Explaining the educational gradient in trust in politicians: a video-vignette survey experiment

Abstract

Many liberal democracies are characterised by strong educational gradients in political trust: less-educated citizens have strikingly less trust in politicians than more-educated citizens. Informed by a wide body of research, this study tests systematically the empirical relevance of three characteristics of politicians that have been theorised as explanations for the educational gradient in trust in politicians: (1) substantive cultural stances (progressiveness/conservatism); (2) (non-)populist rhetoric; and a recently developed explanation, (3) perceived distance to citizens’ life-worlds. This study analyses data from an original, pre-registered, video-vignette survey experiment carried out among a high-quality, representative panel in the strategic case of the Netherlands, where respondents were assigned randomly to one of 16 professionally produced videos of an actor playing the role of a fictitious politician. It was found that the politician’s cultural stances and use of (non-)populist rhetoric were important drivers of the educational gradient in trust in the politician.

Keywords:

• Education
• political trust
• substantive representation
• populism
• survey experiment

In many liberal democracies, there is an entrenched educational gradient in the disengagement from (establishment) politics. While, for instance, more-educated citizens show greater electoral (Hadjar and Beck 2010; Jackson 1995) and non-electoral participation (Spruyt et al. 2020; van Deth 2020; Visser et al. 2023), their less-educated counterparts are more cynical about politics (Agger et al. 1961), more likely to support ‘challenger’ parties and candidates (see, e.g. Bakker et al. 2021b; Mutz 2018), and quite pressingly, report being more distrustful of politics generally and politicians specifically (Hakhverdian and Mayne 2012; Noordzij et al. 2021b, 2021c; van der Meer 2010). Understanding the roots of this educational divide is vital, since distrust in politicians, and politics more widely, could undermine the functioning and legitimacy of the political domain (Citrin and Stoker 2018; Hetherington 1998; Levi and Stoker 2000).

Our study answers the calls for scrutiny of ‘the causes and consequences of citizens’ trust in specific political actors’ (Levi and Stoker 2000: 496), by focusing on the marked educational gradient in trust in politicians in the Netherlands. More specifically, we experimentally test three theoretical approaches formulated in the literature as explanations of this educational gradient. The first of these addresses substantive representation, arguing that people trust politicians they agree with, while positions on substantive issues are strongly linked to the level of education (see, e.g. Aaldering 2017; Schakel and Hakhverdian 2018). The second argues that less-educated citizens in particular trust politicians who use populist rhetoric that agitates against a political elite that, in so-called ‘diploma democracies’ like the Netherlands especially (Bovens and Wille 2017, 2021), is overwhelmingly homogeneously more-educated (see, e.g. Bos et al. 2013), while more-educated citizens prefer pluralist politicians. The third, recently developed, approach, which has received less empirical scrutiny up to now, proffers that citizens are more likely to trust politicians who signal a shared life-world, which is strongly shaped by level of education (see, e.g. Noordzij et al. 2021a, 2021c).

We test the causal hypotheses deduced from each approach systematically. This is achieved using a unique, pre-registered, experimental design incorporated within a survey carried out among the members of a high-quality, representative panel sampled from the official population register in the Netherlands, which is a country characterised by a remarkably strong educational gradient in political trust (Hakhverdian and Mayne 2012; Noordzij et al. 2021b). The respondents (n = 3175) were assigned randomly to a brief, professionally produced video-vignette in which an actor plays the part of a fictitious, white, middle-aged, male politician who introduces himself to the respondents (the actor’s demographic represents the most common type of Dutch parliamentarian; because the actor is male, we employ the pronouns ‘he/him/his’ in what follows). Our factorial design utilises a 2 [cultural progressiveness (conservatism)] by 2 [non-populist (populist) rhetoric] by 2 [a culturally elitist (common) appearance and lifestyle] by 2 [work experience inside (outside) politics] (the last two of which are treatment elements for perceived cultural distance) factorial design.

There are four ways in which our study adds to the field. First, we simultaneously test the relevance of three key approaches to understanding the educational gradient in trust in politicians. Our factorial design is very well suited for doing so because it enables us to empirically pinpoint the contribution of each approach. Second, using a much-called for experimental design (Levi and Stoker 2000) is a step forward in a subfield that relies predominantly on correlational evidence. Third, in contrast to extant studies that introduce a (fictitious) politician to survey participants mostly via written quotes and statements (see, e.g. Bos et al. 2013; McDonald 2021), our treatments mimic citizens’ real-life exposures to politicians using video-vignettes modelled on actual political speeches and introductory videos, adding to the study’s external validity. Moreover, the use of videos is crucial for testing the role of perceived cultural distance, which, being a novel theoretical mechanism for explaining political trust, has not been tested in an experimental research design before. This is because the videos incorporate visual signals of the politician’s life-world that underlie this distance, like his appearance, posture, and way of speaking, for which written quotes or statements are less suited. Last, we employ not only the commonly used single-item measure of trust in the politician but also a newly developed multi-item scale that covers various dimensions of such trust, created based on a systematic inventory of extant research. This type of multi-dimensional measure of trust was already called for decades ago (cf. Levi and Stoker 2000), and its potential relevance to the field has been re-emphasised more recently (cf. Citrin and Stoker 2018).

The study’s findings demonstrate that the politician’s cultural stances were crucial in shaping the educational gradient in trust in the politician. The results also show that, for one of our two outcome measures (detailed below), the politician’s use of (non-)populist rhetoric also shaped that gradient. We conclude by reflecting on the relevance of these outcomes to debates on political trust, as well as on our null findings regarding the treatment used to expose respondents to signals of the politician’s cultural distance. Finally, we propose a research agenda that incorporates insights from our study.

Explaining the educational gradient in trust in politicians: three approaches

Substantive representation

The first approach goes right to the heart of the extant literature on political trust and centres on substantive representation. Here, trust follows from citizens’ rational evaluations of the extent to which their ideological stances are represented or pursued in politics: correlational evidence indicates that those who have greater ideological congruence with the political party they vote for, parliament as a whole, or government show more satisfaction with and trust in politics (Bakker et al. 2021b; Hetherington 1998; Mayne and Hakhverdian 2017; Miller 1974; Noordzij et al. 2021b; Stecker and Tausendpfund 2016). Consequently, the descriptive overrepresentation of specific groups in the political domain is theorised as being linked to an overrepresentation of their substantive stances (see, e.g. Aaldering 2017; Carnes 2012; O’Grady 2019; Pitkin 1969).

In diploma democracies in particular, which are characterised by strong educational gradients in political trust, less-educated citizens face substantive underrepresentation: their positions are advocated for less by politicians, almost without exception existing of more-educated citizens (Bovens and Wille 2017, 2021; O’Grady 2019). Aaldering (2017) highlights that such incongruence exists between less-educated citizens and their government, particularly if no right-wing populist parties are represented. Similarly, Schakel and Hakhverdian (2018) found that their less-educated respondents experienced little congruence with a representative sample of Dutch parliamentarians in relation to all the issues they examined, especially the cultural ones, such as multiculturalism.

Addressing how education is related to citizens’ substantive stances is crucial if we are to understand the views of less- and more-educated citizens in relation to the politicians they perceive to advocate for their stances. Nowadays, cultural issues have come to dominate the political agenda in many liberal democracies (Flanagan and Lee 2003; Inglehart 1997; Kriesi et al. 2006, 2008; Stubager 2010). In northwest Europe, in particular, it is cultural issues rather than economic ones that pit less-educated citizens against their more-educated counterparts (Achterberg and Houtman 2009): the latter are more likely to be culturally progressive or cosmopolitan (in Anglo-Saxon contexts also referred to as culturally or socially liberal; cf. van der Waal et al. 2007) regarding immigration, cultural diversity, and national sovereignty, while the former tend to be more conservative or nationalistic on such matters (Stubager 2010; van der Waal and de Koster 2015). Hence, a politician that voices cultural progressiveness is likely to be trusted more by more-educated citizens and less by their less-educated counterparts, while one that voices cultural conservatism will inspire more trust among less-educated citizens and less among more-educated ones. As a consequence, for our expectation that diverging substantive stances shape the educational gradient in trust in a politician, we hypothesise:

Hypothesis 1: There is a positive interaction effect between education and the politician’s voicing of cultural progressiveness (instead of cultural conservativism) on trust in the politician in the video.

Populist rhetoric

In moving beyond substantive representation, Levi and Stoker (2000: 491) stated that ‘judgments about the trustworthiness of government or of politicians are more than ideological or partisan reactions to specific incumbent administrations. They are generalised judgments that influence whether citizens endorse or reject existing authorities and public policy or institutional reforms’. One such judgement involves perceptions of establishment politicians, in particular.

Populist politicians are the most visible propagators of anti-establishment messages. Populism is commonly defined as ‘an ideology that considers society to be ultimately separated into two homogeneous and antagonistic groups, “the pure people” versus “the corrupt elite”, and which argues that politics should be an expression of the volonté générale (general will) of the people’ (Mudde 2004: 543; see, also, Akkerman et al. 2014). As such, populism serves to address this perceived divide between ‘the elite’ and ‘the people’, claiming ‘to represent the rightful source of legitimate power – the people, whose interests and wishes have been ignored by self-interested politicians and politically correct intellectuals’ (Canovan 2004: 242; see, also, Taggart 2004). A core element of this divide is the attribution of blame to ‘the elite’ for depriving and misrepresenting ‘the homogenous people’ and their needs (Hameleers et al. 2018). Given the primacy of its discursive elements, scholars increasingly emphasise that populism is communicated through political rhetoric, i.e. ‘a rhetoric of appeals to “the people”’ (Canovan 2004: 244; see, also, Bos et al. 2013; De Vreese et al. 2018; Jagers and Walgrave 2007).

Populist rhetoric and its opposite—pluralist rhetoric—are likely to resonate differently among different groups of citizens, whose political attitudes and behaviour are subsequently affected by how dominant their preferred rhetoric is in politics (see, e.g. Marx and Nguyen 2018). In contrast to more-educated citizens, less-educated ones have been found to display greater animosity towards establishment politicians, identify more strongly with the homogenous entity of ‘the people’, and underscore and be more receptive to populist rhetoric (Bos et al. 2013; Noordzij et al. 2021a, 2021c). This rhetoric is thus likely to resonate with the opposition experienced by less-educated citizens towards a more-educated political establishment in many diploma-democratic liberal democracies. As more-educated citizens are, on the other hand, more likely to prefer non-populist (pluralist) rhetoric, we anticipate that a politician using populist rhetoric will be trusted more (less) by less-educated (more-educated) citizens, while a politician using pluralist rhetoric will be trusted more (less) by more-educated (less-educated) citizens. This underlies the second hypothesis:

Hypothesis 2: There is a positive interaction effect between education and the politician’s use of non-populist rhetoric (instead of populist rhetoric) on trust in the politician in the video.

Perceived cultural distance

A third approach shifts the focus towards differences in the descriptive representation of less- and more-educated citizens. Generally, citizens attach great value to politicians’ ‘ability to understand how people live and how they feel about politics’ (McDonald 2021), or to having ‘a certain shared experience or history’ (Arnesen and Peters 2018: 871; see, also, Mora et al. 2021). Additionally, people have a greater sense of satisfaction with (the proposals of) politicians when they feel better represented in descriptive terms or more understood (see, e.g. Arnesen and Peters 2018; McDonald 2021). It is not, therefore, a surprise that research increasingly highlights how members of the lower strata experience political resentment and feelings of ‘being left out’ when politicians are perceived to be indifferent to and distant from their life-worlds (see, e.g. Allen and Sarmiento-Mirwaldt 2015; Cramer 2016; Lamont 2018; Manning and Holmes 2013; Williams 2017). That these feelings of distance are especially widespread among less-educated citizens in diploma democracies is also unsurprising, given that education is one of the crucial factors shaping these life-worlds (Bovens et al. 2014; Bovens and Wille 2017, 2021; Roose et al. 2012).

Nevertheless, studies on the lower social strata’s (feelings of) descriptive underrepresentation have predominantly explored the concept while focusing on substantive representation (see, e.g. Aaldering 2017; Carnes 2012; O’Grady 2019; Pitkin 1969; cf. Heath 2015). Yet the perception of politicians as being distant from one’s life-world is not only likely to go hand in hand with perceptions of them as people with different substantive stances. Qualitative research focusing on less-educated citizens has recently provided in-depth explorations of the meanings they assign to their political positioning, how they experience distance to politicians beyond that in terms of substantive representation, and how this sparks a wide range of political discontents (Noordzij et al. 2021a; see, also, Visser et al. 2023). This research shows that in addition to politicians’ substantive stances, less-educated citizens attribute their political discontents to, among other things, politicians’ insensitivity to their lifestyles and worldviews, their indirect communication styles, and their alleged superiority signalling, each of which affects the perceptions these individuals have of politicians (see, also, Manning and Holmes 2013). Correlational evidence among the population at large, including both less- and more-educated citizens, further illustrates how perceived cultural distance to politicians contributes substantially to the educational gradients in various anti-establishment political attitudes and behaviour, including political distrust (Noordzij et al. 2021c).

In this study, we test the causal implications of this recently developed approach by assessing how the educational gradient in trust in a politician is shaped by signals of distance or proximity to their life-world. Based on extant research, we incorporate two signals: politicians’ a) appearance and lifestyle and b) work experience. While the features of the former indicate politicians’ membership of a ‘culturally elitist’ (or ‘common’) milieu (see, e.g. Roose et al. 2012), the latter signals familiarity with the life-worlds of others inside (or outside) politics (see, e.g. Beauvallet-Haddad et al. 2016; Gaxie and Godmer 2008). As such, politicians’ signalled distance to the life-worlds of less- or more-educated citizens crucially differs from their use of populist or pluralist rhetoric. Cultural distance does not revolve around whether or not politicians claim the pre-eminence of ‘the people’ in politics but is about whether their life-world attributes indicate they are part of ‘the people’ themselves (cf. Mudde 2004).

Tellingly, a less-educated interviewee in recent research by Noordzij et al. (2021a: 572) suggested that a politician in the Dutch parliament who rarely wears a suit and has visible tattoos signals that ‘there’s still a normal one among them’, while another interviewee emphasised that many politicians ‘travel in different circles, they’re confronted with different things’. Similarly, many politicians were said to ‘don’t have a clue about how it works in the world’: only occasionally would they ‘they come to see what it’s like at places where [ordinary people] work’. Consequently, we anticipate that less-educated (more-educated) citizens will report higher (lower) levels of trust in a politician whose life-world herald that he is close to their life-worlds, while more-educated (less-educated) citizens will report higher (lower) levels in case the politician’s life-world matches their own. This leads us to the following hypotheses:

Hypothesis 3A: There is a positive interaction effect between education and the politician’s culturally elitist appearance and lifestyle (instead of common appearance and lifestyle) on trust in the politician in the video.

Hypothesis 3B: There is a positive interaction effect between education and the politician’s work experience inside politics (instead of work experience outside politics) on trust in the politician in the video.

This is the first research to attempt to capture the recently developed notion of perceived cultural distance in an experiment. Moreover, contrary to the notions of substantive representation and populist rhetoric, perceived cultural distance is not wholly identifiable by what the fictitious politician says in the video, but also by his appearance and gestures. An additional analysis was thus performed to validate whether our framing of the politician as someone with (a) a culturally elitist (instead of common) appearance and lifestyle and (b) work experience inside (instead of outside) politics did indeed lead to a greater sense of perceived cultural distance among less-educated citizens than in their more-educated counterparts. We, therefore, formulated two additional validation hypotheses:

Validation hypothesis I: There is a negative interaction effect between education and the politician’s culturally elitist appearance and lifestyle (instead of common appearance and lifestyle) on the perceived cultural distance to the politician in the video.

Validation hypothesis II: There is a negative interaction effect between education and the politician’s work experience inside politics (instead of work experience outside politics) on the perceived cultural distance to the politician in the video.

Data, research design, and analyses

Data and sample

Our study employed a pre-registered, between-subject, experimental design.¹ Respondents were assigned randomly to one of 16 video-vignettes of an actor playing the part of a fictitious politician in a 2 [cultural progressiveness (conservatism)] by 2 [non-populist (populist) rhetoric] by 2 [culturally elitist (common) appearance and lifestyle] by 2 [work experience inside (outside) politics] factorial design.

Our survey experiment was conducted among adult members of the high-quality Longitudinal Internet Studies for the Social Sciences (LISS) panel, which was created via sampling of the official Dutch population register and is administered by Centerdata (Tilburg University, the Netherlands) (Scherpenzeel 2009). Our survey was fielded from 6 to 28 September 2021. Of the 3175 panel members who initially responded (response rate of 63.6%), 3166 completed the survey. This meant that when counting the respondents who were not excluded for the reasons detailed below, between 145 and 168 individuals were assigned to each experimental condition. This roughly corresponds to, or even exceeds, the sample sizes seen in many other experimental designs in political and social science (including those that incorporate interaction terms) (see, e.g. Hameleers et al. 2017; Sulitzeanu-Kenan and Zohlnhöfer 2019; Sorek et al. 2018). Our data, sample, and the ethical considerations pertaining to the survey experiment are described in more detail in Online Appendix 1.

Given our focus on understanding the educational gradient in trust in politicians, we excluded respondents who had not yet completed their educational trajectory (this removed 153 respondents). Moreover, to ensure that we only included those who had watched the full video-vignette assigned to them, we used a variable created by Centerdata that enabled us to determine who had, or had not, played it in its entirety (removing 481 respondents).

Procedure

Respondents were able to complete our survey on a computer or tablet. If they tried to use Safari to do so, they were prompted to re-open the survey in another web browser to ensure the stability of the survey (professional helpdesk support was available, if needed). The respondents were told that we would show them a 40-second (cf. Hameleers et al. 2018) video of a politician, after which we would ask their opinions on several statements. On the next page, they were given the opportunity to test their device’s audio and video player. Before proceeding to the video, the respondents were reminded that they could only watch it once. We also provided contact details of the data collector’s support desk in case they encountered any problems.

We assigned the respondents randomly to one of 16 video-vignettes using forced equal sizes (Alferes 2012). After watching the video, they were asked to answer a total of 30 items, many of which were not used in the present study, but in research with a different substantive focus based on the same experiment. The respondents were also asked to answer a treatment-irrelevant manipulation check.² The survey concluded with a debriefing section, in which we told the respondents that the politician in the video was an actor, that his statements did not reflect his own political stances, and that we were interested in understanding how different people react differently to the videos they had seen.

Stimulus material

Our stimulus consisted of 16 professionally produced video-vignettes in which a professional actor portrays and introduces himself as a Dutch parliamentarian. The vignettes were modelled on real-life political speeches and videos of politicians, used, e.g. during election campaigns, which contributes to the external validity of our research. Our survey experiment crucially differs from extant experiments in its use of video. In contrast to candidate choice experiments in particular, we presented respondents with video-vignettes rather than asking them whether they would vote for or trust a politician by presenting them with written characteristics or pictures of politicians (either in a factorial experiment design or by letting respondents choose between politicians) (cf. Schwarz and Coppock 2022). Our vignettes are closer to how respondents are confronted with politicians in everyday life on, e.g. television or social media. Moreover, while candidate choice experiments have predominantly addressed politicians’ gender, (economic) class position, education, and occupational background (cf. Campbell and Cowley 2014; Carnes 2012, Carnes and Lupu 2016; Schwarz and Coppock 2022; Wüest and Pontusson 2017), our attention to politicians’ life-world attributes through the lens of perceived cultural distance is novel. Our video-vignettes have the advantage of including the visual signals that are relevant for exploring whether the politician is perceived as being culturally distant to the respondent, such as his appearance, posture, and way of speaking.

The four treatment elements central to our hypotheses comprise the politician speaking about his: (1) background and interests (i.e. lifestyle); (2) work experience; (3) opinions on the contemporary political domain, i.e. (non-)populist rhetoric; and (4) cultural stance.³ The video-vignettes also differ in terms of the politician’s appearance (detailed in the stills in Table A5 in the Online Appendices). So, the politician discussing his common (culturally elitist) lifestyle also has a common (culturally elitist) appearance. It should be noted that the actor used to portray the politician is a white, middle-aged male, reflecting the most common demographic profile of Dutch parliamentarians.⁴ Our use of a fictitious politician was not only motivated by the practicalities of using videos specifically designed for our study but also because an experiment showing a real politician would compromise the experimental manipulation due to the associations that respondents would probably already have with this politician before the experiment (cf. Bos et al. 2013; McDonald 2021).

The treatment element cultural attitudes involves the politician talking about his stances on issues of immigration and cultural diversity and traditions, which are currently salient in the Netherlands (Bakker et al. 2021a; Dekker et al. 2020). In this framing, our politician signals either cultural conservatism or progressiveness.

The treatment element (non-)populist rhetoric is based on multifaceted conceptualisations in the literature. It centres on populism as content, which concerns sharing the populist ideal (Bos et al. 2013; De Vreese et al. 2018; Jagers and Walgrave 2007), and taps into populism defined as being anti-elitist (De Vreese et al. 2018). This excludes references to out-groups other than ‘the elite’. The populist treatment comprises five elements: (1) voicing a sense of crisis (see, e.g. Bos et al. 2013; Taggart 2004), for which (2) blame is attributed to a failing, corrupt, and egoistic elite (see, e.g. De Vreese et al. 2018; Hameleers et al. 2018); (3) the need to defend a threatened ‘people’ against this elite (see, e.g. De Vreese et al. 2018; Mudde 2004); and stressing (4) the primacy of ‘the people’ (see, e.g. Akkerman et al. 2014; Canovan 2004), and (5) their common sense (see, e.g. Mudde 2004) in political decision-making. In contrast, our non-populist treatment takes a pluralist view of politics (see, e.g. Akkerman et al. 2014; Mudde 2004; Wirz 2018), and also consists of five elements: acknowledging a complex reality that gives rise to: (1) multiple challenges that are (2) attributable to multiple causes, (3) which affect everyone in society; and (4) stressing the important role played by politicians with different opinions that (5) together need to find a middle-ground.

Third, the treatment elements of appearance and lifestyle and work experience were developed using insights from recent qualitative research on less-educated Dutch citizens’ perceived cultural distance from politicians in the Netherlands (Noordzij et al. 2021a). We used both treatment elements as signals of perceived cultural distance. Our framing of a common appearance and lifestyle consists of the politician: (1) having a name that resonates with the lower social strata (i.e. Rob de Bruin), (2) showing his affinity with their lifestyle preferences (i.e. having a beer in the pub) and his sense of ease in their milieu (i.e. chatting to lots of people there), and (3) being casually dressed. In contrast, our framing of a culturally elitist appearance and lifestyle involves the politician: (1) having a name that resonates with the higher social strata (i.e. Roderick van der Veer), (2) showing his affinity with their lifestyle preferences (i.e. watching new plays at the theatre) and sense of ease in their milieu (i.e. talking to lots of people there), and (3) being formally dressed. In all the conditions, the name and the profession of the politician (i.e. parliamentarian) were displayed in the bottom left corner of the video for 10 seconds.

In relation to work experience outside politics, the politician stresses: (1) his experience outside politics and practical work experience and, as such, (2) his knowledge of the problems that ‘normal citizens’ face in the Netherlands. Work experience inside politics is, in contrast, signalled via (1) the politician’s experience inside politics and political offices and, as such, (2) his knowledge of the issues that are high up on the agenda in parliament.

While some combinations of the conditions above may occur more often in real-life politicians than others (which we further discuss in our conclusion), each treatment element is theorised to shape the educational gradient in trust in the politician in the video via a distinct mechanism. To carefully isolate the effect of each treatment element, our factorial experimental design includes all possible combinations of the four treatment elements’ conditions, resulting in 16 unique video-vignettes.

Measures

Table A6 in the Online Appendices presents the descriptive statistics for all the variables in the analyses. The four treatment elements were all coded as binary variables, in which (0) indicates cultural conservatism, populist rhetoric, a common appearance and lifestyle, and work experience outside politics; and (1) refers to cultural progressiveness, non-populist rhetoric, a culturally elitist appearance and lifestyle, and work experience inside politics.

After respondents watched one of the 16 video-vignettes assigned to them, we asked them to indicate their trust in the politician seen in it. Two measures were included. The first is based on a widely-used, single-item measure of political trust and asks respondents to indicate their trust in the politician in the video (see, e.g. Hakhverdian and Mayne 2012; Marien 2011; Noordzij et al. 2019, 2021b, 2021c). The answer categories range from (0) no trust at all to (10) complete trust. The second measure answers calls to ‘conceptualize and measure judgments of trust and trustworthiness’ (Levi and Stoker 2000: 497), for which ‘scholars have yet to develop a more comprehensive and useful set’ (Citrin and Stoker 2018: 51). Our second measure is a multi-item measure that reflects the multi-dimensional nature of political trust. It was developed especially for this research. Specifically, we created a comprehensive measure of four dimensions of trust that builds on the dimensional approach to political trust as discussed by van der Meer (2010), based on Kasperson et al. (1992), and was informed by extant measures and conceptualisations (Citrin and Stoker 2018; Devine et al. 2020; Levi and Stoker 2000; van Elsas 2015). For an individual to trust a politician, the politician must be perceived as: (1) able to perform according to the citizen’s expectations and the politician’s intentions (competence); (2) caring about the citizen’s wishes (care); (3) being accountable for their actions (accountability); and (4) behaving consistently and acting on their expressed intentions (reliability). We applied these dimensions to trust in the politician in the video using two items for each dimension. The answer categories range from (1) completely disagree to (7) completely agree. Higher values on this multi-item scale indicate greater trust in the politician in the video. As this is a novel measure, we also performed confirmatory factor analyses to test whether all eight items measure a latent construct of trust.⁵ One of the items measuring reliability (‘The politician in the video is someone who doesn’t easily change his opinion’) failed to load on an eight-item scale (0.04) and was subsequently removed, which means that we used a seven-item scale in our analyses. The strong correlation between the single-item and seven-item measure of trust in the politician in the video (r = 0.77; p < 0.001; n = 2541) indicates that the former reflects the multi-dimensional concept of trust and that both of our measures can be used to validly measure trust.

Our independent variable education was measured using the minimum number of years needed to complete the highest level of education our respondents reported to have completed. This ranges from (8) primary education to (18) university degree. We tested our hypotheses by interacting this measure of education with the four binary treatment variables. As our models assess whether the effect of an observed individual-level characteristic is moderated by a randomised treatment, we included relevant pre-measured background characteristics as control variables to allow for a causal interpretation (Kam and Trussler 2017).⁶ These controls are commonly included in other studies on the educational gradient in political attitudes and behaviour (Aaldering 2017; Hakhverdian and Mayne 2012; Mayne and Hakhverdian 2017; Noordzij et al. 2019, 2021c; Spruyt et al. 2016; Stecker and Tausendpfund 2016). These are: gender, age, political interest, income, labour-market position, non-native, religious denomination, attendance at religious services, cohabitation, children, and urbanisation, and were derived from the LISS Core Studies (Elshout 2021a, 2021b, 2021c). Online Appendix 3 contains detailed information on the sources for and operationalisation of our variables.

Confirmatory analyses

As our control variables were derived from separate datasets, respondents could have missing values for some of them. Consequently, to maximise the sample size, multiple imputation by chained equations (MICE) was employed in most of the analyses (see below). Online Appendix 4 contains detailed information on how we employed multiple imputation.

It should be noted that the inclusion of a pre-treatment measure of trust in politicians meant that most of our analyses (see below) employed a quasi pre-test-post-test design. This enabled us to obtain a more precise estimate of the effects of our treatment elements on trust in the politician in the video (Clifford et al. 2021; Lin 2013). These values were taken from the LISS Politics and Values data (Elshout 2021a) and range from (0) no trust at all in politicians to (10) complete trust.

We estimated intention-to-treat (ITT) effects in all our analyses, which could be identified because of randomisation (Alferes 2012; Aronow et al. 2019; Montgomery et al. 2018). Our factorial design (detailed above) meant that we were able to perform multivariate ordinary least squares (OLS) regression analyses in which the four treatment elements were included as binary variables (cf. Auspurg and Hinz 2015; Lin 2013). The confirmatory analyses for trust in the politician in the video involved fitting multivariate ordinary least squares (OLS) regression models for our two measures of trust in the politician in the video, using multiple imputation. Three sensitivity analyses were performed, which entailed either: (1) removing the respondents who failed to answer our manipulation-check question correctly; (2) not imputing the pre-treatment variable; or (3) employing listwise deletion. Online Appendix 4 contains detailed information on (the rationale behind) these analyses, which did not alter the conclusions drawn from our confirmatory analyses.

Validation analysis

For the validation analysis that assessed whether framing our politician as having a culturally elitist appearance and lifestyle and work experience inside politics indeed inspires perceptions of cultural distance among less-educated citizens, we used a three-item scale (Cronbach’s alpha = 0.91) of perceived cultural distance to the politician in the video. It builds on a measure used in prior research (Noordzij et al. 2021c). The three items asked the respondents to evaluate whether the politician in the video-vignette they viewed: (1) is far removed from people like them; (2) does not know what people like them experience; and (3) lives a life completely different from people like them. The answer categories range from (1) completely disagree to (7) completely agree. A higher value on the three-item scale indicated that the respondent perceived the politician in the video to be more culturally distant. Table A7 in the Online Appendices provides details on the items, coding, and factor analysis for this measure.

We performed a validation analysis by fitting a multivariate ordinary least squares (OLS) regression model for perceived cultural distance to the politician in the video, using multiple imputation. We conducted sensitivity analyses that involved: (1) removing respondents who failed to answer our manipulation-check question correctly; or (2) employing listwise deletion. Online Appendix 4 contains detailed information on (the rationale behind) these analyses, which did not alter the conclusions drawn from the main validation analysis.

Results

Table 1 was used to assess our confirmatory analyses. It sets out the results of our ordinary least squares regression analyses for trust in the politician in the video, which we ran using multiple imputation. Each model in Table 1 includes education, the four treatment elements, all the control variables, and the pre-treatment measure of trust in politicians. The panel on the left shows the results for the often-used single-item measure of trust and the one on the right for the newly-developed seven-item measure. Models 1 include the interaction between education and the politician’s voicing of cultural progressiveness, Models 2 encompass education’s interaction with the politician’s use of non-populist rhetoric, and Models 3 cover the interaction of education with the politician’s culturally elitist appearance and lifestyle, as well as his work experience inside politics.

Table 1. Ordinary least squares regression analyses for trust in the politician in the video.

	Single-item measure			Seven-item measure
	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3
Education	−0.19***	−0.12***	−0.07*	−0.07***	−0.03*	−0.01
	(0.03)	(0.03)	(0.03)	(0.01)	(0.01)	(0.02)
Cultural progressiveness	−2.55***	0.99***	0.99***	−1.11***	0.43***	0.43***
	(0.54)	(0.09)	(0.09)	(0.25)	(0.04)	(0.04)
Non-populist rhetoric	0.48***	−0.68	0.47***	0.17***	−0.07	0.16***
	(0.09)	(0.55)	(0.09)	(0.04)	(0.25)	(0.04)
Culturally elitist appearance and lifestyle	−0.43***	−0.42***	0.17	−0.31***	−0.31***	0.20
	(0.09)	(0.09)	(0.55)	(0.04)	(0.04)	(0.25)
Work experience inside politics	0.05	0.05	−0.28	0.05	0.05	−0.16
	(0.09)	(0.09)	(0.55)	(0.04)	(0.04)	(0.25)
Education × Cultural progressiveness	0.24***			0.10***
	(0.04)			(0.02)
Education × Non-populist rhetoric		0.08*			0.02
		(0.04)			(0.02)
Education × Culturally elitist appearance and lifestyle			−0.04			−0.03*
			(0.04)			(0.02)
Education × Work experience inside politics			0.02			0.01
			(0.04)			(0.02)
All the control variables included	Yes	Yes	Yes	Yes	Yes	Yes
Constant	7.57***	6.48***	5.77***	5.32***	4.71***	4.44***
	(0.92)	(0.94)	(0.96)	(0.43)	(0.44)	(0.44)
n	2372	2372	2372	2372	2372	2372

Note: *p<0.05, **p<0.01, ***p<0.001. Unstandardised coefficients; standard errors in parentheses; using multiple imputation; Table A13 in the Online Appendices contains the complete models.

We first assessed the relevance of substantive representation to understanding the educational gradient in trust in the politician in the video, using Models 1 in Table 1. The positive interaction of education with the politician’s voicing of cultural progressiveness (instead of cultural conservatism) for the single-item (0.24; p < 0.001) and seven-item measures (0.10; p < 0.001) indicates that the educational gradient in trust in the politician in the video is larger (smaller) if the politician voices culturally progressive (conservative) stances (corroborating Hypothesis 1).

These findings were further explored as set out in the left-hand panel of Figure 1 and in Figure 2, which report the average predicted levels of trust in the politician in the video for the single-item and seven-item measures, respectively. The figures show the predicted levels of trust in the fictitious politician with either culturally progressive or conservative stances and if non-populist or populist rhetoric is used. The predicted levels of trust for the most- and least-educated respondents are shown across these conditions.⁷ The least-educated respondents were found to have an average predicted level of trust in the politician of 5.08 [single-item measure, ranging from (0) to (10)] and 3.82 [seven-item measure, ranging from (1) to (7)] when the politician voices conservative stances, compared to 4.53 (single-item) and 3.56 (seven-item) when he is culturally progressive. There are even more substantial differences across the conditions for the most-educated respondents: their average predicted level of trust in the politician is 3.26 (single-item) and 3.22 (seven-item) when he voices conservative stances, compared to 5.02 (single-item) and 3.97 (seven-item) when he is culturally progressive.

Figure 1. Average predicted levels of trust in the politician in the video [single-item measure, ranging from (0) to (10)] with culturally progressive or conservative stances (left-hand pane) and using non-populist or populist rhetoric (right-hand pane) for the least and most educated, employing listwise deletion and including 95% confidence intervals.

Figure 2. Average predicted levels of trust in the politician in the video [seven-item measure, ranging from (1) to (7)] with culturally progressive or conservative stances for the least and most educated, employing listwise deletion and including 95% confidence intervals.

Second, we used Models 2 to assess the relevance of populist rhetoric in the educational gradient in trust in the politician. Once again, there is a positive interaction between education and the politician’s use of non-populist (instead of populist) rhetoric, but only for the single-item measure of trust (0.08; p = 0.033), not the new seven-item measure (0.02; p = 0.349). These results show (at least when the trust was determined using the common single-item measure) that the less-educated respondents have more trust in the politician in the video than their more-educated counterparts when populist rhetoric is used (corroborating Hypothesis 2). This is reflected in the right-hand panel in Figure 1, which shows the most- and least-educated respondents’ average predicted levels of trust in the politician in the video for the single-item measure when a populist or non-populist rhetoric is used. Here, the least educated have an average predicted level of trust in the politician of 4.94 when the politician’s rhetoric is populist compared to 4.75 when he uses the opposing rhetoric; meanwhile, the average predicted level of trust among the most educated is 3.75 and 4.50, respectively, for the populist and non-populist rhetoric.

Finally, Models 3 in Table 1 encompass the interaction of education with the two treatment elements that tap into the politician’s distance, or proximity, to the life-worlds of less- or more-educated citizens. The analyses conducted for the single-item and seven-item measures did not find that the educational gradient in trust in the politician is positively shaped by his culturally elitist (instead of common) appearance and lifestyle (−0.04; p = 0.277, and −0.03; p = 0.041, respectively). This means that Hypothesis 3A must be rejected. Similarly, work experience inside politics does not positively shape the effect of education on the single- or seven-item measures of trust (0.02; p = 0.548, and 0.01; p = 0.412, respectively), meaning that Hypothesis 3B is not supported.

Our validation hypotheses help in the interpretation of these null findings concerning the role of perceived cultural distance. Table 2 details the results of the ordinary least squares regression analysis for perceived cultural distance to the politician in the video, obtained using multiple imputation. Contrary to our expectations, we did not identify any significant interaction between education and a culturally elitist appearance and lifestyle (−0.03; p = 0.284), meaning that Validation hypothesis I is not supported, or between education and work experience inside politics (−0.01; p = 0.600), meaning that Validation hypothesis II must also be rejected. This indicates that differences in respondents’ perceptions of the politician in the video are not shaped by the interplay between their level of education and the politician’s characteristics.

Table 2. Ordinary least squares regression analyses for perceived cultural distance to the politician in the video.

	Model 1
Education	0.05*
	(0.02)
Cultural progressiveness	−0.41***
	(0.06)
Non-populist rhetoric	−0.15**
	(0.06)
Culturally elitist appearance and lifestyle	0.95**
	(0.35)
Work experience inside politics	0.21
	(0.35)
Education × Culturally elitist appearance and lifestyle	−0.03
	(0.02)
Education × Work experience inside politics	−0.01
	(0.02)
All the control variables included	Yes
Constant	3.32***
	(0.63)
n	2372

Note: *p < 0.05, **p < 0.01, ***p < 0.001. Unstandardised coefficients; standard errors in parentheses; using multiple imputation; Table A14 in the Online Appendices contains the complete model.

We further explore these findings in Figure 3. This sets out the average predicted levels of perceived cultural distance to the politician in the video for the least- and most-educated respondents when the politician has a common or culturally elitist appearance and lifestyle (left-hand pane) and work experience outside or inside politics (right-hand pane). Both the most-educated (0.491; p < 0.001) and least-educated (0.781; p < 0.001) respondents perceived the politician to be more culturally distant when he is portrayed as culturally elitist, although in case of the least-educated ones substantially stronger so.⁸ Turning to the politician’s work experience, we see that the least-educated respondents perceive more cultural distance if the actor describes having work experience inside politics compared to when he has work experience outside politics. This difference, however, is not statistically significant (0.20; p = 0.267). The distance perceived by the most-educated respondents neither depends on the politician’s work experience (−0.05; p = 0.621). We reflect on the implications of these findings in the discussion below.

Figure 3. Average predicted levels of perceived cultural distance to the politician in the video [ranging from (1) to (7)] when the fictitious politician has a common or culturally elitist appearance and lifestyle (left-hand pane) and work experience outside or inside politics (right-hand pane) for the least and most educated, employing listwise deletion and including 95% confidence intervals.

Conclusion and discussion

How can we explain the educational gradient in trust in politicians as found in many liberal democracies? We tested the causal effects of three characteristics of politicians that a wide body of literature theorises to underlie the educational gradient in trust in politicians: their substantive stances, their use of (non-)populist rhetoric, and their cultural distance, or proximity, to the life-worlds of less- or more-educated citizens.

Our pre-registered survey experiment was set out among a representative sample (n = 3175) of the Dutch population. Respondents were assigned randomly to one of 16 short video-vignettes in which a fictitious politician introduces himself to respondents in a 2 [cultural progressiveness (conservatism)] by 2 [non-populist (populist) rhetoric] by 2 [a culturally elitist (common) appearance and lifestyle] by 2 [work experience inside (outside) politics] factorial design. Our analyses demonstrate that the cultural stances of the politician are strong drivers of the educational gradient in trust in the politician. Moreover, in case of the most widely used single-item measure of trust, we also found that this educational gradient is shaped by the politician’s use of (non-)populist rhetoric. These outcomes suggest that the extent to which people believe that their stances are represented politically, is a crucial driver of the educational gradient in trust in politicians, as is whether politicians employ populist or non-populist rhetoric.

To understand the educational gradient in trust in politicians, it is relevant to note that our analyses indicate that the more-educated respondents in particular are responsive to the substantive stances and (non-)populist rhetoric of the politician in the video. The levels of trust of the less-educated respondents were affected less by these experimental treatments—but still substantially so. This resonates with recent large-scale, cross-national research showing that the cultural liberalism of governing cabinets has a more profound effect on the political trust of the so-called ‘cosmopolitan winners of globalisation’ than that of their ‘nationalist losers of globalisation’ counterparts (Noordzij et al. 2021b). Taken together, these findings indicate that, despite the overwhelming attention paid in the literature to the political discontent of underprivileged social groups, a proper understanding of the dynamics of political discontents in general, and the educational gradient therein specifically, would benefit from a more in-depth examination of their more privileged counterparts. Four ideas retrieved from previous research are relevant for future studies to explore in this regard.

First, more-educated citizens are generally argued to have higher levels of political knowledge, attach stronger to democratic norms, and be more often socialised in milieus in which politics was frequently discussed (cf. Hakhverdian and Mayne 2012; Dinas 2014). This could make them more sensitive to politicians’ signals of distance or proximity to their own political preferences. Second, mirroring studies indicating that supporters of culturally progressive parties report greater dislike of supporters of culturally conservative parties than vice versa (Harteveld 2020), more-educated citizens might harbour stronger negative feelings towards politicians who do not substantively represent them than less-educated citizens do. To the best of our knowledge, the literature on affective polarisation did not test why political opponents inspire more dislike among politically progressives and more-educated citizens.

Two other ideas are potentially relevant to our finding that the more-educated respondents are more responsive to the substantive stances and (non-)populist rhetoric of the politician in the video. These are inspired by sociological literature on social mobility and status signalling. Confrontation with right-wing populist challengers could, first, incite ‘fear of falling’ out of political dominance among more-educated citizens in diploma democracies. This aligns with culturally liberal Dutch citizens who oppose further European integration because of the illiberal turn observed in countries that more recently became members of the European Union (van den Hoogen et al. 2022). This suggests that more-educated citizens resist politicians challenging the culturally progressive status quo they appreciate. Meanwhile, confrontation with politicians more culturally progressive and pluralist than they are themselves might upset less-educated citizens less because this represents ‘business as usual’ to them. Second, it has been well established that less-educated citizens feel less entitled to voice their political opinions than more-educated citizens do (Laurison 2015; Visser et al. 2023). Consequently, confrontation with a politician who voices opposite opinions than their own makes less-educated citizens more likely to respond with docility than more-educated ones, which could underlie the more marked responses among the latter found in our experiment.

Future research is needed to uncover the relevance of these ideas on why the more-educated respondents are more responsive to the substantive stances and (non-)populist rhetoric of the politician in the video than their less-educated counterparts are. It could also uncover whether, as implied by our findings, the rise of right-wing populist parties abates the educational gradient in political trust as found in many, especially West European, countries from the 1970s onwards.

In contrast to our findings on the relevance of cultural stances and populist rhetoric, our study produced null findings in relation to the role of perceived cultural distance. While a recent correlational study on the Dutch case demonstrated that perceived cultural distance contributes substantially to the educational divide in trust in politicians (Noordzij et al. 2021c), the video-vignettes we developed to test this experimentally did not produce similar findings. Our validation analysis suggests this is because those vignettes were not valid in the eyes of our more-educated respondents, as they do not perceive less distance to a culturally elitist politician than their less-educated counterparts do. Although we took great care in creating our video-vignettes, using a wide range of theoretical and anecdotal insights, we assume that this is because the former group perceives that the actor played the role of a culturally elitist politician less authentically—note that the actor used in our vignettes is usually casted to play a working-class persona which, as suggested by our findings, he portrayed more naturally than a culturally elitist one. Also note that our less-educated respondents perceived greater cultural distance to the culturally elitist politician than to the one with common cultural expressions, as expected.

Using the same actor to portray the politician in all the video-vignettes had two clear advantages: (i) respondents did not have any crystallised perceptions regarding that person, while they often have so with regard to a real-life politician; and (ii) using different actors for different treatments makes those treatments vary on more than one relevant aspect for determining trust in politicians, violating the experimental set-up. Hence, using a single actor still comes highly recommended (cf. Druckman et al. 2011; Mutz 2011), but this calls for actors that can play different roles authentically, in the eyes of all relevant societal segments. This is challenging in case it concerns socially shaped appearance, posture, and ways of speaking: because these are deeply ingrained through long-term socialisation, an actor may signal them more naturally if they match their own social background. Future experiments that aspire to include socially shaped physical signals as treatments are therefore advised to take this into consideration in the casting of an actor, and to assure the naturalness of that actor in conveying those signals, by piloting the video-vignettes among different relevant audiences. Limitations in time and resources make that, in practice, however, such additional safeguards are challenging to realise.

Moreover, some societal segments might be more inclined to process a vignette depicting a fictitious politician with a ‘suspension of disbelief’ than others. An in-depth assessment of how ‘seriously’ different groups within the proposed sample take the treatment is, hence, relevant for further securing the internal and external validity of survey experiments in general (Druckman 2011: 44), and of those using videos including fictitious elements in particular. Of course, it could also be explored whether exposing respondents to a longer video-vignette and whether repeating the exposure over an extended period or accompanied by written text (Wittenberg et al. 2021), would lead to greater acceptance of a fictitious persona or a greater sense that this persona is authentic among all groups of respondents. Last, and with regards to our treatment of cultural distance in particular, it would be valuable to examine whether the incorporation of signals of cultural distance different than those included in our video-vignettes, such as communication styles or other lifestyle indicators (Noordzij et al. 2021a), produces different results. Obviously, our video-vignettes were limited in the number of signals that could be included.

A final suggestion for future research concerns how cultural stances, populist rhetoric, and signals of cultural distance may be interrelated in explaining political trust. Our pre-registered study does not include an empirical assessment of such interrelations, which future research could, in detail, assess. For instance, it is commonly held that descriptive representation acts as a catalyst for substantive representation because representatives have a ‘natural’ understanding of the substantive stances of the social group to which they belong (see, e.g. Allen and Sarmiento-Mirwaldt 2015; Mansbridge 2015; Mora et al. 2021; Pitkin 1969). It would thus be worthwhile to explore whether feeling distant from the life-world of politicians also sparks perceptions of distance in substantive terms, or whether experiences of ideological incongruence have a stronger negative impact on trust when the politician is also perceived as distant in descriptive terms.

Moreover, the focus in populism studies is on how populist rhetoric involves ‘displaying closeness to the people simply by talking about the people’ (Jagers and Walgrave 2007: 323; italics original), or ‘the normative distinction between “the elite” and “the people”, not the empirical difference in behaviour or attitudes’ (Mudde 2004: 544). It, therefore, seems relevant to assess whether such populist messages have a greater impact on evaluations of politicians when they also signal a shared ‘common’ background.

Overall, our results provide unique causal evidence of the relevance of substantive representation and populist rhetoric in citizens’ trust in politicians. Generalising from our experimental setting, the findings strongly suggest that the educational gradient in trust that characterises many liberal democracies is driven significantly by the extent to which people experience the representation of their political stances and the presence of politicians who use populist or non-populist rhetoric.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This study was funded by Vidi grants awarded to the second and third author by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO) [grant numbers 016.Vidi.185.207 and 452-17-009].

Notes on contributors

Kjell Noordzij

Kjell Noordzij is an Assistant Professor of Social Inequality in the Department of Public Administration and Sociology at Erasmus University Rotterdam. He studies the social aspect of representative democracy, specifically how different groups of citizens perceive politics and their representation in it.

Willem de Koster

Willem de Koster is a Full Professor of General Sociology, in particular Cultural Sociology, in the Department of Public Administration and Sociology at Erasmus University Rotterdam. He studies how social groups give meaning to social and political issues, how this informs their actions, and how it shapes their responses to new information, policies, and the (urban) environment in which they live.

Jeroen van der Waal

Jeroen van der Waal is a Full Professor of Sociology of Stratification in the Department of Public Administration and Sociology at Erasmus University Rotterdam. His research aims to explain why social stratification is linked to health outcomes and political attitudes and behaviours.

Notes

1 The pre-registration of our survey can be found here: https://doi.org/10.17605/OSF.IO/67MG3.

2 Online Appendix 3 contains extended information on the operationalisation of our manipulation check.

3 Table A5 in the Online Appendices contains the script for all the versions of the video-vignettes.

4 Determined using information on all parliamentarians (see here: https://www.tweedekamer.nl/kamerleden_en_commissies/alle_kamerleden) during the design of this study (in 2020).

5 Online Appendix 2 reports the results and fit-indices of these analyses.

6 Note that the results of exploratory analyses that do not control for these background characteristics are highly similar to those of our confirmatory and validation analyses, and do not inspire different conclusions (see Tables A3 and A4 in the Online Appendices).

7 While our main confirmatory analyses using multiple imputation were employed to test our hypotheses, Figures 1–3 were produced using the results of our analyses following listwise deletion (n = 2079 instead of n = 2372). This is because multiple imputation does not enable average predicted levels of trust to be calculated.

8 To calculate these marginal differences, we used the mlincom command in Long and Freese’s (2014) Stata package SPost13.