tendency toward hypodescent when made to believe that Black people would pose a greater socioeconomic threat in the future (Ho et al., 2013). Individual endorsement of opposition to equality, a construct that Jost and Thompson (2000) argue contributes to SDO, statistically mediates the relationship between political conservatism and hypodescent (Krosch et al., 2013). SDO and political conservatism may impact racial categorization of multiracials because political conservatives or those higher in SDO mentally represent multiracials differently than their counterparts. Evidence for this possibility comes from research on essentialism, the belief that category memberships are innate and immutable (Hirschfeld, 1998). Essentialist beliefs have been shown to promote hypodescent in both children and adults. In one study, White children were asked to memorize a set of computer-generated Black, White, and racially ambiguous Black-White multiracial faces (Gaither et al., 2014). White children who scored higher on race essentialism remembered White faces significantly better than Black or Black-White faces, suggesting that they viewed the racially ambiguous faces as Black. By contrast, children who scored lower on racial essentialism remembered White and Black-White faces significantly better than Black faces, suggesting that they considered the racially ambiguous faces White, and by virtue of the sample, members of the ingroup. Although the study could not fully rule out whether children who scored higher in race essentialism perceived Black-White faces as Black as opposed to "not White," the results demonstrate that essentialism moderates categorization of these racially ambiguous faces. In adults, a similar face-memory paradigm showed that monoracial and multiracial participants who endorsed more essentialist beliefs about human traits showed significantly worse memory for outgroup faces (Pauker & Ambady, 2009). Together, these findings indicate that individual differences may moderate how people separate others into racial categories by tuning the way faces are perceived and represented in the minds of observers. These findings are relevant to our current investigation because essentialism is associated with the endorsement of traditionally conversative policies (Roberts et al., 2017).

A second route by which political orientation may impact the racial categorizations of multiracial people involves downstream cognitive or motivational processes (Chen, 2019;Ho et al., 2020). As an example, Chen and Hamilton (2012) observed that participants (the majority of whom were White) placed under cognitive load were less likely than nontaxed controls to categorize a multiracial face as such. A parallel effect was not observed in categorizations of monoracial targets, suggesting that higher-order cognitive processes (i.e., those that are disrupted under cognitive load) play a greater role in the categorization of multiracial than monoracial people. In a recent meta-analysis, Jost (2017) reported that political conservativism was significantly, albeit modestly, correlated with lower need for cognition. If categorizing multiracial people requires more deliberation, cognitive control, and effort, politically conservative individuals may thus be less inclined to make such judgments. Motivational factors, such as egalitarian motives, may also play a role. Individuals who scored higher on internal motivation to control prejudice (Plant & Devine, 1998), for example, were more likely to classify multiracial faces as multiracial than those lower on internal motivation to control prejudice (Chen et al., 2014; see also Hugenberg & Bodenhausen, 2004). Prejudice and essentialism may also work together to influence multiracial face categorization (Ho et al., 2015). Other research has shown that opposition to equality mediates the link between political conservatism and the categorization of racially ambiguous faces (Krosch et al., 2013). These data underscore the importance of considering both political conservatism and racial prejudice in understanding racial categorization of mixed-race faces. Such a connection arguably manifests in political strategizing (e.g., the "Southern Strategy," in which the Republican Party aligned itself with White voters over anti-Black sentiment to gain a political foothold in the South); ongoing voter suppression that overwhelmingly targets people of color; and political rhetoric through which Republicans covertly communicate racist sentiments (Aistrup, 2014;Combs, 2016;Haney-López, 2015).

Extant data thus reveal two possible mechanisms by which political orientation could relate to multiracial categorization. First, conservatives and liberals could mentally represent multiracials differently. As a concrete example, conservatives may judge Black-Asian multiracial people as more perceptually similar to Black people, whereas liberals may perceive the same targets as more similar to Asian people. Mead et al. (2009, see also Kemmelmeier & Chavez, 2014) provide some support for this prediction within the context of the 2008 Presidential election. In their research, politically liberal participants were more likely to rate a lightened photo of Obama as representative of Obama compared to conservatives who were more likely to rate a darkened photo of him as representative. This effect persisted even after controlling for implicitly and explicitly measured prejudice. These data suggest that there may be a link between low-level perceptual processes related to racial categorization and one's political orientation. Testing whether political orientation impacts categorical representation of multiracial individuals requires perceptually mapping how multiracial faces are psychologically represented by observers, which was one of the goals of the current research.

A second, orthogonal mechanism involves higher-order psychological processes. For example, conservatives and liberals may perceive and mentally represent multiracials similarly, but other motivations or cognitions may cause them to categorize multiracials differently, leading to a disconnect between perception and categorization. Conservatives may be more concerned with maintaining stringent categorical boundaries (Jost et al., 2003) or expend less energy individuating others and think more categorically (Pacini & Epstein, 1999). A study by Kruglanski et al. (2006) found that conservatives were more likely to seek out cognitive closure and preferred simple, unambiguous answers to questions. By contrast, liberals were more likely to prolong cognitive closure and consider alternate viewpoints and perspectives when making social judgements (Jost, 2017;Mccrae & Costa, 1997;Sparkman & Eidelman, 2016). For liberals, this may translate to greater comfort with unconventional categories, such as a multiracial identity, and could also lead to more time spent categorizing racially ambiguous individuals. The current research seeks to explore both low-level perceptual processes and higher-order processes in the mental representation and categorization of Kamala Harris against the backdrop of the 2020 US Presidential election. We bridge recent research mapping the mental representation of multiracials with longer-standing research examining explicit categorization of multiracial faces.

Multiracial face perception research has focused heavily on the explicit categorization of the faces of self-identified multiracial individuals or artificial faces created to appear racially ambiguous (Chen & Hamilton, 2012;Gaither et al., 2019;Ho et al., 2011;Ma et al., 2021;Peery & Bodenhausen, 2008). This work suggests three general possibilities for how perceivers might racially categorize Kamala Harris. The first general finding (referenced above) reveals that multiracial targets are categorized in terms of hypodescent (Gaither et al., 2019;Ho et al., 2011;Young et al., 2020). We note that the prevalence of this finding in the literature may be partly attributable to the fact that many early studies focused on hypodescent, due to the historic treatment of mixed-race individuals in the United States. That said, we could reasonably predict that perceivers would categorize Harris as Black, given her own self-identification as a Black woman (Harris, 2019). A second pattern of data suggests that multiracial individuals are categorized as members of racial or ethnic categories with which the person has no immediate ancestral ties. As an example, Black-White biracial peoples are frequently categorized as Hispanic/Latino or Middle Eastern (Chen et al., 2018;Ma et al., 2021;Maclin et al., 2009;Nicolas et al., 2019). This may be due to perceptual similarities between multiracial individuals and Latino and Middle Eastern individuals, perceived demographic base rates (Chen et al., 2018), or other factors. Finally, there are limited data suggesting that multiracial individuals can be categorized as multiracial, which are referred to as concordant classifications. Although it is reasonable to predict that perceivers may categorize Harris as multiracial, given her unwavering public self-identification as Black and Indian (Harris, 2019), empirical support for this outcome is not overwhelming in the literature. Concordant categorization of multiracials is fairly poor, ranging from 5% to 60% (e.g., Chen et al., 2018;Chen & Hamilton, 2012;Nicolas et al., 2019). In our recently published Multiracial Expansion of the Chicago Face Database (Ma et al., 2020), we took digital photographs of 88 individuals who self-reported mixed-race ancestry. Norming data showed that perceivers only categorized these faces as multiracial approximately 10% of the time; the single face from the database that received the most multiracial categorizations was only judged as such by 45% of perceivers. However, data do suggest that greater exposure to multiracials in everyday life increases concordant multiracial categorizations (Chen et al., 2018;Pauker et al., 2018).

Importantly, Kamala Harris is a widely known public figure; the stimuli in most race categorization experiments are novel to participants. Thus, the effect of familiarity with the multiracial individual being categorized is not well established in the literature. One notable exception was provided by Citrin et al. (2014), who asked participants, "How do you think Obama should have filled out his race on his Census form?" Participants were randomly assigned to three conditions: one in which participants were not given any information, a second in which they were told about Obama's mixed-race ancestry, and a third condition in which they were told about Obama's mixed-race ancestry and were told that Obama identified as Black on his Census form. Across all three conditions, over half of respondents indicated that he should report Black and White. Consistent with this finding, the Washington Post described a study conducted by the Pew Research Center examining perceptions of Obama's race. The majority of respondents reported viewing him as mixed race, but these responses varied significantly by perceiver race. Whites and Hispanics were much more likely to describe him as mixed race (over 50%), while Black respondents were the only demographic to describe him as Black more than half the time (55%). These results are in line with findings by Ho et al. (2017;Study 3), who found that Black perceivers tend to be inclusive of Black-White biracial people in the ingroup, believing that others will categorize and treat them as Black people. We can speculate that the control condition in the Ho et al. (https://www.doi.org/10.4135/97814129851302017) study, which focused on Black participants, may provide an estimate for how Black perceivers might categorize Kamala Harris in the absence of any experimenter-provided primes. That said, Obama has a Black-White biracial ancestry, whereas Harris has dual-minority ancestry as a Black-Asian person. Meta-analytic data (Young et al., 2020) suggest that a multiracial person's racial composition matters for categorization and this same meta-analysis highlighted the real dearth of research on dual-minority multiracials.

Available data suggest that perceivers struggle to categorize multiracials as such. However, explicit categorizations may be limited in terms of what they can reveal about how multiracials are mentally represented by others, in part because they subsume initial perception and additional psychological processes that may not have to do with perception per se (e.g., expectancies, demand, top-down influences, etc.). In an effort to isolate perception of multiracials from downstream mechanisms, our lab recently utilized multidimensional scaling (MDS; Ma, Kantner, Dunn, & Benitez, in preparation; see also Kruskal & Wish, 1978;Shepard, 1962). The MDS paradigm asks participants to judge the similarity/dissimilarity of pairs of stimuli. These pairwise judgments are then used to generate a similarity matrix, which reflects the perceptual similarity of any two stimuli within the judgment set. These scores can be used to place stimuli relative to each other in a physical, n-dimensional space. Stimuli that are highly similar to one another cluster together in this space, while dissimilar stimuli are more distant from one another. The number of dimensions in the space reflects the number of stimulus features spontaneously used by participants in rendering their similarity judgments. Much like in factor analysis, where observed factors must be interpreted and labeled in psychological terms, the latent factors underlying the unspecified dimensions in MDS can be identified from correlating dimension scores with known variables, and/or by visually inspecting stimuli across the range of scores on a given dimension. Whether stimuli are clustered or separable with a categorical boundary also provides an important piece of information in MDS analyses. When an MDS plot displays two clusters that are clearly separable, the case can be made that the stimuli belong to distinct categories. In the current research, we used MDS to determine whether the location of Harris's face in psychological space (relative to Black and Indian monoracial faces) differs between conservatives and liberals.

The current research builds on the existing multiracial face perception literature in two important respects. First, we test the possibility that political orientation moderates multiracial classification due (at least in part) to low-level differences in how multiracial faces are perceived and mentally represented. To test this, we derive and compare perceptual mappings of Kamala Harris relative to Black and Indian monoracial faces in conservative and liberal participants. Second, we explore whether and how perception relates to the categorization of multiracials by bridging recent research on mental representation and explicit categorization. Given her high-profile status, the significant controversy surrounding her racial identity throughout the election, and her significance to contemporary American politics, we focus on the mental representation and racial categorization of Harris in the days leading up to the US Presidential election in November of 2020. We chose to test these hypotheses immediately prior to the election because previous research demonstrates that political orientation is especially salient during election periods, a time at which political partisanship is also most consequential (Kemmelmeier & Chavez, 2014).

We recruited 240 participants from Amazon's Mechanical Turk (mTurk). The targeted sample size was based on research indicating the point at which correlations stabilize (Schönbrodt & Perugini, 2013) and previous analyses characterizing the political orientation of the mTurk population as liberally skewed (Huff & Tingley, 2015). Given the political composition of the sampled

Perceptual mapping: After providing informed consent, participants completed the similarity judgment task. For this task, participants were shown pictures of female faces and asked to indicate their similarity using a nine-point scale (1 = Very dissimilar; 9 = Very similar). Stimuli were selected from the Chicago Face Database (CFD; Ma, Correll, & Wittenbrink, 2015) and Indian Face Set expansion of the CFD (Lakshmi et al., 2021) and included five Indian faces, five Black faces, and one image of Harris. All targets were female and were selected to be approximately the same age as Harris (see Figure 1). These 11 stimuli yielded 55 unique pairs of faces, which were presented to participants in a random order. We did not present same-face pairs to participants. Participants used their personal devices to complete the study, so screen size and image sizing 1 Study 1 was preregistered at the Open Science Framework. Materials and raw data files for all three studies can be found at osf.io/pf3v8. We affirm that all the data and all pertinent measures and conditions of the study are reported and described here. Data were not excluded from our reporting. varied across participants. Image size ranged from 290 pixels wide × 204 pixels high to 800 pixels wide × 562 pixels tall and the aspect ratio was constrained across participants.

Identification and Explicit Categorization: Following the similarity judgment task, participants were shown images of Kamala Harris, Elizabeth Warren, and Condoleezza Rice one at a time and were asked, "Who is this political figure?". These data were used to determine whether participants could identify Harris, but we included Warren and Rice to partly obscure the focus of the investigation. Participants were given the following response options for categorizing Harris: Kamala Harris, Condoleezza Rice, Stacey Abrams, Maxine Waters, and Elizabeth Warren.

We included two explicit categorization measures of Harris, embedded among racial categorizations of Warren and Rice. The first measure asked participants to complete a forced-choice categorization of Harris/Rice/Warren using a set of fixed labels: Asian, Black, Latina, White, or Multiracial. The second categorization measure asked participants to indicate the extent to which they personally viewed Harris/Rice/Warren to be a member of four different groups: Black, White, Indian, and Multiracial. Although Harris has no immediate White ancestry, we showed all the participants the same four sliders across Harris, Rice, and Warren for consistency. To respond, participants placed sliders at any integer between 0 and 100. Responses on each slider were orthogonal and could total more than 100, meaning participants could indicate that Harris was Black, White, Indian, and Multiracial. An image of each politician was displayed above the measure.

Individual Differences: The study included two individual difference measures. The first was political orientation, which was measured with a one-item, 8-point scale asking participants to indicate their political orientation (1 = Extremely Liberal; 2 = Very Liberal; 3 = Moderately Liberal; 4 = Slightly Liberal; 5 = Slightly Conservative; 6 = Moderately Conservative; 7 = Very Conservative; 8 = Extremely Conservative); however, due to a coding error, all conservative responses were coded as Slightly Conservative, which forced us to dichotomize this variable. Participants who responded 1-4 were coded as liberal and those responding 5-8 were coded as conservative. We also included a measure of self-reported prejudice toward Black people with the Modern Racism Scale (α = .910; McConahay, 1980), based on past research indicating its association with hypodescent (Chen et al., 2014).

Before debriefing, we asked participants one additional question as part of a pilot for a study on multiracial categorization. This question asked participants to indicate how much a target's ancestry, physical appearance, and personal identification of their own race/ethnicity impacts their decision to categorize someone as multiracial.

Below, we present the results from all the participants, including those who did not accurately identify Harris (n = 35). Notably, the pattern of results was highly similar regardless of whether we excluded those who did not correctly identify her. Only one test statistic (which was not among the critical analyses) changed in significance. We footnote that test statistic below.

To address the question of whether conservative and liberal participants mentally represented Harris differently from each other, we created separate perceptual maps for those who reported being liberal (n = 116) and those who reported being conservative (n = 108). Within these two groups, we collapsed across participants to create a matrix of average similarity ratings. The similarity matrices were then submitted to multidimensional scaling using the PROSCAL function in SPSS. For conservatives, a two-dimension solution was associated with a dispersion accounted for (DAF; a measure of the variance accounted for in the similarity ratings) of .9896. As shown in Figure 2, the two-dimension solution reveals a distinct cluster of Indian faces and a separate cluster of Black faces, which were separable by a linear boundary. This suggests that conservative participants had distinct mental representations for Black and Indian faces. Harris was placed between the Black and Indian faces on Dimension 1 and was higher than all the other faces on Dimension 2. This indicates that she was perceptually discriminable from both Black and Indian faces. We had no evidence to believe she was perceived in a manner consistent with hypodescent (i.e., she was not represented as more proximal to Black than Indian faces). Critically, the results were very similar for liberal participants (see Figure 2). A two-dimension solution corresponded with a DAF of .9917. As with conservatives, liberals had discrete, separable representations of Black and Indian faces and placed Harris in virtually the same position in face space.

Next, we examined participants' explicit categorizations of Harris using a forced-choice categorization task (see Table 1). To analyze these data, we conducted a multinomial logistic regression.

We regressed participants' categorization on political orientation (0 = Liberal; 1 = Conservative).

Multiracial was specified as the baseline comparison group for the analysis, because it was the modal response and because we sought to compare other classifications against multiracial categorizations. The full model fit was significantly better than the empty model, indicating that political orientation predicted multiracial categorizations, χfoot_0 (4) = 24.022, p < .001, ɸ = .307. Parameter estimates correspond with relative log odds of being categorized as Asian/Black/Latina/White relative to multiracial. Contrary to what we might expect, a significant effect of political leaning only emerged in the categorization of Harris as White relative to multiracial. The relative log odds of categorizing Harris as White versus multiracial corresponded with a 1.617 increase moving from liberal leaning to conservative leaning (p < .001).

The association between categorizations of Harris as White and political orientation was unexpected. To better understand this effect, we conducted an exploratory analysis testing anti-Black prejudice in the model. We regressed participants' categorization on political orientation, prejudice, and the political orientation × prejudice interaction. We again specified multiracial as the reference group. The full model fit was significantly better than the empty model, χ 2 (12) = 109.030, p < .001, ɸ = 0.668. Although the overall model was significant, only prejudice emerged as a significant predictor, χ 2 (4) = 57.587, p < .001, ɸ = 0.486. Neither the political orientation, χ 2 (4) = 5.535, p = .237, ɸ = 0.151, nor the political orientation × prejudice interaction, χ 2 (4) = 6.069, p = .192, ɸ = 0.158, were statistically significant. Thus, political orientation was no longer a significant predictor of relative categorizations. Prejudice emerged as a significant predictor when assessing the relative risk ratio of categorizing Harris as Asian versus multiracial, B = 1.730, p < .001; Latina versus multiracial, B = 1.375, p = .003; and White versus multiracial, B = 2.187, p < .001. We also observed a significant political orientation × prejudice interaction, B = -1.059, p = .045 2 in the relative categorization of Harris as Asian versus multiracial. Higher levels of anti-Black prejudice were associated with greater likelihood of categorizing Harris as Asian relative to multiracial, but this effect was reversed for liberal-leaning participants.

It is important to note that there were few observations in some of the cells (e.g., very few participants categorized Harris as Latina), which can result in unstable estimates in multinomial logistic regressions (Tabachnick & Fidell, 2007). For this reason, we also conducted a binary logistic regression analysis focused only on categorizations of Harris as multiracial and White (0 = multiracial; 1 = White) as a function of political orientation (0 = Liberal; 1 = Conservative). Consistent with the multinomial modeling results, the effect of political orientation was statistically significant, B = 1.617, p < .001. When prejudice was entered into the model at step 2, the effect of political orientation was no longer significant, B = 2.026, p =.312, but prejudice was, B = 2.065, p <.001. The political orientation × prejudice interaction was not significant, B = -0.558, p = .364.

To determine whether the explicit judgment of Harris as White was also observed on the sliding scale measure of racial/ethnic categorization, we regressed participants' 0-100 ratings of Harris as White on political orientation and observed a significant effect, t(242) = 5.112, p < .001, sr = 0.312 (Table 2). The average rating of Harris as White among conservatives was 44.2% (SD = 34.1) compared to 23.0% (SD = 29.9) for liberals. Once we accounted for the effects of prejudice and the political orientation × prejudice interaction, however, the effect of political orientation was no longer significant, t(240) = 1.160, p = .247, sr = 0.064. Prejudice was significant in the model, t(240) = 6.059, p < .001, sr = 0.336, but the political orientation × prejudice interaction was not, t(240) = -0.773, p = .465 sr = -0.041. These findings were thus highly consistent with the results emerging from the forced-choice categorization task.

Study 1 offered an initial test of the effect of political orientation on the perception and categorization of Kamala Harris. We found no evidence that conservatives and liberals differ in how they perceptually represent Harris's face relative to monoracial Indian and Black faces. Judgments in two different explicit racial categorization tasks, on the other hand, did vary as a function of political orientation, but not as predicted. Although participants often categorized Harris as multiracial, more conservatives categorized her as White than multiracial. We thus observed a dissociation between perception and explicit categorization: while liberals' and conservatives' perceptual similarity ratings suggested that they mentally represented Harris similarly, their explicit race classifications of her differed. These results suggest that differences in categorizations between conservative and liberal participants were not perceptually driven. This conclusion was further supported by the fact that differences in categorization as a function of political orientation appeared to be explained by anti-Black prejudice. One alternative explanation for these effects concerns participants' familiarity with Harris as a function of political orientation. Many states employ a "closed primary system" that only permits registered party members to vote on candidates from their political parties. Thus, it is entirely possible that liberal participants in our sample had more exposure to Harris, having observed her in early Presidential debates and throughout the Democratic primary campaign. This familiarity may have translated to more awareness of her ancestry, while the relative lack of familiarity among conservatives may have led conservatives to guess Harris's race. These guesses could conceivably be informed by the base rate (perceived or actual) of White versus non-White politicians in the Federal government. Although this explanation seems plausible, the results were very similar regardless of whether we included participants who did not accurately identify Harris in the analysis. Additionally, this interpretation is also difficult to square with the fact that participants completed the forced-choice categorization and slide ratings while being shown an image of Harris, providing an immediate perceptual basis for racially categorizing her. Moreover, conservatives and liberals did not differ in their mental representations, which were based on similarity ratings using the same image of Harris. A familiarity-based explanation for the observed categorization differences becomes even harder to reconcile with the fact that anti-Black prejudice played a significant role in explaining categorizations. Although it is possible that greater exposure to Harris translated to more positive attitudes toward her (Zajonc, 1968), the measure we included targeted attitudes toward Black people and was not a measure of attitudes toward Harris in particular.

The finding that conservatives were more likely to categorize Harris as White is curious, given she has no immediate White ancestry. Additionally, all participants were presented with her image as they categorized her. Before exploring potential explanations for these results, we first sought to replicate this finding.

The primary goal of Study 2 was to replicate the findings from Study 1 and determine whether conservatives' rate of categorizing Harris as White in that study was an anomaly. We also reasoned that participants may have categorized and rated her as White because they believed she was Middle Eastern. Indeed, previous research has demonstrated that participants sometimes label multiracial participants Middle Eastern when asked to racially categorize targets using a free response format (Nicolas et al., 2019). To address this possibility, Study 2 allowed participants to categorize Harris as Middle Eastern, in addition to the other categories provided in Study 1. Finally, because similarity judgments are heavily influenced by the particular stimuli used, we changed the picture we used of Harris (see Figure 1) in order to determine whether the scaling solution obtained in Study 1 was unique to the image used in that study.

We recruited 263 participants from mTurk (157 males, 105 females, 1 non-binary). The average age of the sample was 36.79 years old (SD = 12.087) with a range of 18-78 years old. Participants included 193 White, 30 Black, 18 Latinx, 13 Asian, and nine Biracial/Multiracial participants. Participants were compensated $1.25. Data were collected in the two days prior to the election.

The procedures and measures were identical to those of Study 1, with two exceptions. First, we changed the image we used for Harris in both the similarity judgment and categorization tasks (see Figure 1). Second, we added the category label Middle Eastern to the forced-choice explicit categorization measure.

As in Study 1, we present the results from all the participants, including those who did not accurately identify Harris (n = 50). Results were consistent regardless of whether these participants were included or excluded from the analyses.

We again dichotomized participants as more liberal leaning (n = 143) or conservative leaning (n = 120). Within these groups, we collapsed across participants to create two matrices of average similarity ratings. The similarity matrices were then submitted to multidimensional scaling using the PROSCAL function in SPSS (see Figure 3). For conservatives, a two-dimension solution was associated with a DAF (a measure of the variance accounted for in the similarity ratings) of .9927. Visual inspection revealed a very similar scaling solution to those seen in Study 1: there were distinct, fully separate clusters for Black faces and Indian faces, suggesting that conservative participants had distinct mental representations for Black and Indian faces. As in Study 1, Harris was located between the Black and Indian faces on Dimension 1 and was higher than all the other faces on Dimension 2. This result indicates that she was perceptually discriminable from both Black and Indian faces and constitutes no evidence for hypodescent. For liberal participants, a two-dimension solution corresponded with a DAF of .9912. As with conservatives, liberals had discrete, separable representations of Black and Indian faces. Critically, as in Study 1, liberals and conservatives placed Harris in the same region of the similarity space.

We first conducted a multinomial logistic regression, regressing participants' categorizations on political orientation (0 = Liberal; 1 = Conservative; see Table 3); however, a number of specious effects emerged from the data, suggesting that the model was unstable, likely due to small numbers of certain categorizations (e.g., Latina and Middle Eastern). We thus conducted a binary logistic regression focusing only on categorizations of Harris as multiracial and White (0 = multiracial; 1 = White). There was a significant effect of political orientation, B = 2.083, p < .001. In a follow-up analysis, we included prejudice (α = .911) and the political orientation × prejudice interaction at step 2. Unlike in Study 1, the effect of political orientation remained significant, B = 7.334, p = .005. Conservatives were still more likely to categorize Harris as White, even after controlling for prejudice and the political orientation × prejudice interaction. Those higher in anti-Black prejudice were more likely to categorize Harris as White, as indicated by a main effect of prejudice, B = 3.007, p < .001. Finally, the political orientation × prejudice interaction, B = -2.212, p = .007 was also significant. Those who were conservative and higher in prejudice were especially likely to categorize Harris as White.

The second measure of racial categorization revealed a conceptually similar pattern of results (see Table 4). Conservative participants' ratings of Harris as White (M = 42.3, SD = 35.0) were significantly higher than liberals' ratings (M = 23.0, SD = 27.9), t(261) = 4.886, p < .001, sr = 0.289. However, this effect was no longer significant, t(259) = 0.248, p = .804, sr = 0.015, once prejudice, t(259) = 6.102, p < .001, sr = 0.355, and the political orientation × prejudice interaction, t(259) = -0.072, p = .943, sr = -0.004, were entered into the model.

Consistent with Study 1, we found that conservative and liberal participants mentally represented Harris as distinct from Indian and Black face exemplars, and separate MDS solutions for conservative and liberal participants bore a striking resemblance to each other. However, when we examined the explicit categorizations of Harris, conservatives and liberals responded very differently. Specifically, we replicated the finding that conservatives were more likely than liberals to categorize Harris as White in forced choice format and rated her as more White in independent slider ratings. These effects were at least partially explained by anti-Black prejudice. When we accounted for anti-Black prejudice in the forced-choice explicit categorization of Harris, the effects of political orientation remained significant; however, the effect of political orientation was no longer significant when anti-Black prejudice was included in the sliding scale ratings of Harris as White.

Studies 1 and 2 provided evidence that participants, regardless of political orientation, mentally represented Black faces, Indian faces, and Harris's face nearly identically. Although the explicit categorization of Harris showed that participants often categorized her as multiracial, conservatives were significantly (and curiously) more likely to categorize her as White. This suggests that differences in categorization according to the political orientation of the observer are not likely to be perceptual in nature, but rather driven by top-down processes. What top-down influence(s) might produce frequent categorizations of Kamala Harris as White by conservative participants? We reasoned that conservatives' motivation to categorize Harris as White could be driven in part by the desire to downplay her racial identity, because conservative participants could perceive her race as a strong asset to her candidacy, enhancing the overall attractiveness of the Biden-Harris ticket. Study 3 tested this possibility by asking participants to rate the perceived value of different candidate features, including candidate race, in increasing the chances of electoral success for the ticket. Of particular interest was whether conservatives generally rated being non-White as valuable to a Democratic presidential ticket.

We obtained data from 245 participants on mTurk (164 males, 81 females; 177 White, 32 Black, 16 Asian, seven Latinx, six Native American, six multiracial, and one other). Participants were compensated $1.25. The sample was slightly liberally skewed (122 indicated they were liberal, 111 indicated that they were conservative, and 12 indicated that they were Other). Data for the study were collected one week following the inauguration of Joe Biden and Kamala Harris.

After providing consent, participants were asked to indicate whether various candidate attributes would harm or help a presidential ticket. Attributes included being White, non-White, male, female, Christian, rich, attractive, heterosexual, homosexual, a career politician, from the East Coast, a lawyer, and a US Senator. We were specifically interested in ratings of being White and non-White, but included other attributes to obscure the purpose of the study. Participants provided separate judgments for a Democratic Presidential ticket and a Republican Presidential ticket using

We excluded the 12 participants who reported Other as their political orientation from the analysis, which left 233 participants in the sample. Our analysis focused on ratings of how harmful versus helpful having a White and non-White candidate was to a Democratic/Republican presidential ticket. One-sample t-tests (comparing scores to a value of 4, the neutral point on the scale) showed that participants, on average, believed that having a White candidate on the ticket enhances electability (M = 5.28, SD = 1.14), t( 232

Study 3 provided some evidence that conservatives and liberals differed in their relative weighting of how candidate race could impact support for a presidential ticket. As predicted, conservative participants believed that having a non-White candidate on a ticket would help both Democratic and Republican tickets more than liberal participants did; however, conservatives rated the positive impact of a non-White candidate higher for Democratic tickets. This perception could help explain why participants in Studies 1 and 2 categorized Harris as White, perhaps in an effort to deny their political opponents this seeming advantage. At the same time, liberal participants also viewed having a non-White candidate as more advantageous for Democrats than Republicans and could be motivated to categorize her as non-White. Although this is a plausible explanation for liberals' tendency to categorize Kamala Harris as White less often than conservatives, it does not fully explain why liberal participants generally categorized her as multiracial and not disproportionately as any of the other non-White categories.

Within the context of a socially and historically significant US Presidential election, we collected data related to perceptions and categorizations of Kamala Harris, the first multiracial individual to receive a major-party Vice Presidential nomination. Our research extended previous work showing that political orientation correlates with racial categorizations of multiracial faces (Ho et al., 2020;Krosch et al., 2013). We explored two possible mechanisms that could account for this relationship. The first explanation posited that conservatives and liberals mentally represented multiracials differently. This possibility was grounded in research showing that memory of multiracial faces differed according to essentialist beliefs (Gaither et al., 2014) and racial bias (Chen et al., 2014;Krosch et al., 2013), as well as work showing that conservatives and liberals differed in their responses to ambiguity (Kruglanski et al., 2006) and category maintenance (Jost et al., 2003). Across two studies, we showed that Harris was perceptually distinct from both Black and Indian individuals and that this was true regardless of perceivers' political orientation. This null effect suggested that low-level perceptual differences in how conservative or liberal participants processed faces were not driving the relationship between political orientation and racial categorizations; put another way, conservatives and liberals did not perceive her face differently from each other. A second mechanism linking political orientation to categorization implicated downstream psychological processes. Given the striking similarity between conservatives and liberals in the perceptual mapping data, downstream processes emerged as more likely candidates for any differences in their racial categorization patterns. Using two separate measures, we showed that Harris was indeed categorized differently by conservatives and liberals. Much to our surprise, conservative participants categorized and rated Harris as White more often than liberal participants, who were more likely to categorize Harris as multiracial. This unexpected effect was explained partially (or fully in some cases) by individual differences in anti-Black prejudice, lending support to the notion that downstream processes can explain the relationship between political orientation and multiracial categorization. The fact that Harris was categorized as White by so many conservatives merits greater attention. At the outset of this research, we could not have anticipated this finding. As we described, some prominent conservatives explicitly denied that Harris is Black (e.g., Benton, 2020), but we anticipated, based on the existing literature, that she would be categorized according to hypodescent (Ho et al., 2011;Young et al., 2020). We can imagine a number of mechanisms that might motivate conservative individuals to categorize Harris as White. Study 3 tested one possibility. Preliminary evidence based on this study suggested that conservatives may view Harris's racial identity as a disadvantage to their political party, which may guide their classifications of her as White. The current data do not allow a direct test of this account, because categorization data were collected in a separate study, but one could envision follow-up research testing whether individual differences in the perceived value of Harris's racial identity correspond with racial categorizations of her. Of course, other factors could play a role in the observed racial categorization patterns of Studies 1 and 2. For example, Harris is highly successful, affluent, and powerful-all traits that are typically associated with being White (Fiske et al., 2002). She is also a politician, which continues to be a White-dominated profession. Whether consciously or not, it is possible that conservative individuals rely on these traits to infer her Whiteness. To us, this poses an interesting possibility, because such an explanation hinges on a cognitive, rather than motivational, mechanism, and is worth additional consideration given that prevailing models of multiracial categorization interpret the influence of political orientation in terms of motivation (Ho et al., 2020). In addition to this potential avenue for future research, better understanding how participants' own racial backgrounds figure into these categorizations merits additional scrutiny. Here, participants who were conservative and especially prejudiced against Black people were more likely to say that a political opponent who identifies as Black was White. Given that conservatives are disproportionately White (Pew Research Center, 2014), this finding is especially counterintuitive and speaks to the possibility that participants were juggling multiple, conflicting motivations in their judgments.

It is important to remember that although we have focused on Harris's multiracial ancestry throughout this paper, Harris strongly identifies as a Black woman. Despite this, relatively few participants categorized her as Black. This discrepancy between her own identification and perceivers' classification of her represents a serious issue for many multiracial individuals. By virtue of their mixed-race status, multiracial individuals straddle multiple racial groups, yet commonly report difficulty gaining acceptance within these different groups because of how others categorize them (Gaskins, 1999). As an example, Chen et al. (2019) found that Asian Americans were more likely to categorize Asian-White biracials as outgroup members, a tendency related to perceived discrimination and expectations on the part of Asian Americans that Asian-White biracials would not be loyal to Asians. At the same time, Albuja et al. (2019) report that Whites also appear to deny and question the racial category membership of mixed-White biracials. Moreover, biracial individuals reporting higher levels of identity denial and identity questioning were also more likely to experience stress and depressive symptoms. In addition to (or perhaps as a result of) these exclusionary categorizations, multiracial people have difficulty developing a strong ethnic identity (Coleman & Carter, 2007;Lusk et al., 2010;Udry et al., 2003), which may lead to negative outcomes, as researchers have established that a strong ethnic identity confers protection to monoracial people of color in the face of discrimination and prejudice (Jones et al. 2007;Sellers et al. 2003). These data may partly explain why multiracial people experience higher levels of negative mental health outcomes (e.g., depressive symptoms, alienation, anxiety; Binning et al. 2009) than both monoracial Whites and monoracial minorities (Cheng & Lively, 2009;Fisher et al. 2014). In the context of the current findings, denying Harris her Black, Asian, and multiracial identities may serve to undermine her personhood more generally and perhaps be a "trolling" tactic.

The current studies offered a first attempt at connecting the perception of a multiracial face with the explicit categorization of the same face. Whereas research has examined how multiracial faces are mentally represented using perceptual mapping (Ma et al., in preparation), past research paradigms did not include participants' explicit race categorization of faces, which precluded testing on whether and how perception relates to categorization within the same perceivers. Although the current studies do not provide evidence that perception informed categorization (at least not for conservative participants), such a relationship could have been eclipsed by Harris's notoriety, leading to results driven by a unique set of parameters (Ho et al., 2020). Future work examining novel targets outside of a highly partisan, contentious context is needed to further explore a potential link between perceptual mapping and explicit categorization in everyday judgment contexts.

There is little doubt that the multiracial population in the United States will continue to grow over time. Some estimate that by 2050, 20% of Americans will identify as multiracial (Goo, 2015). As the multiracial population continues rising, we can expect to see greater representation of mixed-raced individuals in American politics. Understanding how these individuals will be received by the electorate represents an opportunity for theory building and application. Further, documenting how these views shift over time could offer an important window into the American consciousness and direction of our nation.