The Role of Skin Color in Latino Social Networks: Color Homophily in Sending and Receiving Societies

Abstract

How does skin color shape the social networks and integration pathways of phenotypically diverse immigrant groups? Focusing on Dominicans and Puerto Ricans, groups with considerable diversity across the Black-White color line, we explore whether migrants to the United States have greater color homophily in their primary social networks than non-migrants in the sending societies. We analyze egocentric network data, including unique skin color measures for both 114 respondents and 1,702 alters. We test hypotheses derived from ethnic unifier theory and color line racialization theory. The data show evidence of color homophily among Dominicans, but suggest that these patterns may be imported from the sending society rather than fostered by the U.S. context. Further, we find that migrants’ skin color is associated with having ties to White or Black Americans, but with different patterns for each ethnic group. We discuss the implications of these findings for economic mobility and U.S. racial hierarchies.

Latin Americans are the largest immigrant group to the United States and the nation’s largest minority. Unlike many immigrant groups, they straddle the culturally important Black-White color line. Even within nationalities, Latinos can be perceived as White, Black, or a range of colors in between. This raises an important question unaddressed in social science research: How does skin color shape the social networks and integration pathways of phenotypically diverse immigrant groups?

How skin color influences Latinos’ networks is central to ongoing debates about their racial identification, social and political integration in the U.S., and the future of racial stratification in the country. Will Latinos become White as other immigrant groups have, integrating socially and losing power as a distinct political coalition (Lee and Bean 2012; Yancey 2003)? Will they remain a ‘middle race’ and political bloc, speeding us toward a majority-minority society (Jiménez 2010; O’Brien 2008)? Or will Latinos’ skin color determine their place in the social hierarchy (Bonilla-Silva 2004; Frank, Akresh, and Lu 2010)? How skin color influences Latinos’ social ties is a crucial yet unexplored piece in assessing these potential paths.

The social networks that newcomers develop are an important measure of their integration, with their core networks of primary relationships holding particular significance (Gordon 1964). Assimilation theories view primary ties to mainstream society or ethnic communities as indicators of immigrants’ assimilation pathways (Alba and Nee 2003; Gordon 1964; Portes and Rumbaut 2001). Yet due to the lack of available data, little research has considered how migrants’ skin color influences the color composition of their networks – both within and outside their ethnic group. Whether darker Latinos develop more social ties to people with dark skin, and lighter Latinos develop more light-skinned ties – what we call color homophily – can inform whether skin color promotes different integration pathways within Latino immigrant groups. One contribution of this study is its innovative examination of both intra- and inter-group color homophily in migrants’ primary group attachments.

The study also uniquely compares newcomers’ networks to those of similar people in the sending societies who do not migrate. Patterns of color homophily in Latino immigrant networks need not result from the U.S. color line. Many Latin American societies are also stratified along color lines (Dixon and Telles 2017; Telles 2014). Assuming that migrants’ patterns are caused by experiences in the U.S. is a form of “methodological nationalism” that ignores the experiences and concepts migrants bring with them (Wimmer and Glick Schiller 2003). We assess whether those who move to the highly racialized U.S. have greater color homophily than those remaining in their native country.

We use a unique dataset measuring the skin color of social network ties. The data come from a study of Dominicans and Puerto Ricans — two Latino groups with considerable heterogeneity across the Black-White color line (Itzigsohn 2009; Rodríguez 2000). We use ego-centric network data gathered in 2002–03 as part of a qualitative study of Puerto Ricans and Dominicans in the U.S., Puerto Rico, and the Dominican Republic. We analyze data from 114 respondents who together have 1,702 alters in their networks. To correct for potential sampling bias in this non-probability sample, we use inverse propensity score weighting calculated from probability samples of these populations, a technique used in epidemiology with considerable value in social science research (Cole and Stuart 2010; Hernán, Hernández-Díaz, and Robins 2004; Peters, Driscoll, and Saavedra 2015).

We address the following questions in the experiences of these respondents: (1) Is there color homophily in non-migrants’ networks in the sending societies? (2) Does color homophily differ among migrants in the U.S. and non-migrants in the sending societies? And (3) Is migrants’ skin color associated with their likelihood of forming ties to Whites or Blacks in the U.S.¹ To answer these questions, we develop and test hypotheses from two competing theories: ethnic unifier theory and color line racialization theory.

We find evidence of color homophily in both Dominican non-migrants’ networks in the Dominican Republic and Dominican migrants’ co-ethnic networks in the U.S., but not in Puerto Ricans’ networks. Furthermore, we do not find that migrants develop ties across a broader or narrower color spectrum than non-migrants. Rather, the color homophily of migrants’ networks resembles that of non-migrants from sending societies. Color divisions exist in these migrant communities, yet this should not be attributed to racialization in the U.S. However, skin color does contribute to migrants’ ties to Whites and Blacks, albeit differently for Puerto Ricans and Dominicans.

We examine the color composition of migrants’ and non-migrants’ networks to initiate a larger conversation about the role of skin color in immigrants’ primary group affiliations and integration pathways, and to prompt larger, probability-based data collection efforts to contribute to these debates. We conclude by discussing the implications of our findings for contemporary debates on where Latinos fall in American racial hierarchies, and emphasize the importance of considering the racial phenotype of both immigrants and their networks in advancing these debates.

Immigrant Assimilation and Primary Group Attachments

According to Gordon (1964), structural assimilation occurs when individuals form large-scale primary group relationships with the dominant social group and enter into its social networks and institutions. Gordon defines a primary group as one where “contact is personal, informal, intimate, and usually face-to-face, and which involves the entire personality, not just a segmentalized part of it” (1964:31). He argues that most primary relationships occur within an immigrant’s ethnic subsociety, while more impersonal secondary relationships typically occur in larger society. When newcomers shift their primary networks to the dominant group, other stages of assimilation naturally follow.

Gordon’s model addressed European immigrant groups with relatively little diversity in racial appearance. After he wrote, the 1965 Hart-Cellar Act made the face of immigration increasingly non-White. Segmented assimilation theory emerged to account for this new diversity, predicting that skin color is one factor creating vulnerabilities to downward assimilation and integration into a minority group (Portes and Zhou 1993). Scholars have asked, for example, whether second-generation West Indians become seen as Black Americans or whether Mexican-Americans, treated as a homogeneous group, join a reactive subculture that creates barriers to upward mobility (Portes and Zhou 1993; Waters 1999).

However, segmented assimilation theory primarily discusses the role of race on group-level outcomes. It predicts that racial proximity allows some immigrant groups to develop ties to White mainstream society, while prejudice in societal reception leads other groups to form more ties with native-born minorities (Portes and Zhou 1993). It does not address the role of individual-level skin color on primary group attachments, or whether individuals of different skin colors within an ethnic group experience color homophily that distinguishes their networks from those of co-ethnic peers. While recent revisions of assimilation theory highlight historical and institutional factors influencing group-level assimilation (Alba and Nee 2003), they also emphasize that intra-group heterogeneity is frequently overlooked (Alba, Jiménez, and Marrow 2014). Social network composition, particularly the ability to join networks outside one’s group, remains crucial to assessing immigrant incorporation; the role of immigrants’ skin color in shaping those networks is a vital, yet overlooked, aspect of intra-group heterogeneity.

Social Networks and Color Homophily

Although social networks tend to be homophilous – people typically resemble their ties on many characteristics (McPherson, Smith-Lovin, and Cook 2001) – racial and ethnic homophily research has not explored skin color or phenotype.² Marsden (1987), for instance, finds network homogeneity within the categories “White,” “Black,” “Hispanic,” and “Other,” but does not measure phenotypes of respondents or alters. Using U.S. Census classifications, Latinos typically classify their race as White, Black or Other. Yet these classifications are poor proxies for skin color, as Latinos answer these questions in ways that may not reflect their appearance, how they are seen racially by others, or how they see themselves (Dowling 2014; Rodríguez 2000; Roth 2010). Measures of skin color or phenotype are therefore strongly preferable to census race classifications in studying Latinos’ integration pathways. Because skin color, and phenotype more broadly, are often unmeasured axes of social stratification (Roth 2016), we have little evidence of color homophily in Latino networks – i.e., whether darker and lighter Latinos primarily have ties to people with darker or lighter skin colors, respectively.

Compared to the U.S., some Latin American countries – including Puerto Rico and the Dominican Republic – have greater social integration across races (Howard 2001; Loveman and Muniz 2007). This reflects greater “horizontal integration” across race at the level of interpersonal relationships (Telles 2004). Yet many Latin American countries still experience vertical stratification by color (Dixon and Telles 2017; Telles 2014). Both Puerto Rico and the Dominican Republic experience such color stratification, with lighter, more European-looking people at the top of the social hierarchy and darker, more African-looking people at the bottom (Howard 2001; Kinsbruner 1996; Telles 2014). Because of the correlation between color and status and how this structures opportunities to form ties (Blau 1994; Feld 1981), we hypothesize finding color homophily in non-migrants’ networks in these sending societies:

H1: In the sending societies, darker non-migrants have darker social ties than lighter non-migrants.

Explaining Migrant Network Outcomes in the U.S.

Sending-receiving society comparisons are highly valuable for understanding incorporation processes (Levitt 2001; Roth 2012, Smith 2006) . By focusing on whether immigrant groups become similar to the dominant group in the host society, assimilation theory overlooks how immigrants may nonetheless change, even if not coming to resemble the dominant group. Jiménez and Fitzgerald (2007) argue that studying immigrant dissimilation – how they come to differ from those in their sending society – offers another important measure of immigrant change. The challenge of gathering data in multiple national locations, however, often results in omission of sending societies’ data, leading to “methodological nationalism” (Wimmer and Glick Schiller 2003). Here, sending-receiving society comparisons allow us to consider the theoretical mechanisms for why color homophily might exist in immigrants’ social networks.

How might the color composition of Latino migrants’ social networks compare to those of their counterparts back home? Drawing on the literature, we develop two theories with divergent explanations for why the color composition of Latino migrants’ networks might differ from those of non-migrants.

We call the first ethnic unifier theory. For non-migrants in the countries of origin, ethnicity is taken for granted since almost everyone shares it; other characteristics like color or class are more relevant in stratifying individuals’ social experiences, influencing interactions at work, school, or church. However, in an immigrant receiving society, this theory holds, such divisions are overshadowed by immigrants’ common ethnicity and culture. Immigrant groups develop ethnic awareness as they recognize the social distance and negative stereotypes separating them from the dominant group and cleavages within their ethnic group accordingly become less important (Portes 1984). They develop “bounded solidarity” that foregrounds ethnicity and promotes trust and interdependence within the ethnic community (Portes and Sensenbrenner 1993). Confronted with an unfamiliar racialization process, first-generation immigrants often emphasize national origin over race, producing some ethnic closure (Waters 1999). Dividing lines back home, like class or color, become less salient and people form connections with a broader range of co-ethnics. For phenotypically diverse groups, ethnic unifier theory maintains that ethnicity trumps color in first-generation networks, uniting co-ethnics across color lines.

Ethnic unifier theory relates specifically to co-ethnic ties. It predicts:

H2: Associations between respondent color and co-ethnic alter color are weaker for migrants than non-migrants.

H2 predicts that color homophily among co-ethnics is lower in the U.S. as migrants develop ties to co-ethnics regardless of color.

A second pathway for how networks might differ between the U.S. and sending societies is what we call color line racialization theory. This maintains that the U.S. is a highly racialized society where the sharpest social barriers exist along the Black-White color line (Omi and Winant 1994), which in turn shapes migrants’ interaction patterns. As migrants enter the U.S., they tend to develop ties and networks with White or Black communities respectively based on skin color. During the Puerto Rican mass migration after World War II, migrants entered an environment where Blacks and Whites were highly segregated. Because of prejudice and segregation, darker-skinned migrants interacted more with Blacks and lighter-skinned migrants were eventually able to move to the suburbs and socialize more with Whites (Thomas 1967). Scholars predicted that those who looked White would assimilate into White America, while those with dark skin would assimilate into the Black community (see Rodríguez 1990).

Why might these patterns remain today? After all, many Puerto Ricans and Dominicans do not self-identify as Black or White, and many identify racially as Latino or with their nationality (Itzigsohn 2009; Rodríguez 2000; Roth 2012), suggesting a process of ethnogenesis consistent with ethnic unifier theory. Dark-skinned migrants may also distance themselves from Black Americans to avoid being seen as Black and experiencing similar discrimination (Howard 2003; Waters 1999). But research shows that skin color continues to shape Latinos’ lives in the U.S. (Dixon and Telles 2017; Montalvo and Codina 2001; South, Crowder, and Chavez 2005). Color discrimination in housing and labor markets indirectly affects social ties by influencing options for network development (South et al. 2005). These patterns persist, South et al. (2005) find, with darker skin inhibiting Puerto Ricans from moving into neighborhoods with more White Americans. Due to prejudice and discrimination in their opportunities, their color may still shape their social interactions as well.

Color line racialization theory maintains that the color line influences where migrants live, work, and who their friends are. It predicts that Latino migrants develop networks that resemble their own color, both outside and within their co-ethnic community. Residential and occupational segregation by color implies that if light-skinned Latinos are more likely to live and work among Whites, they also are more likely to live and work among light-skinned Latinos who are also in those jobs and neighborhoods (Espino and Franz 2002; Montalvo and Codina 2001) This theory predicts:

H3: The association between respondent color and alter color is stronger for migrants in the U.S. than for non-migrants in the sending societies.

H3 predicts that compared to those back home, migrants to the U.S. develop ties that resemble their color more because of the U.S. color line.

Color line racialization theory holds that the Black-White color line actively shapes all interactions in the U.S., including to Whites and Blacks. This could influence the integration pathways of different individuals within the ethnic group. Since there are few White or Black Americans in the sending societies, we examine this outcome only in the receiving society context. The theory predicts:

H4: In the receiving society, lighter migrants are more likely than darker migrants to have ties to White Americans.

H5: In the receiving society, darker migrants are more likely than lighter migrants to have ties to Black Americans.

Because ethnic unifier theory is tested by comparing co-ethnic ties in both sending and receiving societies, it does not have predictions about ties to Black or White Americans.

Data and Methods

We use ego-centric network data gathered during 114 in-depth interviews with Puerto Rican and Dominican migrants and their non-migrant counterparts. The migrant sample comprises people in the New York metropolitan area who were born in Puerto Rico or the Dominican Republic, came to the U.S. at age 14 or older, and lived in the U.S. for at least 7 years. The non-migrant sample includes those in the San Juan or Santo Domingo metropolitan areas who had not lived outside of their home country for more than 6 months and identified both parents as Puerto Rican or Dominican, respectively. To avoid intertwined networks, we excluded people who knew other respondents. Within each group, sampling quotas provided variation by age, sex, occupational status, and skin color.³ For migrant samples, sampling quotas provided variation in respondents’ age at arrival and amount of time in the U.S. Quotas were filled through a combination of methods including referrals from respondents in a related survey, posted flyers, forwarded emails, professional organizations, and knocking on doors. Migrants were interviewed in 2002–03, and non-migrants in 2003. See Roth (2012) for a full description of data and recruitment methods.

Measuring Respondents’ Networks

Data on respondents’ networks were collected using a name generator (Marsden 2011), which enumerates alters by asking respondents to list people with whom they share a criterion relation. Respondents were asked to list up to 5 individuals who helped them find jobs; their best friend and up to 3 good friends; a current romantic partner and up to 3 previous partners;⁴ up to 3 friends from work or school; and up to 3 friends from organizations, such as clubs, church, or voluntary groups. Name generators are well-suited for testing our hypotheses because they overwhelmingly represent network members who are chosen, rather than family relationships. Respondents could not list the same person twice.

These name generators were followed by name interpreters (Marsden 2011), a series of questions that elicit information about each alter. Questions included alters’ nationality, race, skin color, and relationship with the respondent. Respondents were asked to describe alters’ colors on a scale from 1 (very light) to 10 (very dark), and then to describe their own color using the same scale. Respondents were familiar with this task, as earlier in the interview they used this scale to describe the color of individuals shown in a photographic instrument.⁵ They were told that 1 represented the lightest person they could imagine and 10 represented the darkest. Photograph ratings served to reference the respondents’ ratings of their alters.⁶

Dependent and Key Independent Measures

Our main dependent variable is a three-category measure of alter skin color. Respondents rated their alters on a 10-point scale which we condensed to 3 categories to reduce the measurement error likely to result from variation in respondents’ use of the scale. We coded alters rated between 1 and 3 on the 10-point scale as light (1), between 4–6 as medium (2), and above 6 as dark (3). We treat this variable as ordinal. To test H4 and H5, our dependent variables are dichotomous, coded as: (1) 1 if the alter is non-Latino White (0 otherwise); and (2) 1 if the alter is non-Latino Black (0 otherwise).

Our main independent variable is respondent’s skin color as rated by the respondent. This was coded light, medium, and dark in the same way as alter skin color.⁷ We use self-rated skin color to maximize consistency, as respondents applied the same scale to both themselves and their alters.⁸ Because we treat this as continuous and interact it with other variables, we enter this variable as centered. Other key independent variables include dummy variables for respondents’ ethnicity (1=Puerto Rican) and migration status (1=migrant).

Control Variables

Control variables include respondent’s sex (1=male, 0=female), age (continuous), and education (1=less than high school, 2=high school diploma, 3=BA or more). For models involving only migrants, we omit age and include age at arrival in the U.S. (continuous), years spent in the U.S. (continuous), and self-rated English ability (1=speaks fluently, 2=pretty well, 3=well enough to get by, and 4=not well enough). We omit occupational status because it correlates highly with education.

We tested the models’ robustness by estimating them with different versions of our key independent and dependent variables, including the 10-point measures of skin color, 3-category scale coded with different cut-off points, interviewer-coded skin color for respondents, and with respondent skin color coded as two nominal variables (medium and dark, with light as reference). In all cases, results were substantively the same.

Inverse Propensity Score Weighting

The data analyzed are non-probability data. To allow for statistical inference, we weight the data to match probability data collected from the same populations, which can correct sampling error biases when estimating from non-probability data. Inverse propensity score weights are used to correct sampling bias, most commonly for web-based surveys and in epidemiology where non-probability sampling is common (Cole and Stuart 2010; Hernán et al. 2004; Peters et al. 2015). Sociologists have used inverse propensity score weighting primarily to estimate causal effects (Morgan and Winship 2015), but it can also be used to reduce sample selection bias (Hernán et al. 2004; Peters et al. 2015). We calculate jackknife standard errors for all regression models calculated with weighted data.

We located probability data for each of our four populations: the 2006 Latino National Survey (LNS) for migrant Puerto Ricans and Dominicans, the 2004 Dominican Republic Americas Barometer survey by the Latin America Public Opinion Project (LAPOP) for Dominican non-migrants, and the 2005 Puerto Rico Community Survey for Puerto Rican non-migrants. We recoded key demographic variables to match our data and merged each probability dataset with our corresponding sample.

With each combined dataset we estimated logit models with a dependent variable coded as 1 if a case was from our data and 0 if from the probability data. We weighted probability cases with included survey weights and our cases with the weight 1. When estimating each logit regression, we calculated the estimated probability that each case would come from our dataset. For each case in our data, we took the inverse of this value to create a survey weight. Finally, so the four groups could be combined into the same models, we normalized the survey weights within groups by dividing each weight by the mean weight within that population.

The Online Supplement details the weighting procedure, including logit models. Analyses in Tables 3–5 are weighted using these calculated weights.

Table 3:

Random Effects Ordered Logit Models Predicting Alter Skin Color from Respondent Skin Color, Migration Status, and Ethnicity

	Model 1 All Alters		Model 1 Co-Ethnic Alters
	Coeff.	(S.E.)	Coeff.	(S.E.)
Respondent Color	0.67	(1.10)	1.17	(0.84)
Puerto Rican Non-Migrants	--	--	--	--
Dominican Non-Migrants	0.11	(0.55)	0.17	(0.60)
Puerto Rican Migrants	−0.39	(0.61)	−0.65	(0.76)
Dominican Migrants	−0.29	(0.55)	−0.38	(0.65)
Puerto Rican Non-Migrants * Resp. Color	--	--	--	--
Dominican Non-Migrants * Resp. Color	0.68	(1.24)	0.21	(0.97)
Puerto Rican Migrants * Resp. Color	−0.35	(1.19)	−0.91	(0.97)
Dominican Migrants * Resp. Color	−0.20	(1.17)	−0.24	(0.96)
Male	−0.20	(0.31)	−0.08	(0.38)
Education	0.02	(0.21)	0.26	(0.28)
Age (cont.)	−0.01	(0.01)	−0.00	(0.01)
Cut point 1	−1.12+	(0.63)	−0.62	(0.69)
Cut point 2	1.93***	(0.53)	2.60***	(0.62)
Sigma2	0.50+	(0.28)	0.58	(0.44)
Sigmaµ	0.71		0.76
Log pseudo-likelihood	−1457.72		−1048.61
N (Alters)	1690		1291
N (Respondents)	114		114

⁺p<.10,

^*p<.05,

^**p<.01,

^***p< .001

Table 5:

Odds Ratios for Random Effects Logit Models Predicting Ties to Non-Latino Whites and Blacks

	Model 1 Puerto Rican Migrants Ties to non-Latino Whites		Model 2 Dominican Migrants Ties to non-Latino Whites		Model 3 Puerto Rican Migrants Ties to non-Latino Blacks		Model 4 Dominican Migrants Ties to non-Latino Blacks
	Coeff.	(S.E.)	Coeff.	(S.E.)	Coeff.	(S.E.)	Coeff.	(S.E.)
Respondent Skin	.368**	(.132)	.692	(.247)	.795	(.696)	2.167*	(.689)
Color
Male	.607	(.206)	2.071	(1.334)	.720	(.825)	1.813	(1.541)
Education	1.405	(.292)	1.003	(.471)	.866	(.374)	2.681*	(1.332)
Arrival Age	.858**	(.047)	.748**	(.067)	1.106	(.107)	-- 1
Years In U.S.	1.012	(.017)	1.101*	(.042)	1.049	(.042)	1.062*	(.027)
English Ability	.406**	(.129)	.278	(.369)	.010	(.024)	.580	(.207)
Constant	41.733	(60.25)	4.370	(8.945)	.010+	(.024)	.000***	(.000)
Psuedo-Log	−120.901		−121.646		−56.736		−38.798
Likelihod
N (Respondents)	27		32		27		32
N (Alters)	378		487		378		487

Note:

⁺p<.1,

^*p<.05,

^**p<.01,

^***p<.001

¹Arrival age perfectly predicted ties to non-Latinos Blacks and was thus omitted.

Results

Descriptives

Table 1 shows unweighted descriptive measures for all variables. Respondents were approximately evenly divided between male/female and non-migrant/migrant. Weighted data more closely reflect the characteristics of each population, as shown in the Online Supplement.

Table 1:

Unweighted Descriptives, for all Respondents and by Migrant Status

	All Respondents		Non-Migrants		Migrants
	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.
Model Variables
Male (D)	.47		.48		.45
Education	1.18	.74	1.16	.71	1.19	.78
Age	42.12	15.35	40.88	15.37	43.29	15.37
Years in U.S.	24.69	15.03	--	--	24.69	15.03
Arrival Age	19.49	5.69	--	--	19.49	5.69
English ability	2.03	1.06	--	--	2.03	1.06
3-Cat. Self-Rated Resp. Color	1.87	.67	1.88	.71	1.86	.63
3-Cat. Alter Color	1.65	.67	1.66	.66	1.64	.69
% of Co-ethnic Alters	.77	.42	.95	.22	.58	.49
% of Other Latino Alters	.11	.32	.03	.18	.20	.40
% of White (non-Latino) Alters	.08	.27	.01	.10	.15	.36
% of Black (non-Latino) Alters	.02	.15	.00	.05	.04	.20
Additional Descriptive Variables
Network Size	14.93	2.47	15.21	1.96	14.66	2.88
3-Cat. Interviewer-Rated Resp. Color	1.84	.81	1.91	.84	1.78	.77
10-pt Interviewer-Rated Resp. Color	4.54	2.59	4.79	2.73	4.31	2.44
10-pt Self-Rated Resp. Color	4.61	2.13	4.81	2.23	4.41	2.03
10-pt Alter Color	4.10	2.47	4.20	2.35	3.10	2.58
N (Respondents)	114		56		58

(D) – Dichotomous, coded 0/1

Table 1 shows that non-migrants and migrants are similar in skin-color, as indicated by both self-rated and interviewer-rated color. With both the 3-category and 10-point scales, self-rated and interviewer-rated skin colors have similar means but greater variability in interviewer-rated color (bottom). Non-migrants have slightly larger networks than migrants, although migrants are more likely to have Whites, Blacks, and Other Latinos (i.e. non-co-ethnics) in their networks because they develop these ties in the U.S. Migrants show a weakening of ethnic homophily in their networks from exposure to greater ethnic diversity in the U.S., but we examine whether that translates into weaker color homophily. Non-migrants’ and migrants’ alters are also similar in skin color.

Table 2 shows these descriptives by ethnicity and migration status. Puerto Rican non-migrants have a lower proportion of men and slightly higher education on average than their migrant counterparts, while Dominican non-migrants have a higher proportion of men and lower education than migrants on average. Both non-migrant groups have mean ages of 41 years, but migrant Puerto Ricans are older than migrant Dominicans (50 vs. 37 years old). Among migrants, Puerto Ricans had been in the U.S. longer, reflecting their earlier mass migration (Roth 2012), but lower English ability than Dominican migrants, likely reflecting educational differences.

Table 2:

Unweighted Descriptives by Respondent Group

	Puerto Rican Non-Migrants		Puerto Rican Migrants		Dominican Non-Migrants		Dominican Migrants
	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.
Model Variables
Male (D)	.46		.52		.50		.39
Education	1.39	.63	1.08	.92	.93	.71	1.29	.64
Age	40.79	15.30	50.44	16.08	40.96	15.73	37.07	11.78
Years in US	--	--	31.33	17.28	--	--	18.90	9.80
Arrival Age	--	--	19.37	4.77	--	--	19.58	6.34
English ability	--	--	1.93	.92	--	--	2.13	1.18
3-Cat. Self-Rated Resp. Color	1.93	.72	1.78	.64	1.82	.72	1.94	.63
3-Cat. Alter Color	1.55	.64	1.46	.67	1.77	.67	1.77	.68
% of Co-ethnic Alters	.95	.25	.54	.50	.96	.18	.61	.49
% of Other Latino Alters	.04	.20	.20	.40	.02	.15	.20	.40
% of White (non-Latino) Alters	.02	.12	.21	.41	.01	.08	.10	.30
% of Black (non-Latino) Alters	.00	.05	.04	.20	.00	.05	.04	.20
Additional Descriptive Variables
Network Size	15.29	1.78	14.00	3.15	15.14	2.16	15.23	2.53
3-Cat. Interviewer-Rated Resp. Color	1.79	.79	1.63	.74	1.96	.87	2.04	.88
10-pt Interviewer-Rated Resp. Color	4.38	2.56	3.67	2.42	5.20	2.89	4.87	2.35
10-pt Self-Rated Resp. Color	4.86	2.35	3.82	2.16	4.77	2.14	4.92	1.78
10-pt Alter Color	3.74	2.24	3.31	2.52	4.68	2.38	4.56	2.50
N (Respondents)	28		27		28		31

(D) Dichotomous, coded 0/1

Dominican migrants are described by both themselves and the interviewer as having darker skin than Puerto Rican migrants, on average. Furthermore, the interviewer rated Dominican non-migrants as darker on average than Puerto Rican non-migrants. These observations are consistent with the perception that Dominicans have a greater African contribution to the nation’s mixed ancestry because of the larger African slave population historically brought to the Dominican Republic (Moya Pons 1998).

Table 2 also shows that Puerto Rican migrants rate themselves, and are rated by the interviewer, as lighter in color than Puerto Rican non-migrants, on average. Moreover, while the self-ratings generally resemble the interviewer-ratings of respondent skin colors for all groups, the interviewer ratings are lighter than self-ratings for both Puerto Rican groups and darker than self-ratings for both Dominican groups.

Dominicans also report having alters with darker skin on average than Puerto Ricans, among both migrants and non-migrants. For all groups, the majority of alters are co-ethnics, (Puerto Ricans or Dominicans, respectively). For Puerto Rican migrants, the average network is composed of 54% co-ethnics, 20% Other Latinos, 21% Whites, and 4% Blacks. The average network among Dominican migrants is 62% co-ethnics, 20% Other Latinos, 10% Whites, and 4% Blacks.

Respondent and Alter Skin Color in Non-Migrants’ and Migrants’ Networks

Table 3 predicts alters’ skin color from the respondent’s skin color and migration status, using random effects ordered logit regressions. We use alters as the units of analysis. Our dependent variable is a three-category measure of respondent-rated alter skin color (light, medium, and dark). Because this is not an interval measure, we use ordered logit regressions to correctly fit a model. These models estimate coefficients to linearly predict a latent continuous variable S and provide cutpoints in S, δ₁ and δ_2, for translating this continuous variable to one of the three values of the ordinal dependent variable. The model assumes logit-distributed errors and estimates coefficients and cutpoints to maximize the probability of detecting the observed frequency distribution in the dependent variable. Because alters named by the same respondent are expected to have correlated errors, a random effects model takes this clustering into account. This multi-level model partitions the errors into within-respondent and between-respondent portions and calculates standard errors that account for clustering by respondent. Random effects models are commonly used by researchers studying ego networks and are recommended for models where alters are the units of analysis (e.g., Perry, Pescosolido, and Borgatti 2018).

Coefficients estimated in ordered logit regression models are not easily interpretable, since they estimate a linear effect on a latent variable rather than the actual outcome of interest. Further, because cutpoints vary between models, coefficient sizes cannot be compared between models. To aid interpretation, in Figures 1 and 2 we present predicted percentages of light-, medium- and dark-skinned alters based on each model in Table 3. Significant patterns in Figures 1 and 2 are shaded.⁹

Figure 1:

Predicted Percentages of Light, Medium, and Dark Skinned Alters Estimated from Table 3, Model 1

Notes: Shaded panels are those shown by post-estimation tests to have significantly different predicted values for respondents of different skin colors. Predicted percentages are based on the assumption that sigma=0 and are calculated based on female respondents with HS education, aged 41.75, and alters with HS education.

Figure 2:

Predicted Percentages of Light, Medium, and Dark Skinned Co-Ethnic Alters Estimated from Table 3, Model 2

Notes: Shaded panels are those shown by post-estimation tests to have significantly different predicted values for respondents of different skin colors. Predicted percentages are based on the assumption that sigma=0 and are calculated based on female respondents with HS education, aged 41.75, and alters with HS education.

Our models include both indicator variables and interaction effects, which sometimes means that the comparison required to test a hypothesis is not a comparison to a baseline reference category. When this occurs, we report post-estimation tests of the linear combinations required to test the relevant hypotheses in the text, testing against the null hypothesis that those combinations equal 0.

Table 3 examines the relationship between respondent color and alter color, with respondent color as a continuous variable. Puerto Rican non-migrants are the baseline category, but results do not change with another group omitted. Model 1, which analyzes all alters, provides partial support for H1: we found skin color homophily in the sending society for Dominicans but not Puerto Ricans. Model 1 compares color homophily between the groups, including indicator variables for Dominican and Puerto Rican migrants and non-migrants and interactions between these and respondent skin color.¹⁰ The coefficient for the respondent skin-color variable is positive, reflecting the Puerto Rican non-migrant baseline category. This positive coefficient, predicted by H1, is also reflected in the predicted network compositions for Puerto Rican non-migrants shown in Figure 1 (i.e., lighter-skinned Puerto Rican non-migrants have a greater proportion of light-skinned and smaller proportion of dark-skinned ties than medium-skinned and dark-skinned Puerto Rican non-migrants). However, the coefficient for respondent skin color reflecting this baseline group is non-significant (Table 3), indicating that we cannot reject the null hypothesis of no relationship among Puerto Rican non-migrants. The standard error for the association between ego and alter skin color for Puerto Rican non-migrants is large, however, so we also cannot conclude no relationship between respondent and alter skin color among Puerto Rican non-migrants.

Post-estimation significance tests show that the association between respondent and alter skin color is significant for Dominican non-migrants (.67+.68=1.35, p=.001). This relationship between respondent and alter skin color is reflected in the predicted network compositions in Figure 1 where light-skinned Dominican non-migrants are predicted to have 59% light-skinned and 3% dark-skinned ties compared to 9% and 33% respectively for dark-skinned Dominican non-migrants.

Model 1 does not support H3. Among migrants, no significant association appears in Model 1 between respondent and alter skin color for Puerto Ricans (.67-.35=.32, p=.443) or Dominicans (.67-.20=.47, p=.179). As expected for much smaller coefficients, predicted network compositions in Figure 1 show less dramatic differences in network composition by skin color for migrants than the non-migrant predictions showed. Among Puerto Ricans, light-skinned migrants have 16 percent more light-skinned ties than do dark-skinned migrants. Among Dominicans, the difference is 22 percentage points, compared to 28 and 50 percentage points for non-migrant Puerto Ricans and Dominicans, respectively. For both Puerto Ricans and Dominicans, effect sizes are greater for non-migrants in the sending society (0.67 and 1.35, respectively) than migrants in the U.S. (.32 and .47, respectively), which contradicts the hypothesized direction in H3. Post-estimation tests show that this difference is not statistically significant for either group (p=.528 for Puerto Ricans, p=.356 for Dominicans). Thus we do not find support for H3.

Comparing Migrants’ and Non-Migrants’ Co-ethnic Networks

Ethnic unifier theory applies to co-ethnic ties. To test H2, we examine color homophily among only the subset of co-ethnic alters (Table 3, Model 2). Because non-migrants’ ties are mostly co-ethnic (95% on average), the findings for non-migrants resemble those in Model 1.

The baseline group is Puerto Rican non-migrants; the non-significant coefficient for respondent color thus shows no significant association between respondent color and co-ethnic alter color for Puerto Rican non-migrants. Again, the large standard error means we cannot conclude no relationship exists. Post-estimation significance tests for Dominican non-migrants show a significant association between respondent and co-ethnic alter color (1.17+.21=1.38, p=.003). This is reflected in predicted network compositions in Figure 2 showing that light-skinned Dominican non-migrants are predicted to have 56% light-skinned co-ethnic ties and 3% dark-skinned, compared to 7% and 33% respectively for dark-skinned Dominican non-migrants.

We do not find evidence of color homophily in co-ethnic ties for Puerto Rican migrants; the association between these respondents and co-ethnic alter color is not significant (1.17-.91=.26, p=.599). However, for Dominican migrants, this association is significant with a one-tailed test (1.17-.24=.93, p=.053). Darker-skinned Dominican migrants are predicted to have darker-skinned co-ethnic alters. This is reflected in the predicted percentages in Figure 2; light-skinned Dominican migrants are predicted to have 60% of their co-ethnic alters be light-skinned and 3% dark-skinned, compared to 19% light-skinned and 15% dark-skinned for dark-skinned Dominican migrants.

Our findings point in the direction of H2, or ethnic unifier theory. The coefficient values indicate smaller associations for Puerto Rican and Dominican migrants (.26 and .93, respectively) than non-migrants (1.17 and 1.38, respectively). However, these differences are not significant by post-estimation (p=.396 for Puerto Ricans and p=.271 for Dominicans), so we cannot reject the null hypothesis and do not find support for H2 among Puerto Ricans or Dominicans.

Migrants’ Ties to Whites and Blacks

Table 4 presents descriptive data for H4 & H5, showing the mean ethnic and racial composition of migrants’ networks by respondents’ ethnicity and color. This table is only suggestive, as differences between otherwise equivalent respondents of different skin colors are not statistically significant. Among Puerto Rican migrants, those with medium skin have the greatest proportion of White ties, followed by those with light skin, while dark-skinned Puerto Ricans have a much smaller proportion. Dark-skinned Puerto Ricans have a greater proportion of Black ties, compared with their light- and medium-skinned peers. Dark-skinned Dominican migrants have both more Black and more White ties than their lighter peers. These descriptive patterns suggest very tentative initial support for H5, but not H4.

Table 4:

Mean Ethnic and Racial Composition of Migrants’ Networks, by Nationality and Color

	All Migrants	Puerto Rican Migrantss			Dominican Migrantss
		Light	Medium	Dark	Light	Medium	Dark
Co-ethnic	.62	.56	.54	.75	.65	.66	.39
Other Latino	.22	.28	.21	.12	.23	.24	.26
White (non-Latino)	.10	.13	.20	.05	.10	.04	.31
Black (non-Latino)	.03	.03	.03	.09	.01	.02	.04

Note: Columns do not sum to 100% because alters in other groups (e.g. Asians) are not shown. N=58

Table 5 shows odds ratios from random effects logit regressions that predict migrants’ alters being White or Black based on respondent color. Model 1 shows that alters of medium-skinned Puerto Ricans are 63% less likely to be White than those with light skin and alters of dark-skinned Puerto Ricans 63% less likely than that. This model also shows that those who migrated at older ages and have less proficient English are less likely to have White ties. Model 2 in Table 5 shows that Dominicans’ skin color is not associated with their probability of White ties, although alters of Dominicans who arrived at an older age are less likely to be White, and those of Dominicans who have been in the U.S. longer are more likely to be White. Models 3 and 4 estimate the likelihood that respondents’ ties are Black. For Puerto Ricans, Model 3 shows no evidence that migrants’ color is associated with the likelihood of forming Black ties. Among Dominicans, alters of medium-skinned migrants are more than twice as likely as those of light-skinned migrants to be Black, and dark-skinned migrants are more than twice as likely to be Black as those of medium-skinned migrants. Alters of Dominican migrants with more education are also much more likely to be Black and those who have been in the U.S longer are slightly more likely to be Black.

Discussion and Conclusion

We do not find support for ethnic unifier theory. Although studies show that many U.S. Latinos develop a sense of bounded solidarity to co-ethnics (Itzigsohn 2004; Rodríguez 2000), our data do not show that attachments to ethnicity have weakened color lines within these communities. For Puerto Ricans, this stems partly from lack of evidence for color homophily in the sending society. For Dominicans, we cannot conclude that shared ethnicity has taken on such importance for migrants that color divisions found in the sending society fade away in the U.S.

We also do not find that migrants in the host society experience greater overall color homophily than non-migrants in the sending society, as predicted by color line racialization theory. We find no evidence that the association between migrants’ color and their alters’ color is stronger in the U.S. as a result of racialization.

We observe some evidence of color homophily, as Dominican non-migrants experience color homophily in the sending society. Furthermore, Dominican migrants also show color homophily in their co-ethnic networks in the U.S. Our data do not show that the association between Dominican migrants’ color and their alters’ color is weaker or stronger than for non-migrants, however. Instead, Dominican migrants seem to exhibit similar interaction patterns with co-ethnics to what their counterparts experienced in the Dominican Republic with regard to color, illustrating the importance of sending society comparisons in migration studies.

We find some support for color line racialization theory in the association between migrants’ skin color and their ties to Whites and Blacks. Lighter skin increases Puerto Ricans’ odds of having White ties, while darker skin increases Dominicans’ odds of having Black ties, net of factors like education, English ability, and time in the U.S. that we would expect to be associated with making non-Latino ties. The patterns differ for each ethnicity, emphasizing the need to consider different subgroups. These findings suggest that the U.S. color line does shape migrants’ ties to non-Latino Americans, even if it does not among their co-ethnics or networks overall.

Our study has limitations. Although inverse propensity score weighting can correct for sample selection bias in a non-probability sample, we cannot formally test the effectiveness of these weights in correcting sampling bias. We are also limited by small sample size, and by the variables available in the closest matching probability datasets; for instance, while we can weight by skin color in the migrant samples, we are not able to with the non-migrants samples, where representative skin color data is lacking. Because our study is the first to consider color homophily in ego-centric networks, this paper contributes a model for how color homophily can and should be investigated. We hope this prompts a larger conversation about how skin color influences immigrants’ primary group affiliations and structures social connections both within and outside their community, leading to larger, probability-based data collection on network color composition.

There may also be selection effects in our migrant samples which are not adequately controlled. Those who migrate are likely to have important differences in resources, motivation, or other characteristics (Feliciano 2005). While we do not believe that individuals’ color, or the color composition of their networks, is a major determinant in migration decisions, it may influence opportunities in the home country which could indirectly affect migration decisions. The ideal future research design would be longitudinal, identifying people who are about to migrate and examining the color composition of their ties before and after migration.

Another limitation is our focus on skin color rather than the broader concept of phenotype. This was necessary because our research design did not allow observation of respondents’ alters. Some respondents were initially uncomfortable with the idea of rating their alters’ color, until the purpose of these data was explained (Roth 2012L215–16). To ask for detailed descriptions of their alters’ racialized features would likely provoke greater discomfort or refusals. Creative solutions are needed here, such as using respondent photo albums or social media accounts, asking respondents to photograph their alters, or create images of them with computer software.

The study also draws on data from particular urban regions. In the U.S., we focused on New York because both Puerto Rican and Dominican migrants have historically concentrated there, and even as the communities began to disperse, it remained the largest concentration of each group in the U.S. (Roth 2012). But it is also distinctive in ways that might lead us to expect greater color line racialization there than in other parts of the U.S. Because of mass migration in the 1940s and 1960s, Puerto Ricans and Dominicans settled in New York under conditions of considerable housing segregation. Their numbers also increase awareness of these communities among non-Latino New Yorkers; those settling in areas where Latinidad is more strongly associated with other groups (e.g. Mexicans) might be less likely to be seen as Latino and more likely to be seen as White or Black by locals, facilitating non-Latino ties. Although we hope additional research will investigate these expectations, we see New York as an extreme test of color line racialization theory, and our findings of only partial support for it there lead us to expect even less evidence of it elsewhere.

Our study offers the first attempt to consider how skin color influences migrants’ social networks, and suggests that – with racially diverse migrant groups like Latinos – integration pathways can be shaped both by primary affiliations outside migrants’ ethnic group and by the specific ties they form within their ethnic group. Our study also shows the need to examine whether observed patterns might simply be a continuation of network patterns that occurred in the societies of origin.

The types of social networks migrants develop are important for their future mobility outcomes and those of their children. Networks can be translated into valuable social capital (Portes and Sensenbrenner 1993). Yet we caution against assuming that more Black ties implies vulnerability to downward mobility (Portes and Zhou 1993). Migrants may develop ties to middle-class minorities or otherwise tap into a minority culture of mobility (Neckerman, Carter, and Lee 1999; Smith 2014). Considering whether certain kinds of ties lead to upward or downward mobility is beyond the scope of this paper, but we note that 50% of Black ties formed by our migrant respondents were to people in professional occupations. While most of these professional Black ties were held by migrants with higher education, one-fifth were held by migrants with low education, consistent with findings that disadvantaged groups do not necessarily lack access to resource-rich ties (Fernandez and Harris 1992; Smith 2005). Even if Blacks and dark-skinned Latinos are on average more resource-poor than Whites and light-skinned Latinos, what social capital these migrants yield from their ties is an empirical question.

This study highlights the importance of skin color as a potential dimension of social stratification within Latino populations. Scholars are paying more attention to colorism in a variety of populations (Espino and Franz 2002; Frank et al. 2010; Herring, Keith, and Horton 2004; Monk 2014; Rondilla and Spickard 2007). But color stratification may also occur in subtle and indirect ways, through the social ties people form.

What might our findings imply for the positioning of Latinos within American racial hierarchies? U.S. demographics are shifting toward the arrival of a ‘minority-majority nation’. Our findings suggest that Latinos are unlikely to become fully divided into the White and Black communities, based on their skin color,¹² although some integration along those lines may occur. Internal stratification by color within Latino communities is likely to persist, akin to Blacks and other racial groups (Dixon and Telles 2017; Monk 2014; Rondilla and Spickard 2007). Over generations, some Latinos could exit the group by dropping a Latino identification (Emeka and Vallejo 2011). But continued Latino immigration (Jiménez 2010) and the lack of evidence that skin color trumps ethnicity in the ties they form make it unlikely that Latinos as a group will become viewed as White, or that color alone will determine their place in the social hierarchy. Our findings suggest that Latinos will maintain a distinctive group status in America’s racial hierarchy and political scene, and that the U.S. color line will serve as a source of internal stratification rather than dividing them into White and Black America.

Author biographies:

Wendy D. Roth is an Associate Professor of Sociology at the University of Pennsylvania. Her research focuses on how social processes challenge racial and ethnic boundaries and transform classification systems. She is the author of Race Migrations: Latinos and the Cultural Transformation of Race (Stanford University Press 2012). Her current work focuses on how genetic ancestry testing influences racial and ethnic identities, conceptions of race, racial attitudes, and racial interactions.

Alexandra Marin is an Assistant Professor of Sociology at the University of Toronto. Her research examines how information about job openings travels through social networks. She focuses on the role of information holders who make decisions to share or withhold information from their network members. Her current funded research projects are entitled, Charting the Transmission of Job Information through Social Networks: Contingencies, Mechanisms and Variations and Social Capital and Contingencies: How Labour Markets shape the flow of Job Information and are supported by SSHRC and Connaught.