Immigrant threat and Latino/a disadvantage: disentangling the impact of immigration attitudes on ethnic sentencing disparities

ABSTRACT

The criminal justice system has increasingly been relied upon to address immigration apprehension, resulting in concerns that this institution will be abused in an effort to indirectly address this perceived social problem. The consequences of such an approach will likely extend to Latino/a populations as a result of rhetoric linking ethnicity, immigration, and crime. Despite popular theoretical frameworks suggesting that disadvantage will vary according to the size of the population and the extent of perceived threats toward this minority, many neglect attitudinal measures or fail to measure actual criminological outcomes. This project addresses this oversight by exploring potential mediating effects of attitudes on the relationship between population measures and ethnic sentencing disparities. After fitting multilevel models nesting cases within counties and states, the results indicate that there is significant variation across all levels. While greater disparities in Latino/a sentencing were found in counties with greater Latino/a populations, this relationship was nonlinear. Additionally, state level measures of immigrant threat attitudes appear to be stronger predictors of Latino/a sentencing disparities. These contextual effects are more influential than offender level ethnicity, supporting threat hypotheses and suggesting that measurement of this concept should not be limited to offender ethnicity and population characteristics alone.

KEYWORDS:

• Sentencing inequality
• immigrant threat
• multilevel modeling

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Many criticize the frequently utilized terminology of both Latino and Hispanic as a means of categorizing a large multitude of unique cultures that do not necessarily share language, race, or culture (Hayes-Bautista and Chapa 1987). Of these terms, Hispanic is more narrow in scope, compared to Latino, and is generally limited to nationalities that speak Spanish. Some argue that, due to the colonial origins of language such as this, this label is dated, restrictive, and potentially racist (Gimenez 1989). The term Latino, which is broad in nature and encompasses those of Latin American descent, is criticized for incorporating diverse cultures and neglecting to identify the nuanced uniqueness of varying identities. Although both of these classifications are somewhat misleading standardized terminology measures (Hayes-Bautista and Chapa 1987), this language is useful in acknowledging a shared political delineation frequently experienced within the United States. The term Latino is used throughout this study to acknowledge a broader sociodemographic grouping of various cultures and nationalities falling within panethnic parameters (Alcoff, 2005). Though the authors endeavor to avoid the pitfall of ‘implicit homogenizationʼ (Jones-Correa and Leal 1996), this language is preferable to Hispanic as it acknowledges the variability of cultural backgrounds and ethnic identities at risk of experiencing disadvantage in the United States. The term Hispanic is utilized only when referencing data and previous research that has elected to retain this language in an effort to respect the linguistic preferences of those authors and data collectors.

2. These states include Alabama, California, Colorado, Georgia, Idaho, Indiana, Iowa, Kentucky, Maine, Maryland, Massachusetts, Michigan, Minnesota, Missouri, Montana, Nebraska, Nevada, New Hampshire, New Jersey, New York, North Carolina, North Dakota, Oklahoma, Oregon, Pennsylvania, South Carolina, South Dakota, Texas, Utah, Washington, West Virginia, and Wyoming.

3. The percentage of missing information for those variables with missing values is race (13%), sentence length (3%), number of counts (34%), prior felony convictions (28%), education (23%), male (.01%), determinate sentencing (29%), prior jail time (31%), and prior prison time (26%). As such, these data are appropriate for multiple imputation. While missing values were imputed for measures with missing values, all measures in Table 1 were used as predictors in the imputation analysis (see Schaefer, 1997). The imputation models were also constrained to return an integer within the range of the original data. The sensitivity analysis indicates the missing data have been properly imputed and do not significantly vary from the original dataset with missing values.

4. Varimax rotation was utilized to achieve superior fit and understanding of the underlying structure while maintaining orthogonality and consistency in the eigenvectors without changing the underlying mathematical structure of the data. The two vectors both had eigenvalues greater than 1 (see Kaiser 1960). Additionally, these components also meet other criteria for retention (see Merenda 1997). Specifically, they presented in the step portion of the scree plot and have factor loadings greater than .70 (Mertler and Vannatta 2002). Finally, Cronbach’s alpha reliability statistics were requested for immigrant threat (.870) and fear of immigrant crime (.938) (see Cronbach 1947).

5. The assumptions of multilevel modeling were all assessed and have been met. Multicollinearity was assessed with variance inflation factors (VIF), and all remaining factors were not found to be collinear (VIF < 2). Correlations matrices were inspected at level 2 and 3 to check for highly correlated variables, and standard errors were monitored throughout to check for cross-level collinearity (Raudenbush and Bryk 2002). Finally, residual scatterplots and Q–Q plots indicate that assumptions of linearity and homogeneity of the variance have been met.

6. It may initially appear as though multicollinearity may have caused the change in parameter estimates for immigrant threat from Model 3 to 4 in Table 4. While interaction effects are naturally highly correlated with their main effects, we tested for collinearity beyond those assumed by the interaction effect. VIFs were requested for the variables in question: immigration threat (2.482), percent hispanic (2.595), percent noncitizen (1.108), hispanic offender (1.327) (see Table 5 for correlations matrix). Another symptom of multicollinearity is inflated standard errors and partial parameter estimates, however neither is present when comparing estimates from Model 3 and 4. Thus, this change may be due to suppression.

Additional information

Notes on contributors

Melanie M. Holland

Melanie M. Holland, Ph.D. is an assistant professor of Criminal Justice. Her research interests include courts and sentencing, social inequality, and race and ethnicity. Her prior work has appeared in Race and Justice, Criminal Justice Review, and the Journal of ethnicity in Criminal Justice, among others.

Richard J. Stringer

Richard J. Stringer, Ph.D. is an assistant professor of Criminal Justice. His research interests include drug and alcohol policy, courts and sentencing, policing, and advanced quantitative methods. His prior research has been funded by the US Department of Justice and has appeared in outlets such as the Journal of Drug Issues, Criminal Justice Policy Review, and the Journal of Crime and Justice.