillance. Social trust has a small negative effect, which, as we expand on below, is driven mainly by the Estonian respondents. The effect of privacy concerns in Model 4 is quite strong. As expected, respondents who are sensitive to privacy issues worry more about surveillance. Models 5 and 6 shows that ontological insecurity, i.e. personal worries on an existential level, has a very sizeable and significant effect on the dependent variable. Whereas the earlier models are characterised by rather weak goodness-of-fit, both model 4 and 5 are in the 12-15% range, which is comparatively high considering the "messy" type of survey data used in this study. The effect of ontological security is interesting because it is so strong, andas indicated by the correlation analysispositive. It shows that the less secure people feel in general, the more they worry also about surveillance. In light of earlier research, this is slightly counterintuitive: it means that respondents tend to be more occupied, instead of less, by such worries under circumstances of perceived insecurity. A possible explanation is that this is related to specifics of the post-communist region, where citizens are familiar with abuse of power by intelligence agencies and surveillance may be a cause of distress in its own right. Ontological insecurity is, in any case, a very strong predictor for this dimension of attitudes to surveillance. Country of origin was coded in three separate indicator variables, "Serbian" being the intercept in the regressions. The regressions show the national differences in effect sizes. The large discrepancy between Polish and Serbian respondents with respect to the independent variable tells us that Serbian respondents are less anxious about surveillance than Table 4. Linear regression coefficients and standard errors for models 1-6: Worries. their Polish counterparts. The Estonian responses turn out somewhere in between. To illustrate the country differences, Figure 3 displays marginal effects on the main variables in the full model.
The marginal effects further highlight the relevance of privacy concerns and ontological insecurity in all three cases. In the context of secret surveillance, the country differences may reflect the different experiences with the security apparatuses during communist regimes. As discussed above, the three countries represent different varieties of communism, and the differences seem to fall well in line with the intensity and efficiency of the KGB in Estonia; the highly contested nature of Polish communist rule in the 1970's and 80's, where the Służba Bezpieczeństwa (SB) in Poland became more notorious (Flam 1998;Tang 1983); and, finally, the Uprava državne bezbednosti (UDBA) in Yugoslavia, which was not as repressive as in the other two cases (Flere and Klanjšek 2014). That the effect of ontological insecurity is large throughout all cases thus may tell us something about the post-communist condition and how intelligence agencies are perceived today.
The supplementary materials contain extensive by-country analyses that display similar results (Tables S5-S7). The Estonian data stands out because of the relative importance of social trust in Model 3 in contrast to Poland and Serbia. In Estonia, the effect in model 3 is −0.214 (p < 0.01) whereas in Serbia it is −0.0663 (p < 0.1). In Poland, it is not significant at all. The Estonian models further display the largest effects on institutional trust (−0.425, p < 0.01). We know from the ESS5 that Estonians are more trustful towards the criminal justice system than their Polish and Serbian counterparts, and this is reflected also in our models (Jackson et al. 2011). The Estonian output also features a small but significant effect on both age variables, but not strong enough to conclude the existence of a generational effect.  Finally, the control variables in Table 4 also display interesting effects. Income and age are not significant. Higher education is significant and comparatively sizeable throughout, but it is worth noting that it is mainly driven by Polish respondents (see Table S7). This reflects earlier research, notably by Zureik and Harling Stalker (2010) and Budak, Anić, and Rajh (2012), who discuss the importance of education when modelling attitudes to surveillance. People with higher education thus seem to have greater concerns about secret surveillance than people without. Why this effect is visible only in Poland, however, goes beyond the reach of this analysis. In sum, Table 4 yields that institutional trust, ontological insecurity and privacy sensitivity are the most prominent predictors of this dimension of attitudes to secret surveillance, especially for (well-educated) Poles. The most important result is the large positive effect of ontological security across all cases, despite minor differences in other variables. We now proceed to discuss the second dimension of attitudes to secret surveillance.

Acceptance for intelligence agencies' surveillance practices
We have shown that judging by the LFiW data, a high degree of existential concerns about, for example, war, unemployment etc. tends to explain worries about surveillance. Through the attribution of different institutional restraints to surveillant institutions, respondents were probed for the other dimension of attitudes, namely acceptance. Table 5 shows odds ratios from the ordered logistic regression using acceptance for intelligence agencies' surveillance as dependent variable. Our expectations for this variable are different from Table 4, and we expect even more salience for the institutional trust variable. If our hypotheses are correct, ontological security should play a smaller role, but privacy concerns should remain a key predictor. Table 5 shows odds ratios in which the larger the deviation from 1, the larger the effect. Values below 1 indicate negative effects, while above are positive ones. The ordered logistic regression lies very much in line with earlier research, in that it highlights institutional trust as predictor for acceptance of secret surveillance. The output shows that our models align with e.g. the PRISMS survey, which showed similar effects for these variables (Friedewald et al. 2015). In general, the results meet our expectations. Institutional trust is the main predictor of acceptance for secret surveillance, with increasing values greatly reducing the odds of respondents supporting rigid restrictions of intelligence agencies' practices. The average odds are reduced by around 17% for a unit increase in institutional trust to result in a more positive response than the mean (which is "Yes, court order") in both models 2 and 6. Social trust is insignificant in Model 3 but significant and positive, albeit small (6.6%) in the full model. In the previous regression using worries as dependent variable, this was the other way aroundits effect was consumed by ontological insecurity in Model 6 (see Table 4). This suggests that the more rational dimension of social trust plays a bigger role here and that there is some degree of support for Hypothesis 1b. Privacy sensitivity has a stable but small effect across the models at about 5-10%, which is unexpected only in its small size. Ontological insecurity shows a small but significant positive effect in Model 4, below 5%, but is not significant in the full model. We will return to interpret this effect in the discussion. The by-country analyses show a high degree of uniformity, except a slightly higher effect of ontological insecurity in Estonia (see Tables S10-13 in the supplementary materials). On the overall, however, we find that the data matches our expectations as outlined in the previous section.
The control variables do not show significant effects in the case of acceptance. We are again reminded that the Polish respondents are markedly more critical to intelligence agencies than their Estonian and Serbian counterparts. The odds of replying "No, never" are almost 73% higher in Model 6 as compared to a Serbian respondent. Education, income, and age, however, do not affect the acceptance variable at all, and here we will have to look for other explanations. The generational gap, as mentioned earlier, is thus not relevant in the cognitive evaluation of intelligence agencies' integrity.
In order to test the robustness of these results, the dependent variable was recoded into a three-level nominal scale variable, with the responses "No, never", "Yes, court order" and a combined category for "Yes, senior officer" and "Yes, in all cases". We apply this new dependent variable in a multinominal logistic regression in Table S14 (see supplementary materials). This type of regression has the advantage of showing the likelihood that a respondent will state in favour of the intelligence agency in question, or against it, in relation to the majority response, "Court order". The robustness check corroborates the results presented above: by every unit increase in institutional trust, the relative risk of responding in the "Yes" category increases by approximately 20%. This analysis also substantiates the other factors as well, but they are all fairly Robust seeform in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.
smallabout 2-7% in the full model. We now proceed to discuss the results in more general terms.

Discussion
The aim of the article was to study attitudes to surveillance in post-communist societies, and to test alternative explanations to the institutional trust. In the study, attitudes are conceptualised as worries about and acceptance of secret surveillance by intelligence agencies. These two dimensions bring out different aspects of how people relate to secret surveillance, worries being a more general concern for surveillance, and acceptance a more concrete relation to the intelligence institutions in one's own country. To study attitudes to surveillance, we use a data set created within the research project Like Fish in Water: Surveillance in Post-Communist Societies, which was funded by the Foundation of East European Studies and carried out by Ipsos.
Using earlier research as a starting point, a set of four hypotheses were formulated about institutional and social trust (H1a+b), privacy sensitivity (H2), and ontological insecurity (H3). Of these, only the second on privacy concerns finds support in our data. The other hypotheses are rejected. H1a is rejected because institutional trust does not only predict acceptance but also worries, albeit to a lesser degree than privacy concerns and ontological insecurity. H1b is rejected because social trust is only significant for acceptance but not for worries. Its effect is very small in all country-specific regressions as well, with the exception of worries in Estonia, where it is sizeable in one of the models. Nonetheless, the hypothesis was falsified in our analysis. Finally, H3 is rejected because the expected negative effect of ontological insecurity was not present. Of the demographic controls the presence of higher education additionally has a stable and substantial positive effect on worries about surveillance, particularly in Poland. Finally, there are quite important national differences in the data. The cases were selected on the trust variable, and here we see that this difference is also represented in attitudes to surveillance, where Estonian are the least critical and Poles the most critical. In sum, the majority of our expectations were not fulfilled.
However, rejecting the model's hypothesis does not affect the importance of the results. Our analyses point to the importance of feelings of existential insecurity in the formation of attitudes to secret surveillance. In particular for worries about secret surveillance, it is as an additional factor to institutional trust and privacy concerns. Based on studies mainly in Western societies, the expectation was that insecure respondents would care less for surveillance in light of other stress factors. However, the reverse is the case. Here the post-communist context hypothetically plays a major role in the dynamic between these factors. The post-communist populations probed in this study, despite their vast historical and socio-political differences, all perform rather similarly in the analysis above. Since these societies' shared commonality is the experiences of communist rule, the results suggest that this is a more general phenomenon in post-communist societies.
Theoretically, this may be surprising, but empirically, it shows the specificity of the postcommunist context for attitudes to surveillance. The results have important implications for our knowledge of the European post-communist societies. First, we discovered that worries about surveillance are conditional upon feelings of insecurity, in this case about the contemporary societal threats. It follows that generally anxious people worry more specifically also about surveillance. This explanation thus suggests fear of surveillance is a "spill-over" from a more general state of existential anxiety, which, as we discussed above, has been considered part of the post-communist condition. The rapidly developing security situation in Central and Eastern Europe will therefore most likely make this factor even more important. Second, respondents who worry about surveillance by intelligence agencies and society in general may also do so because surveillance is part of the threat that they perceive. The domestic influence of intelligence agencies renders them something to be concerned about, which would not be surprising due to many surveillance scandals that have plagued post-communist societies after 1989/90 (Berendt 2015;EurActiv 2011;The Economist 2013). Certainly, insecurity is in many cases simply a question of individual personality traits, but, as shown here it seems reasonable to argue that there also is a cultural dimension, in this case defined as post-communism.
This leads us back to the question of communist legacies of surveillance, as discussed in the introductory sections above. An attitudinal legacy can be said to exist if it is contingent upon phenomena or experiences during the previous regime (Wittenberg 2015), but in order to establish it as a fact, further research is necessary. First, similar items should be tested in other countries, and second, other methods should be applied to see if the results still hold. The SurPRISE project used large citizen summits as a means to strike a balance between quantity and quality, and a similar methodology could inform the present issue as well. Nonetheless, the present study yields the hypothesis that attitudes to secret surveillance in post-communist societies, in addition to being a cognitive reaction to institutional performance, are formed on the basis of collective cultural experiences from the communist period. In post-communist studies, the debate on performance versus culture can here be reformulated into performance and culture. The fact that the effects reported here are equal across all three countries, despite their differences, should be a strong indication that shared experiences of secret surveillance and authoritarian rule leave local cultural legacies behind. Even after almost 30 years they still affect perceptions of security. Whether the phenomenon can also be observed in other types of post-authoritarian societies, like Chile, Spain, or Portugal, is an interesting question for future research.
The results are also interesting in light of both surveillance and trust theory. First, our result further weakens the support for linking social trust to attitudes to surveillance. This factor did not predict attitudes to surveillance, despite the instinctive appeal in a post-communist context, where peer relations were often crucial. Hypothetically, the communist security apparatuses were efficient to the extent that they eradicated social trust through the vast informer networks that they employed (Persak and Kamiński 2005). If so, social trust may be more important in contexts where there are no prior experiences of such systems. The PRISMS project would however have discovered this link if it had existed, because it included all EU member states and probed for similar questions as the LFiW survey. Second, earlier quantitative research on surveillance usually probed different types of surveillance technologies in order to deconstruct the nature of attitudes on surveillance. In this vein, the SurPRISE, PRISMS, PACT, and Globalisation of Personal Data projects all looked at forms of surveillance such a as CCTV, smart cards, corporate profiling, and transport monitoring. These projects problematised the concept of surveillance in terms of the means and objectives of surveillance, but respondents' feelings remain understudied. The additional dimension of surveillance allows us to study another aspect of subjects' attitudes to surveillance and arguably get a more complex view of the complexity of beliefs that people hold without losing explanatory power. The fact that this study specifically relates to secret surveillance also poses the question whether ontological insecurity may be specific to covert information gathering. In light of earlier research our results seem to suggest different explanations for open and secret surveillance. Future surveys would be well advised to incorporate a multidimensional view of surveillance in their questionnaires in order to avoid overly deterministic accounts of respondents' attitudes. However, it seems important to further develop the worries dimension in order to better probe this construct. A further development of the survey instrument could probably help us get a clearer picture. Notes research possible. Further, vignettes also have to be translated and thus also pose a methodological question in the local adaptation of an international questionnaire. 13. The wording of the question was "When it comes to privacy, how concerned are you about the following?". 14. There are substantial variations in internet penetration between the three countries, and even bigger differences in use of social media, experiences with online consumption, and online banking (Eurostat, 2017a, b). This is also reflected in the "Don't know/No answer" in our data, with substantial data loss foremost in Serbia and Poland. It therefore cannot be excluded that the sample population may be asymmetrically affected by privacy issues regarding mapping of internet users. 15. Friedewald et al. (2015, 87) make a similar argument in their analysis of the PRISMS data. 16. The rotated factor analysis output is available in the online supplementary materials. 17. Spearman's rank correlation coefficient is considered to be better suited to categorical variables than the more common Pearson's product moment coefficient (see Hauke and Kossowski 2011).