The Partisanship of U.S. Latinos

The aggregation of Latino sub-groups into a single identity category has produced rich insights into racial and ethnic politics, including key differences in partisanship and vote choice between Latinos and other racial groups. At the same time, methods of disaggregation promoted in the past (Alvarez and García Bedolla Reference Alvarez and García Bedolla2003; Hero Reference Hero1992) have seen less theoretical innovation over time. While this approach has been applied to the study of Latino politics, its scope remains limited due to small sample sizes. As such, most research highlights the behavior of Mexican, Puerto Rican, and Cuban populations in the United States. More recent studies that look to interpret recent voting patterns utilize the aggregate approach 2 (Geiger and Reny Reference Geiger and Reny2024; Hopkins, Kaiser and Pérez Reference Hopkins, Kaiser and Pérez2023), although the observed changes in the Latino vote also prompt research into divides in identity and policy issues, outside of the CoO, that could drive this phenomenon (Corral and Leal Reference Corral and Leal2024).

At an aggregate level, the partisanship of Latinos was previously thought to be much less stable than that of White Americans (Hajnal and Lee Reference Hajnal and Lee2011), yet recent work by Hopkins, Kaiser and Pérez (Reference Hopkins, Kaiser and Pérez2023) finds that their partisanship has shown relative stability, at least during the early years of Donald Trump’s presidency, which should be a difficult test of partisan stability, given the anti-immigrant and anti-Latino rhetoric employed by Trump since 2016. In terms of vote choice, Latinos have typically voted Democratic. However, recent elections call this pattern into question. Hopkins, Kaiser and Pérez (Reference Hopkins, Kaiser and Pérez2023) find that, although working in opposite directions, education and income predict shifts in partisanship, with higher education and lower income predicting a shift toward the Republican party for Latinos. This research reveals that the predictors of partisanship for the general population (e.g., income and education) do not drive Latinos’ partisanship in the same direction as for other Americans (Alvarez and García Bedolla Reference Alvarez and García Bedolla2003). In new work, analyzing surveys covering 1989 to 2023, Wakefield, Fraga and Fisk (Reference Wakefield, Fraga and Fisk2025) find that younger Latinos are increasingly identifying as independent, whereas older, US-born Latinos are shifting toward Republican identification. They do find national sub-group differences consistent with ours; however, their data limit them to sub-group analyses of Mexicans, Puerto Ricans, Cubans, and the rest get grouped together, so they are limited in how they study national origin groups. Additionally, issue positions and identities were found to influence support for the Republican candidate in 2016 and 2020. Corral and Leal (Reference Corral and Leal2024) show that being an immigration restrictionist, along with gender and religious divides—particularly among men and evangelical Christians—also helps explain Donald Trump’s success with Latino voters. While this research elucidates predictors of partisanship, the current literature, as well as the media, still questions why a growing number of U.S. Latinos left the Democratic Party over the last decade.

Alvarez and García Bedolla (Reference Alvarez and García Bedolla2003) suggest that inter-generational transmission of party attachment should function the same for Latinos as other identity groups in the United States, “making Latino partisanship less variable over time and across generations” (p. 45). Recognizing the importance of CoO in their study, the authors suggest that studying how the determinants of partisanship are transmitted across generations would be important in the years to come. However, recent studies of immigrant incorporation and acculturation (Hickel Jr. et al. Reference Hickel, Alamillo, Oskooii and Collingwood2020; Jiménez Reference Jiménez2010) may lead to the assumption that the effects of historical political experiences can only explain the partisanship of the immigrants, and subsequently wash away after multiple generations. Given the capacity that researchers now have to answer these questions using resources like the CMPS and other specialized samples of Latinos, we can more accurately assess whether country-of-origin-based disaggregation is a research approach worthy of heavier theoretical investment. As we discuss below, the internal composition of the Latino population in the United States is rapidly changing. Different countries of origin are now contributing more to the population than traditional sources, and research is beginning to recognize the implications of these demographic changes (Ocampo and Ocampo Reference Ocampo and Ocampo2020). How much is this shift responsible for some of the puzzling patterns we observe when analyzing on an aggregate level?

The Role of Pan-Ethnicity in Latino Political Behavior

In making the case for disaggregation by CoO, we do not overlook or dismiss the role of pan-ethnic attachment, which rivals the relevance of national origin identities for many Latinos in the United States. Early studies in political science defined the significance of Latino pan-ethnicity, highlighting that Latino identity is seen as more than just “instrumental,” or solely to build political clout (Calderón Reference Calderón1992; Jones-Correa and Leal Reference Jones-Correa and Leal1996). Pan-ethnic identity represents a significant primary or secondary identity among Latinos, related to political and cultural similarities between country-of-origin groups. Fraga et al. (Reference Fraga, Garcia, Hero, Jones-Correa, Martinez-Ebers and Segura2010a) further support this point in an updated exploration of pan-ethnic attachment among Latinos.

Using National Survey of Latinos data from the early 2000s, Fraga et al. (Reference Fraga, Garcia, Hero, Jones-Correa, Martinez-Ebers and Segura2010a) observed an overall increase in attachment to Hispanic and Latino identifiers across time. These updates to the literature confirmed that attachment to pan-ethnic identifiers was driven not only by political goals but also by the perception of shared commonality among Latino sub-groups. The strong link between perceived cultural and political commonality and pan-ethnicity enhanced the importance of pan-ethnicity in Latino politics research. Its significance also became more evident in the political arena. In recent decades, parties have increased their investment in Latino voters in response to population growth. For example, parties are increasingly targeting Latino voters with Spanish language advertising (Mann, Michelson and Davis Reference Mann, Michelson and Davis2020) and in-person canvassing (Matland and Murray Reference Matland and Murray2012; Valenzuela and Michelson Reference Valenzuela and Michelson2016), resulting in participation gains among these voters, though potentially conditional on identity strength (Valenzuela and Michelson Reference Valenzuela and Michelson2016) and propensity to vote (Matland and Murray Reference Matland and Murray2012). We also know that co-pan-ethnic candidates may indeed be attractive across different Latino-origin groups (Barreto Reference Barreto2010; Zárate, Quezada-Llanes and Armenta Reference Zárate, Quezada-Llanes and Armenta2024).

Latino identity is undoubtedly tangible and consequential in American society. American Latinos often concentrate geographically, and this has been shown to increase group attachment for those living near what Wilcox-Archuleta (Reference Wilcox-Archuleta2018) calls “ethnic stimuli.” Furthermore, the racialization of Latinos in the United States often disregards national distinctions. People of Latin American descent experience the consequences of racial hierarchy in relatively similar ways. The racial hostility that sparked mass immigration protests like those seen in 2006 remains capable of activating Latino group identity and influencing political behavior across Latino sub-groups (Gutierrez et al. Reference Gutierrez, Ocampo, Barreto and Segura2019; Pantoja and Segura Reference Pantoja and Segura2003; Zepeda-Millán and Wallace Reference Zepeda-Millán and Wallace2013).

The importance of pan-ethnicity notwithstanding, we find notable differences in voting patterns across Latino sub-groups and over time. We believe that, parallel to Fraga et al. (Reference Fraga, Garcia, Hero, Jones-Correa, Martinez-Ebers and Segura2010 a), which demonstrates that national origin remains a significant identity for Latinos, our findings indicate that CoO persistently plays a role in voting behavior. This evidence, along with more recent work on the acculturation and sub-group particularities of Latinos, motivates this exploration into national origin identity (Jones-Correa and Leal Reference Jones-Correa and Leal1996; Ocampo and Ocampo Reference Ocampo and Ocampo2020; Wakefield, Fraga and Fisk Reference Wakefield, Fraga and Fisk2025).

Disaggregation and Population Dynamics

How assessments of disaggregation apply to an increasingly native-born population of Latinos is unclear. Based on the acculturation literature, one may argue that generational status should eliminate national origin identification and, with it, the relevance of home country experiences for partisanship and voting behavior. This would negate the argument we make here that grouping people by their national origin can reveal significant implications for Latino political behavior. While some studies find that, by the third generation, descendants of immigrants are significantly acculturated (Fraga et al. Reference Fraga, Garcia, Hero, Jones-Correa, Martinez-Ebers and Segura2010b), more recent literature has deemed acculturation a slow and non-linear process among Latinos (Pedraza Reference Pedraza2014; Pérez and Cobian Reference Pérez and Cobian2024). There also exists a monotonic decrease in attachment to national origin descriptors by generational status (Fraga et al. Reference Fraga, Garcia, Hero, Jones-Correa, Martinez-Ebers and Segura2010a). Do these patterns in attachment to identity translate to the influence of CoO? This question remains unclear as few recent studies explore how changes in the size and concentration of CoO groups influence what knowledge of individual home countries is carried over between generations, and how this impacts vote choice and partisanship.

The increased number of Latinos from all 33 Latin American countries provides a reason to investigate whether population changes sustain the political relevance of one’s CoO. Ocampo and Ocampo (Reference Ocampo and Ocampo2020) explore the relevance of disaggregation as it pertains to Colombians, arguing that this particular sub-group is one of the “new Latinos,” constituting further exploration into the capacity for the growth rate in Colombians to shape what is currently known about Latinos across several political dimensions. In their study of Mexicans, Puerto Ricans, and Cubans, Alvarez and García Bedolla (Reference Alvarez and García Bedolla2003) argue that Latinos are more influenced by social and political factors than economic factors, and they find that CoO is a significant predictor of party identity. They attribute this to commonalities in the political integration process, for which nationality is a proxy. Similar to these approaches, we expect the CoO to be a significant identity cleavage, made more complex and important by recent changes in Latino migration patterns.

De La Garza (Reference De La Garza2004) highlights some instances of politically relevant variation by country-of-origin group, and also affirms that survey researchers should make efforts to incorporate relevant institutional variables that differentiate Latinos’ political experiences, such as exposure to civil conflict, instead of attempting to diversify Latino survey samples by national origin group. He cautions that treating national origin as an independent influence can result in “making ethnicity an unchanging attribute rather than a fluid characteristic, and conceals or distorts historical and ongoing relations between Hispanics and American political institutions” (p. 103). We agree that ethnicity is fluid, but also argue that there are factors from one’s past that influence attitudes and behavior. Institutional influences are only part of the story, and another part can be meaningfully captured by a person’s origin. While we do not explicitly test the specific mechanisms in the home country, we do show that there is something to be gained by its incorporation into the larger story about Latino political behavior. Indeed, we see in the literature plenty of room for an explanation of the variation in vote choice and partisanship that goes beyond institutional factors, and we offer a discussion of what we see as promising potential mechanisms.

The arguments we propose are not of identity and political behavior alone, but of collective histories and the influence of sub-group concentrations across the United States. We agree with Alvarez and García Bedolla (Reference Alvarez and García Bedolla2003) that nationality alone does not move important attitudinal and participatory outcomes. The unique political, social, and cultural experiences in one’s home country should influence political engagement in the United States. This relationship may be sustained by frequent contact with members of their national origin group, given these changes to the composition of Latino communities across space and time (Masuoka Reference Masuoka2008; South, Crowder and Chavez Reference South, Crowder and Chavez2005).

Potential Mechanisms

While we are somewhat limited in our ability to test a direct causal path from CoO to political behavior, here we discuss and propose tests for what we believe are three potential mechanisms that we hope will be more completely explored in future work. Three pathways that we think are worth investigation include (1) how political socialization in the home country influences behavior once in the United States, (2) how it interacts with where they land and their political socialization in the U.S, as well as (3) structural and individual reasons for moving to the United States.

Country of Origin

As we have discussed, studying the Latino population as a monolith may pose disadvantages for some research questions, due to its diverse and distinctive features (Affigne Reference Affigne2000). The consolidation of Latinos as a single identity group is normalized across disciplines and also bolstered by the United States’ conception of the term “Hispanic,” which the Census Bureau recognizes as Mexicans, Puerto Ricans, Central Americans, South Americans, Cubans, and other Hispanics (Read, Lynch and West Reference Read, Lynch and West2021).Footnote ² In response to these issues, we disaggregate Latinos by CoO and, when necessary, by cultural and geographic characteristics. This ensures that smaller groups, which empirically cannot be studied on their own, are still included in this inquiry.

The relatively recent, rapid increase in the study of Latinos as consequential political actors has given rise to better surveying of this population. Five waves of Collaborative Multiracial Post-Election Surveys (CMPS) have featured diverse samples of Latinos, namely respondents from countries outside of Mexico, Cuba, and Puerto Rico, and, in some years, have over-sampled this population. These innovations respond to an increased need to understand minority politics.

The CMPS affords us the ability to create several country-of-origin sub-groups for comparison. It is, again, not lost on us that we advocate studying countries, yet we combine some countries into regions. We are optimistic that future research will include surveys with enough respondents from more countries to draw adequately powered conclusions. Our final sample includes Mexicans, Cubans, Puerto Ricans, Dominicans, and three cultural-based groupings. Latinos who descend from Central America (Hondurans, Guatemalans, Nicaraguans, Costa Ricans, and Salvadorans) comprise one group, the second and third groups include South Americans, who were divided into two subsets: Latinos who descend from Andean Countries (Bolivia, Colombia, Ecuador, Peru, and Venezuela), and Latinos who descend from the Southern Cone (Brazil, Argentina, Chile, Uruguay, and Paraguay).

These groupings are made based on the U.S. classification of countries of origin (National Research Council 2006). In the case of South America, the two subsets of countries were split up based on analogous political histories during colonial times and similar European migration patterns in the late 19^th and 20^th centuries (Centeno Reference Centeno2002). With these seven sub-groups, we disaggregate the Latino monolith in a way that allows us to better understand the political choices of Latinos living in the United States. In Table 2, we report the CMPS sample size by region and year.Footnote ³

Covariates

The covariates in our models represent factors that research has shown to be correlated with political behavior and that are relevant to the study of Latino political behavior in particular. As explained below, we balance the sample of respondents from CMPS over 12 covariates broadly grouped into two categories, as shown in Table 3. Individual socio-demographic characteristics refer to variables correlated with political participation and partisan identification. For example, as age and education increase, and when Americans become homeowners, they are more likely to turn out to vote. Employment, gender, and income are correlated with partisanship. Recent studies show that while Latino partisans are already sorted, lower-income Latinos are more likely to remain independent and be persuaded by economic appeals (Wakefield Reference Wakefield2025). Ideology and religion (i.e., being catholic) are also strong and important factors driving partisan preferences. Recent literature has shown an increased partisan alignment of Catholic and ideologically conservative Latinos with the Republican Party (Fraga, Velez and West Reference Fraga, Velez and West2025).

Table 3.

Dependent and independent variables

* AZ, CA, FL, NM, TX.

** Self-reported (2008, 2012) / share of Latino residents equal to or above state average (2016, 2020).

In terms of contextual factors, given the strong immigrant ties of the Latino population in the United States, extensive previous research has argued that Latinos are more prone to support immigrant rights and pro-immigration policies, resulting in an increase in Democratic support and a decrease in Republican support among Latinos (Barreto and Collingwood Reference Barreto and Collingwood2015; Pérez and Cobian Reference Pérez and Cobian2024). Thus, we want to be able to hold constant observable factors that expose individuals to immigrants, immigration rhetoric, and policy. These variables include whether a respondent lives in a Southern border state (Arizona, California, Florida, New Mexico, and Texas), lives in a highly Latino neighborhood, and lives in a rural area.

Some of the research we mentioned above argues that institutional features explain political differences among Latinos of different countries of origin. One key institutional feature is regime type. Between 1930 and 1990, all countries in Latin America had different democratic experiences and trajectories (Hartlyn and Valenzuela Reference Hartlyn and Valenzuela1994). Some countries, such as Colombia and Costa Rica, experienced long decades of democratic years; others, such as Argentina, Peru, Chile, Brazil, Uruguay, and Paraguay, were under a military dictatorship at some point between the late 1960s and early 1990s. Others, like Mexico, had extended constitutional stability and operated under a one-party authoritarian regime.

While we acknowledge the importance of this variable, we do not include it. The most important reason for this omission is that (with notable exceptions like Venezuela) regime type does not vary much, if at all, in the countries we study in the years the surveys cover. Appendix D in the Supplementary Materials offers data from the Polity5 Project, V-Dem, and Freedom House, demonstrating that these scores change very little, if at all.

Overall, these factors need to be accounted for when we conduct the matching strategy detailed below so that we can be as certain as possible that we match the important characteristics of the respondents and can attribute the differences to the CoO. Finally, a note of caution. While we believe our choice of controls, driven by theory and the existing literature, is exhaustive, we cannot peremptorily claim that we accounted for all possible factors that affect electoral outcomes. One limitation of our approach is that it does not capture potential unobserved confounders (see VanderWeele Reference VanderWeele2019).

Empirical Strategy

In a researcher’s ideal world, we would randomly assign CoO (Rubin Reference Rubin1974) to isolate its causal effects on respondents’ vote choice and partisanship. In the potential outcomes framework, the causal effect of the CoO could be estimated by comparing the observed outcomes produced by a set of individuals from country Z at time $t$ to the outcomes that would have been observed if, all else equal, the same set of individuals had not been from country Z. This is, of course, not possible. We, therefore, attempt to quantify the effect by “matching” survey respondents.

Matching is a popular statistical technique to increase precision and attenuate the bias that stems from the absence of random assignment of treatments in observational studies. Broadly speaking, the fundamental idea of matching is to retrieve the “missing” potential outcomes by imputation, i.e., using the observed outcomes of “paired” units, or by selecting subsets of units from the original data such that the treatment and the other covariates are unrelated (Abadie and Imbens Reference Abadie and Imbens2006; Ho et al. Reference Ho, Imai, King and Stuart2007). Substantively, this amounts to comparing the differences in electoral behavior of two Latino voters who differ in terms of the respective CoO (e.g., one is Puerto Rican, the other is not) but are balanced on other attributes that might influence their behavior (e.g., both are females, college-educated, in their 30s, etc.). Setting up this kind of comparison should bring us closer to estimating the “causal” effect of U.S. Latinos’ countries of origin on political decisions (Rubin Reference Rubin1973).

In practice, the process of achieving this balance and generating robust matching estimators is neither straightforward nor bulletproof. The more basic matching methods do not necessarily (or automatically) enhance balance and might even result in worsened balance and increased bias as a consequence of misspecification (see Drake Reference Drake1993; Greifer and Stuart Reference Greifer and Stuart2021). We employ the “genetic” matching procedure described by Sekhon (Reference Sekhon2011) and Diamond and Sekhon (Reference Diamond and Sekhon2013) using four waves of the CMPS. The main characteristic of this approach is that it relies on an evolutionary search algorithm to check and improve covariate balance, overcoming some of the pitfalls of non-iterative propensity score matching.

Genetic matching operates by minimizing the generalized Mahalanobis distance (GMD) between the ${\bf{X}}$ covariates for treated and control units $i$ and $j$ . We borrow from Diamond and Sekhon (Reference Diamond and Sekhon2013: p. 934) in formalizing this as:

$${\rm{GMD}}\left( {{x_i},{x_j},{\bf{W}}} \right) = \sqrt {{{({x_i} - {x_j})}^T}{{({{\bf{S}}^{ - {{1}\over{2}}}})}^T}{\bf{W}}{{\bf{S}}^{ - {{1}\over{2}}}}\left( {{x_i} - {x_j}} \right)} $$

Where ${\bf{W}}$ is a $k \times k$ positive-definite matrix containing a scaling factor $w$ for each covariate, ${{\bf{S}}^{ - \frac{1}{2}}}$ is the Cholesky decomposition of the sample covariance matrix ${\bf{S}}$ , and ${{\bf{X}}^T}$ is the transpose of ${\bf{X}}$ .Footnote ⁴ Through a pre-specified number of “generations” (500 in our case), the algorithm finds the scaling factors that optimize balance as measured by a given loss function. We employ the default imbalance measure in Ho et al. (Reference Ho, Imai, King and Stuart2011), the smallest p-value in covariate balance tests among the covariates. ${\bf{W}}$ is then used to perform nearest-neighbor matching (with replacement) with the inclusion of a propensity score estimated via logistic regression as a covariate, following Diamond and Sekhon’s (Reference Diamond and Sekhon2013) recommendations.

Although relatively more computationally expensive, an important advantage of genetic matching is the quality of the balance it achieves. In the Supplementary Materials (Figures A1-A7, Appendix A), we plot two covariate balance metrics contrasting the matched and unmatched samples for each country/region (absolute standardized mean differences and Kolmogorov-Smirnov statistics). In nearly all cases, the algorithm produces a balanced matched sample. This makes us more confident that the effect of the CoO we compute on the basis of the matched samples is net of the matching covariates. In the next section, we present our results.