Race/Ethnicity, Perceived Skin Color, and the Likelihood of Adult Arrest

Abstract

Research has long-documented racial/ethnic disparities in criminal justice outcomes. However, despite race/ethnicity being a multidimensional social construct, prior research largely relies on self-identification measures, thereby disregarding research on skin tone stratification within-racial/ethnic groups. The current study extends beyond this by examining the relationship between race/ethnicity and arrest employing both self-identified race/ethnicity and perceived skin color. Using data from the National Longitudinal Study of Adolescent to Adult Health, we explore the main and intersecting effects of self-identified race/ethnicity and perceived skin color on experiencing an arrest in adulthood between- and within-self-identified Whites, Blacks, Latinos, Native Americans, and Asians. We use structural disadvantage as a framework for exploring how social structural factors as well as antisocial behavior mediate the relationship between race/ethnicity/color and arrest. Results suggest that focusing on the racial/ethnic disparities alone masks differences in arrest by color and that the effect of color varies by race/ethnicity. Results also suggest that measures indicative of disadvantage, but not offending, partially explain these associations.

A vast literature examines the relationship between race/ethnicity and police contact and arrest (Kochel, Wilson, & Mastrofski, 2011; Mitchell & Caudy, 2015; Mosher, Miethe, & Hart, 2011; Stewart, Warren, Hughes, & Brunson, 2017; Tapia, 2010; Tonry, 2011). However, despite being a multidimensional social construct that encompasses extra-racial phenotypic characteristics (Dixon & Telles, 2017), few studies on racial/ethnic disparities in arrest use indicators of race/ethnicity beyond self-identification. Moreover, although there is a substantial literature showing that, within-racial/ethnic groups, lighter skinned individuals are more advantaged than those with darker skin (Dixon & Telles, 2017; Hochschild & Weaver, 2007; Hunter, 2013; Monk, 2014), few studies examine the relationship between skin color and police contact—of which the evidence is mixed (Alcalá & Montoya, 2016; Barlow & Barlow, 2002; Branigan, Wildeman, Freese, & Kiefe, 2017; Kizer, 2017; White, 2015). Notably, evidence for the association between color and arrest may be inconclusive given the differences between, and limitations of, previous work.

The current study contributes to existing research by providing a more thorough and robust test of the influence of self-identified race/ethnicity and perceived skin color on arrest. Using the National Longitudinal Study of Adolescent to Adult Health (Add Health), we examine the main and intersecting effects of self-identified race/ethnicity and perceived skin color on arrest in adulthood between and within Whites, Blacks, Latinos, Native Americans, and Asians. We also move beyond previous research by examining the unconstrained effect of color (i.e., we make no assumptions about the linearity of color effects). Further, to understand how macro-level dynamics produce disparities in the criminal justice system (CJS), we use a lens of structural disadvantage to explore how social structural factors act as mechanisms through which darker skinned people are more likely to be arrested. We also consider how offending, drug use, and alcohol use explain this association. We find some, albeit mixed, evidence of a relationship between skin color and arrest. We reconcile these findings with the existing literature and conclude that focusing on the racial/ethnic disparities alone masks differences in arrest by color and that the effect of color varies by race/ethnicity.

Background

Racial Disparities in Arrest

Empirical research largely confirms that Blacks, Latinos, and Native Americans are disproportionally arrested relative to Whites, and Asians are underrepresented among arrestees (Mitchell & Caudy, 2015; Mosher et al., 2011; Snyder, 2011; Tapia, 2010; Tonry, 2011). Kochel, Wilson, and Mastrofski’s (2011) meta-analysis reveal that the likelihood of a Black suspect being arrested is 30% greater than a White suspect, net of legal and extralegal factors. Additionally, Blacks and Latinos are more likely to be arrested than Whites even when accounting for selection into the CJS (Mitchell & Caudy, 2015; Tapia, 2010). The reason for these observed disparities is unclear.

Some speculate disparities in the CJS are due to higher levels of violent or more serious offending by minorities (Beaver et al., 2013; S.W. Perry, 2004; Tonry, 2011). However, self-report data often reveal more similarities than differences across race/ethnicity, particularly among less serious offenses (Mosher et al., 2011) and substance use (e.g., Tonry, 2011). Thus, we contend that more serious offending, as opposed to drug and alcohol use, could explain the association between race/ethnicity, color, and arrest. Given the discretion officers have on the job, some suggest Blacks, Latinos, and Native Americans are arrested more due to racial profiling and discrimination by police (Mosher, 2001; B. Perry, 2006). Others postulate that police focus their attention on minorities due to higher associations of threat and criminality with minority groups (Eberhardt, Goff, Purdie, & Davies, 2004; Johnson & King, 2017; Tonry, 2011). Many Americans, including minorities, hold conscious stereotypes that categorize Blacks (Tonry, 2011), Latinos (Holmes, Smith, Freng, & Muñoz, 2008), and Native Americans (Holmes & Antell, 2001) as unpleasant, violent, dangerous, and criminal, whereas Asians are thought of as nonthreatening (Johnson & King, 2017). Additionally, Eberhardt, Goff, Purdie, and Davies (2004) find that police officers are more likely to perceive Blacks, and more stereotypically Black individuals, as criminal, exemplifying that officers do use racial categorizations on the job. Overall, we contend that research provides some support that racial biases exist in the CJS.

However, although race/ethnicity is a multidimensional social construct (Dixon & Telles, 2017), much of the research on race/ethnicity and CJS outcomes only examines the effect of race/ethnicity based on self-identification questions (i.e., how one classifies their own racial/ethnic background). Often, this is used as a proxy for perceived race/ethnicity (i.e., how one’s race/ethnicity is classified by others), but this is problematic because self-identification does not necessarily correspond with perceived classification (Vargas & Stainback, 2016). Furthermore, this disregards research on skin tone stratification within-racial/ethnic groups.¹

Empirical Research on Skin Color

Despite the transition away from societal practices blatantly indicative of stratification by race/ethnicity and color (Hunter, 2007, 2013; Rondilla & Spickard, 2007), skin tone stratification still exists (Monk, 2014). For example, lighter skinned Blacks (Hughes & Hertel, 1990; Kreisman & Rangels, 2015; Ryabov, 2013) and Latinos (Allen, Telles, & Hunter, 2000; Mason, 2004) have more education, higher incomes, and better employment prospects than their darker counterparts. Moreover, Afro-Latinos (i.e., those with darker skin tones who identify racially as Black) have worse economic and health outcomes than non-Afro-Latinos (Adames & Chavez-Dueñas, 2016). Although scholarship focusing on color among Asians is limited, lighter skin is advantageous for Asians’ body satisfaction, self-esteem, self-rated health, and relationships (Rondilla & Spickard, 2007; Sahay & Piran, 1997). Moreover, Hall’s (2014) historical analysis of skin color discrimination among Native Americans provides evidence that those with lighter skin tones generally experience less difficulty in life than their darker skinned counterparts. As a whole, given that lighter skinned individuals are generally more advantaged than darker skinned individuals of the same race/ethnicity, skin tone stratification is still prevalent today.

Skin Color and the CJS

Scholars have long discussed the possibility that racism, or differences in the likelihood of arrest by racial/ethnic background, exists in the justice system. Fewer studies, however, consider whether colorism (Hunter, 2007), or differences in the likelihood of arrest by skin color within individuals of the same race/ethnicity, exists in the CJS. Yet, in addition to documented racial/ethnic stereotypes, darker skinned individuals are often categorized as violent and dangerous and are presumed to be guilty (Levinson & Young, 2010). Thus, as race/ethnicity is a multidimensional social construct that reflects self-identification and perceived classification, family ancestry, phenotype, and social context (Dixon & Telles, 2017; Roth, 2016), we contend colorism likely exists in the CJS.

Indeed, studies find that darker skin is associated with harsher probation and incarceration sentences among Whites, Blacks, and Latinos (e.g., Gyimah-Brempong & Price, 2006; Hochschild & Weaver, 2007; King & Johnson, 2016; Steinmetz & Koeppel, 2017; Viglione, Hannon, & DeFina, 2011). Related research on Afrocentric feature bias (e.g., Eberhardt, Davies, Purdie-Vaughns, & Johnson, 2006; King & Johnson, 2016; Petersen, 2017; Pizzi, Blair, & Judd, 2005) suggests that stereotypical Black features are associated with harsher treatment in the CJS for White and Black defendants.

Conversely, research investigating the influence of color on arrest is limited and has mixed findings. Particularly relevant to the current study, White’s (2015) research, using Add Health, reveals that color is a predictor of arrest among Latinas, but not among Latino men or Black men or women. Yet some evidence suggests the effect of color among Latinos only exists for second-generation Latinos (Alcalá & Montoya, 2016). Moreover, Branigan, Wildeman, Freese, and Kiefe (2017) find that darker skin is not detrimental for arrests among Black men but is for White men. Further, Kizer (2017), also using Add Health, finds that the effect of color, controlling for race, is a significant predictor of arrest among male nonsiblings and sibling pairs across Black, Latino, Native American, and Asian respondents. The inconclusive findings of previous research may be, in part, due to the variation in the racial/ethnic background of samples, the data sets examined, the measurements of color, as well as variation in examining the effect of color between versus within race/ethnicity. Moreover, existing research has some limitations that may also be contributing to the inconclusive findings.

For instance, some studies (e.g., Branigan et al., 2017) suggest darker skin tones are detrimental among Whites, but this has not been examined using a nationally representative sample. Moreover, previous work using Add Health data has not taken full advantage of the nationally representative nature of the sample. For example, Kizer (2017) examines nonprobability samples from Add Health (a sample of males and male sibling pairs). Additionally, research does not consistently examine the effect of color within-racial/ethnic groups (e.g., Kizer, 2017); this is particularly problematic because it assumes that the association between skin color and arrest is constant across respondents of all races, which previous research suggests is not the case (e.g., Branigan et al., 2017; White, 2015). Moreover, despite evidence for the existence of skin color stratification among Native Americans and Asians, even studies examining the effect of color on arrest within-racial/ethnic groups (e.g., Branigan et al., 2017; White, 2015) exclude Native Americans and Asians. Furthermore, despite that King and Johnson (2016) find evidence that the effect of color on sentencing outcomes among Black defendants is nonlinear, research investigating the influence of color on arrest primarily uses ordinal measures that are treated as continuous and are not transformed to account for the possible nonlinear effects of color (e.g., Alcalá & Montoya, 2016; Kizer, 2017; White, 2015), thereby constraining the effect of color to a single, linear estimate. Finally, previous research has examined the association between color and arrest using only basic control variables, and, surprisingly, researchers have not included measures of substance use in their analyses, despite being available in the data examined (e.g., Kizer, 2017; White, 2015). Given that minorities are more likely to be arrested for drug-related offenses, despite similar patterns of usage as Whites (e.g., Tonry, 2011), one could expect that the association between skin color and arrest may be different if such controls were included. Thus, it is plausible that the influence of skin color on arrest is unclear given the differences between and limitations of previous studies.

Theoretical Framework: Structural Disadvantage

Racial/ethnic and skin tone disparities in arrest are better understood when considering the larger societal context of race relations. Indeed, despite growing awareness of racial issues in America, especially given the increasing attention to police shootings of unarmed Black and Brown men and the Black Lives Matter movement, structural disadvantage remains omnipresent. As Powell (2007) explains, structural racism encompasses macro-level systems, social institutions, and social processes that intersect to generate and perpetuate racial/ethnic inequalities. Notably, individual differences in education, income, and prior criminal record are, in part, products of structural racism (Haney-López, 2007). Since race/ethnicity and color lead to an accumulation of disadvantage within and across domains, examining CJS contact through the lens of structural disadvantage helps to hone in on the macro-level dynamics that produce racial/ethnic disparities in the system (Powell, 2007). Given that disadvantaged minorities are disproportionally represented in the CJS, and contact with the system has consequences that hinder social and economic advancement, the CJS arguably sustains and amplifies structural disadvantage.

Present Study and Hypotheses

The current research addresses the limitations of previous work by examining the effects of self-identified race/ethnicity and perceived skin color on the likelihood of experiencing an adult arrest White, Black, Latino, Native American, and Asian respondents in Add Health. We also expand on previous research by examining: (a) the main effect of color controlling for race/ethnicity (similar to Kizer, 2017), (b) the main effect of race/ethnicity controlling for color (similar to Kizer, 2017), (c) the intersecting effects of race/ethnicity and color, (d) and the effect of color within-racial/ethnic groups (similar to Branigan et al., 2017; King & Johnson, 2016; White, 2015). Utilizing these methods of examination in one paper with a nationally representative sample assists in determining the reliability of these previous examinations. Additionally, this study advances the field by examining the unconstrained effect of color. Finally, our work moves beyond previous research by examining indicators of socioeconomic disadvantage and antisocial behavior (including drug and alcohol use) as possible explanations for the higher likelihood of arrest among minorities and darker skinned individuals. Overall, by addressing the limitations of previous research, the current study provides a more thorough and robust test of the association between race/ethnicity and skin color on arrest.

For all reasons previously discussed, our hypotheses are as follows.

Hypothesis 1: When testing the robustness of the main effect of race/ethnicity, compared to Whites, the likelihood of arrest will be greater for self-identified Blacks, Latinos, and Native Americans and lower for Asians.

Hypothesis 2: The likelihood of experiencing an arrest will be higher for darker skinned than lighter skinned individuals when examining the main effect of color.

Hypothesis 3: Skin color will matter between-racial groups in that lighter skin will be advantageous and darker skin will be detrimental across racial groups (i.e., compared to white-skinned Whites, the likelihood of arrest will not be higher for lighter skinned Blacks, Latinos, Native Americans, whereas the likelihood of arrest for darker skinned Blacks, Latinos, Native Americans, and Asians will be higher than white-skinned Whites).

Hypothesis 4: Color will matter within groups, such that darker skinned Whites, Blacks, Latinos, Native Americans, and Asians will have a higher likelihood of arrest than those with lighter skin.

Hypothesis 5: Characteristics indicative of one’s position in the larger social structure will, in part, explain the association between color and arrest. Offending will, in part, explain the association between race/ethnicity/color and arrest, but drug and alcohol use will not.

Data and Measures

We use data from Add Health. Add Health sampled a nationally representative group of adolescents in the 7th–12th grades in the 1994–1995 school year for its first wave and has since conducted three additional interviews with the participants. The first wave of the in-home interviews had a core of 12,105 adolescent participants plus additional oversampled groups for a total of 20,745 participants. The second interview was conducted in 1996, the third was conducted in 2001/2002, and its most recent (Wave IV) was conducted in 2007/2008 (when the sample was mostly 24–32 years old).

Of the 13,034 respondents who participated in the in-home interviews at Waves I, III, and IV, those who do not have valid data on the dependent/focal independent variables or sample weight are excluded, yielding an analytic sample of 12,160 respondents. Less than 5% of the sample is missing on the foreign-born, offending, substance use, alcohol use, income, neighborhood disadvantage, and interviewers’ race indicators, and less than 11% of the sample is missing on the attained Socioeconomic status (SES) measure; we use multiple imputation to address missing data.

Dependent Variable: Adult Arrest

Adult arrest is based on a retrospective question at Wave IV, which asks how many times respondents had been arrested since their 18th birthday. Given the highly skewed distribution of the continuous measure of the number of arrests, this variable is dichotomously coded (being arrested one or more times is coded as 1 and being arrested zero times is coded as 0). Sensitivity analyses examining the count of arrests using negative binomial regressions reveal results are largely consistent with those presented here.

Independent Variables: Race/Ethnicity and Skin Color

Respondents’ self-identified race/ethnicity is based on questions at Wave III that asked respondents if they were of Hispanic or Latino origin and what race they were (with response options of White, Black or African American, American Indian or Native American, or Asian or Pacific Islander). The Latino category in the current study is all-inclusive in that it is comprised of respondents of all races or multiracial respondents who identified as ethnically Latino. Whites, Blacks, Native Americans, and Asians in the current study identified ethnically as non-Latino and racially as White, Black, Native American, or Asian, respectively. Non-Latino respondents who self-identified as multiracial were asked which one category best described their racial background and we code these respondents into the category they indicated best described their background (e.g., multiracial respondents who indicated that “White” best described their background are coded as self-identified Whites). Thus, we use mutually exclusive categories of non-Latino White (reference), non-Latino Black, Latino, non-Latino Native American, and non-Latino Asian.

Perceived skin color (measured at Wave III) is reported by the interviewer. Specifically, interviewers answered, “What is the respondent’s skin color?” Responses included: white, light brown, medium brown, dark brown, and black.² While some researchers (e.g., Kizer, 2017; White, 2015) treat this ordinal measure of color as a continuous variable, this approach constrains the effect of color to a single, linear estimate as opposed to allowing for variation in the effects of color at different levels. This is problematic since there is evidence for nonlinear effects of skin tone (King & Johnson, 2016; Monk, 2015). Thus, we examine the effects of color on arrest using these indicators as mutually exclusive categories with white skin as the reference.

We also examine the intersecting effects of racial/ethnic self-identification and perceived skin color as focal variables since color, although associated with race/ethnicity, is a distinct physical characteristic that is often excluded in research on racial disparities in the CJS. For self-identified Black and Latino respondents, the skin color categories are partitioned to ensure each category includes at least 15% of the respective racial/ethnic group. As the distribution of skin color is particularly skewed within self-identified Whites, Native Americans, and Asians, we dichotomize the color categories for these groups. Specifically, we create 12 mutually exclusive dichotomous variables to indicate intersecting effects: white-skinned Whites, non-white-skinned Whites, white/light brown/medium brown-skinned Blacks, dark brown-skinned Blacks, black-skinned Blacks, white-skinned Latinos, light brownskinned Latinos, medium brown/dark brown/black-skinned Latinos, white-skinned Native Americans, non-white-skinned Native Americans, white-skinned Asians, and non-white-skinned Asians.

Social Structural Factors

Characteristics indicative of social structural position, which might explain differences in arrest by race/ethnicity or color, are measured at Wave III. Attained SES is based on the sum of the respondents’ highest level of achieved education and occupation, ranging from 1 to 10, where higher values indicate higher SES (following Dennison & Demuth, 2018). Income is a continuous variable measured in 1000s of dollars (Swisher, Kuhl, & Chavez, 2013). Respondents missing on the income variable provided the best guess of their income from a list of categories (e.g., $10,000–14,999; $15,000–19,999); following Mangino (Forthcoming), these respondents were assigned the monetary value equal to the midpoint of the category they chose as their best guess. The binary marital status variable indicates whether the respondent was currently married. Using respondents’ geo-coded residence and the corresponding data reported by the Census Bureau’s American Community Survey, neighborhood disadvantage (α = .62) includes measures for the proportion of female-headed households, the proportion of the population below the poverty level in 1999, the proportion of the population 25 and older with less than a high school diploma, and the proportion of the population that is Black, Latino, or Native American. These indicators are standardized, so that the mean is 0 and the standard deviation is 1 and then summed to create the neighborhood disadvantage variable where higher values indicate higher disadvantage (Elliott et al., 1996).

Antisocial Behavior

To account for the possibility that arrest is based on behavioral risk, we include measures of antisocial behavior from Wave III. Offending (α = .74) ranges from 0 to 14 and is a sum of one’s involvement in criminal behavior over the past 12 months, such as vandalism, stealing, selling drugs, and violence. We construct a scale of drug use (α = .66) with questions that ask about drugs used in the year preceding the interview (range 0–5), with higher values indicating more drug use. Three questions that ask about alcohol use are summed to make a scale (α = .90) that ranges from 0 to 18, with higher values indicating more alcohol use.

Demographics

We include demographic controls, measured at Wave III, for gender (male) and age. Nativity status, measured at Wave I, is dichotomously coded to indicate whether the respondent reported being foreign-born or born in the United States. As prior research finds that interviewers’ self-identified race influences average perceived skin color ratings of respondents (Kreisman & Rangel, 2015), we also control for interviewers’ self-identified race at Wave III.

Analytic Strategy

Using SAS 9.4, we adequately account for Add Health’s complex sampling design with weighted logistic regressions. In the first set of analyses, we examine the main effect of self-identified race/ethnicity, the main effect of perceived color, and the effect of color after controlling for race/ethnicity. The second set of analyses examines the intersecting effects of race/ethnicity and color between-racial/ethnic groups. Next, we add demographic controls. We then explore the possibility that indicators of socioeconomic disadvantage act as potential mediators that undergird structural racism. Next, we explore antisocial behavior, which may account for the relationship between race/ethnicity, color, and arrest. In the last model, we account for both disadvantage and behavior. We then replicate these analyses separately by race/ethnicity to examine the influence of color within-racial/ethnic groups.

Results

Table 1 presents percentages and means for all variables partitioned by race/ethnicity and color. In total, 26% of all respondents report being arrested in adulthood. Moreover, the percent of those who experienced an adult arrest increases as skin color darkens. For example, 23% of white-skinned Latino respondents experienced an arrest, whereas 32% of the darkest skinned Latinos experienced an arrest. Attained SES generally decreases as skin color darkens. Offending is generally higher among those with darker skin tones, but alcohol and drug use are not.

Table 1.

Proportions and Means of All Variables by Self-Identified Race/Ethnicity and Perceived Skin Color.

Race/Ethnicity	White		Black			Latino			Native American		Asian
Sample Size	6,799		2,550			1,917			105		789
Skin Color by Race/Ethnicity	W	N	WLM	D	B	W	L	MDB	W	N	W	N	Total Sample
Sample size	6,546	253	1,140	710	700	903	720	294	43	62	242	547	12,160
Proportion	0.96	0.04	0.45	0.28	0.27	0.47	0.38	0.15	0.41	0.59	0.31	0.69	1.00
Dependent variable
Adult arrest	0.25	0.28	0.29	0.31	0.37	0.23	0.27	0.32	0.25	0.66	0.13	0.18	0.26
Demographics
Male	0.49	0.51	0.40	0.49	0.56	0.48	0.47	0.63	0.50	0.61	0.52	0.51	0.49
Age	21.71	21.42	22.14	21.83	21.84	21.94	21.85	21.81	21.48	21.41	22.02	21.97	21.77
Foreign-born	0.01	0.03	0.01	0.01	0.02	0.18	0.24	0.32	0.07	0.06	0.52	0.40	0.05
Social structural factors
Attained SES	5.02	4.78	4.83	4.76	4.70	4.67	4.50	4.51	3.83	3.83	5.97	5.40	4.93
Attained Income	34.46	30.96	24.20	25.22	19.95	30.97	32.41	32.32	19.59	23.43	42.00	41.36	32.53
Married	0.18	0.16	0.10	0.12	0.10	0.19	0.24	0.24	0.15	0.13	0.06	0.15	0.17
Neighborhood disadvantage	−1.53	−1.59	2.65	2.93	2.78	1.41	1.36	3.03	1.10	7.03	−0.91	−1.03	−0.44
Antisocial behavior
Offending	0.60	0.69	0.68	0.71	0.91	0.61	0.58	0.65	0.81	1.19	0.48	0.59	0.62
Drug use	0.62	0.61	0.35	0.26	0.33	0.49	0.42	0.29	0.76	0.59	0.36	0.38	0.54
Alcohol use	5.45	5.33	2.76	2.65	2.59	4.18	3.86	3.47	3.72	5.59	3.54	4.15	4.80

Note. Skin color categories: W = White; N = non-White; L = light brown; M = medium brown; D = dark brown; B = black; SES = socioeconomic status.

Table 2 reports the odds ratios associated with race/ethnicity and color on arrest. Odds ratios indicate a percent change in the odds of arrest after computing (100 × [odds ratio – 1]). An odds ratio greater than one suggests an increase in the likelihood of arrest, whereas an odds ratio less than one suggests a decrease. In Model 1, which tests the main effect of color, light brown-skinned respondents are not more likely, and those with dark brown skin are only marginally more likely, to be arrested than those with white skin; this exemplifies the importance of examining the unconstrained effect of color, as opposed to treating categorical measures of color as continuous measures. However, consistent with our second hypothesis, relative to white skin, the odds of arrest are 31% and 76% higher for those with medium brown and black skin, respectively. Tests of coefficient differences (proposed by Wooldridge, 2000) reveal that black-skinned respondents are more likely to be arrested than light and medium brown-skinned respondents. Model 2, which provides a robustness test of the main effect of race/ethnicity, reveals that the odds of arrest for Blacks and Native Americans are higher, and the odds of arrest for Asians are lower, than for Whites. Although Latinos are not more likely to experience an adult arrest than Whites, the effects of race/ethnicity largely support our first hypothesis. Examining the effect of color when controlling for self-identified race/ethnicity (Model 3) reveals darker skin tones are associated with a higher likelihood of arrest compared to those with white skin. Specifically, the odds of arrest are 32%, 38%, and 84% higher for those with light brown, medium brown, and black skin, respectively, compared to those with white skin.

Table 2.

Logistic Regression of Adult Arrest on Perceived Skin Color and Self-Identified Race/Ethnicity.

Predictor	Model 1	Model 2	Model 3
Perceived skin color
White (reference)
Light brown	1.17		1.32*
Medium brown	1.31*		1.38*
Dark brown	1.32†		1.39†
Black	1.76***		1.84***
Self-identified race/ethnicity
White (reference)
Black		1.40**	0.94
Latino		1.09	0.93
Native American		2.93**	2.43*
Asian		0.60*	0.49**
Constant	0.33***	0.34***	0.34***
Standardized composite indicators
Perceived skin color			1.88***
Self-identified race/ethnicity			1.64***

Note. n = 12,160. Odds ratios are presented. All models control for interviewers’ race. Standardized composite indicators are created by summing the product of each variable and its standardized estimate.

^†p < .10.

^*p < .05.

^**p < .01.

^***p < .001.

To compare the overall effect of race/ethnicity and color, we calculated standardized composite indicators, which are an exact linear combination of a group of estimates (Bollen & Bauldry, 2011; Heise, 1972). Composite indicators combine the effects of multiple variables into a single composite coefficient by summing the product of each variable and its standardized estimate. That is, the standardized coefficients for light brown, medium brown, dark brown, and black, as well as Black, Latino, Native American, and Asian, are combined into a single composite coefficient that represents the overall effect of color and race/ethnicity, respectively. A benefit of this approach is that standardized composite coefficients can be compared to other standardized composite variables, thus allowing for between-measure comparisons. Testing whether the composite indicators for race/ethnicity and color are different (Wooldridge, 2000) reveals that the composite indicators are not statistically different from one another, indicating that color is as strong of a predictor of arrest as race/ethnicity. Yet, given that examining the effect of color while controlling for race/ethnicity assumes the influence of color is the same for all groups, we also examine the effect of color within each group in subsequent analyses.

Table 3 presents the intersecting effects of self-identified race/ethnicity and perceived skin color on arrest, referencing white-skinned Whites. Model 1 shows that non-white-skinned Whites are not more likely to be arrested than those with white skin. The odds of arrest for the lightest skinned Blacks are marginally higher, and the odds of arrest for dark brown- and black-skinned Blacks are 34% and 78% higher, respectively, compared to white-skinned Whites. Moreover, the odds of arrest for lighter skinned Latinos and Native Americans are no higher than for white-skinned Whites. Yet, the odds of arrest for medium brown-/dark brown-/black-skinned Latinos are 48% higher, and the odds of arrest for non-white-skinned Native Americans are over 5 times higher, than for white-skinned Whites. Further, the odds of arrest for white-skinned Asians are 56% lower than white-skinned Whites, although non-white-skinned Asians do not have lower odds of arrest than white-skinned Whites. This supports our third hypothesis that lighter skin colors are advantageous and darker skin colors are detrimental across racial/ethnic groups (i.e., when referencing white-skinned Whites). Testing for coefficient differences (Wooldridge, 2000) in Model 1 provides evidence that skin color is influential within Blacks, Latinos, and Native Americans in that those with the darkest skin tones are more likely to be arrested than their lighter skinned counterparts. Given that color gradations are influential even when accounting for self-identified race/ethnicity illustrates the importance of considering the role of skin color in the association between race/ethnicity and contact with the CJS.

Table 3.

Logistic Regression of Adult Arrest on the Intersection of Self-Identified Race/Ethnicity and Perceived Skin Color.

Predictor	Model 1	Model 2	Model 3	Model 4	Model 5
Self-identified race/ethnicity and perceived skin color
White
White skinned (reference)
Non-white skinned	1.18	1.17	1.12	1.20	1.15
Black
White/light brown/medium brown skinned	1.21†	1.44**	1.17	1.79***	1.50**
Dark brown skinned	1.34*	1.40*	1.14	1.89***	1.56**
Black skinned	1.78***w	1.72***	1.41*	2 26***	1.92***
Latino
White skinned	0.93	1.02	0.87	1.12	0.95
Light brown skinned	1.14	1.32*	1.12	1.54**	1.28
Medium brown/dark brown/black skinned	1.48*w	1.51*	1.26	1.99***w	1.63*w
Native American
White skinned	0.98	0.99	0.64	1.07	0.70
Non-white skinned	5.80***w	6.00***w	4 07***w	6.45***w	4.07***w
Asian
White skinned	0.44***	0.55*	0.62*	0.63†	0.79
Non-white skinned	0.69	0.79	0.83	0.86	0.95
Demographics
Male		3.80***	3.45***	2 94***	2.66***
Age		0.99	1.03	1.04†	1.07**
Foreign-born		0.53***	0.56***	0.60***	0.63**
Social structural factors
Attained SES			0.79***		0.76***
Attained income			1.00		1.00
Married			0.63***		0.84†
Neighborhood disadvantage			1.02†		1.03*
Antisocial behavior
Offending				1.24***	1 23***
Drug use				1 52***	1.51***
Alcohol use				1.05***	1.06***
Constant	0.34***	0.18***	0.28**	0.04***	0.08***

Note. n = 12,160. Odds ratios are presented. All models control for interviewers’ race. SES = socioeconomic status.

^†p < .10,

^*p < .05,

^**p < .01,

^***p < .001 indicate significant difference from White respondents with white skin (i.e., between-race/ethnicity significance).

^wp < .05 indicate significant difference from the lightest skinned counterparts of that racial/ethnic group (i.e., within-race/ethnicity significance).

Model 2 adds demographic controls and the benefits of light skin color persist for all racial/ethnic groups except self-identified Blacks. Lighter skin color is not advantageous for Blacks since the odds of arrest for Blacks are higher than white-skinned Whites regardless of skin color, suggesting racial discrimination against Blacks persists. However, as expected, lighter skin is still advantageous and darker skin is still detrimental for Latinos, Native Americans, and Asians compared to white-skinned Whites. Tests of coefficient differences suggest the detrimental effects of having dark skin within Native Americans remains.

We explore whether social structural factors act as pathways that can lead to experiencing an arrest in Model 3. Effect sizes for all groups, except Asians, are reduced when including measures indicative of structural disadvantage; however, color still matters for Blacks, Native Americans, and Asians in relation to white-skinned Whites. Lighter skinned Blacks and Native Americans are not more likely to be arrested than white-skinned Whites. Yet the odds of arrest for black-skinned Blacks are 41% higher, and the odds of arrest for non-white-skinned Native Americans are over 4 times higher, than for white-skinned Whites. Further, white-skinned Asians are less likely to experience an arrest than white-skinned Whites, although darker skinned Asians are not. Thus, color is a characteristic that matters between-racial/ethnic groups for Blacks, Native Americans, and Asians even when accounting for socioeconomic factors. Again, the effect of color within Native Americans persists.

Antisocial behavior, including offending, drug use, and alcohol use, is examined in Model 4. Regardless of skin tone, the odds of arrest for all self-identified Blacks are over 79% higher than the odds of arrest for white-skinned Whites. Yet, between-racial/ethnic groups, light skin is advantageous and darker skin is detrimental for Latinos and Native Americans when compared to white-skinned Whites, partially supporting our third hypothesis. Furthermore, white-skinned Asians are marginally less likely to experience an arrest than white-skinned Whites. Tests of coefficient differences (Wooldridge, 2000) reveal that Latinos and Native Americans with dark skin are more likely to be arrested than their lighter skinned counterparts.

Presumably, if the effects of race/ethnicity and color operate indirectly through behavior (i.e., if certain racial/ethnic groups are more likely to be arrested because they engage in more antisocial behavior), then we would expect the effects of race/ethnicity and color to diminish in size after controlling for these measures. This is not the case. Instead, when we control for the differences in offending between groups (comparing Model 4 to Model 2), the odds of arrest for self-identified Blacks and Latinos, generally regardless of color, increase. That is, when accounting for the independent effect that antisocial behavioral has on the odds of arrest, we see that the prospects of arrest among Blacks and Latinos increase considerably. This suggests that the association between race/ethnicity and color is not driven by behavior. Supplemental analyses exploring offending, drug use, and alcohol use separately reveal the likelihood of arrest among Blacks and Latinos decreases when offending is accounted for, but increases when drug and alcohol use are accounted for. That is, Black and Latino respondents are more likely to experience an arrest when accounting for drug use and alcohol use than when not accounting for these indicators. This pattern provides support to our fifth hypothesis that offending, as opposed to drug and alcohol use, explain the association between race/ethnicity, color, and arrest.

Model 5 includes social structural factors and antisocial behavior. Again, non-white-skinned Whites are not more likely to be arrested than white-skinned Whites. All Blacks across the spectrum of skin color are more likely to experience an arrest than white-skinned Whites, which suggests the presence of racial discrimination against Blacks. Yet, compared to white-skinned Whites, color still appears to be a characteristic that matters between-racial/ethnic groups for Latinos and Native Americans. Lighter skinned Latinos and Native Americans are not more likely to be arrested than white-skinned Whites, whereas the odds of arrest for medium-/dark-brown/black-skinned Latinos are 63% higher, and the odds of arrest for non-white-skinned Native Americans are over 4 times higher, than for white-skinned Whites. White-skinned Asians are no longer less likely to experience an arrest than white-skinned Whites. Testing for coefficient differences (Wooldridge, 2000) suggests the effect of color persists within Latinos and Native Americans. Figure 1 displays predicted probabilities of arrest based on Model 5. This illustrates that all Blacks (regardless of color) and darker skinned Latinos and Native Americans have a significantly higher probability of experiencing an arrest than white-skinned Whites, and that darker skinned Latinos and Native Americans are more likely to experience an arrest than their lighter skinned counterparts. These analyses, therefore, illustrate the importance of considering the association between race/ethnicity and arrest in a multidimensional fashion.

Figure 1.

Predicted probability of experiencing an arrest in adulthood by self-identified race/ethnicity and perceived skin color based on Table 3, Model 5 (*p < .05 indicates significantly different from White respondents with White skin [i.e., between race/ethnicity significance] and ^wp < .05 indicates significantly different from the lightest skinned counterparts of that racial/ethnic group [i.e., within-race/ethnicity significance]).

Thus far, analyses have been referencing individuals with white skin (Table 2) or white-skinned Whites (Table 3). To provide a more intuitive comparison group within each racial/ethnic category, we now stratify analyses by race/ethnicity, thereby referencing the lightest skinned counterparts of each respective group (Table 4). This enables us to investigate the possibility of colorism (i.e., to examine the effect of color within groups). Model 1 reveals that color is not influential within self-identified-skinned Blacks are 40% higher than their lighter counterparts and the odds of arrest for the darkest skinned Latinos are 61% higher than white-skinned Latinos. In addition, the odds of arrest for non-white-skinned Native Americans are 6 times higher than for those with white skin. This suggests that colorism exists, particularly among Blacks, Latinos, and Native Americans. Moreover, these analyses illustrate that the influence of having darker skin is not the same for all groups, thereby suggesting that examining the influence of color while controlling for race masks how the effect of color varies by race/ethnicity.

Table 4.

Logistic Regression of Adult Arrest on Perceived Skin Color Within-Racial/Ethnic Categories.

Predictor	Model 1	Model 2	Model 3	Model 4	Model 5
Self-identified race/ethnicity and perceived skin color
Whitea
White skinned (reference)
Non-white skinned	1.19	1.19	1.13	1.22	1.16
Blackb
White/light brown/medium brown skinned (reference)
Dark brown skinned	1.08	0.96	0.96	1.03	1.02
Black skinned	1.40*	1.16	1.12	1.19	1.16
Latinoc
White skinned (reference)
Light brown skinned	1.23	1.29	1.27	1.33	1.31
Medium brown/dark brown/black skinned	1.61*	1.40	1.38	1.57†	1.56†
Native Americand
White skinned (reference)
Non-white skinned	6.00**	6.07*	5.42*	6.68***	6.24***
Asiane
White skinned (reference)
Non-white skinned	1.49	1.35	1.30	1.19	1.10

Note. Odds ratios are presented. These analyses examine the influence of color in separate models for each self-identified race/ethnicity category. Like Table 3, all models control for interviewers’ race; Model 2 adds demographic characteristics; Model 3 examines social structural factors as mediators; Model 4 examines antisocial behavior as mediators; and Model 5 includes social structural factors and antisocial behavior.

^aAnalyses only include self-identified white respondents: n = 6,799.

^bAnalyses only include self-identified black respondents: n = 2,550.

^cAnalyses only include self-identified Latino respondents: n = 1,917.

^dAnalyses only include self-identified Native American respondents: n = 105.

^eAnalyses only include self-identified Asian respondents: n = 789.

^†p < .10,

^*p < .05,

^**p < .01,

^***p < .001 indicate within-race/ethnicity significance.

Model 2, which adds demographic controls, and Model 3, which examines social structural factors, reveal the effect of color only persists within Native Americans. Behavioral indicators are examined as mediators in Model 4, and results further illustrate that lighter skin is not advantageous within Whites, Blacks, or Asians. However, that the odds of arrest for the darkest skinned Latinos are marginally higher than for white-skinned Latinos, and the odds of arrest for non-white-skinned Native Americans are over 6 times higher than for those with white skin. Similar to the analyses examining the effect of color between groups (Table 3), accounting for antisocial behaviors increases the association between color and arrest, particularly for Latinos and Native Americans. Model 5, which includes both social structural factors and behavior, finds evidence that darker skinned Latinos and Native Americans are more likely to be arrested than their lighter counterparts, thereby suggesting that colorism exists in the CJS for these groups. Given the small sample of Native Americans, we are cautious in interpreting the large odds ratios but have confidence in the overall pattern of results. Although the relationship between color and arrest is not identical for all groups, our results exhibit the importance of modeling race/ethnicity in a multidimensional fashion and provide evidence that color is influential between and within groups, particularly for Latinos and Native Americans.

Discussion

This study examines the relationship between race/ethnicity and arrest in a multidimensional fashion with measures of self-identified race/ethnicity and perceived skin color. Examining the main effect of race/ethnicity, we find that Blacks and Native Americans have higher odds, and Asians have lower odds, of experiencing an arrest than Whites. Similar to Kizer’s (2017) finding with Add Health data, we also find that Latinos overall are no more likely to be arrested than Whites. However, analyses also reveal that darker skinned Latinos are more likely to be arrested than white-skinned Whites and Latinos. Hence, only examining the effects of race/ethnicity on arrest masks the underlying differences in arrest among Latinos by skin color.

Regarding the main effect of color, our results parallel prior studies that find darker skinned individuals are more likely to be arrested than lighter skinned individuals (Kizer, 2017), and this remains true even after accounting for self-identified race/ethnicity. Given that color is a primary characteristic used to assess race/ethnicity (Brown, Dane, & Durham, 1998), this is understandable and suggests that examining the relationship between race/ethnicity and arrest should not rely solely on self-identified race/ethnicity and, instead, should consider other dimensions or features associated with race/ethnicity.

When examining the intersecting effects of self-identified race/ethnicity and color on the likelihood of arrest, we find evidence that lighter skin tones are advantageous and darker skin is detrimental between races, particularly when comparing Latinos, Native Americans, and Asians to white-skinned Whites. Conversely, regardless of color, all self-identified Blacks have higher odds of arrest than white-skinned Whites, suggesting that lighter Blacks are not spared the deleterious effects of the one-drop rule (Burch, 2015).

With regard to the effect of color within-racial/ethnic groups, the current research provides evidence that self-identified Latinos and Native Americans with darker skin have higher odds of experiencing an arrest than their lighter skinned counterparts, suggesting that colorism exists in arrest practices. Yet, contrary to some research (e.g., Barlow & Barlow, 2002; Branigan et al., 2017), darker skin is not associated with a higher likelihood of arrest within Whites, Blacks, or Asians, thereby suggesting that not all racial/ethnic groups are equally affected by colorism.

It is likely that Asians, regardless of color, are not more likely to be arrested since Asians are often perceived as the “model minority” and are often regarded as “whiter than white” (Johnson & Betsinger, 2009). Moreover, the limited diversity of our Asian sample likely does not fully capture Asian groups in which colorism may be more common. Other aspects of race/ethnicity, such as facial features or hair texture (King & Johnson, 2016; Petersen, 2017), may also explain why skin color was not influential within Whites, Blacks, or Asians. Given that darker skinned Blacks are more disadvantaged than their lighter skinned counterparts (e.g., Kreisman & Rangels, 2015; Ryabov, 2013), it is surprising that lighter skin is not advantageous among Blacks. It is possible, however, that the skin color category options (e.g., “white” and “black” skin) influence interviewers’ perceptions of color, particularly among respondents who self-identified racially as White or Black. This may have resulted in self-identified Whites and Blacks being perceived as having white and black skin tones, respectively, despite variation in skin color within these groups, which may explain the nonsignificant findings. Yet other studies on arrest also observe the effects of color are not significant within black samples (Branigan et al., 2017; White, 2015). This pattern may reflect an “out group homogeneity effect,” in which out-group members are perceived as all looking alike, such as in the case of White officers stopping and arresting Black individuals (Branigan et al., 2017, p. 2). Supporting Wirth and Goldhamer’s (1944, p. 340) statement that, “in a sense every Negro, whether light or dark, is a marginal man in American society,” we posit that our results suggest that the one-drop rule (Burch, 2015), and, ultimately, racial discrimination against Blacks, persists today.³

Consistent with prior research that finds darker skin is associated with incarceration presumably through limited availability of legitimate opportunities (Gyimah-Brempong & Price, 2006), our exploratory examination of social structural factors suggests that the association between race/ethnicity/color and arrest is, in part, due to darker skinned people being economically and socially marginalized. That is, the disadvantaged conditions that darker skinned Blacks, Latinos, and Native Americans face act as external societal constraints that funnel them into the CJS. The larger implications of these results are that reducing racial/ethnic inequality would assist in dismantling the deleterious role that the CJS plays in perpetuating structural disadvantage. We contend that socioeconomic disparities, resulting from the cumulative effects of disadvantage and discrimination, perpetuate disadvantage among minorities and darker skinned people by increasing the likelihood that they will experience an arrest. Further, given the collateral consequence of involvement in the justice system (Stewart et al., 2017), the CJS acts as an extension of structural disadvantage.

Yet the social structure indicators do not fully account for the relationship between color and arrest for the darkest skinned Blacks or Native Americans. This may be because poverty rates are the highest among Blacks and Native Americans (Macartney, Bishaw, & Fontenot, 2013), and disparities by color exist within these groups (e.g., Allen et al., 2000; Hall, 2014; Hunter, 2007; Kreisman & Rangels, 2015). Further, we posit that the effect of darker skin on arrest is most pronounced for Native Americans given the more extreme conditions of disadvantage that Native Americans face compared to other groups (Tighe, 2014). The effect of color within Native Americans may also persist because those with lighter skin, who are more likely to be bi- or multiracial, are more likely to have moved off reservations and thus do not face as extreme disadvantage as those with darker skin who remain on reservations. Regardless, because the indicators of disadvantage do not fully account for the relationship between color and arrest, this suggests that other unobserved dimensions of structural disadvantage could explain the relationship between color and arrest for these groups.

Moreover, we do not find support that differences in antisocial behavior explain the greater likelihood of arrest among darker skinned minorities. In fact, the effects of color are even more detrimental for Blacks and Latinos when accounting for antisocial behavior. Seeing as minorities are no more likely to use or sell drugs than Whites but are more likely to be arrested for drug-related offenses (Tonry, 2011), and our results suggest that Blacks and Latinos are more likely to be arrested than their involvement in substance use might suggest, this provides evidence of police bias. Arguably, it is plausible that stereotyping Blacks, Latinos, Native Americans, and darker skinned individuals as criminal (Eberhardt et al., 2004; Levinson & Young, 2010; Holmes & Antell, 2001; Tonry, 2011) prompts differential enforcement. We contend that our results as a whole speak to the existence of racism and colorism among police and society, which is plausibly, in part, a result of racial/ethnic stereotypes.

Limitations

This study is not without limitations. First, interviewers were not given systematic training in identifying color and given that gradations of perceived skin color differ across racial groups (e.g., “dark brown” skin tones for Black and Asian individuals differ), this likely introduces variation into these data. However, prior research has documented considerable reliability of skin color categorization among untrained interviewers (Villarreal, 2010). Regardless, this arguably results in conservative estimates of perceived color, and our significant findings are, therefore, especially compelling (Viglione et al., 2011). Nevertheless, it would be worthwhile to conduct analyses with photographs of subjects, or interviewers should be given training to reliably code color (see King & Johnson, 2016).

Despite efforts to ensure proper temporal ordering, the current study is also limited since the social structural and behavioral indicators may be contemporaneous with the outcome, due to the retrospective nature of the arrest measure. Moreover, conceptually, the included mediators could be consequences of arrest. Given the inability to ensure perfect temporal ordering, we are unable to establish the causal effects of color and interpret our results cautiously. Additionally, because our examination of the social structural and behavioral mediators was exploratory in nature, the current study informally assessed mediation. Informally assessing mediation in logistic regressions (i.e., comparing coefficients across models) is inherently problematic because “changes in coefficients across models can depend also on changes in unobserved heterogeneity” (Mood, 2010, p. 72). Thus, future studies should formally test the mediating mechanisms of social structural factors and antisocial behavior.

Research could also explore how this association varies by factors such as gender (e.g., White, 2015), nativity (e.g., Alcalá & Montoya, 2016), and perceived race/ethnicity. Further, the effect of color among Native Americans is an extremely underdeveloped area of study, and research should strive to better understand how color and other dimensions of race/ethnicity influence Native Americans’ experiences in the CJS. Replicating these analyses with other samples of Native Americans, such as from the National Longitudinal Survey of Youth, would be worthwhile. It is also imperative that scholars consider how the criminalization of minorities and darker skinned individuals affect outcomes in the CJS.

Conclusion

This research illustrates the importance of examining the influence of self-identified race/ethnicity on adult arrest in a multidimensional fashion and suggests that focusing on racial/ethnic disparities alone masks differences in arrest by perceived skin color. Our findings provide evidence that color is influential between- and within-racial groups, especially for Latinos and Native Americans. While the relationship between race/ethnicity/color and arrest does not operate the same for all racial/ethnic categories, it is notable that the effects operate in the expected direction. Moreover, this study illustrates how racial/ethnic and skin color disparities in arrest are, in part, better understood through a lens of structural disadvantage. We also find that Blacks and Latinos have a higher likelihood of experiencing an arrest than their involvement in drug and alcohol use might suggest. Thus, we argue that these patterns are suggestive of the existence of structural racism and colorism as well as differential enforcement in the CJS. Given the growing concerns associated with biases in law enforcement and the CJS, it is hoped that these findings will contribute to finding solutions to address these problems.

Biography

Jessica G. Finkeldey is an assistant professor of Criminal Justice in the Department of Sociocultural and Justice Sciences at the State University of New York at Fredonia. Her recent research examines the collateral consequences of contact with the criminal justice system.

Stephen Demuth is an associate professor of Sociology and a Research Affiliate of the Center for Family and Demographic Research at Bowling Green State University. His research focuses on the influence of race/ethnicity, class, and gender in the criminal justice system.