This is an open access article distributed under the terms of the Although many White Americans believe racial minorities act as a cohesive group socially and politically (Knowles et al., 2022), dynamics among minority groups are more complex. A growing body of literature finds members of marginalized groups often show solidarity with other marginalized groups (e.g., Craig & Richeson, 2012; Dixon et al., 2015; Klavina & van Zomeren, 2020; Pérez et al., 2023; Starzyk et al., 2019) but not always (e.g., Capers & Smith, 2016; Craig & Richeson, 2016; McClain et al., 2006; Ramakrishnan et al., 2009).
The common ingroup identity model provides a potential mechanism for intra-minority solidarity. Identification with a common ingroup can improve attitudes toward (Gaertner et al., 1989) and increase willingness to help (Dovidio et al., 1997) former outgroup members included in the common ingroup. Many of the intra-minority solidarity findings suggest common ingroup identification could be the underlying mechanism (Burson & Godfrey, 2020; Craig & Richeson, 2012; Dixon et al., 2015; Glasford & Calcagno, 2012; Pérez et al., 2023).
But as a form of prejudice reduction between majority and minority groups, common ingroup identification could have an unintended consequence of reducing minority group members’ support for collective action or policies that would improve their group’s situation (e.g., Dixon et al., 2010; Ufkes et al., 2016). Whether this “ironic effect” could impact intra-minority solidarity largely remains to be explored.
Additionally, minority groups can share common ingroup identities excluding the majority group (e.g., POC identity; Pérez, 2020) and identities including the majority group. Yet research to date has not explicitly explored this distinction.
The present research uses two 3-wave panel studies to examine how common ingroup identities in the United States that do and do not include the dominant racial group (American identity, and person of color or POC identity, respectively) relate to Asian, Black, and Hispanic/Latino Americans’ political solidarity with their own and other racial minority groups, operationalized as attitudes toward policies associated with these groups. While identification as American might reduce support for policies that benefit racial minority groups, POC identification could increase support.
Intra-Minority Solidarity
The intra-minority solidarity research to date indicates that shared experiences of stigma increase solidarity (Cortland et al., 2017; Craig & Richeson, 2012, 2016; Pérez et al., 2023). However, priming shared values (Burson & Godfrey, 2018) or shared contributions to American culture (Pérez et al., 2022) can also increase solidarity, suggesting that shared identity, not stigma, might be key. But this research does not test identification directly, leaving open whether participants in fact saw themselves as part of a common ingroup with the target groups and whether this common ingroup identification drove solidarity.
What common ingroup identity might drive intra-minority solidarity in the United States? Many Americans who are members of racial minority groups identify as people of color (POC; Pérez, 2020; see also descriptive statistics from Results, below). The term POC was first used by African American activists and later adopted by other activists as a collective identity for Black, indigenous, Latino, Asian, and other minority groups in the United States, though some debate remains about the utility and boundaries of the label (Pérez, 2020; Starr, 2023). Although this history might lead one to expect differences in the degree of POC identification (between and within groups), POC is a readily available collective identity that includes many racial/ethnic minority groups in the U.S. (Starr, 2023).
Empirical evidence suggests something like POC identity contributes to intra-minority solidarity. Among Asian Americans, linked fate with people of color predicted support for the Black Lives Matter movement (Merseth, 2018), and perceived commonality with Black and Latino Americans predicted support for a path to citizenship for undocumented immigrants (Samson, 2015). Although distinct from identification, linked fate and perceived commonality capture the idea that one’s group has something in common with other marginalized groups. Furthermore, Pérez (2020) found a correlation between POC identification itself and support for Black Lives Matter among Black, Latino, and Asian Americans. However, that study did not examine common ingroup identities that include the dominant group.
Ironic Effects vs. Collective Action
The “ironic effects” literature suggests that intergroup contact and common ingroup identity involving the dominant group can reduce subordinate group members’ recognition of inequality (e.g., Dixon et al., 2010; Tropp et al., 2012; Ufkes et al., 2016) and, consequently, their support for collective action or policies that could improve their own group’s status (Carter et al., 2019; Dixon et al., 2010; Saguy et al., 2009; Ufkes et al., 2016).
This contrasts with the role of (sub)group identity in collective action. Under the social identity model of collective action (SIMCA), identification with a disadvantaged group predicts collective action directly and indirectly via injustice (perceptions of unfairness or discrimination and group-based anger) and efficacy (van Zomeren et al., 2008).1
Parts of SIMCA appear as mechanisms in the ironic effects literature. Lower group-based anger and efficacy (Ufkes et al., 2015, 2016), perceptions of ingroup disadvantage/injustice (Cakal et al., 2011; Saguy et al., 2009), and perceived ethnic discrimination (Carter et al., 2019; Tropp et al., 2012) were found to mediate ironic effects of prejudice reduction. These effects could stem from lower identification with the subgroup (Ufkes et al., 2016). But social identity theory suggests collective action also depends on perceiving the dominant group as a relevant outgroup for comparisons (Tajfel & Turner, 1979); common ingroup identification could suppress collective action by removing the dominant group as a comparison outgroup. Empirically, Ufkes et al. (2015) found that both subgroup and superordinate group identification predicted own-group collective action tendencies, in opposite directions. Separately, research on the group-value model of justice suggests that high superordinate group (e.g., American) identification could pacify members of disadvantaged subgroups (e.g., African Americans) even when paired with high subgroup identification (Huo et al., 1996; Smith & Tyler, 1996). Thus, superordinate group identification itself could produce ironic effects.
Hypothesis 1: Identification as American will predict less support for policies or groups that benefit racial minorities.2
Although existing literature demonstrates ironic effects on support for policies benefitting one’s own group, effects may extend to policies benefitting other minority groups, as American identification could also reduce recognition of these groups’ disadvantages compared to the White majority.
But what about a common ingroup that does not include the dominant group? POC identity could fit into SIMCA (van Zomeren et al., 2008) as a disadvantaged group identity that draws attention to POC’s disadvantages compared to Whites. Furthermore, as a shared identity, POC identification could draw attention to not only one’s own group’s disadvantages but also other minority groups’ disadvantages, increasing support for policies benefitting other minority groups. Thus, SIMCA and the intra-minority solidarity literature suggest:
Hypothesis 2: POC identification will predict more support for policies or groups that benefit racial minorities.
Panel Study Design
Although some research has experimentally primed the salience of existing superordinate identities (e.g., Górska & Bilewicz, 2015; Ufkes et al., 2016), most ironic effects research (Reimer & Sengupta, 2021) and much of the collective action and group consciousness literatures (e.g., Cakal et al., 2011; Kessler & Mummendey, 2002; Merseth, 2018; Samson, 2015) have been correlational. Correlational studies on common ingroup identity and intergroup contact tend to be cross-sectional (see Reimer & Sengupta, 2021), with rare (though increasing) exceptions (e.g., Górska & Tausch, 2023; Kotzur & Wagner, 2021). Thus, it is unclear whether group identification precedes attitude change—a key assumption underlying the common ingroup identity and intra-minority solidarity literatures, which suggest that identity causes attitude change. Accordingly, I used 3-wave panel studies to clarify the time-order dynamic of identity and solidarity.
However, identity has been treated as a stable individual difference in parts of the literature (see Huddy, 2001). When variables show trait-like stability, it can be difficult to establish direction of effects (cf. Kessler & Mummendey, 2002), and effect estimates in traditional cross-lagged panel models (CLPM) can be biased because they confound between-person and within-person effects (Hamaker et al., 2015). Hamaker et al. (2015) recommend random-intercepts cross-lagged panel models (RI-CLPM) to distinguish stable, between-person relationships from cross-lagged within-person relationships. But CLPM and RI-CLPM answer somewhat different questions (Orth et al., 2021). The present studies use CLPM and RI-CLPM to answer two questions: 1) Do individual differences in group identification predict changes in policy attitudes (CLPM)? 2) Do shifts in individuals’ group identification predict changes in the individuals’ policy attitudes (RI-CLPM)? The latter is especially important because the intra-minority solidarity and ironic effects literatures theorize that increasing individuals’ common ingroup identification will impact solidarity. Figure 1 illustrates these models for Hypothesis 2.
Figure 1
Hypothesis 2 in a Cross-Lagged Panel Model (CLPM; A) and Random-Intercepts Cross-Lagged Panel Model (RI-CLPM; B)
Note. Hypothesized paths are in bold.
Study 1
Study 13 focuses on Asian Americans, whose political attitudes are relatively understudied (see, e.g., Tran & Curtin, 2017). Asian Americans have historically been “triangulated” with respect to other minority groups and Whites (i.e., their status in society is defined in comparison to both Whites and Blacks; Kim, 1999) and portrayed as opposing policies like affirmative action that benefit minority groups (see, e.g., Wu, 2014). Yet they have also engaged in activism on behalf of Asian Americans (Lee, 2015; Wu, 2014) and in solidarity with other racial minority groups (e.g., Lee, 2015). As a result of this history, Asian Americans may vary greatly in POC identification and support for policies benefitting other racial minorities.
This study examined three issue areas associated with other minority groups: immigration, criminal justice reform, and affirmative action. Although immigration could be an own-group issue for Asian Americans, anti-immigration rhetoric in the U.S. since 2016 has focused primarily on people from Latin America (e.g., the border wall and migrants from Central America). Criminal justice reform has been linked to the Black Lives Matter movement (e.g., Ghandnoosh, 2015), and the death of George Floyd in 2020 would plausibly have made this connection salient. Affirmative action has been portrayed as benefitting other minority groups at the expense of Asian Americans (Wu, 2014), notably in a high-profile lawsuit against Harvard (Students for Fair Admissions, 2019).
Method
Study 1 was administered online through Qualtrics from November 9-11, 2020 (Wave 1), November 30 to December 4, 2020 (Wave 2), and December 28, 2020, to January 2, 2021 (Wave 3). This study was determined to be exempt by the University of Minnesota IRB. The design and analyses were not preregistered.
Participants
Participants were Asian American adults recruited through Prolific (prolific.co). Power analyses for RI-CLPM using Monte Carlo simulation in Mplus (Muthén & Muthén, 1998-2019, 2002) indicated a target sample size of 500-600 completing all three waves. Allowing for 25% attrition across successive waves, Wave 1 target sample size was set at 1000 (see Supplementary Materials and Table S20).
Of the 1006 participants who consented to take part in the study, 17 were excluded from analysis because they did not self-identify as Asian. Excluding an additional underage participant left a final Wave 1 sample size of 988 (488 male, 480 female, 20 nonbinary), of whom 787 returned for Wave 2 (20% attrition) and 651 returned for Wave 3 (17% attrition from Wave 2; 34% overall). See Figure S1.
Most participants self-identified only as Asian, though a small minority (35) identified as Asian and another race/ethnicity. The most common subgroups were Chinese (N = 409), Vietnamese (N = 155), Indian (N = 113), Filipino (N = 109), and Korean (N = 106). Most (98%) were U.S. citizens, and the majority (77%) were born in the U.S. Participants were well-educated (60% had at least a 4-year college degree), with relatively high incomes (median = $70,000-79,000). See Table S21 for additional demographics.
Measures
The study measured 1) identification as Asian American, person of color (POC), and American and 2) attitudes toward immigration, criminal justice reform, and affirmative action.4 Measure order was counterbalanced within each wave.
Identity
Group identification was assessed using the 4-item Huddy, Mason, and Aarøe (2015) scale (e.g., “How important is being [identity] to you?”) adapted for each identity of interest (α = .84-.92).5
Policy Attitudes
Immigration attitudes were assessed using three items (α = .73-.74): 1) whether the participant thinks the number of immigrants allowed into the U.S. should be (1 = decreased a lot; 5 = increased a lot); 2) support for deporting undocumented immigrants versus allowing them to stay (1 = strongly favor deportation; 4 = strongly favor allowing them to stay); and 3) attitude toward detaining undocumented migrants (1 = strongly oppose; 5 = strongly favor; reverse-coded).
Criminal justice reform attitudes were measured using three items (α = .67-.74): 1) whether the participant thinks there are too few (1), about the right number (2), or too many people (3) in prison in the U.S.; 2) attitude toward defunding police and redirecting funds to alternative first-responder services (1 = strongly oppose, 5 = strongly favor); and 3) attitude toward eliminating mandatory minimum sentences for some crimes (1 = strongly oppose, 5 = strongly favor).
Affirmative action attitudes were measured using two items (Ramakrishnan et al., 2018) addressing employment and higher education (1 = strongly oppose, 5 = strongly favor; α = .85-.88).
Demographics
Finally, the Wave 1 survey asked participants to self-report their race/ethnicity, specific Asian ethnic group, gender, age, education level, family income, whether they are a U.S. citizen, whether they were born in the U.S. and if not, in what year they came to live in the U.S., and preference for speaking English versus another language. Participants were asked if they are registered to vote; whether they voted in any past U.S. election; and their political ideology (1 = Extremely liberal, 7 = Extremely conservative) and party affiliation (branching items, collapsed into a 7-point scale; 1 = strong Democrat, 7 = strong Republican).
Exact wording and response options for all items in both studies are in the Supplementary Materials.
Analyses
All items were recoded to range from 0-1. Composite scores were created by averaging the items for each construct in each wave.
For each identity-attitude pair, CLPM and RI-CLPM were fit with and without stationarity of lagged and cross-lagged effects (constraining the path coefficients from Wave 1-2 and their corresponding path coefficients from Wave 2-3 to equality). If stationarity significantly worsened fit, results are reported from the model without stationarity; otherwise, results are reported from the model with stationarity. As robustness checks, additional CLPM were fit with paths from each demographic and political covariate described above to the Wave 2 and 3 identity and attitude variables. See Supplementary Materials (Part 3 and Tables S6-S11) for further details, fit statistics, and results from CLPM with covariates.
Data were analyzed using R, version 4.3.1 (R Core Team, 2023) and the lavaan package (Rosseel, 2012) v0.6-15. Model parameters were estimated using full-information maximum likelihood estimation, with pairwise deletion for missing data.
The anonymized data and analysis code are at https://doi.org/10.17605/OSF.IO/7JDXF
Results
Descriptive Statistics
Descriptive statistics, correlations, and α for the identity and policy attitude variables are in Tables S1-S2. Average American identification was above the midpoint (M = .589-.606), and average POC identification was around the midpoint (M = .490-.504), with some variability in both (SD = .207-.263). POC and American identification were moderately correlated with Asian American identification (POC: r = .35-.39, American: r = .23-.26). Identities (r = .76-.81) and policy attitudes (r = .80-.84) were highly stable.
Hypothesis 1: American Identification
Parameter estimates and fit statistics are in Table 1. For all analyses in both studies, RI-CLPM showed good model fit; CLPM did not.
Table 1
American Identification Panel Model Results (Study 1)
| Immigration | Criminal Justice | Affirmative Action | ||||
|---|---|---|---|---|---|---|
| CLPM | RI-CLPM | CLPM | RI-CLPM | CLPM | RI-CLPM | |
| T1 ID → T2 ID | .783*** (.017) .773 | .058 (.116) .061 | .786*** (.017) .776 | -.050 (.143) -.049 | .788*** (.017) .777 | .118 (.100) .115 |
| T1 attitude → T2 ID | -.039* (.017) -.038 | .071 (.090) .066 | -.022 (.016) -.023 | .212 (.163) .188 | -.017 (.012) -.022 | .077 (.065) .094 |
| T1 ID → T2 attitude | -.031* (.015) -.031 | .047 (.073) .053 | -.019 (.016) -.017 | .088 (.142) .089 | -.006 (.020) -.005 | .184† (.098) .165 |
| T1 attitude → T2 attitude | .809*** (.015) .811 | .047 (.100) .048 | .838*** (.016) .806 | -.193 (.223) -.176 | .819*** (.015) .803 | -.028 (.116) -.032 |
| T2 ID → T3 ID | .783*** (.017) .786 | .058 (.116) .059 | .786*** (.017) .789 | .134 (.098) .126 | .788*** (.017) .792 | .118 (.100) .123 |
| T2 attitude → T3 ID | -.039* (.017) -.038 | .071 (.090) .066 | -.022 (.016) -.024 | -.023 (.092) -.021 | -.017 (.012) -.022 | .077 (.065) .087 |
| T2 ID → T3 attitude | -.031* (.015) -.032 | .047 (.073) .056 | -.019 (.016) -.018 | .065 (.089) .059 | -.006 (.020) -.005 | .184† (.098) .175 |
| T2 attitude → T3 attitude | .809*** (.015) .826 | .047 (.100) .052 | .838*** (.016) .840 | .288** (.089) .253 | .819*** (.015) .837 | -.028 (.116) -.029 |
| ID-attitude trait covariance | -.008*** (.001) -.232 | -.007*** (.001) -.186 | -.005** (.002) -.121 | |||
| CFI | .923 | 1.000 | .934 | 1.000 | .921 | .999 |
| RMSEA | .178 | .000 | .163 | .000 | .179 | .028 |
| SRMR | .041 | .011 | .041 | .000 | .043 | .017 |
Note. Table reports lagged and cross-lagged coefficients, RI-CLPM identity-attitude trait-level covariance, and model fit statistics. Unstandardized coefficients are reported with standard errors in parentheses and standardized coefficients in italics. All models except RI-CLPM for criminal justice assume stationarity of lagged/cross-lagged effects. N = 988.
†p < .10. *p < .05. **p < .01. ***p < .001.
American identification showed no significant within-person cross-lagged effects in the RI-CLPM but had a significant, negative cross-lagged effect on immigration attitudes (b = -.031, SE = .015, p = .038, β = -.031 and -.032), and vice versa, in the CLPM.
RI-CLPM trait-level covariances were negative and significant for American identification and immigration (c = -.008, SE = .001, p < .001, r = -.232), criminal justice (c = -.007, SE = .001, p < .001, r = -.186), and affirmative action attitudes (c = -.005, SE = .002, p = .002, r = -.121), indicating that those who identified more strongly as American tended to hold less racially liberal attitudes.
Hypothesis 2: POC Identification
Parameter estimates and fit statistics are in Table 2.
Table 2
POC Identification Panel Model Results (Study 1)
| Immigration | Criminal Justice | Affirmative Action | ||||
|---|---|---|---|---|---|---|
| CLPM | RI-CLPM | CLPM | RI-CLPM | CLPM | RI-CLPM | |
| T1 ID → T2 ID | .788*** (.016) .793 | .147 (.111) .153 | .805*** (.016) .809 | .122 (.129) .126 | .760*** (.022) .768 | .105 (.089) .112 |
| T1 attitude → T2 ID | .080*** (.021) .063 | -.003 (.098) -.002 | .032† (.019) .027 | .027 (.172) .021 | .110*** (.021) .115 | .109 (.071) .120 |
| T1 ID → T2 attitude | .073*** (.012) .093 | .002 (.059) .002 | .051*** (.013) .059 | .007 (.105) .009 | .106*** (.024) .102 | .177* (.079) .196 |
| T1 attitude → T2 attitude | .782*** (.016) .783 | .043 (.095) .044 | .827*** (.016) .794 | -.125 (.190) -.116 | .751*** (.023) .754 | -.056 (.108) -.063 |
| T2 ID → T3 ID | .788*** (.016) .791 | .147 (.111) .142 | .805*** (.016) .807 | .175 (.110) .165 | .799*** (.025) .796 | .105 (.089) .098 |
| T2 attitude → T3 ID | .080*** (.021) .064 | -.003 (.098) -.002 | .032† (.019) .028 | -.039 (.114) -.030 | .041† (.024) .043 | .109 (.071) .099 |
| T2 ID → T3 attitude | .073*** (.012) .095 | .002 (.059) .003 | .051*** (.013) .059 | .043 (.079) .048 | .042† (.025) .040 | .177* (.079) .185 |
| T2 attitude → T3 attitude | .782*** (.016) .800 | .043 (.095) .048 | .827*** (.016) .829 | .269** (.089) .243 | .832*** (.023) .824 | -.056 (.108) -.056 |
| ID-attitude trait covariance | .019*** (.002) .446 | .014*** (.002) .295 | .027*** (.002) .472 | |||
| CFI | .942 | 1.000 | .946 | .999 | .939 | .999 |
| RMSEA | .161 | .009 | .152 | .063 | .237 | .022 |
| SRMR | .035 | .012 | .035 | .013 | .036 | .012 |
Note. Table reports lagged and cross-lagged coefficients, RI-CLPM identity-attitude trait-level covariance, and model fit statistics. Unstandardized coefficients are reported with standard errors in parentheses and standardized coefficients in italics. All models except criminal justice RI-CLPM and affirmative action CLPM assume stationarity. N = 988.
†p < .10. *p < .05. **p < .01. ***p < .001.
POC identification had a significant positive cross-lagged effect on affirmative action support in the RI-CLPM (b = .177, SE = .079, p = .026, β = .196 and .185).
In the CLPM, POC identification had significant positive cross-lagged effects on immigration attitudes (b = .073, SE = .012, p < .001, β = .093 and .095) and vice versa; criminal justice attitudes (b = .051, SE = .013, p < .001, β = .059); and affirmative action attitudes (b = .106, SE = .024, p < .001, β = .102), and vice versa, from Wave 1-2. Effects remained significant for immigration and affirmative action but not criminal justice when covariates were added (Table S11).
Trait-level covariances were positive and significant for POC identification and immigration (c = .019, SE = .002, p < .001, r = .446), criminal justice (c = .014, SE = .002, p < .001, r = .295), and affirmative action (c = .027, SE = .002, p < .001, r = .472). Figure 2 illustrates the POC identification-affirmative action attitude models.
Figure 2
CLPM (A) and RI-CLPM (B) for POC Identification and Affirmative Action Attitudes (Study 1)
Note. Unstandardized and standardized (β or r) coefficients are provided with standard errors in parentheses. Dotted lines represent nonsignificant effects.
†p < .10. *p < .05. **p < .01. ***p < .001.
Discussion
Study 1 results were consistent with both hypotheses at the level of trait covariances. American identification was associated with less liberal immigration attitudes and less support for criminal justice reform and affirmative action (Hypothesis 1); POC identification was associated with more liberal immigration attitudes and more support for criminal justice reform and affirmative action (Hypothesis 2).
However, Study 1 provides inconsistent evidence of group identification predicting changes in policy attitudes. Hypothesis-consistent within-person effects were only found for POC identification predicting affirmative action attitudes (Hypothesis 2). Otherwise, significant cross-lagged effects were found only in the CLPM, inconsistently across policy areas.
The CLPM results should be interpreted with caution as those models fit poorly, perhaps because the variables were highly stable, leaving little variance to be explained by cross-lagged effects. Stability in the identity variables was anticipated, as those have been treated as stable, endogenous variables in existing literature; however, stable policy attitudes were unexpected, given extensive literature on political non-attitudes among the general public (e.g., Converse, 1964; Zaller, 1992). Perhaps participants in this study, who tended toward high incomes and educational attainment, were unusually politically sophisticated and held stable political attitudes typically found among elites.6
Additionally, by being stereotyped as a “model minority” (Wu, 2014) and contrasted (i.e., triangulated) with both White and Black Americans (Kim, 1999), Asian Americans have been put in a position of non-solidarity with other racial minority groups. Study 1’s findings might not extend to racial minority populations with different histories. Thus, Study 2 attempted to replicate Study 1 in a nationally representative sample of Black and Hispanic/Latino Americans.
Study 2
This study used data from a large, multi-investigator study on the 2020 U.S. Presidential election (University of Minnesota, 2020). I focused on responses from Black and Latino participants, as two racial/ethnic minority groups with distinct histories and experiences.
The data included questions about immigration attitudes and support for Black Lives Matter and the protests after George Floyd’s death. As immigration policy in the U.S. tends to be associated with Latinos (see Study 1) and not African Americans, a largely non-immigrant group (Capers & Smith, 2016), this was expected to be an own-group issue for Latino participants and a common ingroup issue for Black participants. Black Lives Matter and protests support was expected to be an own-group issue for Black participants and a common ingroup issue for Latino participants.
Method
The study was administered online by YouGov, in 3 waves: October 6-14, October 25 to November 3 (Election Day), and November 9-16, 2020. It was determined to be exempt by the University of Minnesota IRB. The design and analyses were not pre-registered.
Participants
Participants consisted of a nationally representative sample of adult United States citizens, plus an oversample of African Americans. This study focuses on 658 participants who self-identified as Black (38% male) and 272 participants who self-identified as Hispanic/Latino (57% male). Of the Black participants, 531 responded to Wave 2 (19% attrition), and 441 responded to Wave 3 (17% attrition from Wave 2; 33% overall). Of the Latino participants, 146 responded to Wave 2 (46% attrition), and 120 responded to Wave 3 (18% attrition from Wave 2; 56% overall). See Figure S2 and Supplementary Materials for additional attrition information.
Participants had lower median incomes (Black participants: $30,000-39,999; Latino participants: $40,000-49,999) and more typical education levels (19% of Black participants and 16% of Latino participants had at least a 4-year college degree) than the Study 1 participants. Most were born in the U.S (Black participants: 9% immigrant citizens; Latino participants: 18% immigrant citizens). See Table S22 for additional demographics.
Measures
The Wave 1 survey asked participants to select racial/ethnic categories that describe them. Participants were also asked their gender, ideology, and party affiliation.
Included in all waves were two items measuring immigration attitudes: the number of immigrants item from Study 1 (1 = increased a lot; 7 = decreased a lot) and an item on ending criminal penalties for crossing the border illegally (1 = strongly oppose; 7 = strongly support). Because these items showed poor internal consistency (Black participants: α = .08-.12; Latino participants: α = .28-.52), they were analyzed separately.
Although this study did not explicitly measure criminal justice attitudes, it asked about attitudes toward Black Lives Matter (1 = very positive; 5 = very negative) and the protests after George Floyd’s death (1 = strongly support; 7 = strongly oppose). Responses to these items were averaged (after 0-1 coding) to form a composite score in each wave (Black participants: α = .75-.82; Latino participants: α = .89-.96).
Racial, POC, and American identification (and identification with groups unrelated to this study) were measured using 2-item versions of the Huddy et al. (2015) scale. Responses were averaged to form a composite score in each wave for POC (Black participants: α = .85-.87; Latino participants: α = .78-.85) and American identification (Black participants: α = .87-.88; Latino participants: α = .87-.90).
Additional information was provided by YouGov, including participants’ birth year, education level, immigrant background, family income, and whether they voted in the 2016 Presidential election.
Analyses
All variables were recoded to range from 0-1, with higher scores indicating stronger group identification or more racially liberal policy attitudes. Data from Black and Latino participants were analyzed separately. CLPM and RI-CLPM were fit with and without stationarity. As a robustness check, additional CLPM were fit with age, gender, education, income, immigrant status, political ideology, and party affiliation as covariates. Data were analyzed using R, version 4.3.1 (R Core Team, 2023) and the lavaan package (Rosseel, 2012) v0.6-15, with full-information maximum likelihood estimation and pairwise deletion. See Supplementary Materials (Part 4 and Tables S14-S19) for additional details, fit statistics, and results from CLPM with covariates.
Analysis code is available at https://doi.org/10.17605/OSF.IO/7JDXF. Data access can be requested from the University of Minnesota Center for the Study of Political Psychology.
Results
Descriptive Statistics
Descriptive statistics, correlations, and α for the identity and policy attitude variables are in Tables S3-S4. Black participants reported high POC (M = .804-.820) and American identification (M = .697-.722); Latino participants reported similarly high American identification (M = .705-.723) and POC identification around the midpoint (M = .471-.523). Racial identification was highly correlated with POC identification (Black participants: r = .83-.88; Latino participants: r = .52-.66) and moderately correlated with American identification (Black participants: r = .29-.32; Latino participants: r = .24-.28). The identity variables were highly stable (Black participants: r = .74-.83; Latino participants: r = .73-.89), as was the BLM/protests variable (Black participants: r = .86-.90; Latino participants: r = .93-.97). The number of immigrants item was also quite stable (Black participants: r = .67-.69; Latino participants: r = .67-.84), and the remaining immigration item was less stable (r = .27-.40).
Hypothesis 1: American Identification
Fit statistics and parameter estimates are in Tables 3 (Latino participants) and S13 (Black participants).
Table 3
American Identification Panel Model Results, Latino Participants (Study 2)
| Number of immigrants | Decriminalizing immigration | BLM/ protests | ||||
|---|---|---|---|---|---|---|
| CLPM | RI-CLPM | CLPM | RI-CLPM | CLPM | RI-CLPMa | |
| T1 ID → T2 ID | .736*** (.045) .803 | -.220 (.211) -.602 | .839*** (.033) .845 | -.354* (.169) -.696 | .719*** (.043) .781 | -.230 (.172) -.407 |
| T1 attitude → T2 ID | -.080† (.047) -.082 | .230 (.236) .555 | -.025 (.027) -.028 | -.015 (.049) -.064 | -.120** (.036) -.158 | -.056 (.189) -.069 |
| T1 ID → T2 attitude | -.003 (.062) -.003 | .528 (.404) .538 | -.199** (.070) -.173 | -.118 (.306) -.052 | -.096** (.036) -.082 | -.030 (.097) -.084 |
| T1 attitude → T2 attitude | .673*** (.065) .664 | -.747 (.512) -.673 | .363*** (.059) .349 | .045 (.150) .044 | .875*** (.031) .905 | -.506*** (.069) -.972 |
| T2 ID → T3 ID | .923*** (.049) .868 | -1.019 (1.165) -.456 | .839*** (.033) .867 | -.354* (.169) -.190 | .918*** (.050) .864 | -.230 (.172) -.153 |
| T2 attitude → T3 ID | -.050 (.046) -.049 | .056 (.373) .067 | -.025 (.027) -.030 | -.015 (.049) -.035 | -.044 (.037) -.052 | -.056 (.189) -.024 |
| T2 ID → T3 attitude | -.124* (.054) -.124 | -.698 (.644) -.319 | -.199** (.070) -.161 | -.118 (.306) -.024 | .045 (.036) .034 | -.030 (.097) -.027 |
| T2 attitude → T3 attitude | .764*** (.051) .790 | .093 (.191) .115 | .363*** (.059) .339 | .045 (.150) .041 | 1.022*** (.027) .977 | -.506*** (.069) -.285 |
| ID-attitude trait cov | -.023*** (.005) -.379 | -.013* (.005) -.260 | -.033*** (.006) -.387 | |||
| CFI | .895 | 1.000 | .913 | 1.000 | .923 | 1.000 |
| RMSEA | .257 | .000 | .131 | .001 | .276 | .000 |
| SRMR | .053 | .012 | .064 | .039 | .027 | .019 |
Note. Table reports lagged and cross-lagged coefficients, RI-CLPM identity-attitude trait-level covariance, and model fit statistics. Unstandardized coefficients are reported with standard errors in parentheses and standardized coefficients in italics. Models for ending criminal penalties for immigration and the BLM/protests RI-CLPM assume stationarity. N = 292.
aRI-CLPM initially had a non-significant negative residual variance for Wave 2 BLM/protests. The reported model constrains that variance to 0.
†p < .10. *p < .05. **p < .01. ***p < .001.
No cross-lagged effects of American identification were significant for Black participants. American identification had a significant, negative trait-level covariance with support for ending criminal penalties for immigration (c = -.006, SE = .003, p = .043, r = -.131).
For Latino participants, no RI-CLPM within-person cross-lagged effects were significant. In the CLPM, American identification had significant, negative cross-lagged effects on both immigration items (criminal penalties: b = -.199, SE = .070, p = .005, β = -.173 and -.161; number of immigrants Wave 2-3: b = -.124, SE = .054, p = .021, β = -.124) and on BLM/protest support (b = -.096, SE = .036, p = .008, β = -.082), and vice versa, across the first time lag. The effect for ending criminal penalties for immigration was not robust to covariates (Table S18). American identification had a significant, negative trait-level covariance with both immigration items (number of immigrants: c = -.023, SE = .005, p < .001, r = -.379; criminal penalties: c = -.013, SE = .005, p = .014, r = -.260) and BLM/protests support (c = -.033, SE = .006, p < .001, r = -.387), indicating that more American-identified participants were less supportive of policies benefitting their own and another racial minority group.
Hypothesis 2: POC Identification
Fit statistics and parameter estimates are in Table 4.
Table 4
POC Identification Panel Model Results (Study 2)
| Black Participants | ||||||
|---|---|---|---|---|---|---|
| Number of immigrants | Decriminalizing immigration | BLM/ protests | ||||
| CLPM | RI-CLPM | CLPM | RI-CLPM | CLPM | RI-CLPM | |
| T1 ID → T2 ID | .786*** (.021) .771 | .207 (.127) .205 | .795*** (.021) .779 | .221 (.139) .218 | .735*** (.023) .728 | .132 (.100) .133 |
| T1 attitude → T2 ID | .065** (.022) .059 | .079 (.077) .076 | .020 (.018) .021 | .010 (.034) .017 | .136*** (.024) .122 | .260* (.106) .178 |
| T1 ID → T2 attitude | .072** (.023) .077 | .158† (.093) .150 | .051 (.036) .045 | -.029 (.132) -.016 | .064*** (.017) .069 | .115* (.056) .130 |
| T1 attitude → T2 attitude | .669*** (.024) .658 | .017 (.101) .016 | .284*** (.032) .271 | -.020 (.069) -.019 | .865*** (.018) .834 | .396*** (.091) .304 |
| T2 ID → T3 ID | .786*** (.021) .792 | .207 (.127) .220 | .795*** (.021) .805 | .221 (.139) .240 | .735*** (.023) .740 | .132 (.100) .131 |
| T2 attitude → T3 ID | .065** (.022) .060 | .079 (.077) .087 | .020 (.018) .022 | .010 (.034) .020 | .136*** (.024) .127 | .260* (.106) .230 |
| T2 ID → T3 attitude | .072** (.023) .080 | .158† (.093) .168 | .051 (.036) .047 | -.029 (.132) -.017 | .064*** (.017) .069 | .115* (.056) .144 |
| T2 attitude → T3 attitude | .669*** (.024) .682 | .017 (.101) .019 | .284*** (.032) .290 | -.020 (.069) -.021 | .865*** (.018) .864 | .396*** (.091) .440 |
| ID-attitude trait cov | .014*** (.003) .273 | .006* (.003) .151 | .030*** (.003) .540 | |||
| CFI | .919 | .998 | .933 | 1.000 | .958 | 1.000 |
| RMSEA | .161 | .031 | .116 | .000 | .146 | .010 |
| SRMR | .053 | .023 | .050 | .013 | .031 | .015 |
| Latino Participants | ||||||
| Number of immigrants | Decriminalizing immigration | BLM/ protests | ||||
| CLPM | RI-CLPM | CLPM | RI-CLPM | CLPM | RI-CLPMa | |
| T1 ID → T2 ID | .674*** (.044) .720 | -.405** (.156) -.554 | .714*** (.042) .756 | -.363* (.172) -.462 | .608*** (.057) .675 | -.463** (.157) -.732 |
| T1 attitude → T2 ID | .152** (.052) .127 | .158 (.134) .207 | .028 (.042) .027 | .099 (.078) -.207 | .116* (.057) .130 | -.236 (.257) -.199 |
| T1 ID → T2 attitude | .051 (.035) .061 | .038 (.095) .059 | .117* (.059) .120 | .430* (.198) .244 | -.009 (.034) -.009 | -.079 (.058) -.268 |
| T1 attitude → T2 attitude | .736*** (.041) .688 | -.320* (.128) -.473 | .384*** (.059) .365 | .070 (.142) .065 | .902*** (.034) .933 | -.509*** (.068) -.915 |
| T2 ID → T3 ID | .674*** (.044) .657 | -.405** (.156) -.253 | .714*** (.042) .697 | -.363* (.172) -.254 | .793*** (.080) .710 | -.463** (.157) -.232 |
| T2 attitude → T3 ID | .152** (.052) .132 | .158 (.134) .087 | .028 (.042) .028 | .099 (.078) .156 | .054 (.069) .052 | -.236 (.257) -.055 |
| T2 ID → T3 attitude | .051 (.035) .063 | .038 (.095) .043 | .117* (.059) .110 | .430* (.198) .188 | .081** (.031) .071 | -.079 (.058) -.098 |
| T2 attitude → T3 attitude | .736*** (.041) .808 | -.320* (.128) -.321 | .384*** (.059) .371 | .070 (.142) .069 | .974*** (.027) .933 | -.509*** (.068) -.294 |
| ID-attitude trait cov | .027*** (.005) .390 | .008 (.007) .144 | .048*** (.007) .480 | |||
| CFI | .843 | .994 | .806 | 1.000 | .903 | 1.000 |
| RMSEA | .200 | .050 | .165 | .000 | .295 | .000 |
| SRMR | .077 | .041 | .075 | .034 | .050 | .034 |
Note. Table reports lagged and cross-lagged coefficients, RI-CLPM identity-attitude trait-level covariance, and model fit statistics. Unstandardized coefficients are reported with standard errors in parentheses and standardized coefficients in italics. All models except Latino participants’ BLM/protests CLPM assume stationarity. Black participants’ N = 658; Latino participants’ N = 292.
aLatino participants’ RI-CLPM initially had a non-significant negative residual variance for Wave 2 BLM/protests. The reported model constrains that variance to 0.
†p < .10. *p < .05. **p < .01. ***p < .001.
Black participants’ POC identification had a significant, positive cross-lagged effect on BLM/protest support in the RI-CLPM (b = .115, SE = .056, p = .039, β = .130 and .144) and CLPM (b = .064, SE = .017, p < .001, β = .069), though the reverse effects were also significant. POC identification had a significant, positive cross-lagged effect in the CLPM on allowing more immigrants (b = .072, SE = .023, p = .002, β = .077 and .080), and vice versa. Cross-lagged effects were not significant for the other immigration item. Trait-level covariances between POC identification and all policy attitudes were significant and positive (number of immigrants: c = .014, SE = .003, p < .001, r = .273; immigration criminal penalties: c = .006, SE = .003, p = .048, r = .151; BLM/protests: c = .030, SE = .003, p < .001, r = .540).
Latino participants’ POC identification had a significant, positive cross-lagged effect on support for ending criminal penalties for immigration, (RI-CLPM: b = .430, SE = .198, p = .030, β = .244 and .188; CLPM: b = .117, SE = .059, p = .047, β = .120 and .110). POC identification and BLM/protest attitudes had inconsistently significant, positive cross-lagged effects on each other in the CLPM that were not robust to covariates (Table S19). POC identification had a significant, positive trait-level covariance with number of immigrants (c = .027, SE = .005, p < .001, r = .390) and BLM/protest support (c = .048, SE = .007, p < .001, r = .480).
Discussion
Study 2 provides inconsistent evidence that identification predicts changes in policy attitudes. Within-person cross-lagged effects of American identification (Hypothesis 1) were not significant, though Latino participants showed significant, negative cross-lagged effects in the CLPM. POC identification (Hypothesis 2) predicted support for own-group issues in the RI-CLPM, but significant cross-lagged effects on other-group issues were found only in the CLPM.
Results were more consistent with both hypotheses at the trait covariance level. Stronger American identification was associated with less liberal immigration attitudes and less support for criminal justice reform (Hypothesis 1) among Latino participants, though only one immigration item showed this association among Black participants. Stronger POC identification was associated with more liberal immigration attitudes and more support for criminal justice reform (Hypothesis 2) among Black participants; among Latino participants, this covariance was significant for one immigration item (interestingly, not the item with significant cross-lagged effects) and BLM/protests support.
A limitation is the smaller sample size of Latino participants. RI-CLPM was likely underpowered for detecting cross-lagged effects, and CLPM could have been as well, given the stability of many of the variables.
General Discussion
Evidence from two 3-wave panel studies indicates that, for members of racial minority groups in the United States, identification with a common ingroup that excludes the dominant group (POC) and identification with a common ingroup that includes the dominant group (American) relate differently to support for policies benefitting racial minorities, a form of solidarity with one’s own and/or other minority groups.
Limited evidence was found for ironic effects of American identification (Hypothesis 1). American identification did not significantly predict policy attitude change among African Americans or Asian Americans. Among Latinos, individual differences (but not within-person changes) in American identification predicted less support for an own-group issue (immigration) and an other-group issue (BLM/protests). Trait-level American identification was negatively correlated with trait-level support for policies benefitting minorities among Asian Americans (Study 1) and Latinos (Study 2), indicating the expected association exists for these groups but may not reflect the theorized ironic effect of common ingroup identity.
POC identification predicted increased support for policies benefitting minorities (Hypothesis 2), as expected from a shared identity in the intra-minority solidarity literature or disadvantaged identity in the collective action literature. Within-person increases in POC identification predicted increased support for affirmative action among Asian Americans (Study 1) and increased support for Black Lives Matter and the George Floyd protests (albeit an own-group issue) among African Americans (Study 2). Individual differences in POC identification predicted increased support for affirmative action and more liberal immigration attitudes among Asian Americans (Study 1), allowing more immigrants among African Americans (Study 2), and increased support for own-group issues among Latino and African Americans (Study 2). Trait-level POC identification was positively correlated with support for policies benefitting minorities across all three participant groups. The overall pattern indicates solidarity, not ironic effects.
These results might partly predict nonwhite Americans’ voting patterns. When anti-immigrant rhetoric and racialized policy issues are prevalent, as in the 2020 and 2024 Presidential elections, individuals/groups who identify more strongly as POC may be more likely to vote against the party or candidate using such rhetoric or advocating such policies. Those who identify more strongly as American might vote for such a party or candidate.
A limitation across studies is that the models included only one identity at a time. Thus, effects of POC identification might be confounded with those of racial identification, particularly among Black participants, for whom these identities were highly correlated, though this correlation was lower among Asian participants. The effects of POC and American identities were unlikely to be confounded, however, especially for Asian and Latino participants, for whom POC-American identity correlations were small and statistically insignificant. Future research could include multiple identities in the same model, though this could require very large sample sizes.
It is also possible that policy attitudes in these studies reflect ideology rather than solidarity, i.e., POC identification predicts progressive policy attitudes, and American identification predicts conservative policy attitudes. Perhaps participants interpreted the issues in such a non-racialized left-right manner. However, one might expect widespread racialized interpretations in the United States in late 2020—the Trump campaign actively linked immigration to Latin America, racial disparities in criminal justice were discussed widely on social media after George Floyd’s death, and the Harvard affirmative action case was moving through the courts. Empirically, controlling for political ideology and party affiliation in the CLPM did not substantially change results, though I could not control for these variables in the RI-CLPM. The present findings regarding POC identity are also consistent with findings from intra-minority solidarity research using other solidarity measures (e.g., Burson & Godfrey, 2018; Craig & Richeson, 2012; Pérez et al., 2023).
More importantly, inconsistent cross-lagged effects do not establish whether group identification precedes political attitudes. Although trait-level correlations were consistently significant, they do not speak to directionality. This pattern of effects could stem from group identification and policy attitudes being stable in adults over short time frames. (Stability may also account for poor CLPM fit.) However, the present findings are consistent with intergroup contact research failing to find significant cross-lagged effects on attitudes (Bohrer et al., 2019), especially within-person (Sengupta et al., 2020), across time lags of up to a year. Taken together, these studies hint that the relationships between contact or common ingroup identification and intergroup attitudes might not be causal for adults under normal circumstances. Identity salience may be a more promising target than identification strength for changing adults’ intergroup attitudes.
Additionally, American identity could have ironic effects not because it includes the dominant group but because it is uniquely associated with the dominant group. Perhaps Latino and Asian Americans associate American with White (cf. Devos & Banaji, 2005), and this explains negative relationships between American identification and support for policies benefitting minorities. American identity might also predict immigration attitudes because immigrants are excluded from the American ingroup, though negative relationships extended beyond immigration policies in the present studies. Alternatively, American identification may predict less support for policies that benefit minorities because American identity is shared with authority (Subašić et al., 2008). Future research could explore these alternative explanations for the effects of American identification.
Finally, this research was done in the United States, using identity labels that are meaningful in the U.S. and policy areas that were racialized in the U.S. While the principles should generalize, different identities and policies may be needed to replicate the findings elsewhere. For example, minority group members in other parts of the world may not identify as POC but may share other identity labels. Similarly, different issues may be associated with minority groups outside the U.S.
The present studies suggest the composition of common ingroup identities matters. For members of racial/ethnic minority groups, identification with a common ingroup that includes the dominant group (American) might be associated with less support for policies benefitting minority groups, but identification with a common ingroup that excludes the dominant group (POC) is associated with more support for these policies. These findings corroborate existing research on intra-minority solidarity and support the common ingroup identity model as a mechanism, while offering the boundary condition that an identity shared with the majority group may not promote solidarity (though it also may not have the ironic effects suggested in previous literature).