The importance of race, gender, and religion in naturalization adjudication in the United States

Materials and Methods

The main source of data for this study is the set of administrative records on N-400 applications filed with USCIS (N-400 Data). The agency produced the N-400 Data in response to two requests filed by the first author in August 2018 and October 2019, respectively, under FOIA.^§ After exhausting administrative appeals, a legal complaint against USCIS was filed in the US District Court for the District of Columbia in July 2020. The data resulting from this litigation contains a variety of information from individual N-400 applications submitted to USCIS. We also collected information on the applicants’ places of residence (Applicant Contextual Data) by matching each applicant’s ZIP Code to their corresponding ZIP Code Tabulation Area (ZCTA). This step was necessary because ZIP Codes relate to mail delivery routes and do not map directly to spatial units (37). We then merged a variety of ZCTA-level information about the applicants’ places of residence from the American Community Survey.

Our analysis examines nonmilitary naturalization applications submitted by individuals 18 y or older that USCIS field offices in the United States adjudicated between the start of the first quarter of fiscal year 2015 (October 1, 2014) and the end of the second quarter of fiscal year 2018 (March 31, 2018). After cleaning the data and implementing listwise deletion, we analyze 2,687,101 nonmilitary applications adjudicated during the study period. For detailed information on our data collection, preparation, and coding strategy, see SI Appendix. SI Appendix, Table S1 summarizes all of the variables described next. A replication package containing all data and code used in this study’s analysis are available online.

Results

Table 1 presents univariate and bivariate statistics on all of the variables used in the analysis. As shown in Table 1, about 92% of all applications in the analytic sample were approved. About 18% of the applicants were White, about 31% were Asian, about 13% were Black, about 0.1% were Other, and Hispanics constituted about 38% of the sample. More than half of all the applicants were female (about 56%). About 12% of the sample originated from a Muslim-majority country. Sample statistics on the control variables provide a more in-depth portrayal of the overall population under study. The average CDI among applicants was 0.455, which indicates that overall, applicants in the sample resided in ZCTAs that were somewhat more disadvantaged than the average ZCTA in the United States. Most of the applicants (about 88%) satisfied the basic eligibility criteria for naturalization by having maintained lawful permanent residency in the United States for 5 y or more. A majority of the applicants were married (about 64%), and just under two-thirds of the applicants had children (about 65%).

The bivariate statistics in Table 1 suggest that certain racial/ethnic groups and applicants from Muslim-majority countries are overrepresented among those who were denied, whereas female applicants are underrepresented. These differences are generally reflected in the baseline model (Model 1a) results shown in Table 2. These results, however, do not adjust for demographic differences between groups that may be systematically related to varying odds of approval. For example, Whites are less likely to reside in socioeconomically disadvantaged communities than all other racial/ethnic groups (we find that Whites have the lowest average CDI score). Socioeconomic disadvantage, in turn, is associated with lower odds of approval. The full model (Model 1b of Table 2) partials out the shared explanatory power of these types of confounders.

Model 1b of Table 2 shows that all else being equal, the odds of approval are consistently smaller for non-White and Hispanic applicants than White applicants. For example, the odds of approval are about 41% smaller for Blacks compared to Whites [(1 − 0.591) × 100]. Table 2 also shows that male applicants are at a disadvantage compared to female applicants and so are applicants from Muslim-majority countries compared to applicants from other countries. Specifically, the odds of approval are about 18% larger for female than male applicants. For applicants from Muslim-majority countries, the odds of approval are about 43% smaller than those of their counterparts from other countries.

Next, we examine the results of a series of regression analyses that each includes two-way interactions described earlier. SI Appendix, Table S2 shows the statistics for model fit for the full model that was presented in Table 2 (see Model 1, “No Interaction Model,” in SI Appendix, Table S2) as well as for each of the interaction models (see Models 2 to 4 in SI Appendix, Table S2). The Akaike’s information criterion (AIC) and Bayesian information criterion (BIC) assess the overall fit of models. All else being equal, the smaller the AIC and the BIC, the better the model fit (45). The results summarized in SI Appendix, Table S2 thus suggest that we should prefer the interaction models. The full set of adjusted predicted probabilities for each subgroup resulting from the interactions are shown in SI Appendix, Tables S3–S5. Consistent with the prevailing method for testing interaction effects in logistic models (42, 46), we estimated the adjusted predicted probabilities conditional on relevant subsamples (e.g., White, Black, Asian, Other, and Hispanic). For example, to calculate the “Muslim Gap” for Whites shown in SI Appendix, Table S4, which is the average marginal effect among Whites of originating from a Muslim-majority country, we estimated the difference in average predicted probabilities among Whites for when Muslim-Majority Country is set to 1 versus when Muslim-Majority Country is set to 0.

Two sets of results are worth highlighting from the interaction models. First, SI Appendix, Tables S3–S5 show that there are statistically significant interactions between the Race/Ethnicity, Female, and Muslim-Majority Country variables (see “contrast” columns in each table). Second, the adjusted predicted probabilities shown in SI Appendix, Tables S3–S5 suggest that there are approval hierarchies among subgroups. To visually depict the nature of these hierarchies, Figs. 1–3 present each subgroup’s adjusted predicted probabilities of approval and respective 95% CIs. The extreme ends (the lowest and the highest probabilities of approval) in these hierarchies are marked in darker shades, and we focus our discussion on those given the large number of subgroups.

Fig. 1 shows that the adjusted predicted probability of approval for Black males is only about 89% compared to that of White females at 94%. Fig. 2 shows even greater subgroup disparities at the extreme ends of the approval hierarchy: Blacks from Muslim-majority countries have only about an 86% predicted probability of approval, whereas Whites from non–Muslim-majority countries have about a 96% predicted probability of approval. The interaction between gender and Muslim-country origin also reveals significant group disparities. Fig. 3 shows that females from Muslim-majority countries have only about an 87% predicted probability of approval compared to 93% for females from non–Muslim-majority countries.

Finally, our supplemental analysis results show that our main analysis results are robust to combining administrative closures with denials. As shown in SI Appendix, Table S6, the results of the logistic regression analysis in which we treated administrative closures as denials are substantially similar to the main logistic regression analysis results presented in Table 2.

Discussion

There is a broad understanding that historically, “American law had long been shot through with forms of second-class citizenship, denying personal liberties and opportunities for political participation to most of the adult population on the basis of race, ethnicity, gender, and even religion” (47). However, over the decades, Congress has sought to eliminate ascriptive inequality in naturalization (48). By ascriptive inequality, we mean stratification based on “traits present at birth (e.g., sex, race, ethnicity, parental wealth, nationality)” that “influence the subsequent social standing of individuals” (49, 50). Yet, our study findings suggest that ascriptive inequalities continue to persist. Specifically, our analysis shows that there are significant disparities in naturalization adjudication outcomes by race/ethnicity, gender, and Muslim-country origin. The intersection of these identities forms hierarchies of various subgroups that experience different levels of advantage in the adjudication process.

Overall, we generally find that applicants of color, particularly men and those from Muslim-majority countries, are at a disadvantage. These results are not only generally consistent with findings of disparate impact in research on immigration enforcement and on intersectionality discussed in the Introduction but also studies examining outcomes in immigration adjudication other than naturalization. For example, Rottman and colleagues have demonstrated that asylum claims of applicants from Arabic-speaking or Muslim-majority countries are more likely to be denied than those of other applicants in the post-9/11 era (51). Although some of the subgroup differences in approval rates in the current study may appear relatively small in the abstract, they translate into a large substantive impact when considered in light of the sheer number of individuals seeking naturalization. For example, a difference of 5 percentage points for Black males means that almost 8,000 additional Black males could have obtained approval during the study period had their approval rate been equal to that of White females. A difference of 9 percentage points for Blacks from Muslim-majority countries means over 3,600 additional Blacks from Muslim-majority countries could have obtained approval during the study period had their approval rate been equal to that of Whites from non–Muslim-majority countries. Finally, a difference of 6 percentage points for females from Muslim-majority countries means that more than 9,600 additional females from Muslim-majority countries could have obtained approval during the study period had their approval rate been equal to that of females from non–Muslim-majority countries.

One possible explanation for the disparities uncovered here might stem from discretionary components of naturalization adjudication. For example, a growing number of scholars have highlighted inconsistent, biased, or discriminatory enforcement of the good moral character provision of the naturalization law (17, 52–54). As Bishop has noted, “In the end, naturalization cases are decided by individual people who arrive at their determination from a whole host of experiences that may lead to intentional or unintentional biases and preferences” (55). Another possible explanation might relate to persistent structural inequalities in other domains of American social life. For example, if higher denial rates are associated with certain types of criminal history (a hypothesis we cannot test given the lack of reliable data on applicants’ criminal history), the group disparities we have presented here may be reflective in part of policing and other criminal justice practices that disproportionately impact certain groups, such as Black and Latino immigrant men (20) and Black Muslims (56).

A comprehensive understanding of the sources of these disparities requires (at a minimum) an analysis of the agency’s stated reasons for the denials. The law requires USCIS to include explanations of the basis for denial in decision notices issued to applicants (57). However, denial reasons are not reliably and systematically captured in the N-400 Data. For example, over 20% of denials in our analytic sample are missing on adjudication reason. Another 7% of denials have the adjudication reason of “Other.” Many of the other categories are equally uninformative. For instance, 7% of denials have the adjudication reason of “Secondary Evidence.” Although it bears noting that officially stated reasons for denial may not offer a full or even an accurate picture of why a particular application may have been denied, information about why an application was denied and at what stage of the adjudication process it was denied is an important first step in investigating why ascriptive inequality continues to exist in naturalization adjudication. That such critical information with clear policy implications is not being collected by USCIS in a comprehensive and standardized way underscores the need for greater transparency and accountability in the naturalization adjudication process.

Another important question raised by this study that warrants future investigation is the extent to which administrative denials are judicially appealed and the outcomes of those judicial appeals. Although naturalization decisions were historically the province of courts, the Immigration Act of 1990 transferred the exclusive power of naturalization from the judiciary to the Attorney General, making the process an almost exclusively administrative system. As one scholar has noted: “With the lesser role of courts in naturalization post-1990, inconsistencies and bias in citizenship adjudication may have become harder to identify and check as adjudication lies in the hands of a single administrative officer” (17). Yet, insofar as judicial review continue to serve as a procedural safeguard for unfair or arbitrary administrative decision-making, we need to understand how often denied applicants engage this process and if not, why not. Other related questions that are also of critical importance include: Which groups are more or less likely to judicially appeal adverse agency decisions, and why? Do the appellants have lawyers, and does legal representation matter in obtaining favorable outcomes in such appeals? If the agency decisions are reversed, what are the bases for the reversals? Addressing these questions will ensure a fuller understanding of the nature and scope of barriers to naturalization and the role of courts in resolving contested questions of citizenship (58, 59).

Finally, this study underscores the need for research on the broader consequences of naturalization denial on immigrants and their family members. A growing body of research has explored what naturalization means for immigrants in terms of their social, economic, and political belonging and integration (60–63). Studies also show that the fates and rights of citizens and noncitizens are closely linked (64, 65), which means that naturalization has far wider and radiating effects than the individual to whom citizenship has been granted. Conversely, naturalization denial likely has far-reaching consequences that go beyond the individual whose application has been denied. Future research on naturalization should attend to these broad short- and long-term impacts of naturalization denials for families and communities and how disparities in naturalization approval rates may be shaping divergent social and economic outcomes for families and communities.

Supplementary Material

Data Availability

Administrative data and code used in this study are posted at openICPSR (https://doi.org/10.3886/E158741V1) (66).