Cultural Heterogeneity in Americans’ Definitions of Racism, Sexism, and Classism: Results from a Mixed-Methods Study

Abstract

This sequential mixed-methods study examines how Americans ascribe meanings to the concepts racism, sexism, and classism. We first conduct interviews (N = 40) using a symbolic boundaries elicitation approach, gathering examples of scenarios that do and do not “count” as racism, sexism, and classism. We then use these examples as vignettes in a nationally representative survey experiment (N = 2,000). Results reveal striking evidence for cultural heterogeneity in how Americans understand and define racism, sexism, and classism. We find that a person’s definition of these concepts depends on their emphasis on intentionality, unequal treatment/outcomes, and power (a)symmetry. We also find that political partisanship, gender, age, and income shape the importance of these three components in their definitions. Finally, we show that Americans’ definitions of racism, sexism, and classism strongly predict their discrimination-related public opinion and policy preferences, such as support for affirmative action and antidiscrimination laws, even after accounting for demographic controls, including political views.

INTRODUCTION

Racism, sexism, and classism are pervasive forms of discrimination faced by many Americans (Pager and Shepherd 2008; Swim and Campbell 2008; Clair and Denis 2015; Krieger 2020; Small and Pager 2020; Doob 2021).² Exposure to these types of discrimination has well established negative effects on life chances, such as a person’s mental and physical health (Major, Dovidio, and Link 2018), educational, occupational, and economic outcomes (Roscigno 2007; Pager and Shepherd 2008), civic and political engagement (Takyar 2019; Oskooii 2020), and experience—or lack thereof—with the criminal justice system (Alexander 2020), to name but a few. In addition, the study of these topics is now flourishing across multiple disciplines, with more than 307,000 peer-reviewed papers published on the topic of discrimination, over 25,000 on racism, and over 5,000 on sexism within the past decade alone (see Dean and Thorpe 2022). Classism has experienced comparatively smaller but growing scholarly engagement with 261 publications over the last 10 years.³

Yet measuring racism, sexism, classism, and discrimination more broadly has been a source of theoretical and methodological debate. A long tradition of scholarship taps into interpersonal forms of discrimination, such as self-reported experiences with daily disparagement on the basis of one’s race, gender, or other social identity (e.g., Williams et al. 1997). More recent work on this topic has advocated that scholars move away from capturing perceived discrimination and move instead toward the use of more implicit (e.g., Quillian 2006; Banaji and Greenwald 2016; Eberhardt 2020), covert (e.g. Chae et al. 2018), and/or structural measures of racism, sexism, and classism (e.g., Pincus 1999; Homan, Brown, and King 2021). Scholars have increasingly argued that this is a key analytic shift, since not all discrimination is perceived and some perceived discrimination may not have actually occurred (Small and Pager 2020).

Building on these two approaches, we argue that it is imperative for sociologists and other social scientists to better understand the perceptual process by which laypersons appraise potential instances of racism, sexism, and classism. A large body of work has demonstrated that the very act of perceiving (or trying to avoid perceiving) discrimination and various isms activates stress processes and leads to other downstream consequences such as hypervigilance, avoidance, and John Henryism (see Allen [2019] for a review). Thus, abandoning the study of these perceptions would leave scholars critically underresourced as they seek to understand discrimination’s effects on stratification outcomes. As the discipline shifts to examine the more structural components of race-, gender-, and class-based discrimination, this extension of previous work is foundational to our ability to disentangle the very real effects of structural discrimination from the very real effects of perceived discrimination. Doing so will ultimately allow us to examine whether disparate outcomes along the lines of race, class, and gender find their roots exclusively in structural causes or whether these disparate outcomes are also shaped by perceptual processes of those who witness and/or experience discrimination.

Furthermore, sociopolitical attitudes regarding the importance of racism, sexism, and classism in the United States, as well as preferences for policies aimed at mitigating discrimination, are highly heterogeneous (Harrison et al. 2006; Daniller 2021). Recent work has shown that Americans’ views of discrimination are tied to their political stances, such as their willingness to support Black Lives Matter protests (Miller, O’Dea, and Saucier 2021). Yet again, scholars are limited in our understanding of these public opinion patterns because of the widely acknowledged polysemy inherent in surveying people’s attitudes and beliefs about topics as culturally fraught as “discrimination” and “racism,” “sexism,” and “classism” (Dunn 2019).

In this study, we bring a cultural, cognitive approach to bear on these issues by directly examining how Americans understand what racism, sexism, and classism mean. Do people perceive these concepts as interpersonal phenomena motivated by hatred, or do they see them as structural phenomena that occur in more diffuse, even unintended ways? As we argue, much of the extant work theorizing racism, sexism, and classism has focused on how academics should define these concepts, leaving these scholars to wonder how laypersons understand these terms and whether their understandings depart from more academic definitions. Discrimination researchers, for their part, have repeatedly called for empirical work that directly investigates this perceptual appraisal process, and this study seeks to fill that important gap. In response to these calls, we examine (1) the basis upon which Americans define racism, sexism, and classism, (2) the demographic patterns in Americans’ definitions of these concepts, and (3) the relationship between these definitions of racism, sexism, and classism and Americans’ sociopolitical attitudes and policy preferences.

To achieve these three aims, we use a sequential mixed-methods approach (Axinn and Pearce 2006). We first gather inductive data from 40 in-depth interviews with a diverse sample of individuals from a large Midwestern city and its surrounding suburban and rural areas. Given the cognitive demands of asking people to define high-level, abstract concepts like racism, sexism, and classism in an interview setting (Vaisey 2009; Jerolmack and Khan 2014), we instead use a symbolic boundaries elicitation approach in which we collect examples and counterexamples about salient cultural concepts in order to observe what people mentally categorize as racism, sexism, and classism (Lamont and Molnár 2002; Sølvberg and Jarness 2019). These data reveal strikingly different understandings of these concepts among our interviewees. Specifically, we find that definitions of racism, sexism, and classism are rooted in different understandings of (1) the action as intentional or unintentional, (2) the situation as a case of unequal treatment or unequal outcomes on the basis of one’s race, class, and/or gender, and (3) the history, power, and position of the group, system, or individual in question.

Next, we deductively test the patterning of these different definitions of racism, sexism, and classism in a nationally representative survey experiment. We do so by asking participants to rate the degree to which various scenarios, derived from the interview data, constitute racism, sexism, and classism. We find strong evidence for cultural heterogeneity: the presence of major sociodemographic cleavages in how Americans define these concepts. Last, we demonstrate that Americans’ definitions of racism, sexism, and classism are predictive of key sociopolitical attitudes and policy preferences, such as support for affirmative action and federal spending increases, even after taking into account individuals’ sociodemographic characteristics (including political views). We conclude that the American public does not have a uniform understanding of terms like racism, sexism, and classism, and as a result, measurement strategies that seek to capture people’s attitudes toward and experiences with these forms of discrimination should be refined accordingly.

RACISM, SEXISM, AND CLASSISM: ACADEMIC THEORIES AND FOLK THEORIES

Sociologists have primarily written about the academic or theoretical way that these concepts should be defined, with much less work examining how individuals actually define them in their everyday lives. This is a critical lacuna because, as we will show, much of the contemporary theorizing on racism, sexism, and classism presumes that these are particularly durable forms of inequality precisely because laypersons hold a very different definition of these concepts than scholars do. Filling this lacuna is vital for accurately capturing personal experiences of discrimination as well as understanding how and why the public holds such divergent attitudes about the importance of racism, sexism, and classism in everyday life and the subsequent need for policies addressing these and other forms of discrimination in the United States.

Considerable scholarship has sought to develop, refine, and critique how sociologists should define racism, sexism, and classism. For instance, work on racial attitudes, race-related beliefs, and racial ideologies has advocated for the existence of racial animus (Bobo and Charles 2009), colorblind racism (Bonilla-Silva 2006; Burke 2017), racial ignorance (Mueller 2017, 2020), tacit racism (Rawls and Duck 2020), systemic racism (Feagin 2013; Ray and Mahmoudi 2022), and institutional racism (McDonald and Wingfield 2008; Hamilton and Ture 2011; Ray 2019; for a systematic review of the differing academic conceptualizations of racism, see Shiao and Woody [2021]). In terms of views on gender and the degree to which those beliefs and institutions constitute sexism, scholars have argued for the existence of concepts like benevolent and hostile sexism (Glick and Fiske 2018) as well as structural sexism (Homan 2019). Finally, classism has received considerably less attention in the scholarly literature. This may be in part because social class is a particularly taboo topic for many Americans (Sanders and Mahalingam 2012), because social class is comparatively less visible in interactions relative to other categorical inequalities like race and gender (Lareau and Conley 2008), or because the law does not currently recognize socioeconomic status as a class protected from discrimination (Peterman 2018). Nevertheless, interest in classism is growing as socioeconomic inequality deepens (see Friedman and Laurison 2020; Laurison and Friedman 2024; Link et al. 2024). Classism has been theorized as occurring through individual beliefs, both explicit and implicit, as well as broader structures of advantage and marginality (Pincus and Sokoloff 2008; Colbow et al. 2016; Shor, Cattaneo, and Alexander 2019).

Importantly, these concepts and terms are academic, derived from theory-building work rather than the terms that laypeople use to define conceptions of racism, sexism, and classism in their own lives. Indeed, much of the aforementioned scholarship is premised on the notion that these forms of discrimination are especially intractable because the general public has a view of these concepts that is distinct from that of academics. As two leading racism scholars recently put it, “most people only see racism as operating in individuals’ minds rather than in structures that facilitate or inhibit movement through institutions” (Ray and Mahmoudi 2022, p. 5; see also Song 2014). Similarly, in framing his theoretical argument for the existence of colorblind racism, Bonilla-Silva argues that “for most Whites racism is prejudice,” a definition of racism that stands in stark contrast to the academic understanding of colorblind racism for which he advocates (2006, p. 18). In fact, there has been surprisingly little scholarship directly examining this empirical question: what are the folk theories and lay understandings (Malle 2006) of racism, sexism, and classism held by the general public? In other words, how do those outside of academia understand these terms?

Although there is a substantive body of work interrogating experiences and perceptions of discrimination as defined through an academic lens (e.g., Anderson [2022] applying the academic framework of symbolic racism to describe the lived experiences of Black Americans regardless of how those individuals might label the incident themselves; or Bolton and Feagin [2004] using the concept of experiential discrimination to understand the prevalence and impact of racism in the lives of Black police officers), limited research has been conducted to date that inductively examines individuals’ own conceptualizations of personal experiences with racism. For instance, Essed (1991) interviewed Black women in the United States and the Netherlands to study how they understood and labeled the racism they encountered. She outlined a multistep process that her interviewees engaged in to make their subjective assessment of potential racism: deciding whether the practice is acceptable, is excusable, occurred due to their race, and is a socially significant event. Lamont and coauthors (2018) also conducted a cross-national interview study to investigate how racial and ethnic minorities interpret their experiences of discrimination and stigmatization. They found that African Americans in the United States generally perceive six types of experiences as racial discrimination and racism: insult or disrespect, being misunderstood, being stereotyped as poor, being stereotyped as threatening, being denied opportunities, and being profiled. Most recently, Wingfield (2019) and Wingfield and Chavez (2020) interviewed Black health care workers, finding that the types of racism they perceived depended on their professional status and location in the organization (i.e., whether they were a doctor, nurse, or technician).

We build on this nascent line of work by broadening the aperture in two key ways. First, we seek to include the views of those who do not identify as racial minorities, since some Whites also believe they have experienced racism and racial discrimination (Fraser and Kick 2000) and because those who are not racial minorities wield important decision-making power via their policy preferences. We also seek to examine appraisals of sexism and classism, as these have gone largely understudied, in spite of their prevalence.

Second, we expand beyond an investigation of personal experiences with racism, sexism, and classism to look at the general process of categorization of these concepts in everyday life. Indeed, as we outline below, our inquiry is primarily focused on the folk theories and lay definitions people use to label and define incidents of potential racism, sexism, and classism that they observe in the world around them—not just those they have personally faced. This shift of focus is vital because, as outlined above, there may be a loose coupling between academic and folk theories of racism, sexism, and classism, with attendant implications for how these folk theories shape Americans’ sociopolitical views, such as intergroup attitudes, redistribution preferences, and support for antidiscrimination laws and policies. Indeed, studying these folk theories and lay understandings is particularly important in social settings where discussions of racism, sexism, and classism are increasingly stigmatized or even silenced by state mandates (Wekker 2016; Beaman 2017; Morning and Maneri 2022; Warren and Valentino 2023; Kramer, Ray, and Bonilla-Silva forthcoming). Furthermore, a focus on appraisals of racism, sexism, and classism as experienced by others avoids some of the known psychological challenges associated with asking people to label and assess their own experiences of discrimination, such as the personal-group discrimination discrepancy (see Taylor et al. 1990).

THE PROBLEM OF POLYSEMY: LIMITATIONS OF CURRENT MEASURES

Social scientists generally rely on surveys such as the General Social Survey (GSS), American National Elections Studies (ANES), and Pew Research Center to understand the perceived prevalence and severity of discrimination and racism, sexism, and classism in American society. These questions often reveal striking variation in public opinion regarding the pervasiveness and role of discrimination in society, both over time and across demographic groups. For example, scholars have found that women tend to perceive sexism as more prevalent in society than men do (Kehn and Ruthig 2013), and people of color tend to perceive racism as more prevalent than Whites (Carter and Murphy 2015; Zell and Lesick 2021). Yet this variation also opens the door to another possibility: that different people may be activating different meanings of “discrimination” or “racism,” “sexism,” and “classism” as they respond to these questions. Our current measures of eliciting public opinion on topics like discrimination, racism, sexism, and classism are limited by the polysemy inherent in asking people to report their level of agreement with statements about the amount of discrimination in society today. Instead, these question wordings tend to rely on the intrinsic assumption that people hold similar—if not identical—views of what these terms mean.

Social scientists are also interested in capturing individuals’ perceived experiences with discrimination in their own lives, because perceived discrimination impacts stratification outcomes beyond those impacts directly resulting from the discrimination itself. For instance, perceived and anticipated discrimination directly shapes life outcomes regardless of the objective reality of discriminatory experiences, as it operates through biological stress processes, selection into (or out of) educational or employment opportunities, and neighborhood choices (Kessler, Mickelson, and Williams 1999; Williams and Mohammed 2009; Allen 2019; Small and Pager 2020). In other words, it is not solely a question of whether discrimination occurred, but whether an individual perceived and assessed discrimination as occurring and whether that perception has shaped their behavior (Doering, Doering, and Tilcsik 2023). Although they do not always explicitly invoke terminology related to discrimination, racism, sexism, or classism in their question wordings, measures such as the Everyday Discrimination Scale (EDS), Experiences of Discrimination Scale (EOD), and the Schedule of Racist/Sexist Events (SRE, SSE) ask survey takers to rate their level of agreement with items such as how often they are treated differently (“with less respect,” EOD; “unfairly,” SRE/SSE) or disparaged (“threatened or harassed,” EDS; “made to feel inferior,” EOD; “called a racist/sexist name,” SRE/SSE) due to their race/ethnicity, gender, and so on (Klonoff and Landrine 1995; Landrine and Klonoff 1996; Williams et al. 1997; Krieger et al. 2005).⁴ Furthermore, the GSS and ANES both ask respondents about the amount of personal discrimination they have faced, both in general as well as in the contexts of workplace and housing specifically.⁵

Yet these measures generally rely on an assumption that individuals perceive racism, sexism, and classism as having occurred when they are disparaged and/or when they are treated unfairly on the basis of their race, gender, or class identity. We argue that these surveys are likely systematically incomplete approaches for capturing the full breadth of how people understand racism, sexism, and classism for two principal reasons. First, disparagement, while often the most egregious form of discrimination, does not encompass many of the other facets of discrimination that scholars have theorized as important aspects of racism, sexism, and classism. Indeed, disparagement on the basis of one’s race, class, or gender best reflects what scholars have called “old-fashioned prejudice” (Swim et al. 1995) and presumes that discrimination emanates from animus or hatred. It does not incorporate newer understandings of racism, sexism, and classism that understand these processes as often rooted in unintentional, even subconscious, cognition like implicit bias. We expect that some—although not all—Americans have become familiar with the concept of implicit bias or unconscious stereotypes and may have therefore integrated this conceptualization of discrimination into their worldview.

Second, a focus on unfair treatment neglects a core component of moral perceptions related to fairness: distributive justice. While unfair treatment relates to procedural justice, a focus on unfair outcomes is what social scientists have referred to as distributive justice (Shepelak and Alwin 1986; Jasso, Törnblom, and Sabbagh 2016). Indeed, much of the academic theorizing around structural discrimination has sought to shift from a focus on disparities in opportunities to disparities in outcomes, or as it is sometimes referred to, the distinction between equality and equity (Cook and Hegtvedt 1983). Furthermore, perceptions of fairness themselves are subject to cultural interpretation. Recent work in social psychology and the study of social morality has shown that people often perceive inequality as fair (Starmans, Sheskin, and Bloom 2017; Heiserman, Simpson, and Willer 2020; Kiviat 2021; Valentino and Vaisey 2022). Thus, the more structural understandings of racism, sexism, and classism—which take unequal outcomes among ethnoracial, gender, and class groups as revealing structural or institutional forms of discrimination—are likely not captured by these measures. Once again, we expect that some—but not all—Americans have gained awareness of these novel understandings of racism, sexism, and classism with the rise of recent discourse around antiracism especially (e.g., DiAngelo 2021; Kendi 2019; see Rozado, Al-Gharbi, and Halberstadt 2021; Dunivin et al. 2022). Thus, existing scales that capture perceptions of discrimination are likely limited in their ability to capture the full breadth of how people understand and make sense of racism, sexism, and classism in the contemporary United States.

In fact, a number of recent studies have shown that there is striking variation in how people respond to scales like the EDS, EOD, SRE, and SSE. For instance, several studies have found psychometric inequivalence in terms of race/ethnicity, age, and education in how individuals self-report items on the EDS (Lewis et al. 2012; Harnois et al. 2019; Bastos and Harnois 2020). Brown (2001) demonstrates that Black respondents report substantively different levels of racial and ethnic discrimination depending on how the question is worded, with distinct impacts on mental health and well-being. Barkan (2018) shows that surveys asking about experiences with “discrimination” as opposed to specific mistreatment situations lead to different levels of reported exposure, and Grollman and Hagiwara (2019) find important differences between how much discrimination people report versus how much unfair treatment they report; troublingly, both studies observed that differences in interpretation vary between ethnoracial groups. As a result, discrimination researchers have repeatedly called for studies that directly investigate the meanings and cultural interpretations individuals use to filter their understanding of these terms (Lewis et al. 2015, p. 413; Grollman and Hagiwara 2017, p. 294; Barkan 2018, p. 252; Krieger 2020, p. 51; Small and Pager 2020, p. 63; Harnois 2023, p. 17). Thus, scholars have become increasingly aware that existing ways of measuring experiences of discrimination likely fall short in capturing the variety of meanings Americans attribute to terms like racism, sexism, and classism, with important implications for how well these measures capture their attendant effects on other stratification outcomes.

SYMBOLIC BOUNDARIES AND CULTURAL HETEROGENEITY

Questions of meaning are fundamental to the social scientific study of culture. In particular, this study’s key goal—mapping the diverse meanings Americans ascribe to concepts like racism, sexism, and classism—first requires an inductive approach that will allow us to identify the boundary of what does—and does not—”count” as racism, sexism, and classism for various individuals. Cultural sociologists have argued that symbolic boundaries are an important aspect of the social processes by which inequalities are produced and reproduced (Lamont and Molnár 2002) and that interviews are uniquely well suited for the delineation of these boundaries (Lamont and Swidler 2014). Lamont and Molnár (2002, p. 168) define symbolic boundaries as “conceptual distinctions made by social actors to categorize objects, people, practices, and even time and space. They are tools by which individuals and groups struggle over and come to agree upon definitions of reality.” Symbolic boundaries therefore enable individuals to assign social groups and concepts to classifications such as deserving and undeserving (Strauss 2002) or even particular racial categories (Abascal 2020). Thus, our analysis will begin by focusing on revealing the conditions under which a person perceives something as racist versus not, sexist versus not, and classist versus not. We will measure the bases upon which they draw symbolic boundaries between what is—and is not—racism, sexism, and classism.

Once we have established the bases upon which Americans define these concepts, we next take a deductive approach to test variation in meaning. Specifically, we test for the presence of cultural heterogeneity. Cultural heterogeneity reflects the idea that there are systematic differences within a single population with respect to particular models, frames, beliefs, symbolic boundaries, and other elements of meaning (Harding 2007; DiMaggio et al. 2018; Corritore, Goldberg, and Srivastava 2019; Goldberg and Singell 2024). Critically, these meaning elements are often patterned along demographic lines, since social position and formative social experiences are thought to drive their divergence (Lamont et al. 2017; Rogers 2019). Cultural sociologists have established cultural heterogeneity across several domains—demonstrating, for example, that liberals and conservatives attribute different meanings to concepts like “poverty” (Homan, Valentino, and Weed 2017; Hunzaker and Valentino 2019), that evangelicals and atheists attribute different meanings to concepts like “religion” (Moore 2017), and that people of different social classes and races attribute different meanings to various scents and perfumes (Cerulo 2018).

We therefore expect similar cultural heterogeneity with respect to how Americans attribute meanings to racism, sexism, and classism. Specifically, we expect that demographic characteristics (e.g., race/ethnicity, gender, age, political views, social class background) will influence the salience of these factors in how people define these terms. First, several studies point toward the importance of ethnoracial identity in shaping Americans’ definition of racism, sexism, and classism. Sommers and Norton (2006) find that racial minorities are more likely to view subtle behaviors and denials of racism as, in fact, instances of racism compared to Whites, who reported racism as occurring only in overt scenarios. Simon, Moss, and O’Brien (2019) find that Black participants were more likely to say racism had occurred when Black individuals were harmed by an incident, whereas White individuals were more likely to say racism had occurred when there was a clear malicious intent. Greenland, West, and van Laar (2022) find that Whites’ definitions of racism are largely confined to acts of hate and deliberate use of offensive language, whereas racial minorities’ definitions also include microaggressions, unintentionally offensive language, and unconscious bias. Harnois (2022, 2023) finds that racial minorities are more likely to interpret survey questions about discrimination through a lens of social inequalities and racism, whereas Whites see them as asking about negative interactions.

Second, a number of studies suggest that people of different genders may hold different understandings of these concepts. Lowe and coauthors (2021) find that women are more likely than men to perceive sexist jokes as harmful. Harnois (2022) also finds that gender may shape the degree to which individuals interpret discrimination items, with women more likely than men to perceive them as questions about social inequalities. Relatedly, Brielle Harbin and Margolis (2022) find that those who identify as feminist are more likely to appraise a hypothetical incident as racial discrimination. Third, at least one study has found that age may be a contributing factor. Weinberg and Nielsen (2017) find that younger individuals are more likely to appraise an event as sexual harassment than older individuals. Fourth, several psychological studies have demonstrated the importance of adherence to conservative-aligned ideologies or dispositions, such as system justification and endorsement of the status quo (Major, Quinton, and McCoy 2002; Stangor et al. 2003; Kaiser and Major 2006), in how people define discrimination. Finally, in terms of social class, Wingfield and Chavez (2020) found status-based differences in perceptions of racial discrimination in health care (e.g., based on one’s status as a physician versus a medical technician), which we expect may be reflective of one’s education level and/or income. Similarly, Harnois (2023) found that some interviewees drew on their formal education or training when deciding how to interpret negative experiences queried as part of the EDS. Thus, we will examine cultural heterogeneity in discrimination definitions with respect to race/ethnicity, gender, age, political views, education, and income.

Last, we note that some social psychologists have articulated a motivational account of group differences when it comes to perceptions of discrimination (Adams, Tormala, and O’Brien 2006, p. 617). This perspective suggests that there are contrasting motivational pressures to appraise and recognize discrimination, depending on one’s own identity. Specifically, this account implies that, for instance, racial minorities’ understanding of racism is distinct from that of Whites’, women’s understanding of sexism is distinct from that of men’s, and the understanding of classism by individuals of lower social class backgrounds (e.g., those below median income, those without a college degree) is distinct from that of individuals of higher social class backgrounds (e.g., those above median income, those with a college degree). Therefore, we will also examine whether there are demographic differences within discrimination type or whether these demographic differences are largely stable across judgments of racism, sexism, and classism.

DATA AND METHODS

Interview Study

For the first, inductive part of the study, we conducted 40 semistructured interviews with a diverse sample of residents from a large Midwestern city and its surrounding suburban and rural areas. We recruited participants from social media platforms such as MeetUp and Facebook. We targeted a wide variety of groups that had shared interests, activities, and demographics. We posted digital advertisements looking for people to share their “experiences, views, and opinions on issues about social life in the US,” avoiding explicit mention of terms like racism, sexism, or classism to mitigate biasing who selected into our sample. Interested participants were directed to a link where they filled out a brief demographic survey, and a total of 1,378 individuals did so. We then used quota sampling techniques to select interviewees to ensure equal representation across race/ethnicity, gender, education, age, and political views (see table 1 for participant demographics). Appendix A further details the recruitment and sampling methods used in the interview study.

Table 1.

Demographic Characteristics of Interview Participants (N = 40).

Variable	Response Option	Percentage or Mean (SD)
Race/ethnicitya	Asian	12.5
	Black/African American	25
	Hispanic/Latinx	12.5
	White	55
	Middle Eastern/North African	5
Gender	Woman	50
	Man	45
	Nonbinary or gender nonconforming	5
Education	No college degree	42.5
	College degree or higher	57.5
Political viewsb	Conservative	42.5
	Liberal	50
	Moderate	7.5
Age	Years	47.42 (15.79)

^aPercentage exceeds 100 due to some participants reporting multiracial identities (N = 4).

^bPolitical views measured on a 1–7 scale where 1 = extremely liberal, 4 = moderate, and 7 = extremely conservative.

Due to the known cognitive difficulties of asking people to explain their reasoning about morally complex topics in interview settings (Vaisey 2009), we instead pursued what we call a symbolic boundaries elicitation approach, drawing on the tradition of qualitative scholarship that compares the existence of symbolic boundaries across categories and groups (Lamont and Thévenot 2000; Lamont and Molnár 2002; Lamont and Swidler 2014; Sølvberg and Jarness 2019). Rather than asking participants to define racism, classism, or sexism directly and then comparing those definitions to each other, we instead asked participants to provide us with examples of racist, sexist, and/or classist incidents with which they were familiar. We then asked them to provide us with negative examples: incidents that others claimed were racist, sexist, and/or classist but with which they disagreed about that appraisal. In probes, we used thought experiments to change key aspects of the interviewee’s example, depending on the specific example (e.g., “What if the police officer had been Black? Would it still be racist then?”) to understand the contours of their definition (Jiménez and Orozco 2021). We then mapped out these boundaries during our analysis phase. This approach allowed us to understand where the boundary exists for each interviewee in terms of these concepts, rather than asking them to explain or justify their boundary. Therefore, our interview guide was mostly focused on observing interviewees as they made real-time categorization decisions about what does and does not count as racism, sexism, and classism, with probes to understand where their definition of these concepts begins and ends (see app. A for the interview guide).

Given the public health restrictions of meeting in person during the COVID-19 pandemic, all interviews were conducted virtually. Interviews lasted between 30 minutes and two hours, with a median length of around one hour. Interviews were recorded and automatically transcribed, after which transcripts were anonymized, cleaned, and corrected for accuracy by members of the research team. After the interview was completed, each participant was compensated with a gift card for their participation in the study.

To analyze the data, we coded all interview transcripts in Dedoose, a qualitative data analysis software program. Through this coding process, we uncovered two sets of patterns. First, we noticed a number of common themes that emerged among interviewees’ examples of racism, sexism, and classism, such as police brutality (racism), high-profile harassment cases in Hollywood (sexism), and ultrawealthy individuals traveling to space (classism). Second, we observed a number of key criteria that define a situation as (not) racist, (not) sexist, and/or (not) classist for interviewees, such as whether the situation was intentional, whether the situation involved unequal treatment or unequal outcomes for the involved individuals or groups, and whether one involved party had more historical, situational, or systemic power than the other. Appendix A contains the analytic codebook developed in this study, documenting the themes we observed for racism, sexism, and classism (table A1), as well as the three main criteria (table A2).⁶ It also includes example quotations that were coded with each criterion in table A2. We used analytic memos and axial coding to examine patterns within and across interviewees.

Survey Experiment

For the second part of the study, we conducted a nationally representative survey experiment that builds on the interview study’s findings about the emerging criteria that define racism, sexism, and classism (see Doering et al. [2023] for a similar inductive-to-deductive approach). We asked participants to assess various scenarios derived from the interviews and then decide whether racism, sexism, and classism had occurred.⁷ The survey experiment contains nine vignettes: three scenarios of potential racism, three scenarios of potential sexism, and three scenarios of potential classism. These vignettes were developed based on the themes that commonly emerged in the interview data (e.g., racial profiling by police for racism, sexual harassment in Hollywood for sexism). Selecting vignettes based on these real-world examples maximized the external validity and verisimilitude of the vignettes. In a pilot study, 81.52% of participants agreed that the vignette was realistic. Appendix B contains the text of the vignettes used in the study.

We sought to manipulate the degree to which the vignettes reflected the three most important defining features of racism, sexism, and classism, as identified in the interview data: intentionality, unequal treatment/outcomes, and power differences. Thus, each of the vignettes varied with respect to whether they were (1) an intentional act of mistreatment based on a person’s race/gender/social class, (2) an unintentional act of mistreatment on this same basis, (3) an act of unequal treatment based on a person’s race/gender/social class that produced equalized outcomes on that basis, (4) an act of equal treatment based on a person’s race/gender/social class that produced unequal outcomes on that basis, (5) an act affecting a person from a traditionally less powerful group (person of color/woman/poor), and (6) an act affecting a person from a traditionally more powerful group (White/man/wealthy).⁸ Figure C1 in appendix C contains a visual representation of the overall experimental design and randomization in the study.

Each of the nine vignettes therefore had five versions, and a participant was presented with a random draw of which vignette version they saw (see fig. C2 in app. C for an example vignette and appraisal rating from the survey experimental task). These vignette versions were balanced across participants to ensure that an equal number of participants rated each vignette version. In the pilot study, manipulation checks ensured that participants saw the intentional condition as more purposeful (F = 319.83, df = 5, P < .001), the unequal treatment condition as affecting one party more than the other (F = 157.56, df = 5, P < .001), and the individual in the less powerful condition as having less power (𝜒² = 51.23, P < .001). The pilot study also ensured that the vignettes were easily understood by participants; in open-ended responses where participants summarized the vignette text, 94.09% of the summaries were coded as accurate by research assistants who were blinded to the study’s goals.

After each vignette, participants were asked to provide a rating on a scale of 0–100 indicating the degree to which they believed that the scenario was a case of racism, sexism, or classism, respectively, where 0 represents “strongly disagree” that the scenario was racism/sexism/classism and 100 represents “strongly agree.” In the pilot study, we found that the majority of participants reported they were not familiar with the term “classism,” and so we provided the following definition to participants: “For the next set of scenarios, please keep in mind that some people define classism as being discriminated against because someone is poor or middle class or rich.”⁹

We collected the data in partnership with YouGov during October 2022. YouGov is a survey firm providing access to national samples that include sampling weights to ensure representativeness of the data to the noninstitutionalized US adult population. All results presented here include sampling weights.¹⁰ We also collected basic sociodemographic information about participants in the survey. Table 2 shows the descriptive statistics for the survey experiment sample.¹¹

Table 2.

Survey Experiment Descriptive Statistics (Unweighted)

Variable Category	Response Option	Mean (SD) or Proportion
Political partisanship	Republican	.323
	Independent	.184
	Democrat	.493
Gender	Man	.462
	Woman	.522
	Nonbinary or gender diverse	.017
Race/ethnicity	Non-Hispanic White	.653
	Non-Hispanic Black	.123
	Hispanic	.134
	Other	.086
Age	Years	48.5 (17.7)
Education	Less than high school	.054
	High school graduate	.318
	Some college	.199
	Associate’s degree	.098
	Bachelor’s degree	.208
	Graduate degree	.124
Household incomea	Dollars	68,916.53 (69,454.41)
Nativity	US born	.928
	Foreign born	.073
Appraisal	Response scale 0–100	52.6 (35.8)

Note.—N = 2,000.

^aMidpoint values used.

After reading and rating the vignettes, participants were also asked standard survey questions used in prior research about discrimination, sociopolitical attitudes, and policy preferences pertaining to discrimination. These included public opinion items such as how much discrimination they think various groups face, their level of support for antidiscrimination laws and affirmative action, and whether they believe federal spending should be increased in various domains. Following existing question wording for these items, respondents were asked separately about their level of (dis)agreement with each policy-related statement for race/racial minorities, gender/women, and social class/the poor, depending on the question. Responses to these three questions were combined into a single standardized index for each outcome in order to ease interpretation. Cronbach alpha values for these indices range from .810 to .949, suggesting overall high levels of reliability. Table 3 shows specific question wording for all of these sociopolitical questions and displays the Cronbach alpha values for each index.

Table 3.

Sociopolitical Outcome Variables

Outcome Variable	Question Wording	Cronbach’s Alpha
Perceived prevalence of discrimination	“How much discrimination is there in the United States today against each of the following groups?” Groups: racial minorities, women, the poor. Response options: 1–5, where 1 = “a great deal” and 5 = “none at all” (reverse-coded).	.846
Discrimination resentment	“When the following groups complain about discrimination, how often do they cause more problems than they solve?” Groups: racial minorities, women, the poor. Response options: 1–5, where 1 = “always” and 5 = “never” (reverse-coded).	.810
Media attention	“How much attention should the news media pay to discrimination against the following groups?” Groups: racial minorities, women, the poor. Response options: 1–5, where 1 = “a lot more attention” and 5 = “a lot less attention” (reverse-coded).	.869
Issue salience	“In terms of issues facing the nation right now, how important would you rate discrimination against the following groups?” Groups: racial minorities, women, the poor. Response options: 1–5, where 1 = “extremely important issue” and 5 = “not at all an important issue” (reverse-coded).	.884
Support for anti-discrimination protection	“Do you favor or oppose laws to protect the following groups against job discrimination?” Groups: racial minorities, women, the poor. Response options: 1–5, where 1 = “strongly against” and 5 = “strongly for.”	.944
Support for affirmative action	“Some people say that because of discrimination, disadvantaged groups should be given preference in hiring and promotion. Others say such preference in hiring and promotion of disadvantaged groups is wrong because it gives them advantages they haven’t earned. What about your opinion—are you for or against preferential hiring and promotion of the following groups?” Groups: racial minorities, women, the poor. Response options: 1–5, where 1 = “strongly against” and 5 = “strongly for.”	.949
Spending preferences	“Should federal government spending on the following areas be increased, decreased, or kept the same?” Areas: Improving the conditions of racial minorities, supporting women in the workplace, aid to the poor. Response options: 1–5, where 1 = “decreased a lot,” and 5 = “increased a lot.”	.878

Note.—Groups/areas were presented to respondents in random order.

FINDINGS

Interview Findings

Through our analysis of the interview data transcripts, field notes, and analytic memos, we identified three key evaluative criteria that figured in most interviewees’ definitions of racism, sexism, and classism: (1) whether the action was intentional or unintentional, (2) whether it resulted from unequal treatment or unequal outcomes, and (3) whether social groups are perceived to be equally powerful or whether some social groups are seen as having more power (either historic/systemic or situational/individual) than others. Some aspects of these evaluative criteria mirror existing academic definitions of racism, sexism, and classism (e.g., the importance of unequal outcomes in identifying instances of racism/sexism/classism, the role of historic power dynamics in shaping the direction of racism/sexism/classism, and the shift from an interpersonal focus to a structural or systemic focus), whereas others stand in stark contrast to accepted academic understandings of these concepts.

Intentionality

We first observed a key division between those who believed racism, sexism, or classism could occur unintentionally versus those who believed it requires intention—typically, malice or hatred. Meg is a 27-year-old White woman who does not have a college degree and identifies as moderate politically. As an example of something others saw as racist, but she does not, Meg mentioned depictions of racialized animals in historic children’s literature, such as The Jungle Book. She said, “To me, I don’t think that’s racist. I think that’s historical. I don’t think they’re mocking somebody. They’re not being derogatory about it. … I don’t think it would be racist unless they were making derogatory comments.” When probed about whether any historical depictions in children’s literature could be racist, she stated that Mickey Mouse making a Nazi salute would be racist to her: “Because it is mimicking something horrible that happened. But in The Jungle Book, it was more so them singing songs and stuff. They weren’t mimicking bad things happening. They weren’t saying bad things.” Absent a visible intent to cause harm to someone on the basis of their race, gender, or social class, interviewees like Meg do not see racism, sexism, or classism.

However, some interviewees believed that something could be racist, sexist, or classist even if one’s intentions were not to cause harm. Several mentioned the role of unconscious or subconscious processes, such as implicit bias, in creating racism, sexism, and classism. This represents a prime example of academic definitions of discrimination permeating lay conceptualizations. Ryo, a 61-year-old Asian man with a college degree who identifies as conservative, described an instance of sexism in which a female professor was denied tenure at a university despite being well qualified for promotion. He noted that the chair of the woman’s department was also a woman. For him, that fact made this scenario not an instance of gender-based animus but a likely case of subconscious, internalized sexism: “You know, you can be a woman. The feminist movement likes to say that sometimes, women are unconsciously or consciously … they’re their own worst enemy, because they believe they have consumed the patriarchy and they act upon that. And whether it’s sexism, racism, or whatever it is, you yourself can be a participant in that and act on things without knowing it. So I think there might have been something like that going on.” Other interviewees for whom intentionality did not matter in their appraisal simply espoused a worldview in which some people do not realize that they are behaving in a racist, sexist, or classist way. Daniel, an 18-year-old Hispanic man who is currently in college and identifies as liberal, noted, “I think there’s unfortunately a lot of people that grew up in a household where they learn [racism] from their parents or the people they hang out with, and then maybe they don’t know right and wrong. And then to them, it’s not maybe racist or they don’t understand that concept.”

Some interviewees’ symbolic boundary was not influenced by intentionality, as evidenced by their response to thought experiments in which the interviewer ascribed good intentions to an actor in a situation the interviewee had described as racist, sexist, or classist. These interviewees often mentioned two reasons for discounting intentions. First, many distrusted intentions because they worried that these could be used as a fig leaf for people who sought to behave in a discriminatory way without repercussion. Andrew, a 39-year-old Asian man with a college degree who identifies as conservative, recounted how he has noticed that people in online spaces often engage in what he calls “subtle racism”: “It’s really easy to get banned on social media because they have the algorithms where they will flag inappropriate words. So if you’re using a racial slur, that’s an automatic ban without a moderator looking at it. So yeah, I mean people are more careful now because they probably figured it out through time. So they’re a bit more subtle, a bit more sly about it.” Similarly, Jessica, a 46-year-old White woman without a college degree who identifies as liberal, highlights how good intentions or “joking around” are often invoked as an excuse to sidestep condemnation: “If you like put on a kimono or something when they came to your home. Like that’s … things that are kind of clowning and playing around … yeah, I can see how that’s offensive. And they do try to write it off as ‘oh we’re just being inclusive.’” As Jessica saw it, regardless of intention, an excuse of joking or “being inclusive” does not absolve someone of racism, especially when the excuse is provided as a means to avoid accountability.

Second, some interviewees—especially those whose understanding of racism, sexism, or classism included unequal outcomes, as discussed below—felt that reasonable people should be aware of their own implicit biases at this point in time. They were suspicious of people who had not examined their own subconscious. As a result, they did not believe that lack of awareness about these issues exempted a person or their behavior from racism, sexism, or classism. Maya, a 36-year-old mixed race Black and White woman without a college degree who identifies as liberal, provided the Confederate flag as an example of racism. When probed as to whether the people flying the flag are aware that they are causing harm—even if they live in an isolated place like rural Appalachia where they may not have access to the kind of information that others have—Maya said: “That’s a hard question to [answer] because a lot of my Black male friends will be like, ‘Yeah whatever, Confederate flag, it is what it is,’ because they don’t have the luxury. ... I feel like I can speak up more. I’m lighter skinned and I can draw a harder line. When you have to deal with all different types of people, especially in [state] when you can just go twenty minutes out and see a Confederate flag, darker people can’t get as worked up. My partner will say—he’s White—my partner will say, ‘anybody knows by now.’… If you’re under a certain age group, you know, you’re choosing to stay ignorant. … I have bias and everybody has bias. You have to work on that and acknowledge that. And if you think like ‘I’m not racist because—’ [and] you have that flag, you’re supporting slavery to me.” As in Mueller (2017, 2020), interviewees like Maya do not see ignorance or lack of awareness as precluding discrimination from occurring, so long as the discriminatory action still leads to unequal outcomes or a disproportionate impact shouldered by the person or group in a lesser position of power.

Unequal Treatment Versus Unequal Outcomes

Many interviewees provided examples of racism, sexism, and classism that were unequal treatment—particularly when that treatment was intended to disparage a person on the basis of their social identity. Someone who consciously targets someone or acts out of maliciousness or hatred toward a person because of their race, gender, or class background was commonly seen as having committed racism, sexism, or classism. Frequent examples included slurs or other highly visible instances in which a person was disrespected on the basis of their background. Dilip, a 71-year-old Asian man with a college degree who identifies politically as liberal, centered his examples of sexism and racism on gendered and racialized slurs: “People use too many slurs when talking about each other. … It basically says that you think less of a whole class of people when you use a slur. … Most of the slurs I’ve heard talked about women or people of different cultures and ethnicities.” However, many interviewees also generalized beyond slurs to include disrespectful treatment toward particular groups. Chris is a 27-year-old White man with a college degree who also identifies as liberal. He described what he saw as sexism in his wife’s workplace: “Just not treating the younger women that were on staff in this organization with the same sort of respect or deference that they would be treating other people.” He felt that male employees were treated more favorably compared to the female employees, including his wife.

While many interviewees espoused a definition of racism, sexism, or classism where there is unequal treatment in the form of conscious disparagement, cleavages appeared in two important ways. First, some interviewees perceived unequal treatment that endeavors to equalize outcomes as racist, sexist, or classist, while others did not. Angela, a 52-year-old Black woman without a college degree who identifies as conservative, articulated an example of classism as means-tested social programs that aid the poor, which she saw as classist because they discriminate against the rich. We can compare this view to that of Ashley, a 36-year-old White conservative woman without a college degree, who disagreed with those who see means-tested social programs—like the income-limited COVID stimulus checks—as classist against wealthier Americans. As she explained, “The stimulus check was used to get people in need money for necessities and stuff. If somebody makes $100,000 a year, obviously they don’t need $1,000.”

Similarly, some interviewees perceived unequal treatment on the basis of gender as sexism, such as different physical standards for men and women in the military or in athletic events. Meanwhile others, like Matt, a 41-year-old White man with less than a college degree who identifies as liberal, did not perceive different standards for men and women as sexism: “There can be different sets of qualifiers just based off of—you have this type of body. So some guys will be like ‘well why is it different like this?’ You have to accommodate the way that a body grows. … Hormones change how bodies are formed. That’s going to have an impact. … You should be willing to say okay, these are just different things. It’s okay to be different. … You can accommodate things differently. It’s okay to understand one of these is not exactly like the other. I think it’s not sexism. That’s just accounting for differences in people.” Here we see how some interviewees also relied on unequal outcomes as an important criterion in their definition of racism, sexism, and classism.

In general, interviewees who were attentive to unequal outcomes saw disparities in outcomes as sufficient evidence to know that racism, sexism, or classism had occurred. This definitional boundary is reflective of the equity versus equality perspective among academic definitions (Cook and Hegtvedt 1983), with interviewees typically identifying unequal outcomes as a clue to the presence of structural forms of racism, sexism, or classism. Critically, these assessments of unequal outcomes were often the result of ostensibly equal treatment à la colorblind racism (Bonilla-Silva 2006) that did not account for historic forms of unequal treatment. For example, Denise, a 57-year-old Black woman with a college degree who identifies as conservative, provided an example of racism from her workplace. She recounted how some years ago there was a position in her department to which very few Black people applied and none made it to the final round of interviews. This disparity alone was enough information for her to consider it racism, but as she explained to the interviewer, the position required many years of state highway patrol experience. However, Denise noted that African Americans were barred from serving in the highway patrol for many decades, so they had no way of accruing the necessary years of experience to qualify for the position. As she explained, “You were never going to get an African American to meet those qualifications. … To me that’s … racism because there may have been people with great investigative experience that didn’t work in the highway patrol. … You’re ruling out a whole other people.” Denise was agnostic as to whether unequal treatment at the moment of hiring produced the racism; indeed, this scenario was one in which ostensibly equal treatment produced unequal outcomes between racial groups due to the legacy of racial exclusion from certain occupations. Yet for her, these unequal outcomes were indicative of a clear and present case of racism.

For some respondents, determining whether an action results in an unequal outcome was predicated on the target’s view or position. For instance, Pete is a 53-year-old White man who identifies as liberal and does not have a college degree. He noted how he had recently shifted his understanding of sexism to be more attentive to how women might perceive or label otherwise equal treatment. Drawing on hugs as an example, he described how he is now aware that a woman’s perception of a hug might be different than his perception of a hug as a man: “After the Me Too [movement], everything came out. ... I was becoming more aware that they [women] may not be comfortable with that [unsolicited hugs]. So before, I would just reach out and hug them. [Now] I say, ‘can I give you a hug?’ That kind of stuff changed my opinion.” An understanding of racism, sexism, and classism that includes unequal outcomes thus attends to (1) disparate outcomes between social groups and/or (2) how other people label the interaction or situation, especially, but not exclusively, if those people are in lesser positions of interactional/individual or historic/systemic power—a point to which we now turn.

Power Differences

For interviewees who saw unequal outcomes as evidence of racism, sexism, or classism, they needed a way to parse whose outcomes matter in a given situation. Thus, for many of these interviewees, they observed power asymmetries and opted to attend to outcomes for the person or group in what they perceived to be the lesser position of power. For some, their measure of power was situational (e.g., who was the boss and who was the employee). For others, their measure of power was historical or structural/systemic (e.g., the economic legacy of slavery and exclusion of Black Americans in the United States), hewing closer to academic definitions of racism, sexism, and classism.

Often, power asymmetry manifested when interviewees were probed to reverse their proffered scenarios—considering whether, for instance, men could experience their sexism example, Whites could experience their racism example,¹² or the rich could experience their classism example. Isabel, a 30-year-old Hispanic woman with a college degree who identifies as liberal, could not “reverse” her classism example because she did not believe that wealthy people in the United States could suffer from classism: “I feel like they have all the means to not feel that way. For the lower class, it’s an everyday struggle. You don’t know what your next meal is going to be. And wealthy people don’t really have to go through that. They have friends. Some of them have connections and friends that would help them out.” Finn, a 26-year-old nonbinary White liberal with a college degree, articulated a similar view when asked to reverse their racism example: “An act can only be racist when it flows with this existing power dynamic. Black people can’t be racist to White people.” For many of the interviewees who perceived power asymmetries between social groups, these imbalances were seen as rooted in historical realities. For instance, several interviewees mentioned that they believed police officers—and sometimes the larger policing system—were racist specifically toward African Americans, and not other racial groups, due to what they described as police departments’ historical origins as slave patrols.

However, other interviewees simply believed that some groups wield more power than others without having to reference to history. Tom, a 72-year-old White man with less than a college degree who identifies as conservative, perceived power asymmetries between men and women such that women wield more power in society. As he saw it, this gender imbalance stems from the numerous false accusations that women can level against men, which he saw as acts of sexism against men: “The things that I’ve read—even asking a woman on a date can be considered harassment. Where the hell did that ever come from? Why is that? I mean women ask men out. But a lot of the sexism discussion still leans on men as the Big Bad Wolf and the woman is Little Red Riding Hood carrying a picnic basket. And that’s false. That’s a fantasy. It really is. And as long as we keep those kinds of beliefs, it actually increases the amount of sexism.” It is important to note that the notions of power being (a)symmetrical are perceptions. Like all elements of cognition, they are subject to bias and motivational thinking, and may or may not reflect reality. Indeed, it matters little whether these perceptions of power imbalances are “really” right or wrong—instead, what matters is that the interviewees perceive them as such (see Shweder [1992] for more on this problem in cultural analysis). Of course, this perceptual component applies to all interviewees whose understanding of racism, sexism, and classism relies on notions of power asymmetry.

This perspective stands in stark contrast to interviewees who perceived power symmetries between people and groups. For these interviewees, they saw all groups as equally likely to be both instigator and target of potential racism, sexism, or classism. For instance, Sharon, a 65-year-old White woman with a college degree who identifies as liberal, did not see police violence as inherently racist because she believes that police officers often act out of fear or dislike of the unknown: “I think sometimes racism might be applied too quickly when it’s not necessarily racism, but it’s actually someone just hating people, and just anybody different from them. It’s not necessarily racism. It’s just ‘I just don’t like you’ type of thing.” By her reasoning, Black police officers could be racist toward White civilians because they’re “different,” a sharp contrast to the view of those like Finn (mentioned above) who do not believe African Americans can behave in a way that is racist toward Whites. Interviewees who perceived power symmetries also tended to perceive equity-based initiatives—in which unequal treatment equalizes otherwise unequal outcomes—as racist, sexist, or classist. Because they did not perceive a power differential—as in the case of Angela, mentioned earlier, who saw means-tested social programs as classist against the rich—they saw these programs (like the income-limited COVID stimulus checks) as instances of racism, sexism, or classism.

To assist in identifying power differentials, some interviewees saw their examples of racism, sexism, or classism as isolated incidents, separating individuals from the systems or structures of which they might be part by atomizing blame. For example, Karl, a 55-year-old conservative White man with a college degree, described an incident of racism in which he recounted how an affluent Black man was pulled over by the police for no reason. However, Karl did not see this incident as part of the historical context of racialized policing. Instead, when probed as to the source of the racism in his example, he set blame at the foot of a racist individual rather than a racist institution: “Well, I think racism always has to do with people. I think it was absolutely a result of—I mean, the information I had, it sounded like it was very much a racist policeman, and I don’t—I guess I have a difficult time understanding how a situation can be racist without people with racist motivations making it happen.” Similarly, Joyce, a 69-year-old Black woman with a college degree who identifies as liberal disagrees with those who see the modern education system in the United States as racist—not because of a lack of racism related to education in this country, but instead because she did not believe systems can be racist: “This woman was giving a speech, and essentially her argument was they had done the [Kenneth and Mamie Clark] doll study in her elementary school and the results were the same in 2018. And what that meant was the educational system was just as racist in 2018 as it was in, I think 1954. I understand what she’s saying. I understand the fact that they did the study, and they got the same results. But I guess I questioned the leap from there to the conclusion that the educational system is fraught with racism.”

Other interviewees framed their assessment of power differentials within the context of systems that enabled and facilitated instances of racism, sexism, and classism. For example, Mark, a 68-year-old White man with a college degree who identifies as liberal, described how his university had a sexist policy that resulted in female faculty being paid less, a case of unequal outcomes: “As a general rule, at this university, [female professors] were getting paid less than the male professors. And the reason was that male professors were more likely to be willing to move and leave the university, whereas the women … they had families, they didn’t want to leave, blah blah blah. The men were single. … This was just the university us[ing] this rule to be violating [against the women].” In this case, Mark’s retelling of this scenario does not focus on specific individuals at the university taking action against the women but instead understands unequal outcomes as resulting from a system’s (the university’s) policy.

Survey Experiment Results

We now turn to the survey experiment that was designed to examine the demographic predictors and public opinion consequences of these definitions of racism, sexism, and classism in a nationally representative sample. We begin by considering the cultural heterogeneity perspective, which postulates that the factors we found in the interview data—intentionality, unequal treatment versus outcomes, and power differences—matter in diverging ways for how Americans define these concepts, depending on their sociodemographic characteristics. We test this prediction by comparing two sets of models for each of the three sets of factors: one model that presumes participants make these judgments identically (regardless of their sociodemographic background) and one model that presumes that participants’ judgments depend on their political party, gender, age, race/ethnicity, education, and income.¹³ Rather than simply rely on the presence of a significant interaction term, we compare the cultural homogeneity and cultural heterogeneity models using parsimonious measures of model fit (Akaike and Bayesian information criteria, referred to as AIC and BIC, respectively; Kuha 2004). Results for these model comparisons, shown in table 4, demonstrate that the cultural heterogeneity model is a better fit by AIC standards for all three experimental contrasts and by BIC standards for two of the three experimental contrasts.

Table 4.

Model Fit for Cultural Homogeneity and Cultural Heterogeneity Specifications

Model	N	AIC	BIC
Intentionality
Cultural homogeneity	6,310	60153.77	60268.52
Cultural heterogeneity	6,310	60149.04	60358.28
Unequal Treatment Versus Outcomes
Cultural homogeneity	6,359	62890.27	63005.15
Cultural heterogeneity	6,359	62624.96	62834.45
Power Differences
Cultural homogeneity	6,361	62962.54	63077.42
Cultural heterogeneity	6,361	62864.18	63073.67

Holistically, these model fit results indicate the presence of cultural heterogeneity with respect to how Americans understand racism, sexism, and classism. In other words, Americans define racism/sexism/classism differently, depending on their sociodemographic background. We next provide a detailed exploration of these specific sociodemographic differences.

Sociodemographic Differences in Definitions of Racism, Sexism, and Classism in the U.S.

We found that three of the six demographic variables varied significantly at the P < .05 level with respect to at least one dimension of racism, sexism, and classism, and one additional variable exhibited suggestive evidence of variation at the P < .10 level. The top half of table 5 summarizes these results (see table F1 in app. F for detailed marginal effects and table F2 for full regression results).

Table 5.

Summary of Survey Experimental Results

Definition of Racism/Sexism/Classism Shaped by …
Sociodemographic Characteristic	Intentionality	Unequal Treatment/Outcomes	Power Differences
Politics	Suggestive	Yes	Yes
Gender	No	Yes	Yes
Race/ethnicity	No	No	No
Age	Yes	No	No
Education	No	No	No
Income	No	No	Suggestive
Definition of Racism/Sexism/Classism Shapes …
Sociopoliticalattitude	Intentionality	Unequal Treatment/Outcomes	Power Differences
Perceived prevalence	No	Yes	Yes
Resentment	No	Yes	Yes
Media attention	No	Yes	Yes
Issue salience	No	Yes	Yes
Antidiscrimination laws	Suggestive	Yes	Yes
Affirmative action support	Yes	Yes	Yes
Funding increase	No	Yes	Yes

Beginning with political views, we find that two of the three dimensions of individuals’ definition of racism, sexism, and classism vary significantly between political partisans—unequal treatment/outcomes (F(2, 1738) = 77.34, P < .001) and power differences (F(2, 1736) = 31.01, P < .001), with intention showing suggestive evidence of differences (F(2, 1744) = 2.92, P = .054). Predicted values for self-identified Democrats, Republicans, and independents from the cultural heterogeneity model (in which all six sociodemographic variables are allowed to vary) are shown in figure 1.

Fig. 1.—

Differences in definition of racism/sexism/classism by political partisanship.

In terms of intention, Democrats and Republicans are somewhat more sensitive to intention than are independents: although all partisans are more likely to rate intentional vignettes as racist, sexist, and classist compared to unintentional vignettes, Republicans increase their appraisal the most in intentional scenarios, by 11.880 points on the 0–100 scale. Democrats’ appraisal of racism, sexism, and classism is on average 9.469 points greater in intentional vignettes, and Independents show only a 5.312-point difference in how they rated intentional and unintentional scenarios.¹⁴

For unequal treatment/outcomes, political partisans hold inverted definitions. Republicans are less likely to say racism, sexism, or classism occurred in cases of unequal outcomes compared to cases of unequal treatment. They rate vignettes that are unequal treatment as 14.969 points higher compared to vignettes in which unequal outcomes occurred. Democrats do the opposite, rating vignettes that are unequal outcomes as 14.119 points higher compared to vignettes in which unequal treatment occurred. Independents are in the middle, neither significantly more nor less likely to appraise unequal treatment or outcomes as racism, sexism, or classism.

In terms of power, Republicans are power symmetric: they are neither significantly more nor less likely to say racism, sexism, or classism occurred if the person affected is a member of a traditionally more powerful group. If anything, Republicans show evidence of increased agreement that racism, sexism, or classism has taken place when it affects the more powerful group, although this difference of 2.078 points is not statistically distinguishable from zero. By contrast, Democrats, and to a lesser extent Independents, show evidence of power asymmetry. They are significantly less likely to appraise racism, sexism, or classism when it affects a member of a more powerful group than a less powerful group: Democrats were more likely to say the vignette was racism, sexism, or classism when it affected the less powerful group by 15.193 points, whereas for independents this difference was 6.453 points on the 0–100 scale.¹⁵

Turning now to gender differences, figure 2 shows predicted values of these judgments for men, women, and gender nonbinary individuals from the cultural heterogeneity models for each of the three experimental contrasts. We observe that people of different genders do not differently appraise racism, sexism, or classism on the basis of intentionality (F(1, 1744) = 2.22, P = .109), but they do significantly differ with respect to the role of unequal treatment/outcomes (F(1, 1738) = 11.04, P < .001) and power (F(1, 1736) = 4.16, P = .016).¹⁶

Fig. 2.––

Differences in definition of racism/sexism/classism by gender.

Men are on average 1.747 points less likely to appraise racism, sexism, and classism in cases of unequal outcomes compared to unequal treatment, although this difference is not statistically distinguishable from zero. Women have the opposite pattern and are 2.661 points more likely to appraise racism, sexism, or classism in vignettes depicting unequal outcomes compared to unequal treatment, although this difference is also not significant. This pattern is particularly pronounced for gender nonbinary individuals, who are 32.462 points more likely to say racism, sexism, or classism has occurred when the vignette is a case of unequal outcomes compared to unequal treatment, a statistically significant difference. People of all genders are less likely to appraise racism, sexism, or classism in cases where the more powerful group is affected. Nevertheless, nonbinary individuals are very power asymmetric, perceiving a large difference in racism, sexism, and classism when it affects less powerful groups as compared to more powerful ones (24.923 points on the 0–100 scale). Women also strongly differentiate between these scenarios (exhibiting a 9.213-point difference), with men rating them the least different (only a 4.933-point difference).

Next, we consider age differences in how people define racism, sexism, and classism. Figure 3 shows predicted values for people of different ages for each of the three experimental contrasts from the cultural heterogeneity models. The figure depicts representative values for age in our sample that show variation across life course stage and cohort—18, 49, and 75 years old.

Fig. 3.—

Differences in definition of racism/sexism/classism by age.

We observe that a person’s age shapes their likelihood of differentiating between unintentional and intentional behavior when making these appraisals (F(1, 1744) = 7.03, P = .008), but not unequal treatment/outcomes (F(1, 1738) = 0.11, P = .736) or power (F(1, 1736) = 0.00, P = .982). For intentionality, we find that this effect is largely driven by older individuals’ increased view that intentional acts are racism, sexism, or classism relative to younger individuals. For example, those who are 75 years old rate the intentional vignettes, on average, 13.032 points higher on the 0–100 scale compared to vignettes depicting unintentional behavior. By contrast, those who are 18 years old rated intentional vignettes only 5.688 points higher than they rated unintentional vignettes, on average.

Finally, we examine differences with respect to income. Figure 4 depicts the predicted values of representative income amounts (25th, 50th, and 75th percentiles) in our sample, using results from the fully interacted cultural heterogeneity models for each experimental contrast.

Fig. 4.—

Differences in definition of racism/sexism/classism by income.

We do not find differences with respect to intentionality (F(1, 1744) = 0.80, P = .373) or unequal treatment/outcomes (F(1, 1738) = 2.62, P = .106) in terms of how individuals of various incomes make appraisals. However, we do find suggestive evidence that income shapes the importance of power differences (F(1, 1736) = 2.72, P = .099). Lower-income Americans are less likely relative to higher-earning Americans to appraise racism, sexism, or classism when it affects a member of a more powerful group. For instance, those earning $15,000 a year rated vignettes affecting the less powerful 9.428 points higher on the 0–100 scale than they did vignettes affecting the more powerful. By comparison, those earning $135,000 a year only rated vignettes affecting the less powerful 5.542 points higher compared to vignettes in which the more powerful were affected.

Critically, we do not find differences in how Americans define racism, sexism, or classism with respect to race/ethnicity or education once other sociodemographic factors are accounted for. Individuals of different ethnoracial backgrounds do not vary with respect to the importance of intentionality (F(3, 1744) = 0.07, P = .977), unequal treatment/outcomes (F(3, 1738) = 0.27, P = .846), or power difference (F(3, 1736) = 1.35, P = .257) in how they make these appraisals. Nor do we find that individuals with various levels of educational attainment exhibit different patterns of appraisals with respect to intentionality (F(5, 1744) = 1.07, P = .377), the role of unequal treatment/outcomes (F(5, 1738) = 0.26, P = .933), or power difference in judgments of racism, sexism, or classism (F(5, 1736) = 1.75, P = .121). Figures depicting these patterns are shown in appendix H.

Differences Between Judgments of Racism, Sexism, and Classism

To what extent do demographic patterns in these definitions change, depending on the type of discrimination being considered? The motivational account of group differences suggests that individuals may define discrimination in a way that aligns with their group if it is the one affected (e.g., women define sexism differently from men). In figure 5, we display marginal effects of the three experimental contrasts for partisanship, gender, age, and income from cultural heterogeneity models that are segmented by discrimination type (racism, sexism, and classism).¹⁷

Fig. 5.—

Marginal effects for experimental contrasts, segmented by discrimination type.

Note.—The x-axis represents the difference between the two experimental contrasts on the 0–100 scale. For example, positive values for “Intentional—Unintentional” indicate that a participant was more likely to appraise an intentional vignette as racism/sexism/classism compared to an unintentional vignette. Positive values for “Uneq. Outcomes—Treatment” indicate that a participant was more likely to appraise a vignette with unequal outcomes as racism/sexism/classism compared to a vignette with unequal treatment. Positive values for “More—Less Powerful” indicate that a participant was more likely to appraise a vignette with a more powerful individual affected as racism/sexism/classism compared to a vignette with a less powerful individual affected.

For most of the demographic differences reported above, we find that these patterns are consistent across discrimination type. However, we do find two key exceptions: the pattern in which Independents and Democrats as well as women rate vignettes of unequal outcomes higher than those depicting unequal treatment is true for sexism and classism but not racism.¹⁸ We also observe that older Americans’ perception that intentional vignettes are more likely to be seen as racism, sexism, or classism relative to unintentional ones is largely driven by judgments of sexism, although the overall pattern holds across all three types of discrimination.

To be clear, we do not interpret this lack of demographic differences in how Americans make judgments of racism versus sexism versus classism as evidence that people perceive these types of discrimination identically. As discussed above, the interview data revealed that people offer categorically different types of examples of, say, racism, as opposed to, say, sexism, which are reflected in the domain differences of the vignettes (e.g., a racism vignette about racial profiling in traffic stops by police versus a sexism vignette about an actor being sexually harassed on a film set (see app. B for vignette text; see also Maxwell [2015])). Furthermore, figure 5 shows that the demographic patterns we found with respect to how Americans define these concepts are not driven by discrimination type, as predicted by the motivational account of group differences. Instead, we observe that political partisanship, gender, age, and income represent key cleavages along which definitions of racism, sexism, and classism occur, and this is generally true regardless of discrimination type.

Relationship Between Definitions and Discrimination-Related Public Opinion Items

For the last set of analyses, we investigate how these different definitions of racism, sexism, and classism relate to key beliefs, sociopolitical attitudes, and policy preferences about discrimination, net of a person’s demographic characteristics that are known to shape these outcomes and—as shown above—that are also related to their definition (i.e., political partisanship, gender, age, income, race, education, and nativity). Results from models predicting these public opinion outcomes, including all sociodemographic controls, are shown in table 6 and are summarized in the bottom half of table 5.¹⁹

Table 6.

Marginal Effects of Racism/Sexism/Classism Dimensions on Sociopolitical Outcomes

	Unintentional	Intentional	Unequal Treatment	Unequal Outcomes	Less Powerful	More Powerful
Prevalence of discrimination	.005*** (.001)	.004*** (.001)	−.003 *** (.001)	.004 *** (.000)	.004 *** (.000)	−.000 (.001)
Discrimination resentment	–.004*** (.001)	–.004*** (.001)	.004 *** (.000)	.001 (.001)	.001 (.001)	.003 *** (.001)
More media attention to discrimination	.005*** (.001)	.006*** (.001)	–.003 *** (.001)	.003 *** (.000)	.003 *** (.000)	–.001 (.001)
Issue salience of discrimination	.007*** (.001)	.006*** (.001)	–.002 *** (.001)	.005 *** (.001)	.005 *** (.001)	.000 (.001)
Support for anti discrimination laws	.005*** (.001)	.007*** (.001)	–.005 *** (.001)	.001 (.001)	.001 (.001)	–.003 *** (.001)
Support for affirmative action	.005 *** (.001)	.003 *** (.003)	–.004 *** (.001)	.004 *** (.001)	.004 *** (.001)	.000 (.001)
Support for increased federal funding	.006*** (.001)	.005*** (.001)	–.004 *** (.001)	.003 *** (.000)	.003 *** (.000)	–.000 (.000)
N	6,310		6,359		6,361

Note.—Bold indicates marginal effects for which there is a significant difference (P < .05) in the two slopes. All models include control variables (political partisanship, gender, age, income, race/ethnicity, education, and nativity).

^***P < .001.

^**P < .01.

^*P < .05 (two-tailed tests).

For all seven outcomes, we find that two of the three dimensions of individuals’ racism, sexism, and classism definition—unequal treatment/outcomes and power differences—are significantly related to discrimination-related public opinions and attitudes. Crucially, these relationships occur even while controlling for an individual’s demographic characteristics, including their political partisanship. We also find that a third dimension of the racism/sexism/classism definition, intentionality, matters in the case of affirmative action support and shows suggestive evidence that it predicts support for antidiscrimination laws (F(1, 1744) = 3.63, P = .057).

Specifically, we first find that Americans who are more likely to report racism, sexism, and classism has occurred when vignettes depicted both unintentional and intentional action were more likely to believe racial minorities, women, and the poor face a lot of discrimination. They are less likely to believe that people cause more harm by complaining about the discrimination they experience.²⁰ They believe the media should pay more attention to discrimination. They are more likely to see discrimination as a major problem facing the United States. They report increased support for antidiscrimination laws as well as affirmative action.²¹ Finally, they are more likely to support federal spending increases to support racial minorities, women, and the poor. In most cases, however, we do not observe that a heightened likelihood of appraising racism, sexism, or classism in unintentional scenarios as opposed to intentional scenarios is related to these sociopolitical outcomes.

Second, we find that those who are more likely to see racism, sexism, and classism when vignettes illustrated unequal treatment are less likely to hold the aforementioned attitudes. By contrast, those who are more likely to see racism, sexism, and classism when vignettes illustrated unequal outcomes are generally more likely to espouse these views. Third, we find that those who are more likely to appraise racism, sexism, and classism in vignettes where more powerful groups (Whites, men, and the wealthy) are affected are overall less likely to hold these views, whereas those who are more likely to appraise racism, sexism, and classism in vignettes where less powerful groups (racial minorities, women, and the poor) are affected are more likely to hold these views. The differences in the relationship between racism/sexism/classism definitions and the seven public opinion outcomes are all significant at the P < .05 level or higher (represented by coefficients in bold in table 6).

DISCUSSION AND CONCLUSION

To summarize results from this sequential mixed-methods study, the interview data inductively revealed three key aspects of how people make appraisals of racism, sexism, and classism: whether they believe it can occur unintentionally as well as intentionally, whether they are attuned to unequal treatment or unequal outcomes, and whether they perceive differences between social groups in access to power and resources. In the follow-up survey experiment, we found that members of different political parties, genders, ages, and income groups vary greatly with respect to how these aspects of a scenario factored in to whether Americans defined it as racism, sexism, or classism. We showed that these definitions of racism, sexism, and classism matter because they are predictive of a variety of beliefs, intergroup attitudes, and policy preferences pertaining to discrimination, even after controlling for an individual’s sociodemographic characteristics (including their political partisanship).

These findings challenge the notion that certain groups or worldviews are more sensitive to discrimination. It is a common claim that there are people who are eager to acquire the status of “victim” by labeling almost anything as racism, sexism, or classism (e.g., Campbell and Manning 2018; McWhorter 2021). Furthermore, psychologists who study appraisals of discrimination have been largely focused on whether certain groups are more or less sensitive to or vigilant about the potential discrimination they observe or encounter (Major et al. 2002; Stangor et al. 2003) or have a narrower or broader definition of discrimination (Greenland et al. 2022). The present research clearly shows that Americans do not define racism, sexism, and classism in such a unidimensional way. Instead, our interviews repeatedly revealed cases where a person was unwilling to acknowledge racism, sexism, or classism has occurred in one instance but acknowledged its existence in another—and did so in ways that are inconsistent with a sensitivity/vigilance account. Recall Finn, the 26-year-old nonbinary White liberal, who noted that White people cannot experience racism from Black people but offered examples of Black people experiencing racism at the hands of Whites, or Angela, the 52-year-old Black conservative woman who offered means-tested government programs as an example of classism but did not see poor people’s overall worse outcomes as classism. These cases cannot be understood as narrow versus broad definitions of racism, sexism, and classism. The survey experiment further underscores this point. Recall that Democrats generally perceive cases of unequal outcomes and those in which less powerful individuals are affected as racism, sexism, and classism but do not perceive cases of unequal treatment and those in which more powerful groups are affected this way—and that, critically, Republicans exhibit the opposite pattern. Thus, it is Americans’ definition of racism, sexism, and classism, rather than overall willingness to appraise racism, sexism, and classism, that forms the main difference across sociodemographic groups and that predicts discrimination-related intergroup attitudes and policy preferences.

Implications

The findings from this study have important implications for the way we study and measure experiences of and attitudes toward discrimination and racism, sexism, and classism. Our results indicate that disparagement is a very narrow slice of the many meanings people attribute to concepts like racism, sexism, and classism. Indeed, existing items from the EDS, EOD, and SRE/SSE that ask people whether they were treated with less respect or insulted on the basis of their social identity are likely tapping into a conceptualization of discrimination that is largely intentional, leaving out those who also see unintentional acts as discriminatory. Similarly, scale items that ask people to identify whether they were “treated unfairly” rely on an unequal treatment conceptualization of discrimination, excluding appraisals rooted in unequal outcomes—which, ironically, is the prevailing paradigm that many scholars now use to define these concepts in academic work. To that end, we suggest that these scales should incorporate items about having experienced both unintentional behavior and unequal outcomes on the basis of some aspect of the participant’s identity in order to better reflect the discrimination experiences of those who define racism, sexism, and classism this way. Results from our survey experiment suggest that experiences of Democrats, women, and gender nonbinary individuals in particular are likely systematically missed by existing survey questions.

Our findings also point toward some previously unexplored determinants of individuals’ sociopolitical views and policy preferences, such as how likely they are to support affirmative action or their willingness to increase federal funding for certain groups. Our findings suggest that Americans define racism, sexism, and classism in discrepant ways and that these are tied to policy stances that reflect those definitions. We found that intentionality is less connected to these political stances and preferences. For those scholars and policymakers who work to increase support for policies and initiatives that combat discrimination, we therefore recommend expending less energy educating the public about unconscious bias and other forms of unintentional discrimination. Instead, they should focus on shifting the public’s understanding of the other two dimensions that we found are linked to these stances. For instance, they could continue to highlight the ways that unequal outcomes, in the form of gender, racial, and class-based disparities, serve to impede opportunities and life chances. They could also better educate the public about the historical roots and situational imbalances that lead to power asymmetry in how racism, sexism, and classism manifest.

In addition, this study demonstrates the methodological utility of symbolic boundaries and cultural heterogeneity for understanding the multiple cultural meanings individuals attribute to concepts, even within the same population. In other words, it helps solve the problem of polysemy. We expect that this symbolic boundaries elicitation approach—of gathering inductive interview data in which participants offer examples and counterexamples—could be useful for scholars interested in other cultural concepts that are highly contested, such as who is considered “elite” or what is deemed “healthy.” Inspired by Lamont’s (1992) method of “illuminat[ing] the structures of thought through which … people organize (i.e. select and hierarchize) the ‘raw data’ they receive on others” (p. 4), the symbolic boundaries elicitation approach revealed the role of both socioeconomic and moral boundaries in how Americans decide what does and does not count as racism, sexism, and classism. This technique can be a useful first step toward understanding how people think about a particular topic, especially when existing survey measures do not already exist or when cognitive interviewing techniques have revealed that new survey items should be created. This inductive approach can then be combined with more deductive ones, such as survey experiments like the one featured here, to examine cultural heterogeneity in a broader population and to pinpoint sociodemographic differences in meanings.

Furthermore, this study has two important implications for scholars who study discrimination and those who are interested in the disconnect between academic theories of racism, sexism, and classism and Americans’ folk theories of these concepts (Bonilla-Silva 2006; Gelman and Legare 2011; Ray and Mahmoudi 2022). First, our study contributes to the small but burgeoning literature on classism (Friedman and Laurison 2020; Laurison and Friedman 2024; Link et al. 2024) showing that many Americans do recognize this form of discrimination and in ways that resemble that of racism and sexism. This suggests that there may be more popular support than previously expected for discrimination protections on the basis of social class (Peterman 2018), despite Americans’ general lack of fluency around the topic and its lowered visibility vis-à-vis other identities, such as race and gender.

Second, our work suggests that research studying differences between groups in terms of the experience, prevalence, or importance of discrimination and various isms is likely capturing the polysemy inherent in these terms. Scholars should be aware that asking a younger, high-earning man who is a Republican about his experiences with or views of discrimination will most likely evoke an entirely different baseline definition of this term than it does for an older woman who is a Democrat with a low income. When measuring personal experience with discrimination, we recommend analysts use survey items that define specific experiences (e.g., “You were called a racial slur”) rather than using more vague questions about “fairness” that are open to cultural interpretations (e.g., “You were treated unfairly by your coworkers”) or that invoke terms like discrimination or racism/sexism/classism (e.g., “You were called a sexist name”) to avoid this problem of cultural polysemy. Similarly, public opinion questions about discrimination attitudes and beliefs, such as the importance of discrimination or racism, sexism, and classism as a problem facing the country today, should first query individuals’ own definitions of these concepts with regards to intention, unequal treatment/outcomes, and power differences.

Avenues for Future Research

Future work can gain more purchase on the causal direction of the relationship between definitions of racism, sexism, and classism and policy attitudes by examining the development of these definitions and subsequent policy attitudes over the life course. Recent research has shown that deep-seated beliefs and policy attitudes are likely inculcated at a relatively early period in a person’s socialization and remain relatively stable once someone reaches adulthood (Vaisey and Lizardo 2016; Kiley and Vaisey 2020). Thus, it is vitally important to examine the conditions under which young people acquire these definitions of racism, sexism, and classism, with special attention toward the potential role of homophilic networks, given our findings that these definitions are patterned along the lines of political partisanship, gender, age, and income.

Our interview data do provide us with some clues as to whether and how these understandings change over time. We observed a number of interviewees whose boundaries were in flux or who recounted that their understandings of racism, sexism, and classism had changed over time. For instance, several interviewees mentioned how social movements (Black Lives Matter, #MeToo, Occupy Wall Street) or recent participation in book clubs sparked by these highly publicized movements had caused their conceptualization of these terms to evolve from unequal treatment to unequal outcome–based definitions. One interviewee mentioned how participating in a youth program for ethnoracial minorities as a child exposed her to information about the historical legacy of racial inequality, leading her to become more attuned to power asymmetries between social groups. A handful of interviewees also talked about how having contact with friends or family members of other races or genders had opened their eyes to the kinds of unintentional discrimination that their loved ones endured. Thus, another important area for future research is to examine how these definitional boundaries develop and change over the life course. This will prove vital for pinpointing interventions (e.g., deep canvassing; Demetrious 2021) that seek to effectively educate the broader populace about the nature and prevalence of discrimination. In addition, future scholarship can examine whether Americans perceive and define discrimination differently from racism/sexism/classism. For instance, a recent interview study found some parallels to at least one of the major definitional cleavages we observed, noting that some people seem to see discrimination as a type of “differential treatment” (what we refer to as unequal treatment) and others see it as “social inequality” (what we refer to as unequal outcomes; Harnois 2023). Thus, additional survey experiments could straightforwardly assess the degree to which these concepts overlap or diverge by varying the question wording of the outcome variable scale.

Finally, our work, by methodological necessity, treated racism, sexism, and classism as separately occurring phenomena. Intersectionality theory tells us that these forms of discrimination often co-occur due to interlocking systems of oppression (Crenshaw 1991; Collins 2002; see also Heiserman 2023). Some work has established that the burden of experiences of discrimination falls heaviest on those who hold multiply marginalized identities (Harnois 2014; Armstrong, Gleckman-Krut, and Johnson 2018), while other work has established the impact of victim identity in shaping third-party assessments of discrimination (Warren and Valentino 2023). Future research in this vein would benefit from an explicitly intersectional lens, examining how definitions of these concepts shift in response to considerations of the overlapping aspects of racial, gender, and class-based discrimination. For example, vignettes could systematically vary both the race and gender of the target to assess whether appraisals of racism and sexism are dependent on gender- and/or race-neutral framing. Further, participants could be asked to appraise both racism and sexism in response to a vignette, thereby establishing to what extent laypeople apply intersectional lenses in their own assessments of discrimination. Finally, scholars should expand the lens from how Americans define racism, sexism, and classism to how Americans define other forms of discrimination such as those rooted in religion, weight, or marital status.

Conclusion

Decades of empirical scholarship and theorizing have led to the development of a wide-ranging and robust literature on the nature and prevalence of discrimination and racism, sexism, and classism in the United States. This study has examined the on-the-ground definitions and understandings of these concepts among Americans who do not study these topics for a living. Our findings can provide a bridge between scholars’ understandings of these concepts and those of laypersons. Doing so is a necessary first step in educating the broader public on topics like racism, sexism, and classism and garnering public support for policies that rectify these injustices, as well as better capturing and understanding the deep and enduring impacts of discrimination on life chances.