2.1. Norms in social philosophy

One of the most influential accounts of social norms in the philosophical literature is that of Bicchieri (Reference Bicchieri2006). Bicchieri’s analysis of social norms is developed in a rational choice framework and, as such, it shares the same underlying approach with institutional economics (see sections 2.2 and 2.3).

According to Bicchieri, social norms transform mixed-motive games into coordination games, this way providing incentives for individuals to contribute to the common good. To see how this works, consider a standard prisoner’s dilemma game. When there are no social norms prescribing to contribute to a common good, the incentives of the game are such that rational players prefer to defect from cooperation. However, when social norms are present, the payoffs of the game change: the incentives to defect decrease by a factor that measures the psychological (or material) cost of deviation. This cost arises when a player expects that others cooperate and thinks that others expect her to cooperate. As it was briefly mentioned before, these expectations are called empirical and normative expectations.

More formally, a behavioural rule is a social norm when the following three conditions hold for individuals that are part of a community or group:

Only when a sufficiently large part of the community – a subset – adheres to these three conditions, it can be inferred that a social norm exists in that community.

The first condition entails that an individual knows that the rule applies to a certain context. This condition implies that for a behavioural rule (for example that you should not litter in public spaces) one should know that the rule prescribes you to act in a certain way when you find yourself in such a situation (e.g. to throw your rubbish in a bin when you find yourself in a public space). Condition (2a) specifies that an individual expects that others will conform to the norm in similar situations, e.g. that others will throw their rubbish in a bin when they find themselves in a public space. This expectation may arise through inference: when observing that others do not litter, it is plausible to assume that their behaviour remains constant over time and that they will therefore not litter in the future. ^{Footnote 2,Footnote 3} The third and final condition (2b) refers to the belief that an individual has about others expecting him/her to conform to the norm, e.g. that others expect you to put your rubbish in a bin when you are in a public space. For some people this belief is motivated by observing that a form of punishment, such as shame or social ostracism, is enforced. Normative expectations have a non-social counterpart that specifies ‘personal beliefs about what should be done’, so-called normative beliefs (Bicchieri et al. Reference Bicchieri, Lindemans and Jiang2014: 3).

We highlight an additional feature of Bicchieri’s model. As Bicchieri shows, normative and empirical expectations motivate norm conformity. When individuals are asked why they followed a certain norm, they typically report both kinds of expectations. However, there may be cases where normative and empirical expectations are not both simultaneously present for the same individual. For example, consider a group where recycling is not a common practice. In this group, there may be an individual who does not have empirical expectations (she thinks others typically do not recycle) but may hold normative expectations (she thinks that others believe that she ought to recycle). Vice versa, an individual may hold empirical expectations (she thinks others typically do recycle) but lack normative expectations (she thinks that others believe that she ought not to recycle). This theoretical distinction is examined in an experimental study, in which Bicchieri and Xiao (Reference Bicchieri and Xiao2009) analyse the relative importance of empirical and normative expectations for pro-social behaviour. In particular, the study shows that empirical expectations tend to overrule normative expectations when the former are present but the latter are not. In other words, it is more likely that someone will follow the norm when only empirical expectations are present than when only normative expectations are present. An intuitive explanation for this is that, given that social norms prescribe to contribute to a common good, if an individual believes that others do not contribute to a common good but believes that others expect her to do so, she might decide to defect as she realizes she would not benefit from the advantages that derive from mutual cooperation.

2.2. Norms in economics

In the institutional economics literature, we encounter the notion of civic norms. Civic norms have been defined in a variety of ways. As was briefly mentioned in the introduction, Knack and Keefer (Reference Knack and Keefer1997) define civic norms as constraints on narrow self-interest that give people incentives to contribute to the public good. Guiso et al. (Reference Guiso, Sapienza, Zingales, Benhabib, Bisin and Jackson2011: 3) define civic norms as ‘those persistent and shared beliefs and values that help a group overcome the free-rider problem in the pursuit of socially valuable activities’ (see also Nannicini et al. Reference Nannicini, Stella, Tabellini and Troiano2013). Related views are found in the work of Owen and Videras (Reference Owen and Videras2006: 814), who state that civic norms arise when ‘people cooperate and contribute to public goods more than is expected given their incentives to free-ride’. Other studies refer to civic virtues or civic morality, which are informed by ethical theories such as virtue ethics (Letki Reference Letki2006). These accounts stress the habit of forming civic virtues such as trust, honesty and reciprocity. It is assumed that such virtues generate a sense of responsibility for the public good that fosters cooperation and deters people from corruption and free-riding. Other, more general, definitions describe civic norm compliance as acts in favour of the community (Aghion et al. Reference Aghion, Algan, Cahuc and Shleifer2010) or as an individual’s perception of how citizens should behave in relation to the state (Kotzian Reference Kotzian2014).

Oftentimes, civic norms are related to social capital (Letki Reference Letki2006; Guiso et al. Reference Guiso, Sapienza, Zingales, Benhabib, Bisin and Jackson2011; Kotzian Reference Kotzian2014): a phenomenon that generates trust and cooperation (Knack and Keefer Reference Knack and Keefer1997; La Porta et al. Reference La Porta, Lopez-de-Silanes, Shleifer and Vishny1997), fosters economic growth and financial development (Knack and Keefer Reference Knack and Keefer1997; Guiso et al. Reference Guiso, Sapienza and Zingales2004), and contributes to well-functioning institutions (Letki Reference Letki2006; Nannicini et al. Reference Nannicini, Stella, Tabellini and Troiano2013). ^{Footnote 4} The social capital literature analyses the impact of civic norm conformity on various aspects of society. In their seminal paper, Knack and Keefer (Reference Knack and Keefer1997), for example, find a significantly positive relationship between civic norm conformity, trust and economic development. Other studies have related civic norm conformity to citizens’ participation in society (Dalton Reference Dalton2008), to citizens’ perception of their neighbourhood, and to citizens’ trust in others (Tabellini Reference Tabellini2008; Aghion et al. Reference Aghion, Algan, Cahuc and Shleifer2010). Also the roots of civic norms have received attention (see for example the work of Letki Reference Letki2006; Dinesen et al. Reference Dinesen, Nørgaard and Klemmensen2013; Kotzian Reference Kotzian2014).

To operationalize civic norm conformity, scholars in institutional economics typically use social survey databases (ESS, EVS and WVS among others). ^{Footnote 5} They extrapolate survey items that inform us about one’s individual beliefs (Knack and Keefer Reference Knack and Keefer1997; Owen and Videras Reference Owen and Videras2006; Guiso et al. Reference Guiso, Sapienza, Zingales, Benhabib, Bisin and Jackson2011). In particular, survey items are selected that ask whether behaviour, such as avoiding taxes or claiming government benefits to which one is not entitled, is justified. Such types of behaviour are illustrative examples of instances in which civic norms play a role; tax avoidance as well as claiming benefits to which one is not entitled are both acts that can be considered as free-riding, since individuals choose to maximize their own private gain instead of public gains. The selected survey-items ask respondents to rate their beliefs regarding behavioural rules on a 10-point scale, where 1 is ‘never justifiable’ and 10 is ‘always justifiable’. Depending on the aim of the analysis, scholars sum the individual items for each individual (Knack and Keefer Reference Knack and Keefer1997), recode the items using only the extreme responses (Owen and Videras Reference Owen and Videras2006), or reduce the items to an underlying factor using factor analysis (Letki Reference Letki2006; Guiso et al. Reference Guiso, Sapienza, Zingales, Benhabib, Bisin and Jackson2011). Moreover, analysis is conducted at the individual level (Dalton Reference Dalton2008), national level (Knack and Keefer Reference Knack and Keefer1997) or at multiple levels (Letki Reference Letki2006; Kotzian Reference Kotzian2014).

Notice that the operationalization of civic norms considers whether an individual believes that the civic norm in question is justified, but not what an individual believes others will do. This is striking for various reasons. Accounts within the institutional economics literature incorporate the interaction between the individual and the group in their definitions, by taking civic norms as rules of behaviour that the members of a group follow so as to contribute to the common good and to overcome free-riding behaviour. As we saw before, also Bicchieri considers the inter-subjectivity of norms to be a crucial aspect of norm compliance. Moreover, within cross-cultural psychology and cross-cultural economics group-norms and their influence on individuals’ behaviour is increasingly taken into account. Perceived normative beliefs have been found to affect behaviour, and are in turn affected by (one’s perception of) group behaviour (Fishbein and Ajzen Reference Fishbein and Ajzen1975; Cialdini et al. Reference Cialdini, Kallgren and Reno1990). ^{Footnote 6} Fischer (Reference Fischer2006) highlights the difference between self- and culture-referenced value ratings at the individual level. He argues that social norms regulate the latter. Relatedly, other scholars have argued that it is one’s perception of group behaviour, instead of the actual group behaviour or personal beliefs (Shteynberg et al. Reference Shteynberg, Gelfand and Kim2009), that directs norm conformity (Leung and Morris Reference Leung and Morris2015). The importance of social ties is also examined via social networks, in which individual beliefs are modelled as being affected by one’s connections with others (Bandiera and Rasul Reference Bandiera and Rasul2006; Videras et al. Reference Videras, Owen, Conover and Wu2012). Norms are one of the several mechanisms that are proposed as explanation of social linkages.

Furthermore, among others, Leung and Morris (Reference Leung and Morris2015) note that an ‘inter-subjective approach’ that accounts for interaction between individuals is not put forward in the institutional economics literature. ^{Footnote 7} Inglehart and Welzel (Reference Inglehart and Welzel2005) provide more detailed evidence that indicates that civic norm measures are not as ‘civic’ as the literature suggests. Instead of measuring norm conformity, they capture a resistance to conformity pressures (Delhey et al. Reference Delhey, Newton and Welzel2011), which is, according to Bicchieri (Reference Bicchieri2006), an individual psychological parameter that might modulate norm compliance but does not determine it.

2.3. Comparing the approaches

At this point, we may want to take stock and reflect on the sense in which social norms and civic norms can indeed be compared in a meaningful way. After all, Bicchieri’s operationalization requires behavioural, empirical and normative expectations; survey measures, on the other hand, capture normative beliefs. The respondent’s expectations regarding others and his or her beliefs regarding the expectations of others are not taken into account.

One option would be to say that civic and social norms are simply different and their measurement procedure reflects such differences. Remember, however, that the concept of civic norms largely overlaps with the definition of social norms by Bicchieri (Reference Bicchieri2006). Both civic norms and social norms indicate rules that impose constraints on self-interest so as to solve social dilemma games, i.e. situations where the interest of a single individual is in conflict with the interest of the group. Furthermore, both economists and philosophers refer to behavioural rules that can be captured by one’s knowledge and beliefs.

Given the analogies, it would be surprising if different properties are operationalized in order to test such similar outcomes. Moreover, institutional economists seem to agree that the role of the group is crucial for civic norms, but they simply do not include it in their measurement. In this respect, it is important to notice that the operationalization of a definition is affected in important ways by the empirical method that is used for testing a related hypothesis. It may be the case that Bicchieri’s definition is too demanding when working with social survey data. Scholars in institutional economics are constrained by the availability of survey items, which have not been originally collected with the purpose of testing civic norm conformity. Other accounts that highlight the inter-subjectivity of norms may be less demanding than Bicchieri’s definition, yet they do not offer a theoretical framework in which expectations are conceptualized as essential conditions for norm conformity. While perhaps more demanding, Bicchieri’s account formalizes the role of the group in such a way that we can gain more insight into people’s motivation for norm conforming behaviour (as motivated by NEs and EEs). To examine this further, in the next section we suggest a possible way to operationalize Bicchieri’s three conditions while using social survey data, which can be used for standard examples of civic norms.

3.1. Operationalization of conditions

In order to assess ways to operationalize civic norm conformity while using social survey data, we take Bicchieri’s three conditions as point of departure. To start with, we are interested in measuring conformity to, for instance, the following behavioural rule: to pay a fare when travelling by public transport. In an ideal scenario, following Bicchieri’s framework, we would devise survey items that reflect each condition. We would ask for the behavioural condition ‘When travelling by bus or train, is there a rule requiring passengers to pay a fare?’. ^{Footnote 10} Upon a positive answer to the previous question, the empirical question would ask: ‘When travelling by bus or train, to what extent do you expect that others pay their fare?’. Finally, the normative question would ask: ‘To what extent do you think that others expect you to pay your fare?’. Given Bicchieri’s framework, conformity to the norm is more likely when all three conditions are present, in other words, when the respondent answers positively to the three questions: ‘I know that, when travelling by bus or train, there is a rule requiring passengers to pay a fare’ – ‘I expect that (a sufficient number of) others pay their fare’ – ‘I expect that (a sufficient number of) others expect me to pay my fare’.

Following this line of thought, we looked for survey items that match the ideal scenario. To this purpose, we translated Bicchieri’s three conditions (behavioural, empirical and normative) into keywords that reflect aspects of each respective condition. In a next step, the keywords were used to extrapolate items from the EVS, and as such to operationalize each condition. We briefly describe our approach below.

For the behavioural condition, the EVS does not contain items that ask directly if a respondent is aware of the existence of a rule in a certain situation. However, we take the answer an individual gives about how he or she would behave in a situation as the closest indication of the behavioural condition. To do so, we translate the behavioural condition into keywords that ask for the respondents’ behaviour, for example: whether and how often they engage in certain activities, whether they are willing to engage in certain activities in the future, and whether they have done so in the past. ^{Footnote 11}

The empirical condition is represented by keywords that ask the respondents questions regarding other people’s behaviour, using words such as: most people, people in general, or compatriots. Who these ‘other people’ are is determined by the country of residence of the respondent. Survey items are adapted accordingly. In this manner the expectations towards other citizens are captured. For the operationalization of the normative condition, we note that the EVS does not contain items that ask whether the respondent expects that others believe that they should act in a certain way. ^{Footnote 12} We thus follow the approach of existing measures and use normative beliefs as first-best operationalization of normative expectations. ^{Footnote 13} We select keywords that are normatively laden, for example: moral rule, must, need to, one should or duty. In doing so, we assume that if an individual believes that x should be done or is justified, it is plausible to expect that this individual believes that others think that x should be done. Table A1 gives an overview of all keywords.

Table 1. Relation NE and EE conditional upon socio-demographic characteristics.

The nine columns show the regression results for nine separate regression models, in which the dependent variable is the NE for a particular behavioural R and the independent variable of interest is the EE for the same behavioural rule R. The table depicts clustered standard errors at the country level in parentheses. The regression models are estimated with country fixed effects. *** p < 0.01, ** p < 0.05, * p < 0.1.

3.2. Item selection

We use the keywords to search for items that operationalize the three conditions for social norm conformity. The search yields a large number of items (> 600). From this list, we find that there are 18 items that capture normative (NE) and empirical expectations (EE) with respect to the same behavioural rule. ^{Footnote 14} For these rules, there are no items that can be used to test the behavioural condition together with the normative and the empirical conditions. Omission of the behavioural condition is, however, not as problematic as omitting the empirical or normative condition. The latter two conditions together define a conditional preference, which is, as argued in section 2.1, a prerequisite for social norm conformity.

The 18 extrapolated items ask for nine behavioural rules denoted as R: ‘Do you think it is justified to do [R]?’ (NE) and ‘According to you, how many of your compatriots do [R]?’ (EE). R captures civic behaviour that is related to the common good and is situated in the public sphere, such as cheating on your taxes or avoiding a fare when using public transport. The nine behavioural rules and accompanying items are listed in Table A2. Answers to the NE items are given on a 10-point Likert scale, where 1 reflects ‘never justified’ and 10 ‘always justified’. EE items are measured on a 4-point Likert scale. An answer of 1 reflects the belief that ‘almost all compatriots do it’ and 4 that ‘almost none do it’. We inverse the scale of the NE items such that their interpretation is in line with the interpretation of the EE items: a higher score means a greater likelihood that expectations are present. ^{Footnote 15} Consequently, a lower score should be interpreted as a lower likelihood that expectations are present.

Table 2. NEs and EEs as two dimensions of civic norms.

Factor outcomes are rotated using an oblique rotation. N = 13,902.

Recall that norm compliance requires the respondent’s normative and empirical expectations to be both present (Bicchieri Reference Bicchieri2006). For the extrapolated items, this entails the following. Taking the rule claiming government benefits to which someone is not entitled as an example, an individual that answers the NE item with ‘never justified’ and the EE item with ‘almost no compatriot does it’ believes that claiming government benefits to which you are not entitled is never justified and that almost no compatriot claims government benefits. For the individual in question, both expectations are present and it is therefore likely that this individual conforms to the civic norm and contributes to the common good by, in this case, not claiming government benefits to which he or she is not entitled. Alternatively, when she answers the NE item with ‘never justified’ and the EE item with ‘almost all compatriots do it’, empirical expectations are absent while normative expectations are present. Following Bicchieri, the likelihood that this individual conforms to the civic norm is lower than in the previous case where both EEs and NEs are present. Similar interpretations hold for the other eight paired items.

While the NE items are available across multiple waves, the EE items have only been asked in wave 3 of the EVS (1999–2001). ^{Footnote 16} We therefore use data from this wave in our analyses. Descriptive statistics can be found in Table A3. These statistics show that the mean values of the nine matched NE and EE items differ. T-tests of the difference in (normalized) means for each matched pair show that the differences are indeed statistically different as Bicchieri’s model would predict: it suggests that the expectations are empirically distinct and thus the likelihood to conform varies across individuals. In the next section we further examine the relation between the two types of expectations.

Table 3. The impact of NEs and EEs on trust.

Clustered robust standard errors at the country level in parentheses. Estimated with country fixed effects.

*** p < 0.01, ** p < 0.05, * p < 0.1.

5.1. Response-style biases

We assess whether survey responses are affected by factors that are unrelated to the survey item’s content (Baumgartner and Steenkamp Reference Baumgartner and Steenkamp2001; Beugelsdijk and Klasing Reference Beugelsdijk, Klasing, Costa-Font and Macis2017). Such response-styles may bias our results and therefore warrant attention. Previous research has shown that extreme response styles (ERS) can be explained by country-level extraversion (Harzing Reference Harzing2006), and are more likely to be encountered in countries with high uncertainty avoidance and high power distance (Hofstede Reference Hofstede1980). ERS refers to responses that are at the extreme ends of an answer-scale (negative or positive). In contrast, people may also have a tendency to give responses that lie in the middle of the answer-scale, so-called middle response styles (MRS). Such styles are argued to be more prevalent in collectivist societies (Hofstede Reference Hofstede1980). For further details regarding these hypotheses we refer to the work of Harzing (Reference Harzing2006) and Beugelsdijk and Klasing (Reference Beugelsdijk, Klasing, Costa-Font and Macis2017).

We test for ERS and MRS biases in the following manner: negative ERS is the fraction of people within a country that has given either the response 1 (EE) or 9/10 (NE). Positive ERS is defined as the fraction of people within a country that has given the response 4 (EE) or 1/2 (NE). MRS on the other hand focuses on the middle range responses: the fraction of people that has answered 2/3 (EE) or 5/6 (NE). We run a set of regressions at the country level with negative ERS, positive ERS and MRS as dependent variables and the degree of extraversion (Lynn and Martin Reference Lynn and Martin1995; Hemert et al. Reference Hemert, van de Vijver, Poortinga and Georgas2002), individualism, power distance and uncertainty avoidance as independent variables (Hofstede Reference Hofstede1980, Reference Hofstede2001). Significant effects of the four independent variables indicate we should be concerned for the consistency of the main findings. However, we find no significant relations between the dependent and independent variables: all coefficients are not significantly different from zero (see Table A4). This suggests that the results are not liable to response style biases.

Table 4. The impact of NEs and EEs on trust (sensitivity analysis).

Clustered robust standard errors at the country level in parentheses. Estimated with country fixed effects.

*** p < 0.01, ** p < 0.05, * p < 0.1.

5.2. Operationalization of NEs and EEs

NEs and EEs are measured on two different scales, i.e. a 10-point and 4-point scale. We check whether the results in Table 3 change if we rescale the NE items such that both NE and EE items are measured on a 4-point scale: for all NE items we construct a measure that attributes the values 1, 2 and 3 to a value 4; the values 4 and 5 to a value 3; the values 6 and 7 to a value 2; and the values 8, 9, and 10 to a value 1. ^{Footnote 24} We take the average value of the nine survey items to reconstruct our NE measure and use this measure in our regression model (eq. 1–4). For matters of brevity, Table 4 displays the estimation results of eq. (3) in column (1) (Table A5 part (a) displays the full results). The main findings are corroborated: the coefficients on NEs and EEs are of the same size and significance as those reported in Table 3. This gives us no reason to think that the results are driven by methodological considerations.

Table 5. Multilevel analyses of trust (sensitivity analysis).

Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Additionally, we examine whether the results in Table 3 are affected by the operationalization of NEs and EEs as approximated latent factors. As explained in section 4.2, we approximate the two latent factors by calculating the average value of the nine items that load highly on the respective factor even if the respondent has not answered all items. While these measures perfectly correlate with predicted factor scores, we assess whether this decision has affected the main results. We therefore approximate the two latent factors by the average values for only those respondents that have answered all 18 items. This restricts our sample from 38,278 respondents living in 33 countries to 13,902 respondents that live in 12 countries. ^{Footnote 25} We re-estimate the regression model (eq. 1–4) with the newly constructed approximated latent factors. Table 4 displays the results for equation (3) in column (2) (Table A5 part (b) displays the full results). When comparing the tabulated results in Table 4 with those tabulated in Table 3, we can observe that the EE coefficient is of comparable size and significance. Interestingly, the NE coefficient is significantly different from zero. While the coefficient is small, especially considering the effect size of EEs, we explore this finding further.

A subset of the 18 items has been surveyed in a very limited set of countries. ^{Footnote 26} We expect that the limitation of the country sample explains the findings displayed in column (2). We therefore construct a dummy ( ${\delta _{ij}}$ ) that indicates whether respondents are in the restricted country sample ( ${\delta _{ij}}$ = 1) or in the unrestricted sample ( ${\delta _{ij}}$ = 0). We add this dummy to our estimation model (eq. 1–4) as well as an additional interaction term with either NE, EE or both. The displayed results (columns 3–7 in Table 4) show that the interaction-effect is only significantly different from zero when interacted with NE (columns 3, 5 & 7). The effect of EEs on generalized trust, conditional upon controls, remains significantly positive (columns 4–6). These results suggest that country sample matters especially for the effect of NEs on trust and not so much for the effect of EEs on trust. This is in line with other studies that have shown that norm conformity at the country level is highly diverse and that there are various institutional factors that affect this (Gelfand et al. Reference Gelfand, Nishii and Raver2006, Reference Gelfand2011).

5.3. Multilevel analysis

The individuals in our data set live in various European countries. Observations are therefore likely to be clustered within countries, which is why we have estimated equations 1–4 in Table 3 using clustered standard errors at the country level. Another way to account for the nested structure of the data set is to use hierarchical modelling. Such a model differs from our previous analysis in that it assigns a statistical model to each hierarchical level (in our case, individual versus country level) and accounts for the varying group sizes (in our case, country sizes). Therefore, when working with nested data, it has been argued that such models yield more efficient estimates (Snijders and Bosker Reference Snijders and Bosker2012).

We estimate an empty hierarchical model that accounts for the nested structure. We thus remain agnostic as to the explanation of any country level variation. Similar to our previous analyses, we use a fixed effects logistic regression model to account for the binary nature of our dependent variable. Moreover, we include the same set of socio-demographic factors as done in equations 1–4 but do not display these results (see online appendix for results). The results in Table 5 indicate that the key findings are robust to the specification of our model. As expected, the estimates of the hierarchical logistic model are more efficient than those of the logit model as shown by the smaller standard errors.

5.4. Disaggregating the effect of NEs and EEs

Based on the factor analysis in Table 2, we have constructed approximated latent variables of NEs and EEs. We examine whether the findings tabulated in Table 3 are robust at the item level. We conduct the same analysis as in section 4.3, yet instead of including the average value of NEs and EEs, we relate the separate behavioural rules to generalized trust. We report the coefficients on NEs and EEs for equations 1–4 in Table 6 for the first behavioural rule, namely Claiming state benefits to which one is not entitled. Results for the other items resemble those displayed in Table 6 and are tabulated in Table A6 in the appendix. The results indicate that the positively significant relation between EEs and trust also holds at the item level, keeping other factors constant. We find that the presence of EEs strengthens trust, even when NEs are controlled for. In contrast to the results in Table 3, the relation between EEs and trust becomes insignificant when moving from specification (3) to (4) for the majority of behavioural rules. It may be the case that the variation that is unique to each rule obscures the relationship that we found at the dimensional level. Factor analysis is not liable to such variation, since it relies on the variation that is common to all input variables. We therefore recommend the use of a data reduction technique, such as factor analysis, in similar applications.

Table 6. The impact of NEs and EEs on generalized trust (item level).

Clustered robust standard errors at the country level in parentheses. Estimated with country fixed effects. *** p < 0.01, ** p < 0.05, * p < 0.1.