Political Opinion Leaders in High-Choice Information Environments: Are They More Informed Than Others?

ABSTRACT

One implication of the transition to high-choice media environments is that what information people are exposed to depend less on journalistic curation and more on social, algorithmic, and personal curation. This has sparked a renewed interest in the concept of political opinion leaders, who are often assumed to be more interested in and informed about politics and society. Theoretically, political opinion leaders could hence help disseminate information to less interested and informed. At the same time, there are theoretical reasons for why political opinion leaders may be more prone to politically motivated reasoning, which may lead them to believe in and disseminate misinformation. Thus far, there is only limited research on whether political opinion leaders are more informed than others that also takes into consideration that some facts are contested, whereas other facts are uncontested. Hence, this paper seeks to investigate the relationship between political opinion leadership and knowledge, distinguishing between (a) uncontested and (b) contested facts. Among other things, findings show that those who score high in political opinion leadership traits in general are not more knowledgeable about contested and uncontested facts.

The increasing diffusion of digital and social media has created a greater abundance of information than ever. While only a small portion of this information pertains to politics and society (Hindman, 2009), from a democratic perspective, the development is Janus-faced (Van Aelst et al., 2017). On the one hand, it has made it easier for citizens to find high-quality news and information about virtually all issues. On the other hand, it has also lowered the entry barriers for those who want to disseminate false and misleading information (Benkler et al., 2018; Kavanagh & Rich, 2018; Strömbäck et al., 2022).

One important aspect of this process is that the role of news media as the traditional gatekeepers to the public sphere has weakened (Bro & Wallberg, 2015; Shoemaker & Vos, 2009; Vos & Heinderyckx, 2015). More specifically, journalistic curation has become supplemented with and at times replaced by social curation, algorithmic curation, and personal curation (Thorson & Wells, 2016). The implication is that what information people are exposed to depend more than ever on people’s social networks, algorithms, and own motivations (Messing & Westwood, 2014; Prior, 2007; Soffer, 2021; Strömbäck et al., 2013; Thorson & Wells, 2016).

This development has sparked renewed interest in the position of opinion leaders, first proposed by Lazarsfeld et al. (1948). Originally, the concept referred to those “people who are most concerned about [an] issue as well as most articulate about it” (p. 49), and a key function is that they serve as intermediaries in the flow of information from different media to other, less interested and engaged, people (Lazarsfeld et al., 1948; Ognyanova, 2017). Since then, the concept has evolved, and different operationalizations have been proposed (e.g., Childers, 1986; Gnambs & Batinic, 2011). In terms of political opinion leaders, the concept usually refers to people who self-report that they are often talking to others about politics, often share information about political matters with others, and often try to persuade others about their political views (e.g., Jungnickel, 2018; Katz, 1957; Mangold & Bachl, 2018; Ognyanova, 2017). Important to note, though, is that contemporary research suggests that political opinion leadership should be conceptualized as a continuous trait rather than as discrete types, where some are, and some are not opinion leaders, and that opinion leadership is usually domain specific (Gnambs, 2019). When henceforth referring to opinion leaders, we thus refer to those who score high in political opinion leadership traits.

It is commonly assumed in research on opinion leaders that those leaders are more informed than those they influence (Katz, 1957; Lazarsfeld et al., 1948; Trepte & Scherer, 2010). To date, however, there has been surprisingly limited research on whether political opinion leaders actually are more informed than others. As noted by Trepte and Scherer (2010, p. 122), most studies on the role of opinion leaders “concentrate on the flow of information but neglect the question of what exactly is flowing. In particular, the question of knowledge has not been addressed.”

Beyond the need to investigate whether political opinion leaders are more informed than others, there are theoretical reasons to distinguish between knowledge that relates to uncontested issue domains and knowledge that relates to issues that have been subject to ongoing public controversies (Damstra et al., 2023). So-called uncontested facts pertain to, for example, the number of seats a party has in parliament, whereas contested facts are the object of political battles, in spite of expert and scientific consensus. An example is whether there is a process of anthropogenic climate change. In the former case, we can expect that opinion leaders will either be informed and know the correct answer or uninformed and not know the correct answer. In the latter case, however, political opinion leaders may also be misinformed, that is, firmly believe in the wrong answer (e.g., Kuklinski et al., 2000) and be knowledge resistant, rejecting available evidence (Glüer & Wikforss, 2022; Klintman, 2019). The key reason is that psychological mechanisms such as politically motivated reasoning and confirmation bias (Kahan, 2016; Kunda, 1990; Lodge & Taber, 2013; Strömbäck et al., 2022) may lead them to select and accept information that is consistent with their priors rather than the information that is factually most correct. In an age in which social curation has become an important gatekeeping mechanism, misinformed political opinion leaders may hence potentially contribute to the dissemination of misinformation instead of the spread of knowledge. To fully understand and assess the role of political opinion leaders in contemporary political information environments (Strömbäck et al., 2022; Van Aelst et al., 2017) and for public knowledge, it is thus key to investigate the relationship between political opinion leadership and knowledge about both uncontested and contested facts.

Against this background, the purpose of this paper is to investigate the relationship between political opinion leadership and knowledge, distinguishing between (a) uncontested and (b) contested facts. In terms of contested facts, we will investigate this in relation to five issue domains that are subject to political contestation across the world: global warming, immigration, vaccines, GMO, and crime.

Implications of the transition to high-choice media environments

While the transformation into high-choice media environments (Prior, 2007; Strömbäck et al., 2022; Van Aelst et al., 2017) has been consequential in many respects, in the context of this study, three aspects are of particular importance. First, the total media supply has increased dramatically. Not only can people get information from traditional news media on different platforms, but they can also get information directly from an ever-increasing supply of political alternative media, interest groups, businesses, and political organizations. In addition, people get informed through social media, video-sharing sites, message apps, and discussion boards. In the latter cases, the information may be posted by the original sources, but also by others, whether these are peers, people known only through social media, or anonymous sources. Although mainstream news media remain highly important sources of information, research shows that audiences are getting more fragmented and that people increasingly get news and other information from social media (Newman et al., 2021; Walker & Matsa, 2021).

Second, the increasing media supply implies that people have to be increasingly selective in terms of what media content, media channels, and media platforms they use (Prior, 2007; Van Aelst et al., 2017). Simply put, the greater the media supply, the more selective people have to be, and the more selective people have to be, the more important their personal preferences become (Luskin, 1990; Prior, 2007). This implies that some people will attend to more news and other types of information, while others might instead avoid news (Damstra et al., 2023; Prior, 2007; Skovsgaard & Andersen, 2020; Strömbäck et al., 2013). While this does not preclude that people are incidentally exposed to news (Bode, 2016; Kümpel, 2020), the routes through which people get informed have multiplied.

Third, as pointed out by Thorson and Wells (2016), this development has made curation and curated flows more important for understanding what people are exposed to and how they orient themselves in current media environments. This curation pertains to “the production, selection, filtering, annotation, or framing of content” (Thorson & Wells, 2016, p. 310). In contrast to the mass media era, such curation is no longer only undertaken by journalists and news media but also by “actors such as friends and social contacts, computer algorithms, and individual media users themselves” (Thorson & Wells, 2016, p. 310). Based on that, Thorson and Wells (2016) distinguish between journalistic curation by journalistic media, strategic curation by strategic communicators, personal curation by users themselves, algorithmic curation by algorithms, and social curation by friends, family, and others to which users are connected. Over time, it is evident that journalistic curation has lost ground, while other types of curations have become more important, that gatekeeping processes have changed in form and structure, and that the number of involved actors has multiplied (Bro & Wallberg, 2015; Hameleers, 2021; Vos & Heinderyckx, 2015).

Opinion leadership and political knowledge

Taken together, the changes described above have made political opinion leaders more important as gatekeepers and intermediaries in the flow of political information. As such, they may contribute to the dissemination of political information, helping to inform the public, but that presupposes that they are better informed than others. In the literature, political opinion leaders are often assumed to be more avid users of high-quality media, leading them to be well informed, but in contemporary high-choice media environments, there is quite limited research on whether this is actually (still) the case. While there are several recent studies that investigate the characteristics of opinion leaders in terms of their media use (Mangold & Bachl, 2018; Schäfer & Taddicken, 2015), information-seeking behavior (Karlsen, 2015), motivations (Winter & Neubaum, 2016), activities on social media (Karlsen, 2015; Weeks et al., 2017), and engagement (Nisbet, 2006), there are few studies investigating the association between political opinion leadership and knowledge.

Two exceptions are worth noting. In the first, Schenk and Rössler (1997) investigated the linkage between influentials and knowledge. Following Noelle-Neumann (1985) and Weimann (1991, 1994), they defined influentials in terms of personality strength. Among other things, they found that “those with great personality strength are generally better informed about current daily political affairs” (Schenk & Rössler, 1997, p. 15). In a more recent study, Trepte and Scherer (2010) investigated the linkage between four opinion leadership scales and knowledge, tapping into uncontested facts such as “When did Switzerland join the UN?” The results showed that most scales correlated with knowledge. In a follow-up, they investigated the linkage between the Childers scale (1986) and 10 knowledge items. This scale consists of six items on the role of topic-related communication, such as questions about information on education or healthcare. Based on a cluster analysis, they differentiated between informed opinion leaders and uninformed opinion leaders (alongside silent experts and others). The results thus suggest that “not all opinion leaders are well-informed” (Trepte & Scherer, 2010, p. 135).

Opinion leaders as disseminators of information vs misinformation

Since Brexit and the election of Donald Trump as President of the United States in 2016, research on misinformation and misperceptions has virtually exploded (Benkler et al., 2018; Flynn et al., 2017; Hameleers, 2021; Kavanagh & Rich, 2018). The backdrop is that these events were characterized by widespread dissemination of misinformation and evidence suggesting that misinformation leads to misperceptions (Hochschild & Einstein, 2015; Flynn, 2016; Flynn et al., 2017; IPSOS, 2018; Sides & Citrin, 2007). However, the role of opinion leadership in this context has largely been neglected.

To assess the role of opinion leadership in terms of counteracting or spreading misinformation, it is key to learn more about whether political opinion leaders are more informed than others. Theoretically, there are reasons to expect that the answer hinges on whether or not information relates to facts and issue domains that are subject to political controversies. Most importantly, when facts and issues are contested and subject to political controversies, cognitive biases such as confirmation bias, politically motivated reasoning, and identity-protecting mechanisms are likely to be triggered. As shown by Kunda (1990), people are often not motivated just to get the facts right. Instead, they are often (unconsciously) more motivated to arrive at particular conclusions that confirm their already held beliefs, attitudes, and values, and to protect their group identity (Flynn et al., 2017; Kahan, 2016; Kahan et al., 2011; Kraft et al., 2015; Lodge & Taber, 2013). This influences what information they expose themselves to (Garrett, 2009; Knobloch-Westerwick & Meng, 2009; Stroud, 2011) and how information is processed and interpreted (Gaines et al., 2007; Glüer & Wikforss, 2022; Kunda, 1990; Nickerson, 1998). Once a set of beliefs have become “a badge of membership within identity-defining affinity groups” (Kahan, 2016, p. 2), people might align their beliefs to what identity-relevant groups advocate, even when that runs counter to scientific evidence.

Research also shows that politically motivated reasoning is particularly likely among groups that are politically engaged (Flynn, 2016; Flynn et al., 2017; Lodge & Taber, 2013; Miller et al., 2016). Importantly, these are the same groups where political opinion leaders are likely to be found. The politically engaged have a stronger political interest and are often more opinionated (Flynn et al., 2017; Zaller, 1992), and they also have a strong motivation to protect their (more salient) political identities (Lodge & Taber, 2013). It furthermore requires a certain level of political engagement and sophistication to detect whether some facts are congenial with other attitudes and values or not. When factual information is congenial, people tend to accept and remember it better than when the information runs counter to their other attitudes and values (Flynn et al., 2017; Jerit & Barabas, 2012; Lodge & Taber, 2013). In terms of uncontested facts, such as the name of a party leader, such biases are less likely to occur. Then, it is rather a matter of knowing or not knowing, i.e., how much and what information people have stored and can recall.

Although prior research on the relationship between political opinion leadership and knowledge is limited, evidence suggests a positive relationship (Katz & Lazarsfeld, 1955; Schenk & Rössler, 1997; Trepte & Scherer, 2010). With respect to uncontested facts, there are also theoretical reasons to expect a positive relationship, as political opinion leaders are more likely to be interested, engaged, and sophisticated media consumers (Karlsen, 2015; Katz, 1957; Mangold & Bachl, 2018; Trepte & Scherer, 2010), which are all factors that tend to be associated with higher levels of knowledge (Aalberg & Curran, 2012; Delli Carpini & Keeter, 1996; Luskin, 1990; Shehata & Strömbäck, 2021). Based on this, we hypothesize that:

H1:

There will be a positive relationship between having political opinion leadership traits and correct knowledge about uncontested facts.

Concerning facts relating to contested issue domains, the situation may, however, be different. On the one hand, facts are facts—regardless of people’s beliefs. For the same reasons as we hypothesize a positive relationship between political opinion leadership and knowledge of uncontested facts (H1), political opinion leaders might be more correctly informed about topics that have been subject to political controversies. They may, however, also be more engaged in politically motivated reasoning. Depending on whether their beliefs align with the best available evidence or not, this could either result in a negative or a positive relationship between political opinion leadership and correct beliefs related to contested facts. To address this, we ask the following research question:

RQ1:

What is the relationship between having political opinion leadership traits and holding correct beliefs about politically contested facts?

Case selection, methodology and data

When investigating our hypothesis and research questions, we focus on the case of Sweden. Sweden is a parliamentary democracy with proportional elections and eight parties in parliament. In terms of its media system, Sweden is considered a “media welfare state” (Syvertsen et al., 2014) and belongs to the democratic corporatist model of media and politics (Hallin & Mancini, 2004). As such, it is characterized by strong public broadcasting, rather widespread newspaper readership, comparatively high media trust, and prevalent digital and social media use (Nord & Grusell, 2021; Newman et al., 2022). To our knowledge, there are no previous studies on how common opinion leadership is within in Sweden or in comparison to other countries.

With respect to the empirical data, we used data from the second wave of a web-based panel survey with Swedish citizens conducted with the primary purpose to assess why people resist knowledge on questions that enjoy academic or expert consensus. The data was collected by the SOM Institute at the University of Gothenburg, Sweden, between February 25 and March 30, 2021. For the first wave, a probability-recruited sample with the net sample size of 5,223 Swedish residents aged 18 years or older and pre-stratified by gender, age, and education was invited to participate. Of the initially invited, 3,433 completed the survey in Wave 1 and 2,337 completed the Wave 2 survey with an attrition rate of 31.9%. All individuals that were invited to participate in Wave 1 (excluding those that had chosen to opt-out of the panel) were re-invited to participate in Wave 2, amounting to 4,607 individuals. Of these, 2,535 participated in the second panel wave, giving a response rate of 55% in this panel. Individuals that had missing values in any of the variables in our statistical models were excluded from the analyses, leading to a total sample of 1,759 individuals.¹ Breakdowns of the sample on age, gender, and education revealed a higher share of younger and highly educated individuals in the final sample compared with the full sample (before excluding respondents), but the differences are relatively small, as shown in Table A in the third section of the online supplementary materials. Education, age, and sex are controlled for in the statistical analyses. The data collection has ethics approval from the Swedish Ethical Review Authority (1016–18) in December 2018.

Measures

The key variable political opinion leadership is constructed based on three items building on previous theory and operationalizations (e.g., Childers, 1986; Mangold & Bachl, 2018; Schäfer & Taddicken, 2015) and following the most common approach of investigating opinion leadership through self-assessment (Jungnickel, 2018). This approach acknowledges that political opinion leadership is a continuous trait rather than a matter of type where some are and some are not opinion leaders (Gnambs, 2019; Katz, 1957), and that at the heart of opinion leadership is talking to others, sharing information, and trying to convince others (e.g., Jungnickel, 2018; Katz, 1957; Mangold & Bachl, 2018; Trepte & Scherer, 2010). More specifically, respondents were asked how often they “… discuss politics with others,” “… share information about political issues with others,” and “… try to convince others of their political views.” Response options ranged from 1 = never to 7 = very often. The three items were loaded on one factor (in Principal Components Analysis with Varimax rotation, explaining 72.2% of the variance) and were added to form an index ranging from 3 to 21 (α = .81). The index was normalized to range from 0 = low in opinion leadership to 1 = high in opinion leadership (M = .37, SD = .22). To validate that the political opinion leadership scale is distinct from related concepts such as political interest and news media exposure, analyses reported in the first section of the online supplementary materials demonstrate discriminant validity vis-à-vis political interest and news media exposure, and as evidence for construct validity, positive and significant and rather robust correlations with political interest. In other words, political opinion leadership is both theoretically and empirically distinct from political interest and news media exposure.

To measure knowledge related to politically uncontested facts, we created an index based on questions tapping into surveillance knowledge and demographic trends/statistics. Surveillance knowledge consisted of five multiple-choice questions about people or events covered in the Swedish news media in the two months preceding the second panel wave (e.g., the name of the newly elected party leader of the Green party). Knowledge of demographic trends and statistics comprised six multiple-choice questions about the current demographic composition in Sweden on variables such as the share of unemployed and the share of the population older than 65 years. For each question, there were four substantial response options and a “don’t know” option to reduce guessing on closed-ended knowledge questions (Luskin & Bullock, 2011). By providing respondents with the opportunity to express their lack of knowledge, DK options can thus help limit the number of lucky guessers that are wrongly coded as correct, hence reducing the risk of overestimating the share of knowledgeable (Lindgren et al., 2022). To minimize the possibility for respondents to search for the correct answer before responding, they were automatically forwarded to the next question after 20 seconds (which they were instructed about in the introduction of the survey). In the analyses, respondents that chose the right answer were coded as correct, while the ones that responded incorrectly, chose the don’t know option, or withheld answering were coded as incorrect. Table A in the fourth section of the online supplemental materials provides an item response model for the 10 items (when coded 1 = correct and 0 = otherwise). Item response theory models knowledge constructs based on test items while accounting for item difficulty (van der Linden, 2010). Given that the Item-Response-Model-generated uncontested knowledge scores were highly correlated with a simple additive index (r = .89), the simple index approach was utilized in the analyses. In other words, the 10 knowledge questions were combined in an additive index, normalized to range from 0 = low knowledge to 1 = high knowledge (α = .56, M = .52, SD = .20).

The measure of knowledge pertaining to politically contested facts includes questions on five issue domains that vary in their public salience and degree of political contestation: climate change, vaccination, crime, immigration, and GMO. For each domain, respondents were asked to indicate whether four factual statements are correct, with response options of 1 (very certain it’s false), 2 (rather certain it’s false), 3 (uncertain whether it’s true or false), 4 (rather certain it’s true), and 5 (very certain it’s true). In line with previous researchers (e.g., Damstra et al., 2023; Flynn et al., 2017), we define knowledge or correct perceptions about contested facts as perceptions that align with the best available evidence or expert consensus. We therefore only included statements that are supported by empirical evidence, and/or on which a high degree of expert consensus existed at the time when the survey was conducted. Each topic includes both statements that are correct based on the best scientific evidence or official statistics and statements that are incorrect according to the best available evidence. Furthermore, to avoid systematic ideological biases in the statements (i.e., that items would systematically support either one or another political position on the issues), we strived to include variation both in terms of what general perception was supported (e.g., whether society is becoming more crime-ridden or not, and whether vaccines are generally harmful or not), and in terms of what answer is correct. For example, on knowledge about crime, some questions were formulated to incorrectly support a perception that society is becoming more crime-ridden (e.g., the number of homicides has increased in the past years), whereas other questions were formulated to correctly support the same view (e.g., the number of rapes reported to the police has increased in the past years). Items were recoded so that the response categories of every item run from 1 = confidently incorrect to 5 = confidently correct. For each respondent, the scores of the 20 items were summed in an index, normalized to range from 0 = confident, low accuracy to 1 = confident, high accuracy (α = .78, M = .61, SD = .15).² To enable tests of the robustness of the results across issue domains, we also created five issue-specific indices for each issue domain. A list of all items included in the uncontested and contested knowledge indices are presented in Tables A and B in the second section of the online supplementary materials.

While we strived for including variation both in terms of what general perception is supported (e.g., whether society is becoming more crime-ridden or not) and in terms of what answer is correct to avoid systematic ideological biases, we cannot ensure that the items are balanced on ideology. Hence, we also include ideological left–right orientation as a covariate in the analyses. Here, we rely on the standard question “In politics, people sometimes talk about left and right. Where would you place yourself on this scale?” The response options ranged from 0 = far to the left to 10 = far to the right (M = 5.02, SD = 2.42) and normalized to range from 0 = far left to 1 = far right. Because research on motivated political reasoning indicates that individuals with strong political positions are less easily affected by any type of argument or information that runs against their already existing views (e.g., Kunda, 1990), we also included a measure of ideological left–right extremity as a covariate in all models including ideological left–right placement. Ideological extremity was calculated by multiplying the left–right measure with itself (left–right^2) (M = 3.10, SD = 2.47). By adding the squared version of left–right ideology, we are able to account for a possible non-linear relationship between left–right ideology and knowledge, where knowledge is associated with ideology mainly among those who place themselves toward the left and right extremes, respectively.

In addition to ideological predispositions, we included several individual background factors in our analyses as covariates: education, political interest, news exposure, news media trust, age, and sex. Education was measured by asking respondents what level of education best described their educational level, with responses ranging from 1 = pre-upper secondary school shorter than 9 years to 9 = Ph.D. degree. The responses were coded in three categories of Low level (completed high school or less), Medium level (post-high school/non-university studies/University, less than 3 years), and High level (University studies, 3 years or longer/Ph.D. degree). Political interest was measured by the question: “Generally speaking, how interested are you in politics.” The response options (reversed) ranged from 1 = not at all interested to 4 = very interested and were normalized to range from 0 = low interest to 1 = high interest (M = .65, SD = .25). News media trust was measured by the question: “Generally speaking, to what extent do you trust information from the news media in Sweden?” The response options ranged from 1 = do not trust at all to 7 = trust completely and were normalized to range from 0 = low trust to 1 = high trust (M = .65, SD = .22). To control for news exposure, we created five variables measuring, respectively, mainstream news media use, left and right political alternative media use, exposure to news on social media, and intentional news avoidance (Damstra et al., 2023). To measure mainstream and political alternative media use, we created additive indices based on two multi-item questions asking how often in a typical week one uses, respectively, each of all the major national newspapers as well as television and radio news broadcasts in Sweden as a mainstream news index, and 10 of the larger Swedish partisan alternative media/media sites as one left-wing and one right-wing political alternative media index.³ All outlet-specific questions were measured on 8-point scales ranging from 0 = less than one day per week to 7 seven days per week; the indices were normalized to range from 0 = low use to 1 = high use (Mainstream: α = .55, M = .32, SD = .18; Left: α = .55, M = .02, SD = .07; Right: α = .80, M = .03, SD = .10). Exposure to news on social media was measured by a multi-item question asking how often, in a typical week, one comes across news or discussions about politics and society through, respectively, each of eight different social media including Facebook, Twitter, Instagram, YouTube, Telegram, WhatsApp, and Flashback. The response options ranged from 1 = never to 7 = several times a day. The items were added in an index, normalized to range from 0 = low exposure to 1 = high exposure (α = .61, M = .15, SD = .13). Intentional news avoidance was measured by the question: “In a typical week, how often do you actively try to avoid the news?” The response options ranged from 0 = never to 7 = several times a day and were normalized to range from 0 = low avoidance to 1 = high avoidance (M = .23, SD = .29). Age, finally, is coded into categories based on the respondents’ year of birth: 1 (under age 30), 2 (30–39), 3 (40–49), 4 (50–59), 5 (60–69), 6 (70 years or older).

Analytical strategy

The hypothesis and research questions are analyzed using standard OLS regression models. We start by regressing the knowledge indexes on opinion leadership to assess the bivariate associations between knowledge and opinion leadership traits. In a second step, we test whether the bivariate associations hold under control for demographic and socio-demographic variables, by including educational level, left–right predispositions, political interest, news exposure, news media trust, age, and gender as covariates.⁴

Results

Starting with H1, we predicted a positive relationship between opinion leadership and correct knowledge about uncontested facts. The results from two OLS regression models are presented in Table 1. Model 1 in Table 1 presents the bivariate association between opinion leadership and knowledge about uncontested facts, and Model 2 presents the relationship when covariates are included for control. In accordance with H1, the results suggest a bivariate association between opinion leadership and uncontested knowledge, with individuals with higher scores on opinion leadership being more knowledgeable (Model 1). When controlling for background factors, however, the association disappears (Model 2). Instead, Model 2 shows that several of the covariates predict knowledge. More specifically, those with higher political interest, education, and mainstream media use, lower exposure to news on social media, of older ages, and men tend to be more knowledgeable than individuals with low education, lower political interest, lower mainstream media use, with higher exposure to news on social media, that are under age 30, and women. There is, however, no relationship between knowledge and left–right political predispositions, news media trust, political alternative media use, and intentional news avoidance. Since the association between opinion leadership and knowledge did not hold under control for the covariates in Model 2, the null hypothesis of no relationship cannot be rejected. Hence, we conclude that scoring high in opinion leadership per se does not mean higher levels of knowledge with respect to uncontested facts. It seems possible that scoring high in opinion leadership covaries with (higher) knowledge on uncontested facts because opinion leaders are more likely to, for example, be politically interested and avid mainstream news users (strong predictors of uncontested knowledge in our analytical model).

Table 1. Knowledge about uncontested facts predicted by opinion leadership.

	Model 1	Model 2
Knowledge on Uncontested Facts	B (SE)	B (SE)
Predictors
Opinion leadership	.19*** (.02)	.01 (.02)
Left-right predispositions		−.00 (.07)
Left-right predispositions - squared		−.01 (.07)
Political interest		.22*** (.02)
News media trust		.01 (.02)
Mainstream news use		.20*** (.03)
Alternative left media use		−.06 (.06)
Alternative right media use		.05 (.04)
Social media news use		−.09* (.04)
Intentional news avoidance		−.02 (.01)
Man (ref = Woman)		.05*** (.01)
Educational level (ref = Low)
Medium		.03** (.01)
High		.08*** (.01)
Age group (ref = under 30)
30–39 years of age		.03 (.02)
40–49 years of age		.07*** (.02)
50–59 years of age		.09*** (.02)
60–69 years of age		.09*** (.02)
70 years or older		.12*** (.02)
Constant	.45*** (.01)	.18*** (.03)
Number of observations	1,759	1,759
Adjusted R-squared	.04	.26

Entries are unstandardized beta coefficients (B), with standard errors (SE) in parentheses. ***p < .001, **p < .01, *p < .05

Turning to the relationship between opinion leadership and knowledge about politically contested facts (RQ1), the results of our regression analyses are presented in Table 2. The results from the first model (Model 1) reveal a positive bivariate association between opinion leadership and our knowledge construct, indicating that the opinion leaders are more knowledgeable about contested facts. In contrast to the results for knowledge about uncontested facts, the results from the second model (Model 2) also show that, although the size of the coefficient is much reduced, the association between opinion leadership and knowledge on contested facts remains statistically significant when controlling for demographic background. Looking at the covariates in Model 2, some of the results resemble those for uncontested facts, in that the highly educated, men, and those with higher political interest were more knowledgeable than lowly educated, women, and those with lower interest in politics. In the case of knowledge about contested facts, there is also a positive relationship between news media trust and higher levels of knowledge, and a negative relationship between knowledge and intentional news avoidance, exposure to news on social media, exposure to right-wing political alternative media, and being older than 30 years of age. There are no associations between knowledge and left-wing political alternative media use, or between knowledge and left–right predispositions. The answer to RQ1 is thus that individuals scoring high on opinion leadership are somewhat more likely to hold correct knowledge about contested facts than are individuals scoring lower on this trait.

Table 2. Knowledge on contested issue domains predicted by opinion leadership.

	Model 1	Model 2
Knowledge on Contested Facts	B (SE)	B (SE)
Predictors
Opinion leadership	.13*** (.02)	.04* (.02)
Left-right predispositions		.00 (.05)
Left-right predispositions - squared		−.03 (.05)
Political interest		.12*** (.02)
News media trust		.15*** (.02)
Mainstream news use		−.02 (.02)
Alternative left media use		.07 (.05)
Alternative right media use		−.08* (.03)
Social media news use		−.05* (.03)
Intentional news avoidance		−.04*** (.01)
Man (ref = Woman)		.05*** (.01)
Educational level (ref = Low)
Medium		.02** (.01)
High		.09*** (.01)
Age group (ref = under 30)
30–39 years of age		−.01 (.01)
40–49 years of age		−.03* (.01)
50–59 years of age		−.04* (.01)
60–69 years of age		−.04** (.01)
70 years or older		−.06*** (.01)
Constant	.56*** (.01)	.42*** (.02)
Number of observations	1,759	1,759
Adjusted R-squared	.04	.25

Entries are unstandardized beta coefficients (B), with standard errors (SE) in parentheses. ***p < .001, **p < .01, *p < .0

However, in contrast to when it comes to knowledge on uncontested facts, it should be noted that those who are correct or not on contested factual claims may be sensitive to the nature of the claim as well as the issue in question (for example, the extent to which the issue and/or claim is politically polarized). As a further check, we therefore repeated the regression analyses using indices of knowledge for each issue-specific domain (climate, GMO, immigration, crime, and vaccines) as dependent variables. The results of these analyses reveal that, when controlling for covariates, the relationship between opinion leadership and knowledge was statistically significant only in one case, when the factual claims pertained to immigration. Furthermore, also in the case of immigration, the association between knowledge and opinion leadership was much reduced when controlling for background factors. These additional tests underscore the possibility that the relationship between having opinion leadership traits and being knowledgeable on contested facts are likely conditional on the issue in question and/or the nature of the claims and the general context. Hence, findings may not be a generalizable across issues and claims. The results from the robustness analyses are presented in Tables A–E in the sixth section of the online supplemental materials.

Discussion

While research on political opinion leadership dates back to the 1940s, fairly little scholarly attention has been paid to empirically investigate the assumption that those who score high in opinion leadership are more knowledgeable about politics and society than others. In addition, previous research that has addressed this question has not differentiated between knowledge about uncontested versus politically contested facts. This is problematic, as it is on politicized issues that political opinion leaders may be the most likely to have an invested interest and try to influence others. If they are misinformed, they may thereby spread misinformation rather than knowledge. To understand the role of political opinion leaders for public knowledge, it is hence crucial to investigate, and distinguish, the relationship between opinion leadership and knowledge about uncontested and contested facts. In this study, we have addressed this question using data from an empirical survey with Swedish citizens. Based on the results, our study offers three main takeaways.

The first takeaway concerns the relationship between opinion leadership and knowledge about uncontested facts. Here the bivariate but not the multivariate analyses suggest that opinion leaders are more informed than others. This opens up for two different interpretations. On the one hand, it might imply that the higher knowledge scores are not due to their opinion leadership per se, but rather be a result of these individuals scoring high on other factors that in turn influence knowledge. Considering that political opinion leadership was strongly correlated with political interest—which proved to be a strong predictor of knowledge on uncontested facts—one plausible interpretation is that political interest is the decisive factor. On the other hand, it is also possible that opinion leadership traits indirectly influence knowledge, for example, if opinion leadership traits lead people to become more interested, follow mainstream news more closely, and become more engaged in politics. Unfortunately, the causal mechanism explaining the pattern of results cannot be deduced from the current data and theoretically, the possibility of reciprocal relationships cannot be negated. We ran a series of mediation analyses, which suggested that mediation of this sort is possible and consistent with the data. However, given the cross-sectional nature of the data, we cannot formally test causality. To address these possibilities, future research could employ experimental studies and/or panel studies, in which the order of the variables can be controlled, to test potential mediated relationships and indirect effects of opinion leadership traits on knowledge. This would also be important to disentangle the relationship between political opinion leadership and related constructs such as political interest.

The second takeaway relates to the relationship between opinion leadership and being informed about politically contested facts. Here, we did not formulate a hypothesis but asked a research question. While we expected that opinion leaders would be more informed than others on uncontested facts, on contested issues, there are factors that speak for the possibility that opinion leaders may also be more likely to hold incorrect knowledge; perhaps they are better equipped to counterargue information and facts that challenge their political identities and beliefs. However, we found some indications that people scoring high on opinion leadership, in some cases, may be more knowledgeable about contested facts. When controlling for covariates, those scoring high on opinion leadership were somewhat more likely to be correctly informed about contested facts related to immigration—but not other issues—than those with lower scores on opinion leadership.

The third takeaway relates to the importance of distinguishing between contested and uncontested facts when seeking to understand the role of opinion leadership. The finding that, when controlling for background factors, opinion leadership was associated with knowledge of contested—but not with uncontested facts—implies that these are separate constructs. At the same time, our findings suggest that the issues at hand might be important. Given that opinion leadership was associated with higher knowledge about contested facts on immigration but not on other issues, further investigation of the difference between contested and uncontested knowledge, including the discriminant validity of these constructs, is warranted. In any case, another avenue for future research is to further explore when and why opinion leadership is associated with more knowledge about contested facts.

The last takeaway highlights some limitations of this study that are related to the importance of context. This study took place in one country, at a particular point in time between elections, and focused on certain topics. This implies that the findings may not necessarily hold in other countries, in politically more intense periods such as election campaigns, and with respect to other topics. It might, for example, be the case that the degree of polarization matters in the sense that the difference between political opinion leaders and others are greater in countries, contexts and with respect to topics that are more politically polarized. To explore this, future research should compare whether political opinion leaders are or are not politically more informed with respect to both uncontested and contested knowledge across countries, election campaigns, and non-election campaigns and with other topics where the degree of polarization or contestation differ.

At the same time, it should be noted that it can be difficult to empirically distinguish those who are uninformed (i.e., lack information and knowledge) from those who are misinformed (i.e., believe in the wrong answer), in surveys (Graham, 2022; Lindgren et al., 2022). Future studies are therefore encouraged to include different measures of knowledge, which could separate between the informed, uninformed, and the misinformed, to investigate the relationship between opinion leadership and political knowledge about uncontested and contested facts, respectively.

Summing up, while this study found only limited empirical support for the common assumption that those who score high in opinion leadership are more informed about politics and society than others, and therefore contribute with knowledge and information, there are strong reasons for further research on the relationship between opinion leadership and knowledge and to disentangle whether, and if so what (mis)information different types of political opinion leaders may help disseminate in what contexts. Social curation will probably only get more important, and hence also the role of political opinion leaders.

Disclosure statement

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

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15205436.2023.2281311.

Additional information

Funding

The work was supported by the Riksbankens Jubileumsfond [M18-0310:1].

Notes on contributors

Jesper Strömbäck

Jesper Strömbäck is professor in journalism and political communication at the Department of Journalism, Media and Communication, University of Gothenburg.

Elina Lindgren

Elina Lindgren is a post-doctoral scholar at the Department of Journalism, Media and Communication, University of Gothenburg.

Yariv Tsfati

Yariv Tsfati is a professor at the Department of Communication, University of Haifa.

Alyt Damstra

Alyt Damstra is a post-doctoral at the University of Amsterdam and member of the research staff at the Netherlands Scientific Council for Government Policy.

Rens Vliegenthart

Rens Vliegenthart is professor and chair holder of Strategic Communication at the Wageningen University & Research.

Hajo Boomgaarden

Hajo Boomgaarden is professor for Empirical Social Science Methods at the Department of Communication, University of Vienna.

Elena Broda

Elena Broda is a doctoral candidate at the Department of Journalism, Media and Communication, University of Gothenburg.

Noelle Lebernegg

Noelle Lebernegg is a doctoral candidate at the Department of Communication, University of Vienna.

Sebastian Galyga

Sebastian Galyga is a doctoral candidate at the Department of Communication, University of Vienna.

Notes

¹ The reduction in observations after excluding participants with missing values can be explained by a combination of the fact that some variables (media use and knowledge on uncontested facts) have some notable item non-response, and that we have a fairly high number of covariates in our models. The number of observations for each of all variables included in the analyses are presented in Table C in the second section of the online supplementary materials.

² Results from an item response model for the 20 items (when coded 1 = correct and 0 = otherwise) is reported in Table B in the fourth section of the online supplementary materials. The IR analysis indicated that some of the questions that were designed with the intention to challenge different ideological perspectives to avoid ideological biases in the contested knowledge construct did not function well in terms of difficulty and discrimination parameters. However, since we do not have theoretical reasons to assume that responses to some contested knowledge items should correlate strongly with responses to other contested knowledge items (in particular those that are ideologically charged), together with the fact that the Item-Response-Model-generated contested knowledge scores were highly correlated with a simple additive index (r = .84), we use the simpler additive index approach also for this knowledge construct.

³ The mainstream news index included: Aftonbladet, Expressen, Dagens Nyheter, Svenska Dagbladet, Rapport SVT, Aktuellt SVT, Nyheterna TV4, and Ekonyheterna i Sveriges Radio. The alternative media index included outlets on the right—Fria Tider, Samhällsnytt, Nyheter idag, Ledarsidorna, Kvartal—and outlets on the left— Arbetaren, Arbetet, ETC, Expo, Inte rasist men.

⁴ Because we assume count-type variables to be underlying the empirically derived scores on the two knowledge constructs, we re-ran all analyses using Poisson regression models. Overall, the results—reported in Tables A and B in the fifth section of the online supplementary materials—confirm the results from the standard OLS regression models presented in the main manuscript.