Introduction
The notion that a well-functioning democracy is predicated on an informed and knowledgeable citizenry is a key tenet in the scholarly literature (Bartels, Reference Bartels1996; Berelson et al., Reference Berelson, Lazarsfeld, McPhee and McPhee1954; Delli, Delli and Keeter, Reference Delli, Delli and Keeter1996). But with the rise of misinformation rapidly increasing in the United States, hopes of achieving such a citizenry are at grave risk (Flynn, Nyhan and Reifler, Reference Flynn, Nyhan and Reifler2017). Given the highly polarizing nature of misinformation against historically and socially marginalized groups in the United States (Scheufele and Krause, Reference Scheufele and Krause2019), unpacking the determinants of misinformation about them has never been more pressing.Footnote 1
Socially marginalized groups comprise a politically, socially, and economically vulnerable segment of the United States. Currently, they constitute a growing share of the nation’s populace at roughly 40.3%. Latinos comprise the largest share at 18.3% of the population, followed by Black Americans at 13.4%, Asian Americans at 5.9%, and multiracial Americans at 2.7%. Another vulnerable group with a highly minoritized and racialized status is Muslim Americans, who comprise about 1.1% of the population. And in a mere 25 years, Census demographers predict these groups will together surpass white non-Hispanic/Latino Americans to become the majority in the United States. Latinos will increase to 24.6% of the U.S. population, Black Americans remaining approximately the same at 13.1%, Asian Americans 7.9% and multiracial Americans will be 3.8% of the U.S. population; the U.S. Muslim population is projected to grow to be the second largest religious group by 2050.
In this paper, we define socially marginalized groups as racial/ethnic minorities (e.g., Black Americans, Latinos, Asian Americans) and Muslim Americans. Our decision to include Muslim Americans stems from their increased visibility in the American public as a marginalized, salient, and increasingly racialized group over the past two decades. Studies affirm this racialization (Aziz, Reference Aziz2021; Kalkan, Layman and Uslaner, Reference Kalkan, Layman and Uslaner2009; Lajevardi, Reference Lajevardi2020, Reference Lajevardi2021; Nacos and Torres-Reyna, Reference Nacos and Torres-Reyna2007; Nguyen, Reference Nguyen2019; Oskooii, Reference Oskooii2016; Oskooii, Dana and Barreto, Reference Oskooii, Dana and Barreto2019; Sediqe, Reference Sediqe2020; Yazdiha, Reference Yazdiha2023). Though U.S. Muslims are often conflated with MENA (Middle East and North African) Americans, it is crucial to appreciate that the global Muslim community is incredibly diverse, spanning various ethnic, racial, national origin, and cultural backgrounds. In the United States, the racialization of Muslim Americans is shaped by distinctive domestic dynamics, informed by prevailing narratives, geopolitical contexts, and media portrayals that often merge Muslim identity with MENA ethnicity. Scholarship has shown that both the American public and the media regularly conflate Middle Easterners and North Africans with Muslims. This conflation suggests an implicit racial undertone in the perception of Muslims (Aziz, Reference Aziz2021; d’Urso, Reference d’Urso2022; Maghbouleh, Schachter and Flores, Reference Maghbouleh, Schachter and Flores2022) and that white Americans, in particular, often overlook the nuanced differences between these two groups, highlighting the depth of the conflation (d’Urso and Bonilla, Reference d’Urso and Bonilla2023).
This study departs from the existing research by focusing on the predictors of misinformation toward a sizable share of the public—socially marginalized groups. Thus, we draw a distinction between measures of misinformation that focus on rumors or conspiracy theory, such as death panels and the Affordable Care Act and “Pizzagate” (Lopez and Hillygus, Reference Lopez and Hillygus2018; Nyhan, Reference Nyhan2010; Schaffner and Luks, Reference Schaffner and Luks2018). Our rationale for studying this type of misinformation stems from the reality that the vast majority of the most current and controversial policies that have risen to the forefront of American politics are now highly racialized in nature. That is, the frames and discussions surrounding these policies, such as immigration, criminal justice policies, climate change, welfare, affirmation action, terrorism, and national security, as well as health care, often invoke explicit racial language, especially as it pertains to the recipient of these policies (Abrajano and Hajnal, Reference Abrajano and Hajnal2017; Gilens, Reference Gilens1996; Pérez, Reference Perez2016; Tesler and Sears, Reference Tesler and Sears2010).
We examine three potential contributors to misinformation: racial context, partisan context, and interpersonal contact. We define racial context as the percentage of racial/ethnic minorities residing in a given county. Partisan context, meanwhile, is defined as the Clinton 2016 presidential vote share percentage in a given county. Past research has similarly used party vote share in a presidential election as a proxy for geographic variation in partisanship and has demonstrated the importance of partisan presidential vote share on political identity (Feinberg et al., Reference Feinberg, Tullett, Mensch, Hart and Gottlieb2017), changes in economic consumption (Gerber and Huber, Reference Gerber and Huber2009), and immigration attitudes (Newman, Reference Newman2013). Finally, extant research has long stressed the importance of interpersonal contact in shaping perceptions of racial threat and political attitudes (McLaren, Reference McLaren2003; Walker, McCabe and Matos, Reference Walker, McCabe and Matos2021; Yadon and Piston, Reference Yadon and Piston2019). Interpersonal contact, especially with a more diverse array of people, has been shown to reduce prejudice toward minoritized populations in ways that improve their democratic incorporation in meaningful ways, by enabling individuals to perspective take and challenge stereotypes (Allport, Reference Allport1954; Enos, Reference Enos2014; Jordan, Lajevardi and Waller, Reference Jordan, Lajevardi and Waller2022; Tropp et al., Reference Tropp, Okamoto, Marrow and Jones-Correa2018).
Taken altogether, we test three hypotheses. First, we expect those who reside in counties with higher concentrations of minorities to have more information about them as a result of the informal interactions that residents routinely have with one another (such as at the grocery store, post office, in schools, etc.). In turn, we would expect individuals residing in more diverse counties to be less misinformed about socially marginalized groups. Second, we would expect those individuals residing in areas with larger Democratic vote shares in 2016 to reside in contexts that would render them be less informed about minoritized social groups. We also offer two additional hypotheses focusing on subgroup variations based on race and partisanship. In light of the recent research by Abrajano and Lajevardi (Reference Abrajano and Lajevardi2021) who find that misinformation about socially marginalized groups is the most acute among Republicans and white Americans, we similarly hypothesize that these patterns will be pronounced among white and Republican respondents. Building on past research that finds that whites and Republicans are more influenced by contact with members of out-groups (Jordan, Lajevardi and Waller, Reference Jordan, Lajevardi and Waller2022), despite being more prone on average to stereotyping, we posit that these two groups are more likely to be positively affected by contact with more diverse groups. Last, we posit that more diverse interpersonal contact with minoritized people is linked to being less misinformed about them, given that more first-hand contact with members of this group can educate individuals about these populations.
We use three original surveys with the same measure of misinformation about minoritized groups to address these hypotheses. We then merge this data with contextual data at the county level. Across the three surveys, our findings challenge the conventional wisdom that geographical context inevitably shapes racial attitudes. Rather, the findings reveal that a diverse network more robustly shapes individuals’ misinformation levels than does racial and partisan context. More specifically, when we evaluate racial and political context and interpersonal contact on their own, we find that they are each negatively correlated with levels of misinformation, though the relationship between having a more diverse personal network and misinformation is most robust. When taken together, our analyses reveal that diverse interpersonal contact continues to be the most powerful and consistent indicator of the three: those with more diverse contact with non-white populations are less likely to be misinformed about minoritized groups. And for white Americans (and somewhat less consistently, for Republicans), having more diverse personal contacts is associated with lower levels of misinformation as well across the three surveys.
This research contributes to the understanding of factors that shape misinformation in the United States and departs from existing work in two key ways. First, our study investigates the determinants of misinformation specifically pertaining to minoritized social groups. We focus on how the diversity of contact that one has with members of stigmatized groups affects one’s misinformation levels about these populations. Second, our study explores the influence of partisan and demographic contexts on misinformation. Our results indicate that engaging with members of minoritized groups holds some promise in mitigating misinformation. Understanding the factors that are associated with misinformation about minoritized groups is important in voter decision-making as well as in their electoral choices.
Misinformation and Marginalized Groups
It is no secret that the freedoms of speech and of the press are pillars of democracy. Traditional sources of news—namely, print, television, and radio—are supplemented by social media in a digitized world. Access to news, therefore, can be considered as useful to a functioning democracy only to the extent which one might find accurate information. The growing digitalized age has made it easier for inaccurate sources of information to be shared and to spread online with little to check its veracity, creating the circumstances where misinformation gains power over reputable information.
Kuklinski et al. (Reference Kuklinski, Quirk, Jerit, Schwieder and Rich2000) make a significant distinction between being uninformed (those holding no belief about the correct answer to a factual question) versus being misinformed (holding a false or unsubstantiated belief about the correct response). As Flynn et al. (Reference Flynn, Nyhan and Reifler2017) note:
“While scholars have long lamented public ignorance about politics, misperceptions (i.e., being misinformed) may be an even greater concern. In particular, misperceptions can distort people’s opinions about some of the most consequential issues in politics, science, and medicine.”Footnote 2
Our definition of misinformation is in line with Flynn at al.’s (Reference Flynn, Nyhan and Reifler2017) definition of a misperception: “factual beliefs that are false or contradict the best available evidence in the public domain” (p 128). As such, our measure does not exclusively focus on conspiracy theories or rumors, which has been the focus of some existing studies of misinformation (Miller, Saunders and Farhart, Reference Miller, Saunders and Farhart2016). This measure of political misinformation helps us understand the public’s basic and general understanding about social groups. For instance, the American public is largely unaware that most immigrants in the United States are here legally.Footnote 3 A 2018 Pew study found that while 75% of immigrants in the United States are here legally, only 45% of Americans know that most immigrants living in the United States are here legally. What’s more, only 33% of Latinos knew this fact followed by 43% of Black Americans. Thus, we also included several survey items asking respondents about their knowledge on the major trends in demographic patterns in the United States, as this has been a widely publicized and discussed phenomenon by the news media. Another area where misinformation arises relates to issues where marginalized social groups are the main beneficiary or target. That being the case, several survey items focus on policies that directly relate to social groups (e.g., immigration). And in light of extant research on the infamous “birther” question and its Islamophobic origins (Jardina and Traugott, Reference Jardina and Traugott2019; Tesler, Reference Tesler2018, Reference Tesler2022), we also include it in our survey.
Edelman (Reference Edelman2001) outlines how one of the most pressing outcomes of the spread of misinformation is the chronic resentment against groups labeled as others, often receiving blame for situational imbalance compared to another group. This resentment is not an uncommon finding in other scholarly work, particularly held by white Americans toward their non-white others (Feldman and Huddy, Reference Feldman and Huddy2005; Huddy and Feldman, Reference Huddy and Feldman2009), perhaps even without exposure to misinformation. Although beyond the scope of this paper, developing a deeper understanding of how misinformation might interact with previously held racial resentment would be a significant contribution to the literature.
Recently, misinformation has taken on a partisan context. The rallying cry of “fake news” as tauted by former President Donald Trump was an influential factor in his 2016 electoral victory. It is well-documented, however, how misinformation was actually a tool in galvanizing the former president’s base: the misinformation used by his campaign was not only successful in getting him elected to office but it also showed to be powerful in fueling negative rhetoric toward political opponents, marginalized groups and even encouraged acts of violence (Dwoskin and Timberg, Reference Dwoskin and Timberg2021; West, Reference West2021). It is evident that misinformation has been used successfully as a powerful tool in politics and that compounding effects of sharing among one’s echo chamber play a role in spreading misinformation—and it has the potential to be particularly dangerous toward marginalized groups.
Finally, misinformation is associated with racialized resentment toward marginalized groups (Edelman, Reference Edelman2001; Jaiswal, LoSchiavo and Perlman, Reference Jaiswal, LoSchiavo and Perlman2020), and previous studies have paid much attention to existing racial attitudes against marginalized groups in the American public (Ditonto, Lau and Sears, Reference Ditonto, Lau and Sears2013; Grigorieff, Roth and Ubfal, Reference Grigorieff, Roth and Ubfal2018; Hopkins and Washington, Reference Hopkins and Washington2020). As we discussed previously, contact theory is most prevalent in the debate on racialized perceptions: the prevailing argument lies on findings that show how intergroup contact alleviates negative racialized perceptions toward the out-group (Allport, Reference Allport1954; Dixon and Rosenbaum, Reference Dixon and Rosenbaum2004; Emerson, Kimbro and Yancey, Reference Emerson, Kimbro and Yancey2002). Contact theory, therefore, matters for prejudice formation: is contact experienced through geographic context linked with individual levels of misinformation? This critical oversight has yet to be addressed in the literature, yet it is one that carries implications in a larger political context.
Why Context and Personal Contact Matters
By nature, humans are social creatures. While political science research favors analyses at the unit of an individual, it is important to recognize that we are not shaped in a vacuum: where we go to school, who we grow up around, and who we directly and indirectly interact with on a day-to-day basis shape our understanding of the world. An individual born and raised in a predominantly white, Midwestern town might not carry the same lived experiences as someone with the same education, income, and racial background raised in the suburbs of southern California. Context matters in creating a narrative for how we view the world—and, by extension, how we process our political understandings.
The scholarly literature highlights the importance of contextual factors in explaining a whole host of political attitudes and behaviors. Broadly, the existing research reveals how neighborhoods are able to produce civically engaged citizens through political socialization (Burbank, Reference Burbank1997; Cho, Gimpel and Dyck, Reference Cho, Gimpel and Dyck2006; Huckfeldt, Reference Huckfeldt1979; Straits, Reference Straits1990). Contextual factors can also explain depressed turnout levels, as evidenced by findings on negative engagement with the criminal justice system (Burch, Reference Burch2013). Contextual interactions further exercise influence in the political realm, given that stereotypes, prejudice, and exclusionary attitudes all factor into the political decision-making process (Enos, Reference Enos2014; Foladare, Reference Foladare1968; Straits, Reference Straits1990). Residing in relatively heterogeneous and diverse areas would be likely to have more direct and indirect interactions with both people who are different from themselves, and by proxy, the cultural background that diverse groups bring to a place’s identity. If racial context and having greater social diversity in one’s personal networks are influential in determining positive attitudes toward different groups, we would expect the case to apply to information levels (and, conversely, misinformation levels) of marginalized groups. Based on research by Key (Reference Key1949), we also know that the opposite dynamic could also occur. In terms of contextual partisanship, exposure to the political right—and, by extension, the campaigns targeting and information shared by those in the political right—may be associated with higher levels of misinformation among individuals.
We expect levels of misinformation to be affected by the political context in which they reside in for several reasons. On a daily basis, individuals move anywhere between 5 and 15 miles for work, school, recreation, shopping, and more (Earls, Reference Earls2017; Friedrich, Reference Friedrich2021; Omnibus Household Survey, 2003). This means that individuals are not constrained to their neighborhood, and can be exposed to politics and current events on a larger geographic scale. Gravelle (Reference Gravelle2016) argues that “it is important to stress that counties are politically consequential ‘containers’ in the American context”; elections and the provision of certain public goods take place at the county level (Branton and Jones, Reference Branton and Jones2005; Glaser, Reference Glaser1994). Given that important political decisions are being made at the county level, we decided to measure political context at the county level. We expect individuals living in counties with a large percentage of Republicans will be more misinformed about social groups than are Democrats and Independents, which would be consistent with the previous research (Ahler and Sood, Reference Ahler and Sood2018; Hochschild and Einstein, Reference Hochschild and Levine Einstein2015 a; Jardina and Traugott, Reference Jardina and Traugott2019).
Moreover, contemporary U.S. politics exhibit liberal and conservative political coalitions that are firmly sorted along the lines of race. Take the fact that partisanship is highly polarized by race; a mere 14 percent of the registered Republicans identify as non-whites, while 43 percent of Democratic registered voters consider themselves to be non-whites. The same divides are also reflected when looking at the racial composition of Republican versus Democratic elites, particularly in the U.S. Congress. In turn, Republican officials uphold and advocate for policy positions that respond to the preferences of their constituents, primarily conservative white Americans, whereas Democrats push for more progressive policies that are more closely aligned with socially marginalized groups. While we recognize that other contextual factors could also help to predict levels of misinformation about minoritized groups, we contend that both partisanship and race capture a great deal of explanatory power, based on findings from previous research.
Personal and regular interactions with individuals, especially those from out-groups, have also been shown to be relevant in understanding racial attitudes. Contact theory, as developed by Allport (Reference Allport1954), stipulates that contact with a group different from one’s own has the potential to form positive associations between the two groups. The differing dynamics of contact theory and one’s contextual environment thus set the stage for a similar relationship between information and contextual factors, particularly those which are directly related to one’s knowledge of minoritized groups.
A significant body of research also reveals that racial animus plays an important role in explaining American’s political attitudes, beliefs, and behaviors (Jardina and Traugott, Reference Jardina and Traugott2019; Tesler and Sears, Reference Tesler and Sears2010), and this is particularly the case for white Americans. For instance, both Hoschshild and Einstein (Reference Hochschild and Levine Einstein2015 a) and Jardina and Traguott (Reference Jardina and Traugott2019) find that white Americans who hold negative assessments of Black Americans are the ones most likely to accept the birther rumor. Moreover, where Americans choose or are able to live is highly segregated along the lines of race, particularly for whites (Reardon, Fox and Townsend, Reference Reardon, Fox and Townsend2015). As such, most individuals do not regularly interact with those outside of their own racial group (Reardon, Fox and Townsend, Reference Reardon, Fox and Townsend2015). Thus, in the case of white Americans, those who live in diverse areas and know individuals from minoritized groups could serve to diminish their feelings of racial animus and thereby be associated with their levels of misinformation. In our analyses, we therefore test the predictive power of social and political context, as well as knowing someone from a minoritzed group on misinformation about socially marginalized groups. Specifically, we pose the following hypotheses:
-
H1: Individuals who reside in areas with larger shares of socially marginalized individuals, reside in areas with larger shares of Democratic voter support, and have more diverse personal contacts will be less misinformed about minoritized groups.
-
H2: Residing in areas with larger shares of socially marginalized individuals, residing in areas with larger shares of Democratic voter support, and having more diverse personal contacts with members of minoritized groups will differentially impact white Americans and reduce their misinformation levels more than their non-white counterparts.
-
H3: Residing in areas with larger shares of socially marginalized individuals, residing in areas with larger shares of Democratic voter support, and having more diverse personal contacts with members of minoritized groups will have greater impacts on Republicans and will reduce their misinformation levels more than Democrats and Independents.
Our rationale for H2 and H3 is guided by the existing research which documents that both Republicans and white Americans more likely to be generally misinformed than their counterparts. Research has shown that white Americans are more likely to be exposed to fake news (Grinberg et al., Reference Grinberg, Joseph, Friedland, Swire-Thompson and Lazer2019), and there is also some evidence that Republicans are more likely than Democrats to be susceptible to misinformation at baseline (Abrajano and Lajevardi, Reference Abrajano and Lajevardi2021; Lee and Hosam, Reference Lee and Hosam2020). As such, for these two groups, perhaps contact with a more diverse network, residing in more Democratic-leaning areas, and residing in areas with more diverse populations may be able to have greater impacts on them and move them more.
We therefore expect our three independent variables of interest to exert stronger predictive power on their misinformation levels. Further, those with a diverse network of personal contacts may not operate in the same way for non-white respondents, who on average live in more diverse counties, reside in areas with greater 2016 Democratic vote shares, and report more diverse personal networks than white respondents (see Table 1). Non-white Americans likely already have extensive personal networks with people from minoritized backgrounds, especially within their own communities. Consistent with existing research (e.g., Jordan, Lajevardi and Waller, Reference Jordan, Lajevardi and Waller2022), it could be that incremental benefit of each additional minoritized contact might be less pronounced than it is for white Americans.
Table 1. Key explanatory variables among respondent subgroups
Note: The figures presented above are weighted averages.
Data and Measures
We assess these questions through three surveys. Each of these includes geographic information on respondent residence, allowing us to estimate contextual contact, as well as detailed questions about contact with members of minoritized groups and misinformation about such populations. Survey 1 was fielded on Amazon Mechanical Turk (MTurk) from September 20–21, 2019 on 410 respondents, Survey 2 was fielded on Lucid Academia (Lucid) from October 23 to November 12, 2019 on 2,851 respondents, and Survey 3 was fielded on Amazon Mechanical Turk (MTurk) from January 28 to 29, 2019 on 1,036 respondents.
Importantly, our surveys did not recruit participants using probability sampling. Surveys 1 and 3 both rely on MTurk convenience samples, which are not nationally representative. Survey 2 is also not nationally representative, as it was fielded on oversamples of Black, Asian, and Latino Americans, in addition to a sample of white Americans. To render the results more robust and to ensure that the demographic makeup of the respondent population in each survey better corresponds to the U.S. population, we weight each of the surveys to population benchmarks for race, gender, and education. The analyses we present include survey weights using the 2018 American Community Survey (ACS) 5-year estimates for sex, race, and education. Summary statistics for the unweighted data can be found in Tables A1–A3.
Outcome Variables
The main dependent variable of interest is misinformation about socially marginalized groups in the United States. All three studies operationalize misinformation by using the same measure as in Abrajano and Lajevardi (Reference Abrajano and Lajevardi2021), which is based on 14 separate questions.Footnote 4 Respondents were asked factual questions about socially marginalized groups that are prone to misinformation from the media, ranging from their knowledge about the size of the immigrant population in the United States to questions about the primary beneficiaries of public housing. Higher values on this additive scale indicate greater levels of misinformation, while lower values indicate lower levels of misinformation. Respondents who selected the incorrect answer to these questions were coded as being misinformed (e.g., value 1). Nonetheless, as Kuklinski et al. (Reference Kuklinski, Quirk, Jerit, Schwieder and Rich2000) note, a significant distinction exists between being uninformed (those holding no belief about the correct answer to a factual question) versus being misinformed (holding a false or unsubstantiated belief about the correct response). As such, those who indicated “Don’t know” in response to this measure were not coded as being misinformed (e.g., received value 0).
Table A4 details the precise questions used to create the misinformation measure, the percent misinformed for each question, and the percent who answered “Don’t know” across each survey. On average, respondents scored 5.93, 6.10, and 5.97 in Surveys 1, 2, and 3, respectively.Footnote 5 For ease of interpretation, this measure was normalized to range from 0 to 1, with the highest value indicating more misinformation.Footnote 6
Explanatory and Control Variables
Our three key explanatory variables measure racial and political context at the county level, as well as the diversity of personal contact with minorities. Figure 1 presents the raw distributions for these main outcomes and explanatory variables across the three surveys. As can be seen, the distributions of these four variables take similar shape across each of the survey instruments.
Figure 1. Distributions of main variables
First, County Percent Non-white denotes the measure of racial context and evaluates the contact with racial/ethnic minorities that a respondent might where they live. It is measured as the share of non-white Americans who reside in a respondent’s county. On average, respondents resided in counties where 25% of the residents were non-white in Survey 1, 30% in Survey 2, and 25% in Survey 3.
Next, County Democratic Vote Share 2016 is our measure of political context and measures contact with Democrats that a respondent might have in the county wherein they reside. This measure uses county-level data from the MIT Election Data and Science Lab on the percentage of votes cast for Clinton in the 2016 presidential election. On average, respondents in Survey 1 resided in counties where Clinton won 49% of the vote share in 2016, 53% in Survey 2, and 49% in Survey 3.
Our final key explanatory variable is Diverse Network, an additive variable ranging from 0 to 4. This variable quantifies the diversity of an individual’s social contacts based on their personal acquaintance with people from four key groups. Specifically, this variable is an indicator of whether the respondent reports personally knowing a Latino American, Asian American, Black American, or Muslim American. Like the main independent variable in Jordan, Lajevardi and Waller (Reference Jordan, Lajevardi and Waller2022), this variable measures the diversity of contact that an individual reports having with people in their lives across the three surveys. Because we probed respondents with dichotomous questions that asked about their personal contact with each of the four groups, this aggregate variable ranges from 0 to 4, with 4 indicating a highly diverse network and 0 suggesting a rather homogeneous network. On average, respondents reported knowing people from 2.9 of the 4 groups asked about in Survey 1, 2.8 people in Survey 2, and 3 people in Survey 3. Over one-third of respondents in each survey (38.29%, 38.58%, 42.47%) reported personally knowing someone from each socially marginalized group that we inquired about. Together, the “Diverse Network” variable reflects the breadth of an individual’s interpersonal interactions across diverse social groups and serves as an empirical indicator of the richness and social diversity of one’s personal networks. As such, it offers insights into the potential influence of diverse social interactions on individual perspectives and behaviors.
Table 1 displays summary statistics for each of the three explanatory variables among different subgroups in each survey. The results from this table suggest two important takeaways with respect to the insulation of the political and racial subgroups examined. First, beginning with racial subgroups, key differences emerge between white and non-white respondents. In each survey, non-white respondents were significantly more likely than their white counterparts to reside in counties with greater proportions of non-white residents, to reside in counties with greater Democratic 2016 vote shares, and to report having more diverse personal networks. Second, turning to partisan subgroups, though these differences are not consistently statistically significant, Democrats were more likely than Independents and Republicans to reside in counties with a greater share of the percent of the population identifying as non-white and in counties where Democrats had a larger proportion of the vote share in the 2016 presidential election. They were also more likely to have more diverse personal networks. Republicans were also socially and politically insulated: across the three surveys, on average, they resided in counties where Democrats lost the 2016 election (e.g., the Democratic vote share was on average under 50%), resided in areas with lower shares of non-white residents, compared to Democrats, and consistently reported having more homogeneous friend networks than did Democrats.
Finally, our analyses account for control variables that were queried across all three survey instruments. These include standard demographic controls (sex, race, education, income, and age), as well as party identification. Please note that while each of the variables we employ does not all range from 0 to 1 in their original forms, in the analyses that follow, each of our explanatory and control variables has been rescaled to range from 0 to 1 for ease of interpretation.
Results
To test our three hypotheses, we estimated a series of OLS regressions to evaluate whether a statistical link exists between our three key explanatory variables and levels of misinformation about socially minoritized groups.Footnote 7 While these analyses are not causal, by evaluating our research question across multiple studies that include fairly large samples of partisans and racial subgroups we are interested in studying, we can determine whether the statistical relationships we observe generalize to Americans regardless of partisan stripes and racial group membership.
We begin by testing the association between misinformation and each of our three variables of interest independently. To this end, we estimated ordinary least squares (OLS) models where the full misinformation scale was regressed on County Percent Non-white, County Democratic Vote Share 2016, and Diverse Network (Additive). Table 2 presents our findings across each of the three datasets.Footnote 8 Models 1, 4, and 7 regress misinformation on County Percent Non-white, Models 2, 5, and 8 regress misinformation on County Democratic Vote Share 2016, and Models 3, 6, and 9 regress misinformation on Diverse Network (Additive).
Table 2. Testing the statistical relationship between contact and misinformation
Standard errors in parentheses.
These weighted models include control variables for gender, race, education, income, age, and partisanship.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Two takeaways emerge from this analysis. First, there is a negative relationship in every model, apart from model 8, between the key explanatory variable of interest and misinformation. Second, Diverse Network (Additive) is statistically associated with misinformation across the three surveys. This suggests that one additional contact with a minoritized person is associated with a reduction of misinformation (rescaled to range from 0-1) by 0.098, 0.035, and 0.059 points (or a reduction of 0.148, 0.055, and 0.094 standard deviations) in Surveys 1, 2, and 3, respectively. Nonetheless, the coefficients for County Percent Non-white and County Democratic Vote Share 2016 do not consistently rise to traditional levels of statistical significance, indicating that the relationship between misinformation and contextual contact does not reliably shape misinformation.
Outside of personal contact, partisanship is the only other consistent and strong predictor of misinformation about socially minoritized groups. For instance, a one unit increase in Democrat party identification is negatively linked to misinformation by between 0.051 and 0.113 points, across each of the nine models. Age is also strongly and negatively correlated with misinformation; in six of the nine models it is negative and statistically significant.
As a robustness check, Table A6 conducts a similar analysis as in Table 2, but instead regresses the explanatory variables on the abbreviated misinformation scale. The findings from this second analysis are consistent with those in Table 2. Here, again we find that the only consistent relationship between our key explanatory variables and misinformation that exists is for the Diverse Network (Additive) variable, with coefficients being slightly larger in this analysis. Specifically, one additional contact with a minority is associated with a reduction of misinformation (using the alternative scale) by 0.134, 0.0.49, and 0.084 points (or a reduction of 0.126, 0.054, and 0.087 standard deviations) in Surveys 1, 2, and 3, respectively. One other finding of note is that County Democratic Vote Share 2016 is at times (in Surveys 1 and 3) positively related to misinformation about socially minoritized groups, though statistically insignificant. In Survey 2, this coefficient is negative and statistically significant. The differences in the coefficient’s direction, size, and significance across the three surveys render us hesitant to draw any conclusions about partisan context one way or another. Finally, the coefficient for County Percent Non-white is negative across each of the models (1, 4, and 7), though remains statistically insignificant.
Thus far, our results indicate that multiple types of contact are negatively related to Americans’ misinformation levels about socially minoritized groups, though the effect of personal contact appears to be the most consistent among the three key explanatory variables examined across the three surveys. Such associations lend some initial support for H1. Nonetheless, it remains unclear whether these findings persist when each of the explanatory variables is tested alongside one another. Thus, in order to test H1, we include our three independent variables of interest in the same regression models and test whether the findings from Table 2 persist. We again estimated ordinary least squares (OLS) models, which include controls and survey weights.Footnote 9 Models 1, 2, and 3 present the results for Surveys 1, 2, and 3, respectively. Each of the models regresses misinformation on County Percent Non-white, County Democratic Vote Share 2016, and Diverse Network (Additive).
Across our analyses, we again find a persistent, negative, and significant association between having more diverse personal networks and misinformation about socially minoritized groups (at p < 0.05). Consistent with Table 2, personal contact with one additional minoritized person is linked to a reduction of misinformation (on a 0-1 scale) by 0.100, 0.034, and 0.068 points (or by 0.151, 0.054, and 0.109 standard deviations) in Surveys 1, 2, and 3. County Democratic Vote Share 2016 again presents conflicting results. In Survey 2, it is negatively and significantly associated with misinformation, though in Survey 3, the coefficient is positive and significant. County Percent Non-white does not also consistently predict misinformation; while the coefficient never reaches statistical significance, in Surveys 1 and 2 the coefficient is positive, and in Survey 3 the coefficient is negative. These results offer some support for H1, such that having a diverse network is consistently associated with reductions in misinformation, though our expectations regarding social and political context are more uneven.
These findings persist when we employ the abbreviated misinformation scale in Table A8. The results for County Democratic Vote Share 2016 and County Percent Non-white are inconsistent in size, direction, and significance across the three surveys, indicating that these contextual variables are not reliable predictors of misinformation about socially minoritized groups. Diverse Network (Additive), however, remains significant, negative, and relatively large in substantive size. A one-unit increase in increasing the diversity of one’s personal network results in a reduction of misinformation on the abbreviated scale by 0.130, 0.047, and 0.091 points (or 0.121, 0.052, and 0.094 standard deviations) in Surveys 1, 2, and 3.
Together, these results challenge conventional understandings about the link between geographical context and prejudicial attitudes. While extant research suggests that our physical environment (e.g., where we live) largely shapes our perceptions and beliefs, this study underscores the importance of interpersonal contact and ties. The results here demonstrate that the diversity of our personal networks plays a much more powerful role in dispelling misinformation about socially minoritized groups. Merely residing in a diverse or politically slanted area is not a solution against harboring misinformed beliefs. Instead, it is the active, interpersonal engagements with a diverse set of individuals that emerges as a much more powerful catalyst in challenging and reshaping misconceptions.
Findings on the Relationship Between Contact and Context Among Racial and Partisan Subgroups
Thus far, the results offer some support for our first hypothesis: having a diverse group of personal contacts predicts lower levels of misinformation about socially minoritized groups, but that the impact of geographical racial and partisan context is much weaker and more mixed. Next, we test H2 and H3 to evaluate whether our three independent variables of interest offer more predictive power for whites and Republicans, relative to non-whites and Democrats/Independents. In order to test these two hypotheses, we use the same model specification as those in Table 3, but disaggregated the sample by racial group (in Table A9) and party identification (in Table A10).Footnote 10 Figure 2 plots the coefficients from these models for our three core explanatory variables for subgroups of whites, non-whites, Democrats, Republicans, Democrats, and Independents.
Table 3. Testing the joint statistical relationship between contact and misinformation
Standard errors in parentheses.
These weighted models control for gender, race, education, income, age, and party.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Figure 2. Exploring the relationship between contact and misinformation among racial and partisan subgroups
As we hypothesized in H2, we expected that having a diverse network would be an important predictor of misinformation reduction for white respondents, given that they tend to have less exposure to minoritized groups in the areas where they live (Emerson, Kimbro and Yancey, Reference Emerson, Kimbro and Yancey2002; Rugh and Massey, Reference Rugh and Massey2014).Footnote 11 Results from Figure 2 reveal that this is indeed the case; among our white respondents, the most consistent predictor of lower levels of misinformation about socially minoritized groups across the three surveys was a more diverse personal network. Specifically, a one-unit increase in increasing the diversity of one’s personal network results in a reduction of misinformation by 0.108, 0.041, and 0.061 points in Surveys 1, 2, and 3 among white Americans.
Additionally, we find that a one-percentage-point increase in Clinton’s 2016 vote share within a county where a white participant lives led to a notable decrease in their misinformation levels—by 0.237 and 0.180 points in the first and second surveys, respectively. However, in the third survey, the coefficient shifted to being positive, though was not significant. This implies that while partisan context might influence misinformation levels among white Americans, the results from Survey 3 fail to provide enough evidence to firmly establish this correlation. Finally, the results offer no evidence to suggest that the proportion of non-white residents in a county is associated with white respondents’ levels of misinformation; the coefficient for County Percent Non-white is insignificant across the three studies and in mixed directions. In contrast, none of the three explanatory variables we examined were significantly associated with misinformation about socially minoritized groups for non-white respondents. In fact, no variable among our suite of controls consistently predicted misinformation for this group, perhaps due to the small sizes of minority respondents in the two MTurk studies. As such, we do not draw any firm conclusions with respect to minority respondents. Turning to the results testing H3, we find no consistent support for our expectation regarding partisan differences in the predictive power of context on diverse personal networks on misinformation. Though we hypothesized that Republicans would be more consistently associated by exposure to diversity, either through interpersonal contacts or through geographical partisan or racial context, the results here suggest otherwise. For full models among partisan subgroups corresponding to Figure 2, see Table A10.Footnote 12 In terms of the racial context, the coefficient for County Percent Non-white is positive and insignificant aside from one instance for all partisan groups in Surveys 1 and 2. We note that it is negative for all groups in Survey 3. Turning to the partisan context, the coefficient for County Democratic Vote Share 2016 is negative among all partisans in Surveys 1 and 3, but is positive for all groups in Survey 3. Though the coefficient rises to levels of statistical significance in several instances, it does not consistently predict misinformation in the expected direction in these instances (e.g., models 3, 4, and 7 in Table A10).
The most consistent of our support for H3 pertains to the explanatory variable, Diverse Network (Additive), though even this variable is not significantly linked to misinformation across all models. Among Republicans and Independents, the Diverse Network (Additive) is negative across all three survey instruments and is statistically significant in models 3, 4, and 8 (see Table A10). For Democrats, this coefficient is negative in Surveys 2 and 3, but not in Survey 1. As such, we cannot reliably draw any conclusions about the association between each of our three core explanatory variables and misinformation among partisans.Footnote 13 In sum, we find inconsistent support for H2 and H3; only among white respondents do we find evidence of a consistent pattern of a more diverse personal network being significantly linked to lower levels of misinformation across all survey instruments.
Discussion and Conclusion
This research adds to the existing research on the factors that contribute to the rising tide of misinformation in the United States and moves beyond existing research in two ways. First, it examines the determinants of misinformation specifically about minoritized social groups, which has yet to be fully explored in the existing literature. Namely, as suggested by the existing literature, we focus on the diversity of personal contact with individuals from minoritized groups to understand whether it is associated with greater knowledge about them. Second, our research also considers whether living by those who are part of this minoritized group is correlated with being less misinformed about them. We also test the possibility that those living in communities where there is a significant percentage of individuals predisposed to this misinformation (in this case measured by their partisanship) factors into their misinformation levels. Taken altogether, our findings indicate that knowing more individuals from marginalized backgrounds is the most consistent predictor of our three explanatory variables of interest: those with more diverse contacts with non-white populations are less likely to be misinformed about minoritized groups. And while some may quibble with the magnitude of our effects, given that a one-item shift on a 14-item scale may appear to be modest, it is arguably significant when viewed in terms of misinformation’s cumulative impact. Reducing even a single piece of misinformation can potentially alter a person’s understanding or perspective on a critical policy. When extrapolated across a population, this change can have a substantial cumulative impact on public opinion and policy preferences.
This finding suggests that individuals who regularly interact with minoritized groups can be one way to combat the racial divide on important current events. Recent studies have revealed that when white Americans are informed of the fact that racial groups are disproportionately affected by COVID-19 infections and deaths, their support for increased COVID-19 safety precautions diminished (Skinner-Dorkenoo et al., Reference Skinner-Dorkenoo, Sarmal, Rogbeer, Andre, Patel and Cha2022; Stephens Dougan, Reference Stephens Dougan2022). Moreover, our results challenge the conventional wisdom underscoring the relationship between geographical context and prejudicial attitudes. While the scholarly research has concluded that our physical environment (e.g., where we live) largely shapes our attitudes, our study demonstrates that the diversity of our personal relationships plays a more predictive role in dispelling misinformation about socially minoritized groups. Merely residing in a diverse or politically homogeneous area is not sufficient in combating inaccurate beliefs. Rather, it is the active, interpersonal engagements with individuals from either the same or different minoritized backgrounds that is positively associated with altering misconceptions. Our work is not without limitations. Future work would be well served to examine the prevalence of misinformation about other minoritized groups, not presently examined here, to evaluate whether these patterns extend to additional groups and contexts. Moving forward, research should validate our measure of diversity of contact alongside other measures of contact. We fully recognize that our findings do not identify a causal relationship, but we highly encourage other scholars to head in this direction. It could also be the case that in communities that experience substantial demographic change, where minoritized groups may begin to surpass whites, this shift could also be associated with misinformation.Footnote 14 Understanding the causes of misinformation toward minoritized groups could be quite consequential in shaping individuals’ policy beliefs and decisions, and ultimately in their electoral decisions.
Appendix
Table A1. Summary statistics: Survey 1 (weighted data)
*Note: All variables have been normalized to range from 0 to 1.
Table A2. Summary statistics: Survey 2 (weighted data)
*Note: All variables have been normalized to range from 0 to 1.
Table A3. Summary statistics: Survey 3 (weighted data)
*Note: All variables have been normalized to range from 0 to 1
Table A4. Misinformation measure questions (full sample)—Surveys 1–3
Note: Question numbers in blue form the basis of the abbreviated misinformation scale employed in robustness analyses. Percent “Don’t know” is included in parentheses. The percentages above are unweighted averages.
Table A5. Full model corresponding to Table 2 (DV: full misinformation scale)
Standard errors in parentheses.
These models include survey weights and correspond to Table 2. The dependent variable is the full misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Table A6. Full model corresponding to Table 2 (DV: abbreviated misinformation scale)
Standard errors in parentheses.
These models include survey weights and replicate Table 2 using the abbreviated misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Table A7. Full model corresponding to Table 3 (DV: full misinformation scale)
Standard errors in parentheses.
These models include survey weights and correspond to Table 3. The dependent variable is the full misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Table A8. Full model corresponding to Table 3 (DV: abbreviated misinformation scale)
Table A9. Full model corresponding to Figure 2, among minorities (DV: full misinformation scale)
Standard errors in parentheses.
These models include survey weights and correspond to Figure 2. The dependent variable is the full misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Table A10. Full model corresponding to Figure 2, among partisans (DV: full misinformation scale)
Standard errors in parentheses.
These models include survey weights and correspond to Figure 2. The dependent variable is the full misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Table A11. Full model corresponding to Figure A2, among minorities (DV: abbreviated misinformation scale)
Standard errors in parentheses.
These models include survey weights and correspond to Figure A2. The dependent variable is the abbreviated misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Table A12. Full model corresponding to Figure A2, among partisans (DV: Abbreviated Misinformation Scale)
Standard errors in parentheses.
These models include survey weights and correspond to Figure A2. The dependent variable is the abbreviated misinformation scale.
+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Figure A1. Misinformation items by race in Surveys 1, 2, and 3
Figure A2. Exploring the relationship between contact and misinformation among racial and partisan subgroups (Using the Abbreviated Misinformation Scale)
Open access
