Multiracial self-identification is frequently portrayed as a disproportionately female tendency, but previous research has not probed the conditions under which this relationship might occur. Using the 2015 Pew Survey of Multiracial Adults, we offer a more comprehensive analysis that considers gender differences at two distinct stages: reporting multiple races in one's ancestry and selecting multiple races to describe oneself. We also examine self-identification patterns by the generational locus of multiracial ancestry. We find that females are more likely to be aware of multiracial ancestry overall, but only first-generation females are more likely than their male counterparts to self-identify as multiracial. Finally, we explore the role of racial ancestry combination, finding that multiracial awareness and self-identification are likely gendered differently for different segments of the mixed-race population. This offers a more nuanced picture of how gender interacts with other social processes to shape racial identification in the United States.
Mixed racial ancestry plays a prominent role in the biographies of several pioneering American men, from historical figures such as Frederick Douglass to recent president Barack Obama. Yet, as philosopher Naomi Zack puts it, “most mixed-race symbols in the mass media are women” (1994:xi). From British royal Meghan Markle to the computer-generated images of social media influencer Lil Miquela and of TIME's iconic “New Face of America” cover, young women play a prominent role in representing multiraciality to the American public.
The gendering of multiraciality suggested in the media, however, has deeper and wider roots. Harkening back to nineteenth century “tragic mulatta” narratives, Bost (2003:2) contends that “debates about the nature of mixed-race identity are mapped out on the body of a woman because thinking about racial mixture inevitably leads to questions of sex and reproduction.” The historical feminization and sexualization of mixedness also underpin its enduring association with physical beauty (Rhodes et al. 2005; Sims 2012; Zackodnik 2004). In addition, the late twentieth century multiracial movement had a decidedly female cast, as White mothers played a major role in its leadership (DaCosta 2007; Ibrahim 2012; Williams 2006).
Against this backdrop, social scientists have discerned markedly gendered patterns in individuals' self-identification as multiracial. Women tend to make up a disproportionate share of self-identified multiracials both in the United States and abroad (Davenport 2016; Renn 2012). Yet despite demographers' increasing attention to skewed sex ratios at birth (e.g., Guilmoto and Tovey 2015), this puzzling demographic imbalance has gone largely unexamined.
To explore the gendering of mixed-race enumeration, we draw on the Pew Research Center 2015 Survey of Multiracial Adults and introduce three innovations to previous investigations. First, building on Goldstein and Morning (2000), we distinguish two processes that may differ by gender: possessing an awareness of multiracial ancestry on one hand, and self-identifying as mixed-race on the other. Existing commentary has largely focused on the latter to the exclusion of the former, but we argue that both are at work. Second, we account for individuals' generational locus of mixed ancestry (Morning and Saperstein 2018) to investigate whether gender skews are most pronounced among people with recent or distant multiracial roots. Third, we examine a broader range of racial ancestries—not only the Black-White and Asian-White combinations most commonly discussed in U.S. scholarship—to determine whether gender differences are particularly salient for certain segments of the mixed-race population.
Our analytical distinction between multiracial ancestry awareness and self-identification reveals a striking and original finding: although American women are more likely than men to report a racially mixed family tree, women who report multiracial ancestry are less likely than men to identify personally with multiple races. Moreover, we find wide variation in rates of multiracial self-identification depending on both generation and racial ancestry: from highs above 50% among first-generation men and women with mixed Asian ancestry to a low of less than 20% among higher-generation women with mixed Black ancestry. We conclude that previous studies portraying multiraciality as a disproportionately female phenomenon did so primarily because they relied on self-identification without accounting for reported ancestry but also because many were limited to either first-generation multiracials, people of particular ancestry combinations, or both.
According to our data, the population of Americans with multiracial ancestry is second only to the monoracial White population in terms of size; thus, a full accounting of its diversity and sources is in order. As we show, analyses that rely solely on a subset, such as first-generation individuals with Black and White ancestry, lead to conclusions that are not borne out when we reckon with the broader population that reports mixed-race ancestry. Our more comprehensive analysis also raises important questions that can help future research better illuminate the workings of racial self-identification in the demographically heterogeneous society that is the contemporary United States.
Gendered Constructions of the Multiracial Population
Research to date has often maintained that women are more likely than men to label themselves as mixed-race and that women outnumber men in the multiracial population as a whole (Davenport 2016; Liebler and Zacher 2016; U.S. Census Bureau 2010). However, Liebler (2010) found that part-American Indian men are more likely to identify with multiple races than are women, and some studies revealed no gender difference in mixed-race labeling (Khanna 2004; Rockquemore 2002; Saenz et al. 1995).
A skew in research designs may itself convey the impression that women are more likely to identify as multiracial. Most studies of the psychological processes behind multiracial identity have focused on women's identity development (e.g., Henriksen and Trusty 2004; Kelch-Oliver and Leslie 2007; Mahtani 2014). Although social scientists have begun to attend to men's perspectives (Davenport 2018; Newman 2019; Sims and Joseph-Salisbury 2019), overall, little empirical effort has been made to understand why enumerating the mixed-race population is not a gender-neutral exercise.
To better understand the gendered mechanisms of multiracial identification, we build on Goldstein and Morning (2000) and Chandra (2012) by decomposing multiracial racial identification into a two-step process. First, one must possess an awareness of the genealogical history of racial mixture in one's ancestry (Goldstein and Morning 2000; Woo et al. 2011). In Chandra's (2012:14) terms, people must realize that they belong to the nominal pool of people who possess the “descent-based attributes” required for multiracial membership. One activates the nominal category by self-identifying as multiracial. Awareness of multiracial ancestry is thus a prerequisite for self-identifying as multiracial yet is distinct from it, and both processes could be gendered.
We expect that gender is associated with knowledge of mixed-race ancestry for two reasons. First, women primarily take on the kin work of documenting and sharing family histories and have been shown to possess more knowledge about them than their male relatives (di Leonardo 1987; Hackstaff 2010; Nash 2002). Second, gendered family socialization may also influence awareness of multiracial roots. Some scholars have suggested, for example, that women have stronger ties to their female relatives' racial identity and culture, and men have stronger ties to their male relatives' (Padilla 2006; Saenz et al. 1995), which could facilitate the differential transmission of genealogical knowledge.
Gender differences in the propensity to self-identify as multiracial also may be due to several factors. For one thing, men and women may experience different degrees of validation—that is, legitimation or acceptance by others—of their asserted multiracial identities (Brunsma 2006; Rockquemore and Brunsma 2002). For example, Rockquemore (2002) and Strmic-Pawl (2016) found that Black-White biracial women experience rejection and negative comments from monoracial Black women, but men do not pay an interpersonal penalty for identifying as mixed. On the other hand, scholars have argued that feminization of mulatto status (Giraldo et al. 2014) may lead women to embrace a mixed-race identity more than men. Finally, gender-specific experiences with racial discrimination may also contribute to the differential appeal of a multiracial identity (Phoenix and Tizard 2005; Renn 2012).
To better understand gendered multiracial identification, we also consider the genealogical locus of an individual's mixed-race heritage (Morning 2000), which critically organizes knowledge of and identification with multiraciality (Bratter 2007; Morning and Saperstein 2018; Song 2017). First-generation multiracials are more likely to be aware of and identify with their mixed ancestry, given that the members of the first interracial union in their family tree are immediate kin (Song and Gutierrez 2015). Moreover, distinctions by gender and generation could amplify one another: the more distant one is from the original interracial union(s), the more feminized genealogical work could shape one's awareness of multiracial ancestry (Hackstaff 2010; Nash 2002). Such gendered generational differences remain underexplored because previous research on multiracial identification has tended to rely on sampling respondents who report having parents of different races (e.g., Davenport 2016) or studying children born to parents who racially self-identify differently (e.g., Xie and Goyette 1997).
Racial ancestry combination is similarly salient in analyses of both awareness of multiracial roots and self-identification with multiple races, for at least two reasons. First, when it comes to developing awareness, people with different racial ancestry combinations likely hold different levels of interest in pursuing genealogy. For example, Asian Americans express relatively high rates of ancestral certainty of their family origins and are less likely than other groups to express interest in genealogical research (Horowitz et al. 2019). In contrast, African Americans may be particularly interested in tracing ancestry as a way to reclaim family history lost during the transatlantic slave trade (Nelson 2016), whereas many White Americans find a panethnic European identity to be bland and embrace genetic estimates of non-European heritage that make them feel distinctive (Roth and Ivemark 2018). Second, particular ancestry combinations are likely related to gender imbalances in multiracial identification because diverse gendered (and sexualized) connotations arose from the varied historical contexts that produced interracial unions (Charmaraman et al. 2014). For example, the contemporary meaning of Black-White multiraciality continues to be shaped by nineteenth century stereotypes associating the “tragic mulatta” with beauty and sexual transgression (Bost 2003:2; Mahtani 2014). Such race- and gender-specific “controlling images” may contribute to the availability and desirability of a multiracial identity for men and women, depending on their particular racial ancestry combination (Vasquez 2010:46).
In summary, existing scholarship points to a number of reasons why awareness of mixed-race ancestry and subsequent multiracial self-identification may differ by gender, generational locus, and racial ancestry. Although investigating specific mechanisms is beyond the scope of our data, we can derive and test some basic hypotheses. First, research on genealogy suggests that women are more likely than men to be aware of multiple-race ancestry. Second, existing scholarship has led many to expect women are more likely than men to self-identify as multiracial. Extrapolating from the feminized nature of genealogical work and kin-keeping, we also hypothesize that gender skews in multiracial self-identification could be amplified among higher-generation, mixed-race ancestry individuals. Finally, building on prior research suggesting that ancestry awareness is both gendered and racialized, we expect that asymmetries between awareness and self-identification might manifest differently in degree or direction for different ancestry combinations.
Data and Methods
We use data from Pew Research Center's 2015 Survey of Multiracial Adults, the first nationally representative sample to include self-reports of multiple generations of ancestry and measures of self-identification. Pew initially sampled 22,719 U.S. adults from an online research panel1 and asked for the races or origins of (1) themselves, (2) their mothers, (3) fathers, (4) grandparents, and (5) great-grandparents or earlier ancestors. Respondents selected one or more responses from the following categories: (1) White; (2) Hispanic, Latino, or Spanish origin; (3) Black or African American; (4) Asian or Asian American; (5) American Indian; (6) Native Hawaiian or Other Pacific Islander; and (7) some other race or origin.2 Pew used this self-identification and ancestry data to identify a smaller sample of multiracial adults for further questions.
Our analyses use data from the screening sample to distinguish genealogical awareness from self-identification, a crucial step to understanding the links between gender, generation, and claiming multiracial identity.3 We include only those respondents who completed all the preceding items (N = 22,616) because incomplete ancestry data prevents us from drawing conclusions about the presence and timing of mixed-race heritage.4
Outcome Measures: Multiracial Awareness and Self-identification
We identify individuals as aware of multiracial ancestry if they report more than one race across all ancestors (mother, father, grandparents, and great-grandparents or earlier ancestors).5 In the Pew data, nearly 20% of respondents reported at least two different races in their family trees (N = 4,330), significantly higher than the 2% to 3% of Americans categorized as multiracial by census self-identification (Jones and Bullock 2013). Descriptive statistics for the full sample and this subsample of respondents with multiracial ancestry are shown in Table 1.
We consider individuals to self-identify as multiracial if they chose more than one category in response to the question, “What is your race or origin?” Additionally, respondents who selected “some other race or origin” for themselves could specify more information, and we manually reviewed these write-ins (1.2% of the full sample). In some cases, the respondent seemed to signal a multiracial identity, such as “White and Black.” In other cases, when a respondent chose “some other race” as their second category, a manual review suggested that the respondent was not signaling multiracial identification. We recoded these cases as necessary.6
For both outcomes, we count “Hispanic, Latino, or Spanish origin” as a racial category, given that Americans tend to view the “Hispanic or Latino” label as commensurate with categories such as Black and White (Compton et al. 2010; Dowling 2014; Pew Hispanic Center and Kaiser Family Foundation 2002; Rodriguez 2000). Moreover, most Latino respondents in our sample stated that being Hispanic was part of their racial identity (see Gonzalez-Barrera and Lopez 2015). This coding decision yields a larger self-identified multiracial population than is typically reported by the U.S. Census Bureau (approximately 5% in the full sample), but we believe it also more accurately captures how Hispanic origins are racialized in the United States.7
Although this data set is the best available source of information on multiracial ancestry and identification in the United States, we acknowledge several limitations. First, the survey is cross-sectional and thus cannot speak to changes in multiracial identification over time (e.g., Doyle and Kao 2007; Hitlin et al. 2006). However, previous research has not found significantly different rates of race response change between women and men (see Liebler et al. 2017; Penner and Saperstein 2013), so we do not expect racial fluidity to explain the gendered patterns in our results. Second, we recognize that the meaning of race and multiraciality has changed over time; how a respondent classifies an ancestor may not match how the ancestor self-identified or was classified at an earlier point in time. Although we expect that respondents relied on present-day conceptions of multiraciality when responding, we also attempt to historicize these processes by attending to generational and age differences in our analyses. Finally, some respondents' ancestral knowledge may be incomplete because parts of family history can be intentionally omitted in genealogical accounts. It is also possible that some respondents were aware of having multiple racial origins but did not report all of them. In other words, these data show only who articulated awareness of multiracial ancestry; we cannot observe people who have multiple-race ancestry but are unaware of or do not articulate it. Even so, these data provide unprecedented analytical leverage for distinguishing multiracial ancestry awareness from self-identification.
Identifying Gender and Multiracial Generation
Our primary independent variable is gender. In these data, gender is measured by self-identification as male or female. Although binary sex categories do not necessarily correspond to people's gender identities, data limitations require us to assume that all males in the sample identify as men and all females as women, and we use these terms interchangeably throughout the paper. In the total sample, females slightly outnumber males, consistent with the U.S. adult population.
For respondents reporting mixed-race ancestry, we use the first interracial union in their family trees to ascertain the generational locus of multiracial ancestry (see Morning and Saperstein 2018). Respondents are categorized as first generation if they identified their parents with different single races (e.g., White mother and Black father) and associated their earlier ancestors with those same races. Everyone else who reported mixed-race heritage is considered second generation or higher; these respondents are not the first multiracial people in their families. In the full sample, just 2.7% of American adults are estimated to be first-generation multiracial, but 16.4% can be classified as second generation or higher.
We cannot clearly disaggregate the higher generations or track gendered patterns in the intergenerational transmission of genealogical knowledge because Pew did not ask about specific gender and race configurations of ancestral couples beyond the respondents' parents.8 Nevertheless, even separating the first from higher generations underscores the selection bias in previous research on multiracial identity that focused on people with different-race parents; first-generation multiracial Americans account for just 14.3% of all respondents who report multiracial ancestry (see Table 1).
From Racial Ancestry to Classification Regimes
Pew's national sample allows us to examine gendered patterns of multiracial identification across a broader range of ancestry combinations than most research in this area, which tends to focus on a few specific combinations. However, it includes hundreds of different ancestry combinations, many of which were selected by only one or two respondents. To aggregate the detailed reports into categories amenable to statistical analysis, we assign respondents to one of four nested and mutually exclusive racial classification regimes. The overall distribution of respondents across regimes can be found in Table 1.9
We use the term “racial classification regimes” (Gullickson and Morning 2011) to capture both structural and cultural dynamics associated with different ancestry combinations. Much of the salience of specific racial ancestry combinations for multiracial awareness and identification stems from the fact that they connote past social arrangements, even as these arrangements may be remade in the contemporary era. The possibility and prevalence of interracial unions, as well as the treatment of their offspring, were shaped by historical circumstances; policies, laws, and other institutions consolidated particular social rules for categorizing mixed-ancestry individuals by race, and changing political and economic conditions fostered certain types of interracial union at some times and not others. With this concept, we aim to encapsulate ways of both thinking about and acting on racial status with reference to mixed-race people of varying ancestral combinations.
Specifically, we identify four racial regimes by their focal ancestries, and we order them along a spectrum of historical classificatory supercession or recession (see Table 2). At one end, Blackness has traditionally been understood as dominant, as in the proverbial “one drop rule” (Davis 2010). At the other, indigeneity has been treated as vanishing, overpowered, and rendered invisible when combined with other racial heritage (Davis 2010; Forbes 1993). We use mutually exclusive coding to reflect this history (and for ease of interpretation), but nonmutually exclusive coding (e.g., assigning respondents who report Black and Hispanic ancestry to two regimes) produces similar results (see the online appendix).
What sets apart U.S. categorization regimes for part-Asian, part-Latino, and even part-White people is harder to establish because they have received much less scholarly attention to date. The position of individuals with some Hispanic ancestry is also complicated by the fact that Latinidad oscillates between being understood as a distinct race (Compton et al. 2010; Hollinger 1995; Rodriguez 2000) and “fundamentally multiracial” (Emeka and Vallejo 2011:1553). In the absence of clear predictions, we order the classification regimes for partial Hispanic or Asian ancestry, respectively, as “mostly dominant” and “mostly vanishing” along our gradient, with the rationale that Latinidad is the more stigmatized of the two racialized statuses (Bonilla-Silva 2004). Respondents who report only White or “some other race” ancestry (or both) comprise a fifth regime that serves as the reference group in our models. The end result is a regime variable that efficiently and comprehensively groups the myriad racial combinations our respondents reported in a way that reflects their historical import.
Table 3 presents the distribution of awareness and self-identification within each racial regime and demonstrates that the regimes capture key differences in the categorization of mixed-ancestry individuals. For example, respondents with Black or Indigenous ancestors are the most likely to be aware of mixed-race ancestry, but among those who report such awareness, they are the least likely to self-identify with multiple races. By contrast, fewer respondents with Hispanic or Asian ancestors reported mixed-race heritage, but those who did were more likely to self-identify as multiracial. Moreover, reading down the columns of Table 3, there is a decreasing percentage of aware respondents who identify with the race by which the regime is anchored. For example, 67% of aware respondents in the Black regime self-identify as only “Black or African American,” as a logic of “dominant” ancestry would predict, but just 4% of aware respondents in the Indigenous regime self-identify as “American Indian” or “Native Hawaiian.”10 These patterns echo previous research and underscore the importance of accounting for differences in racial ancestry in studies of multiracial identity (Gullickson and Morning 2011; Morning and Saperstein 2018).
We begin with descriptive results that reveal broad patterns of multiracial awareness and self-identification by gender and generation. Then, we use logistic regression to estimate the associations between gender and both ancestry awareness and self-identification while controlling for racial regime and other demographic factors. We use the full sample to model multiracial ancestry awareness (N = 22,616), but we fit self-identification models on the subsample of respondents who reported multiracial ancestry (N = 4,330). Our self-identification analyses also include the generational locus of one's multiracial ancestry, the interaction between gender and generation, and a three-way interaction between gender, generation, and racial regime.
Control variables were measured before the Pew survey as part of the annual demographic profile data collected for the online research panel. Region of residence is one of four census-designated categories: Northeast, Midwest, South, and West. Respondents' age is self-reported in years. Household income was presented in 19 categories on the questionnaire; we create a continuous measure by calculating the midpoint for each category, dividing by 1,000, and applying a log transformation. Educational attainment is an indicator differentiating respondents with and without a bachelor's degree. We also account for whether the survey was presented in Spanish.
We report unweighted counts throughout because Pew's post-stratification weights account for self-reported race and because our models control for other demographic factors included in weight calculations. However, results with weighted data are substantively similar to those shown. Including controls for respondent's nativity and marital status also does not change the main results (see the online appendix).11
Conventional accounting of multiraciality centers on people who self-identify with two or more races. However, this approach obscures the segment of the multiracial population aware of multiracial ancestry that does not self-identify as such. Figure 1 contrasts the conventional accounting (left) with our strategy of distinguishing between multiracial awareness and self-identification (right). Consistent with recent research on multiracial self-identification (e.g., Davenport 2016; U.S. Census Bureau 2010), the conventional tabulation yields a slight female skew, although it is not statistically significant in the Pew sample. Our approach reveals a submerged population of Americans who report mixed-race heritage but self-identify as monoracial (cf. Roth and Ivemark 2018), which is more than double the respondents who report multiracial ancestry and self-identify as multiracial. Our tabulation also shows that women are approximately 3 percentage points more likely to report multiracial ancestry than men, a difference that is statistically significant. Thus, rather than finding that women are more likely to self-identify as multiracial, we find that women are more likely to be aware of multiracial ancestry.
Our approach theorizes multiracial ancestry awareness as a prerequisite for multiracial self-identification. In other words, only people aware of mixed-race ancestry are “at risk” of self-identifying with more than one race. Under this model, the denominator for the rate of multiracial self-identification should not be the entire U.S. population but rather the subpopulation that reports having multiracial ancestry. When we change our denominators from all men and women to men and women aware of having multiracial ancestry, our perspective on the gender balance in these populations also changes (see Figure 2). Among men who report multiracial ancestry, 27.2% simultaneously self-identify with two or more races, compared with 24% of aware women. Notably, even as these relative proportions shift, the count of female respondents who self-identify as multiracial is still slightly larger than the count of male respondents who self-identify as multiracial because the number of women who are at risk of selecting multiple races is larger than the number of men. Thus, although women are more likely to be aware of multiracial ancestry, men are more likely to self-identify as multiracial conditional on awareness. Conventional accounting of multiracial self-identification alone misses this male skew.
Accounting for Generation
When previous research on multiraciality has considered ancestry, it has usually done so by studying the children of interracial unions. However, as noted earlier, just 14.3% of respondents who report mixed-race ancestry are first-generation multiracial; the rest trace their locus of mixed-race ancestry further back in the family tree. Previous research found that these generational differences have important implications for self-identification in general (Morning and Saperstein 2018), and we find that accounting for generation also clarifies previous inconsistent results regarding men and women's different propensities for multiracial self-identification.
As expected, first-generation women and men are more likely to self-identify with two or more races than their higher-generation counterparts. Contrary to expectation, however, we find that gender differences in self-identification run in opposite directions for first- and higher-generation individuals (Figure 2). First-generation women are more likely than first-generation men to self-identify as multiracial (44.6% vs. 36.6%; p = .052); comparatively, in higher generations, men are more likely to multiracially self-identify than women (25.6% vs. 20.7%; p < .001). That is, women exhibit greater generational attrition in multiracial self-identification. Rates of multiracial identification for women drop by about 54% (more than 20 percentage points) after the first generation, and rates for men drop by around 30% (10 percentage points). The larger female skew in the smaller first-generation population and the smaller male skew in the larger higher-generation population partially offset, yielding slightly more females among adults who self-identify as multiracial, consistent with findings from previous research. However, a focus on first-generation biracials in most previous self-identification research has obscured the differential attrition by gender across multiracial generations.
Modeling the Role of Gender in Multiracial Ancestry Awareness and Self-identification
We turn to multiple regression to assess whether the associations between gender and both multiracial ancestry awareness and self-identification might result from compositional differences in other sociodemographic factors. We first regress multiracial awareness on gender and our control variables (Table 4, Model 1a). Results indicate that the odds of reporting multiracial ancestry are about 9% higher for women than men, holding these demographic factors constant. However, when we also account for racial classification regime, we find that the female skew in multiracial awareness remains positive but is no longer statistically significant (Table 4, Model 1b). In fact, the other controls also are no longer significant predictors of awareness, with the exception of survey language. (Respondents who took the survey in Spanish are significantly less likely to report multiracial ancestry.) Regime estimates mirror the descriptive patterns (Table 3), including that people with Indigenous ancestry have by far the greatest odds of being aware of having kin of different races. This finding suggests that compositional differences across classification regimes account for the bulk of the observed variation in awareness of multiracial ancestry.
Next, we model the relationship between gender and multiracial self-identification among those who are aware of having multiracial ancestry (Table 4, Models 2a and 2b). Here, a significant male skew in self-identification remains after accounting for other measured factors: all else equal, women have about 17% lower odds of multiracial self-identification than men. We also find that respondents in the Hispanic and Asian regimes have the highest odds of self-identifying as multiracial, with all other measured factors being equal (echoing Table 3). Finally, estimates for age, region, and survey language indicate significant associations with self-identification. Older respondents and Spanish survey-takers have lower odds of self-identifying as multiracial net of other measured factors, whereas Southerners have lower odds compared with residents of the Northeast and the West. In sum, we do not find statistically significant gender differences in multiracial ancestry awareness once we account for racial classification regimes, but we do find significant gender differences in multiracial self-identification even when we control for regime and standard demographic factors.
Interactions Between Gender and Generation in Multiracial Self-identification
We now add generation to the previous model to assess whether a greater propensity for multiracial self-identification in the first generation can be explained by other compositional factors. After we account for generation, estimates for gender and racial regimes remain statistically significant and in the same directions (Table 5, Model 1). Consistent with Figure 2, we find that, all else being equal, odds of multiracial self-identification are about 42% lower for higher-generation multiracials compared with the first generation.
Descriptive patterns also suggest the overall male skew in multiracial self-identification is driven primarily by higher-generation men (Figure 2). Thus, we add a gender and generation interaction to assess whether the gender balance varies by generation holding other measured factors equal. We find that the interaction term is statistically significant, although the base terms for gender and generation cannot be distinguished from 0 in this sample (Table 5, Model 2). According to multiple goodness-of-fit statistics, accounting for generation improves model fit, and allowing gender differences to differ by generation explains even more of the observed variation in self-identification (Table 6).
To visualize the full range of gender and generation comparisons, we use coefficient estimates to simulate gender differences in the expected probability of multiracial self-identification by generation and racial regime, holding all controls at their median or mode (Figure 3).12 When the expected probability difference is less than 0, men are more likely to self-identify with multiple races; when it is greater than 0, women are more likely to self-identify as multiracial. Figure 3 shows that across our four regimes, expected probabilities of multiracial self-identification are 6–8 percentage points higher for first-generation women than first-generation men. By contrast, expected probabilities are about 4–7 percentage points higher for higher-generation men than higher-generation women. These gender differences are statistically significant for higher generations, but confidence intervals for estimates among first-generation respondents include 0 because of the larger error variance from smaller cell sizes. Nevertheless, the expected probabilities illustrate a consistent female skew in multiracial self-identification in the first generation and an opposite pattern for higher generations.
Racial Regime Differences in Gender and Generational Patterns
The previous models account for average differences by racial regime but do not estimate separate gender and generation coefficients for each regime. Yet, we hypothesize that the degree and direction of gender differences could vary by classification regime. To explore this possibility, we specify a final model that includes a three-way interaction for gender, generation, and regime (Table 5, Model 3). None of the three-way interaction estimates are statistically significant, likely because the data are too sparse to provide the necessary statistical power.13 Nonetheless, we believe that expected probabilities from this model merit consideration given their theoretical justification, their considerable variation, overall model fit, and the novelty of this data set.
When we allow gender differences to vary across regimes and generations, we observe a range of magnitudes for gender gaps in multiracial self-identification (Figure 4). For example, in the first generation, the female skew in multiracial self-identification ranges from about 14 percentage points for the Black regime to just 5 percentage points for the Hispanic regime. Among higher generations, the male skew in multiracial self-identification ranges from about 9 percentage points for the Hispanic regime to less than 1 percentage point for people with any Asian ancestry.14Figure 4 also illustrates how generational attrition in multiracial self-identification varies by regime and gender. For women with Black ancestry and for men and women with Asian ancestry, the probability of multiracial self-identification drops by more than one-half between the first and higher generations. In contrast, rates of multiracial self-identification increase slightly for higher-generation men in the Hispanic regime.
We urge caution in generalizing these results given the uncertainty of the estimates. Although we are confident in the finding that higher-generation men as a whole are more likely to self-identify as multiracial than higher-generation women, we are less confident about how this plays out differently for individuals with various ancestry combinations. At a minimum, though, our results indicate that a better understanding of the differential transmission of racial self-identification will require that future quantitative and qualitative data collection be designed with these potential differences by gender, generation, and racial ancestry in mind.
In this study, we highlight previously unobserved patterns shaping racial identification among U.S. adults by using novel data that allow for a key analytical distinction between awareness of multiracial ancestry and self-identifying with multiple races. Moreover, by studying a broader sweep of the multiracial population—beyond first-generation, “biracial” individuals of just a few ancestry combinations (e.g., Black-White, Asian-White)—we explore claims about links between gender and multiracial identity in unprecedented fashion.
In contrast to a generalized association between womanhood and multiracial self-identification, we show that this relationship obtains strongly under certain circumstances but not at all under others. Most importantly, we find that among Americans who are aware of having mixed origins, women are overall less likely than men to identify themselves with more than one race. Although first-generation women self-identify with multiple races at greater rates than first-generation men, among the bulk of the multiracial population that is rooted in more distant interracial unions, men are most likely to choose multiple races for self-identification.
We also show that multiracial ancestry awareness is shaped most strongly by racial ancestry and its associated classification regimes. We find that women are more likely to report mixed-race heritage than men overall, but these differences are statistically indistinguishable from 0 when we account for racial ancestry. Because there is no reason to expect gender skews in the composition of specific racial ancestries in the United States, this differential ancestry awareness points to potential gendered mechanisms—such as kin-keeping or the influence of controlling images—that make information about ancestors of different races more salient or accessible for some more than others. Of course, our measures of ancestry awareness and self-identification were collected at the same time point. Future research should examine whether multiracial ancestry reports are more stable than multiracial identification and whether ancestry response changes differ by gender.
Potential Explanations for Key Findings
With this more nuanced picture of the gender-multiraciality connection in mind, we now speculate about some causes of these relationships. We highlight four empirical questions that emerge from our analyses and offer preliminary hypotheses to explain the observed patterns. As described in the following paragraphs, these relate to differences in (1) awareness, (2) first-generation self-identification, (3) higher-generation self-identification, and (4) racial regimes.
Awareness: Why Do Men Underreport Multiracial Ancestry Relative to Women?
Previous research suggests that mixed roots are underreported because of the loss of genealogical information across generations (Song 2017). As a result, we assume that the overall gender differential in multiracial awareness reflects not an overreporting by women but rather underreporting by men. We suspect that this difference is grounded in the gendering of kinship work more broadly (di Leonardo 1987) and family history research more specifically (Hackstaff 2010). The feminization of family tree-keeping might be further amplified by gendered channels of transmission, reinforced through the transmission of genealogical information from mothers to daughters (Brunsma 2005; Padilla 2006; Zack 1994).
However, a person's racial identity also may influence how they report their ancestry. Our data cannot speak to this possibility because the Pew survey asked about self-identification before racial ancestry. Thus, the overall gender imbalance we observe in multiracial ancestry awareness could be attributed to men being more likely to omit aspects of family history that are not salient to their racial identities. Future study of the transmission, storage, and reception of genealogical knowledge—with special attention to both the gender and race of participants—would be a welcome resource in pursuing this question further.
First-Generation Self-identification: Why Are First-Generation Women Particularly Likely to Identify With More Than One Race?
Previous research speculates that multiracial self-identification is gendered because the consequences of phenotype and external judgments differ for men and women (Davenport 2016; Doyle and Kao 2007). This hypothesis is consistent with experimental evidence on racial classification (Ho et al. 2011) and qualitative research suggesting that appearance is a topic that biracial women confront more frequently than their male counterparts (Root 1994). The enduring association of racial hybridity with feminine beauty (Sims 2012) may also play a role. These factors could be especially salient in the first generation because phenotypical signs of mixed-race ancestry may become less visible with each passing generation. Future research should consider whether appearance and external judgments (e.g., Norman and Chen 2020) matter differently for mixed-race men and women, as well as how phenotype relates to self-identification for higher-generation multiracials.
A less explored explanation for first-generation women's tendency to self-identify with multiple races is the organization of their kinship ties. Morning and Saperstein (2018) hypothesized that the first generation is more likely to have social connections with the kin of multiple racial groups, whereas higher generations may have lost links to different-race kin. Given that women are disproportionately responsible for maintaining kinship ties (di Leonardo 1987), it also seems plausible that first-generation mixed-race women are the most likely to be active in family networks that cross racial boundaries.
Higher-Generation Self-identification: Why Are Higher-Generation Men Generally More Likely Than Higher-Generation Women to Identify With More Than One Race?
We hypothesized that gender differences in genealogical work might compound over generations, amplifying the female skew in multiracial self-identification, but our data do not support this supposition. The predominance of people of Indigenous descent among higher-generation multiracials suggests that imagery of American Indians might be particularly influential for second-plus-generation individuals' embrace of a mixed identity. Liebler (2010:138) referred to the gendering of such stereotypes by juxtaposing “Indian princess” versus “warrior” images, for example (see also Jacobs 1991), although it is unclear why one would be more appealing than the other. Here an inquiry into “ethnic options” (Waters 1990) or “racial appraisals” (Roth 2018) examining how higher-generation multiracial people evaluate their choices—and why men and women might perceive them differently—would be of real use.
Racial Regimes: Why Might Racial Ancestry Combinations Explain Differences in Awareness and Self-identification?
In many respects, the Black and Indigenous racial classification regimes that we employ exhibit comparable outcomes in self-identification, which differ from both Hispanic and Asian regimes. For example, the Black and Indigenous regimes yield relatively lower rates of multiracial self-identification and similar gender and generation patterns, whereas the latter two show the highest overall rates of multiracial self-identification. We expect that the distinction between Black and Indigenous outcomes and the patterns observed in other regimes stems from the role of large-scale immigration from Asia and Latin America in the wake of the 1965 Immigration and Nationality Act. Asian and Hispanic populations in the United States have greater foreign-born shares and shallower pools of higher-generation multiracials. Indeed, prior research using these data found that 62% of individuals reporting Asian-White ancestry were first-generation multiracials, as were 36% of people reporting Hispanic-White ancestry and 27% of those reporting Black-Hispanic ancestry; conversely, just 2% of individuals reporting Black-Indian ancestry, 8% of people reporting White-Indian ancestry, and 13% of people reporting Black-White ancestry identified their parents with different single races (Morning and Saperstein 2018).
We posit that the relative novelty of Asian—and to a lesser extent, Hispanic—mixture in the United States may subject multiracial Americans with those origins to different racial classification logics and concomitant gender stereotypes. Our findings align with Gullickson and Morning (2011) in suggesting a meaningful divide rooted in U.S. demographic history between this newer mixedness and Black, White, or Indigenous multiracialism. Rather than established classification regimes shaped by early and legally enforced norms of “mulatto” (Black-White) hypodescent and “mixed-blood” (Indian-White) hyperdescent, the treatment of Asian and Hispanic mixture is less clear-cut and more emergent.15 The differing degree of institutionalization for these regimes could impact how much mixed-ancestry matters for racial categorization, everyday interactions, and life chances.
For example, emerging and established regimes may differ in the valuation of mixed-race people. For starters, the newer regimes seem to allow for—and even privilege—multiracial self-identification, whereas the established regimes historically stigmatized mixed origins. The newer regimes' distinct treatment of multiraciality coincides with a cultural devaluation of Whiteness, which some younger people equate with mediocrity (Jiménez and Horowitz 2013). Equally important, the growth of U.S. Asian and Hispanic populations coincides with an era ushered in by 1967 legalization of interracial marriage nationwide, where mixed-race Americans have increasingly perceived recognition of their multiple origins as an ideal and a right (DaCosta 2007; Hernández 2018; Skerry 2002).
Stereotypes arising from emergent regimes also may be gendered differently than those associated with established regimes. We contend there is no part-Asian historical figure comparable to the (tragic) “mulatta” in the U.S. cultural imaginary, and the mestiza of yesteryear is not identical to the half-Latina of today (Bost 2003; Joseph 2013; Zackodnik 2004). To be sure, Asian and Hispanic Americans were not spared anti-miscegenation laws (Karthikeyan and Chin 2002; Menchaca 2008) and continue to confront racial stereotypes that are gendered and weighted with sexual inferences (Balistreri et al. 2015; Chou et al. 2015; Kim and Chung 2005; Lu and Wong 2013; Prasso 2005; Rodríguez 2008). However, those stereotypes do not invoke mixed-race people specifically, unlike images that circulated in French and Dutch Southeast Asian colonies (Saada 2011; Stoler 1989). In the absence of explicitly gendered symbols of Asian or Latina mixedness, there is little reason to expect multiracial self-identification would be more appealing to women than men (or vice versa).
Indeed, it is too soon to say what racial classification norms may form for Americans with mixed Asian or Hispanic ancestry—or whether any such complexes will consolidate. Yet the distinction we find between their self-identification patterns and those of Americans with mixed African or Indigenous ancestry speaks to the existence of different multiracialisms, with distinct logics and experiences (Strmic-Pawl 2016), grounded in the history of varied racialized groups' incorporation in the United States.
The Sociopolitical Context and Consequences of Multiraciality's Feminization
Multiraciality today is often interpreted in the public sphere as both proof and catalyst of racial progress. Mixed-race people have been cast in a role of bridging racial divides (Corrin and Cook 1999; Strmic-Pawl 2016), co-opted by the media to symbolize a post-racial society (Strmic-Pawl 2014), and even held up as proof that color-conscious anti-discrimination laws are no longer necessary or desirable (Hernández 2018). Less noted are the roles that the feminization and sexualization of multiraciality play in this post-racial narrative, even though some scholars have argued that femininized representation serves to soften the blow of White demographic decline (Bost 2003:2).
More generally, rendering multiraciality as feminine and beautiful seems to defang it as a potential political force. For example, recent interviews with mixed-race women documented a desire to find a middle ground that “actively works to avoid creating discomfort” stemming from “issues deemed political” (Buggs 2017:544). Others have suggested that the White mothers who led the multiracial movement of the late twentieth century depoliticized its agenda, prioritizing recognition of interracial families and multiracial children over Civil Rights–era antiracism that relied on mutually exclusive racial categories for identifying patterns of discrimination (Nakashima 2001; Williams 2006). Similarly, Ibrahim (2012:25) perceived gender at work in “a maternal multiracialism that neutralizes identity politics. . . . Civil rights enforcement is somehow reconfigured, through maternity, as antithetical to state recognition of personal liberty.” Thus, the portrayal of multiraciality by both popular media and its academic critics has been disproportionately feminine.
Still, the findings presented in this article suggest that the gendering of multiraciality is at once a more complex and more limited phenomenon than previous research has suggested. Prior associations of femininity with mixed identity are partially attributable to less comprehensive data but also may reflect a culturally circumscribed scope of inquiry wherein first-generation, Black-White individuals—often women—somehow seemed the natural choice for exploring multiraciality. We contend that the general exclusion of people with Indigenous or genealogically distant mixed-race ancestry reflects not only something about the American historical imaginary, obsessed with Black-White difference, but also something about contemporary multiracial symbolism. In other words, how mixed-race identities have been empirically investigated is revealing of the political and social work that feminized and sexualized multiracial imagery does today.
The gendering of multiraciality, by definition, is not a neutral narrative or iconography. Broader forces clearly shape what demographers, sociologists, and psychologists have so often approached as a matter of individual identity choices. The political exigencies and social resonance of a feminized multiraciality likely have repercussions not only for individuals' awareness of mixed-race ancestry and multiracial self-identification but also for how social scientists construct inquiries into such processes.
This article has greatly benefited from comments by Lauren Davenport, Taylor Hargrove, Catherine Lee, Wendy Roth, and Torsten Voigt. We would also like to thank Anna Boch for providing research assistance. Finally, we are grateful to staff at the Pew Research Center for discussions of the data set.
Respondents were drawn from GfK’s KnowledgePanel, which recruited through probability sampling (Pew Research Center 2015).
This measurement strategy is unique in asking sequentially about self-identification and racial ancestry along with using a combined-question format that includes Hispanic origin as a race response. The latter differs from recent census data recorded in a separate-question format. The former differs from the college surveys used by Davenport (2016), in which self-identification and parents’ backgrounds were recorded simultaneously in a grid format and the responses were described as “ethnic.” We do not expect the “race or origin” wording to affect our results because recent testing by the U.S. Census Bureau found that such measurement effects are small (usually on the order of 1–3 percentage points) and statistically insignificant (Mathews et al. 2017). Differences between combined- and separate-question formats also are not known to be related to respondents’ gender.
Data and code necessary to reproduce the analyses reported here are available on Dataverse (https://dataverse.harvard.edu/dataverse/MRgenerations).
Of the 103 omitted respondents, 48.5% were male. The full sample is 49.0% male.
This includes respondents who reported that their great-grandparents or earlier ancestors were a different race/origin but did not specify which (0.7% of sample) and excludes the small number of respondents (0.2% of sample) who reported multiple races for themselves but only one race among their ancestors.
Manual recoding changed 34 monoracial “other” cases to multiracial identification and 66 multiracial cases to monoracial. Common reasons for the latter recoding were that the “other” response was (1) some variation of “human” or “I don’t believe in race” or (2) a nationality (e.g., French, Mexican) encompassed by federal definitions of the first racial category they selected (OMB 1997). These verbatim write-in responses were provided to us in a restricted-use data file available to advisors on the Pew Study and are not part of the public-use data file.
The results are substantively similar when based on an alternative coding that counts “Hispanic or Latino” as a race only for respondents who explicitly stated that Hispanic was part of their racial—as opposed to ethnic—identity (see the online appendix).
We cannot distinguish the second and third generations with certainty because when respondents report multiple races for their grandparents collectively, they could be signaling that they had monoracial grandparents of different races (making them second generation) or any of their grandparents could be mixed race (making them third generation or higher). However, we can distinguish these second- and third-generation multiracials from fourth-generation-and-higher multiracials who reported the same single race for their parents and grandparents but indicated different-race earlier ancestors. This alternative coding also supports our conclusions (see the online appendix).
Our regime coding scheme is based on self-reported racial ancestry, so its distribution differs from the U.S. racial composition presented in census statistics based on self-reported race.
Office of Management and Budget (OMB) standards identify Native Hawaiian or Other Pacific Islanders (NHOPI) as a distinct category, but we could not create a NHOPI regime from the small number of respondents who report NHOPI ancestry. We group NHOPI-ancestry respondents with American Indian–ancestry respondents based on their shared indigeneity; Native Hawaiians previously petitioned OMB to be counted with other Indigenous people (Espiritu and Omi 2000). However, our main conclusions are similar if we use alternative categorization schemes, such as placing NHOPI-ancestry respondents with Asian-regime respondents (results available upon request).
We do not include these factors in our main models because nativity data are available for only a nonrandom subsample of respondents (about 14%).
We use the expected probability of multiracial self-identification rather than the predicted probability to show the average change in the dependent variable given change in a particular explanatory variable, rather than show how far the outcome could deviate from expectation due to unmodeled random factors. In practice, the average expected value from simulations is very similar to the predicted value, but the former’s variance is lower (King et al. 2000).
We posit this explanation because the estimates for which we have the largest samples (e.g., higher-generation Indigenous) also have smaller standard errors.
The gender difference for higher-generation, Black-regime respondents is smaller if we classify people with Black-Hispanic ancestry in the Hispanic regime instead of the Black regime. The small male skew among higher-generation individuals with Asian ancestry is larger if NHOPI-descent individuals are included in the Asian regime (results available upon request).
Collectively, our results suggest the Hispanic regime not only differs from the Black and Indigenous regimes but also differs somewhat from the Asian regime. Latinidad today is associated with both the oldest and the newest roots in North America, with both a very long and a very recent experience of racial mixture. This might help to explain why, for example, higher-generation Hispanic multiracials have the lowest generational attrition in self-identification (whereas those in the Asian regime have the highest).