"The Myth of the Zero Sum: Rethinking Acculturation in American Politics" (under review, Politics, Groups, and Identities) 

Political science has long assumed that acculturation operates as a zero-sum tradeoff, where adopting American culture requires abandoning heritage culture. This assumption has never been empirically tested. Using the 2006 Latino National Survey, which uniquely measures heritage and American cultural attachments and identity as independent dimensions, I conduct the first direct test of the binary model against a bidirectional alternative. Comparative cluster analysis across five algorithms with different mathematical foundations reveals four distinct acculturation orientations, with three algorithms converging on the same structure. Over 76% of Latino voters fall into hybrid categories, bicultural or demicultural, that the binary framework structurally cannot detect. The orientations the binary model does recognize, assimilationist and culture-affirming, account for less than a quarter of the population. These findings demonstrate that the dominant framework in political science misclassifies the vast majority of Latino voters and call for a fundamental shift in how the field conceptualizes and measures immigrant incorporation. 

Replication materials and code are available at: https://github.com/jagrijalva/bam-scale 

"Hidden in Plain Sight: Latino Political Diversity Before the Trump Era" (under review, Political Psychology) 

Latino political behavior has long confounded scholarly expectations. Frameworks built on linked fate and collective threat response predict unified political responses to shared marginalization, yet Latinos consistently demonstrate political diversity that existing approaches cannot explain. This paper argues that the source of this diversity lies in how individuals navigate cross-pressures between heritage and American cultural demands. Binary acculturation models, which force a zero-sum trade-off between cultures, recognize only two orientations and thereby misclassify 76% of Latino voters. Using the Bidimensional Acculturation Model, I identify four distinct orientations among 4,785 eligible voters in the 2006 Latino National Survey: culture-affirming, assimilationist, bicultural, and demicultural. Each orientation represents a different resolution of competing cultural demands, and each produces a distinct political profile. A two-phase non-parametric analysis reveals that immigration attitudes show the strongest associations with acculturation orientations, significantly exceeding ideology and partisanship. This finding is theoretically coherent: immigration is the issue that most directly implicates questions of belonging and exclusion, the very tensions that acculturation orientations navigate. Latino political diversity is not puzzling variation but the predictable consequence of different strategies for managing the cross-pressures inherent in navigating American society as members of a racialized immigrant-origin population. 

Replication materials and code are available at:
https://github.com/jagrijalva/acculturation-politics

"Stable Drivers of Campaign Effects: Mapping Latino Support for Trump" (accepted to MPSA 2026, 2016 CMPS analysis is complete, 2020 CMPS analysis is underway)

What explains Latino support for Donald Trump? While post-election commentary often emphasizes single-factor explanations such as economic anxiety, masculinity, or immigration enforcement, these accounts rarely systematically assess the relative importance of competing drivers or examine whether these factors remain stable across electoral contexts. Using data from the Collaborative Multiracial Post-Election Survey (CMPS) in 2016 and 2020, this paper employs random forest classification with SHAP (SHapley Additive exPlanations) value interpretation to identify and rank the key predictors of Latino Trump support. Unlike traditional regression approaches, this method captures non-linear relationships and complex interactions without requiring pre-specification, while SHAP values provide interpretable measures of each variable's contribution to individual predictions. I estimate models with and without partisanship to distinguish proximate from underlying drivers. Preliminary findings from 2016 CMPS reveal that immigration attitudes emerge as the most prominent predictor beyond partisanship, and notably, the importance of immigration attitudes does not diminish when partisanship is removed from the model. This suggests that immigration is not simply operating through partisan attachment but exerts an independent influence on Latino vote choice. This approach offers a more comprehensive and less assumption-dependent picture of Latino political behavior than existing work.

Replication materials and code for the 2016 analysis are available at: https://github.com/jagrijalva/ml-latino-vote-2016

"The Power-Sharing Index: Measuring Democracy Beyond Institutions" (preparing for submission to V-Dem Working Paper Series, subsequent submission to Journal of Democracy) 

Standard democracy indices measure whether democratic procedures function; they cannot detect whether those procedures function for everyone. This paper introduces the Power-Sharing Index (PSI), a measure of cross-group power distribution built from five V-Dem indicators capturing whether political power can transfer across demographic boundaries: social group power, gender power, civil liberties equality, freedom from torture, and freedom from political killings. The index employs a hybrid aggregation method that combines a multiplicative penalty for exclusion in any dimension with a procedural democracy cap, ensuring that power-sharing is measured only within functioning democratic institutions. Validation demonstrates excellent internal consistency (Cronbach's α = 0.96), unidimensionality (87.4% variance on PC1), and robust construct validity: the index responds appropriately to known historical events including the 19th Amendment, the Voting Rights Act, and Shelby County v. Holder. The critical discriminant validity test reveals what I call the "Herrenvolk Paradox": during 1789-1899, PSI shows a negative correlation with V-Dem's Electoral Democracy Index (r = -0.25), demonstrating that procedural democracy expanded among white men while cross-group power-sharing remained near zero. This divergence confirms that PSI captures variation that existing indices have been unable to detect because they focus on institutional/procedural aspects. Sensitivity analyses show that substantive conclusions are invariant to aggregation method, component weighting, and jackknife specifications. 

Replication materials and code are available at: https://github.com/jagrijalva/psi-scale

"By Design: Herrenvolk Democracy and the Founding of the United States" (preparing for submission, PS: Political Science & Politics special issue on 1776) 

Scholars of American political development have documented racial exclusion extensively, typically framing it as a contradiction of liberal principles, one tradition among several, or a failure to realize founding ideals. This paper advances a different claim: that racial exclusion defined the regime type itself. Drawing on statutory evidence from colonial legislation through the early Republic, the paper demonstrates that the United States was founded as a herrenvolk democracy, a regime characterized by robust democratic procedures for the included racial group and systematic legal exclusion of others. The 1790 Naturalization Act, which restricted citizenship to "free white persons," codified at the federal level a racial order that colonial assemblies had deliberately constructed over the preceding century. The classificatory argument developed here builds on existing scholarship while extending it in a critical direction: from treating race as an attribute of American democracy to recognizing it as constitutive of the regime. This reframing carries implications for how scholars understand both historical development and contemporary politics. If the founding regime was herrenvolk, then the persistent challenges facing American democracy concern not only institutional design but the boundaries of the demos itself. 

Book Project: The Herrenvolk State: Power and Exclusion in American Democracy (under consideration with Cambridge University Press and Princeton University Press) 

This book argues that American democracy was designed as a herrenvolk democracy and that the current crisis is not a departure from a healthy democratic past but the latest manifestation of this enduring regime type. Standard democracy indices cannot detect this because they measure whether democracy functions, not democracy for whom. Drawing on Tocqueville's observations about group-level power tensions, I develop a novel Power-Sharing Index that captures variation in whether and how non-white groups have been systematically excluded throughout American history. This index reveals four distinct eras in American political development: the Herrenvolk Era (1776-1865), the first real possibility of multiracial democracy during Reconstruction (1865-1877), Partial Inclusion (1877-1965), and the Power-Sharing Dilemma (1965-present). The book traces how citizenship and immigration law operated as primary mechanisms of exclusion and how conflicts over power-sharing have driven political development from the 1790 Naturalization Act to the contemporary democratic crisis.

Replication materials and code are available at:
https://github.com/jagrijalva/psi-scale

"Building Inference in Cluster Analysis with Multi-Algorithm Validation" (preparing for submission, Political Science Research and Methods) 

Cluster analysis is widely used in political science to discover latent structure, yet it is typically dismissed as "merely exploratory" because results depend heavily on algorithmic choice. This paper demonstrates how multi-algorithm comparison transforms cluster analysis into a rigorous framework for hypothesis testing. The core insight is that different clustering algorithms encode conflicting assumptions about data geometry and distribution. When algorithms with divergent mathematical foundations converge on the same solution, that convergence constitutes evidence that recovered patterns reflect genuine data features rather than methodological artifacts. I formalize this approach through a comparative validation framework with explicit evaluative criteria for cluster recovery, stability, and theoretical alignment. An application to competing models of acculturation demonstrates the framework's capacity to adjudicate between theoretical predictions: five algorithms were tested, three converged on a four-cluster solution that falsified the dominant binary model in political science. The paper provides a template for building inference in unsupervised learning more broadly, offering political scientists a principled approach for evaluating when discovered structure warrants theoretical conclusions.

Replication materials and code are available at: https://github.com/jagrijalva/bam-scale

"Interpretable Machine Learning for Political Behavior Research" (accepted at AAPOR Annual 2026 Conference; preparing for submission, Political Science Research and Methods)

Machine learning methods have gained traction in political science for text analysis, forecasting, and institutional research, but their application to individual-level political behavior remains limited. This paper argues that interpretable machine learning, specifically random forest models paired with SHAP (SHapley Additive exPlanations) value decomposition, offers significant advantages for understanding vote choice and political attitudes. I demonstrate this approach using Latino support for Trump as a case study, drawing on CMPS data from 2016 and 2020. The method addresses three limitations of traditional regression: (1) it captures non-linear effects and high-order interactions without manual specification; (2) SHAP values provide both global feature importance and individual-level explanations; and (3) it enables principled model comparison across time periods or subgroups. I provide a practical guide for implementation in Python/R, discuss best practices for variable selection (including handling tautological predictors), and illustrate how to interpret SHAP summary plots for substantive inference. The goal is to equip political behavior researchers with a powerful but underutilized toolkit for exploring complex, heterogeneous effects in survey data. 

Replication materials and code for the 2016 analysis are available at: https://github.com/jagrijalva/ml-latino-vote-2016