In many Latin American countries, census data on race and skin color are scarce or nonexistent. In this study, we contribute to understanding how skin color affects intergenerational social mobility in Mexico. Using a novel data set, we provide evidence of profound social stratification by skin color, even after controlling for specific individual characteristics that previous work has not been able to include, such as individual cognitive and noncognitive abilities, parental education and wealth, and measures of stress and parenting style in the home of origin. Results indicate that people in the lightest skin color category have an average of 1.4 additional years of schooling and 53 % more in hourly earnings than their darkest-skinned counterparts. Social mobility is also related to skin color. Individuals in the darkest category are 20 percentile ranks lower in the current wealth distribution than those in the lightest category, conditional on parental wealth. In addition, results of a quantile regression indicate that the darkest group shows higher downward mobility.
Racial stratification is a key component in understanding a country’s inequality. In the United States, for example, where racial characteristics are included in official surveys, it has consistently been found that blacks have less of a chance to climb the socioeconomic ladder than whites (Chetty et al. 2018; Jencks and Phillips 2011; Mazumder 2014; Sawhill et al. 2012). In most Latin American countries, such analyses are not possible because official surveys do not include racial or skin color identification.1 This practice results from an ideology of mestizaje, or racial mixing, which promotes the idea that stratification is driven only by class and wealth and not by skin color (González Casanova 1965). However, this assumption has recently been challenged: new data sources that include skin color show its relevance to life outcomes throughout Latin America, including in Mexico (Aguilar 2011; Arceo-Gomez and Campos-Vazquez 2014; Flores and Telles 2012; Telles 2014; Telles and Bailey 2013; Telles and Sue 2009; Telles et al. 2015; Trejo and Altamirano 2016; Villarreal 2010). Building on this recent work, we use a novel data source in Mexico to understand the effect of skin color on life outcomes, such as schooling and earnings, and also its relationship to social mobility.
These relationships are an important object of study for Mexico. The country’s Spanish colonial experience left a society highly stratified on the basis of racial features. The colonial period (1521–1821) was one of a race-based and highly unequal social, economic, and political order. It was not until the country achieved independence in 1821 that the legal basis for discrimination by race (the casta system)2 was eliminated and a new social hierarchy was imposed based on wealth and education (Lomnitz 1993). However, racial inequality was deeply rooted in the society, and preferences for white skin and the European phenotype continued to play an important role (Villarreal 2010). After the Mexican Revolution (1910–1917), the government promoted the ideology of mestizaje as the foundation of national identity in the mixture of races. Post-revolutionary Mexico no longer recognized ethnic or racial categories.3 Contemporary official criteria classify the population only as indigenous or nonindigenous on the basis of the self-reported ability to speak an indigenous language (Villarreal 2010).
Recent studies of Mexico show the persistence of skin color distinctions in daily life. Although Mexicans describe themselves as mestizos, they also express a preference for European features and lighter skin colors, and they acknowledge the presence of discrimination against ethnic and racial minorities (Telles 2014; Telles and Bailey 2013; Telles and Sue 2009; Trejo and Altamirano 2016; Wade 2009). Anthropological studies have shown that the epithets describing people with darker skin colors—for example, indio, negro, and prieto—are often pejorative terms or insults and are assumed to be negative values and qualities. Comparatively, terms such as güera mean both “light-skinned” and “pretty.” The ordinary use of these meanings reflects a deeply rooted cultural structure (Feagin and McKinney 2002; Oehmichen 2006) that could affect socioeconomic outcomes through discrimination or unconscious biases favoring light-skinned individuals. Indeed, recent research has shown that darker-skinned individuals attain lower educational levels and earnings (Flores and Telles 2012; Telles 2014; Trejo and Altamirano 2016; Villarreal 2010).
A key question, and a neglected one, is whether skin color affects intergenerational mobility—that is, whether darker-skinned individuals progress socially and economically at the same rate as their lighter-skinned counterparts across the distribution of parental socioeconomic status. Our study contributes to the literature with an empirical evaluation of the relationship between skin color and social mobility in Mexico. We first assessed the impact of skin color on social mobility. In particular, we compared the degree of persistence in socioeconomic status for similar individuals with different skin colors. We controlled for a rich set of observable characteristics (controls that have not been performed in previous studies). If skin color is irrelevant, then social outcomes should not differ across skin color categories. We then corroborated the effect of skin color on life outcomes (educational attainment and hourly earnings) described in recent literature (Flores and Telles 2012; Trejo and Altamirano 2016; Villarreal 2010) but using additional information and introducing new variables, including cognitive and noncognitive skills, parental education and wealth, and measures of stress and parenting style in the home of origin.
Conducting this analysis required detailed information at the individual level for current and past outcomes. Such data became available only recently, in the 2015 Survey of Social Mobility in Mexico (SSM-2015), whose design and implementation was directed by the first author of this study. The survey, which is representative at the urban level (>100,000 inhabitants), is based on interviews with a parent and an adolescent aged 12–18 in each of 2,600 households; it provides detailed and innovative information for research on intergenerational social mobility. Along with questions related to current dwelling, employment, and sociodemographic characteristics, it asked respondents for retrospective information on their parental home, parental educational level, and home environment when they were 14 years old, thus providing data for a calculation of social mobility that compares respondents’ current socioeconomic status with that of their parents in the past. Key aspects of the survey are the inclusion of measures of cognitive and noncognitive abilities for each respondent and, for the first time in a survey of this size in Mexico, the inclusion of a reliable categorization of skin color, using the Project on Race and Ethnicity in Latin America (PERLA) scale (Telles 2014). The PERLA measure is not self-reported; it is recorded by trained interviewers.
Our results provide empirical evidence of profound social stratification by skin color in Mexico. Individuals with darker skin color have significantly lower levels of education and hourly wages than those with lighter skin. Those with the lightest skin color classification have an average of 1.4 more years of schooling and earn 53 % more in hourly wages than their darkest counterparts. Social mobility is also affected. Individuals in the darkest category are 20 percentile ranks lower in the current wealth distribution than those in the lightest, conditional on parental wealth. In addition, results of a quantile regression indicate that the darkest group shows higher downward mobility. These results are in line with previous studies showing that skin color in Mexico has an important effect on economic outcomes, and they advance our knowledge of the mechanisms through which stratification is reproduced. They are also in line with the recent findings for the United States of Chetty et al. (2018), who showed that blacks in the United States face greater downward mobility than whites and that family characteristics do not explain the difference in mobility between whites and blacks.
Few studies have highlighted the role of discrimination based on appearance in Mexico. Using a data set of 2,395 adults that was initially designed to study political behavior prior to the 2006 elections and also included information regarding skin color and socioeconomic characteristics, Villarreal (2010) examined the association between these variables. His findings highlighted a strong association between skin color and educational attainment, occupational status, and the probability of living in poverty. Individuals with darker skin have substantially lower educational attainment and work in less-skilled occupations, such as domestic work, manual labor, and jobs as drivers and security guards; they also have a higher probability of living in poverty. Flores and Telles (2012) replicated Villarreal’s analysis but included controls for class origin and a more detailed measure of skin color. They analyzed information from 1,562 respondents, compiled by the 2010 Americas Barometer of the Latin American Public Opinion Project (LAPOP), and found general support for the negative effect of skin color in Mexico. However, controlling for class origin, they found less of a disadvantage than Villarreal did.
Skin color affects other aspects of life in Mexico, such as the provision of public goods, job application callbacks, self-reported health, and electoral preferences. Trejo and Altamirano (2016), also employing the LAPOP survey data, showed that access to public goods and services in Mexico is strongly conditioned by race and skin color. Dark-brown indigenous people and mestizos with indigenous phenotypic features systematically have less access to clean water and education than Mexicans with lighter skin. Using an audit field experiment, Arceo-Gomez and Campos-Vazquez (2014) found that Mexican employers showed higher callback rates toward white job applicants than toward their dark brown counterparts with the same resume qualifications. Perreira and Telles (2014) studied the relationship between skin color and health disparities in Mexico and other Latin American countries and found a negative association between darker skin color and self-reported health in Mexico. Aguilar (2011) used a laboratory experiment to show that racial appearance plays an important role in contexts in which social evaluation is important. In that experiment, Mexicans tended to evaluate European-looking individuals and political candidates more positively than those who appeared to be indigenous or of mixed race.
Other Latin American countries facing the same lack of official data on skin color have also received attention.4 Researchers have found that racial mixing and the embracing of multiracial identities in the region have not led to an absence of racism, that racial mixing coexists with a racial hierarchy and ideologies of “whitening” (Telles 2004; Telles and Sue 2009), that skin color and racial self-categorization correlate strongly with perceived discrimination (Canache et al. 2014), and that darker skin color is negatively and consistently related to educational attainment, socioeconomic outcomes, and self-rated health (Marteleto 2012; Perreira and Telles 2014; Telles 2004; Telles et al. 2015).5
If skin color-based discrimination is present in Mexico and other Latin American countries, we should expect not only life outcomes (e.g., wage level and educational attainment) but also the possibility of social mobility to be directly related to skin color. However, to our knowledge, only one published work addressed the relationship between skin color and social mobility in Latin America. In this study of Brazil (published in Portuguese), Costa (2006) concluded that lighter skin color is associated with greater social mobility in wealthier households but not in poorer ones.
Building on the literature described here, our study contributes to the analysis of racial stratification in two ways. First, we provide empirical evidence regarding the relationship between intergenerational social mobility and skin color. In the study of this relationship, we also include measures of stress and parenting style in the home of origin that studies in psychology and economics have found essential in the developing skills of children (Hart and Risley 1995; Putnam 2015). Second, we corroborate earlier studies by using additional information and controlling for specific individual characteristics, such as individual cognitive and noncognitive abilities, that are considered central in explaining life outcomes but that previous studies have not been able to include. In the following section, we describe in greater detail each of our variables as well as their association with life outcomes and social mobility.
Data and Descriptive Statistics
Data for this analysis come from the Survey of Social Mobility in Mexico 2015 (SSM-2015), whose design and implementation was directed by the first author of the present study.6 The survey is representative at the urban level7 in Mexico and consists of 2,616 households in 23 states and 86 municipalities. In each household, one parent and one adolescent aged 12–18 were interviewed. The adult sample is thus restricted to households inhabited by at least one teenager aged 12–18.8 The large sample size allows for a fine-grained analysis of social mobility by skin color.
The main objective behind the survey on social mobility is to understand the underlying factors that affect social mobility in Mexico. To meet this objective, it includes detailed and innovative information for research on intergenerational social mobility. The adult sample includes educational and demographic data for all household members; information on employment status, income, and occupation for the interviewee; and data on household assets. In addition, the questionnaire solicited retrospective information on respondents’ parental homes when they were 14 years old, including household assets, parental educational attainment, and home environment.
Another innovative aspect of the survey is its inclusion of measures of cognitive and noncognitive abilities as well as information on stress and parenting style in the interviewees’ parental homes. Cognitive skills were measured with tasks such as remembering the names of animals after 30 seconds, a digit-span test (5 questions), and 10 questions from the Raven’s Progressive Matrices test. Noncognitive abilities were assessed using the Big Five trait taxonomy (10 questions), a locus of control test (10 questions), and the Grit Scale (8 questions). Stress and parenting styles were measured through retrospective questions about parenting characteristics, such as time spent with children, shouting and insults, and the presence of consistent rules when respondents were 14 years old.
A key aspect of the survey is that it includes data on the skin color for each individual surveyed, a first for a survey of its size in Mexico. The survey employed the scale used by PERLA (Telles 2014), which has been used to analyze a wide range of ethnic and racial issues in Latin America. The PERLA is based on a color scale from 1 to 11, with 1 representing the lightest skin color and 11 representing the darkest.9
The SSM used the PERLA color palette described in Telles (2014). Each interviewer received detailed training in rating respondents’ skin color according to the palette of 11 skin tones (1 = lightest, 11 = darkest) and was instructed not to take into account any other factor in doing so. Interviewers discreetly registered respondents’ PERLA classification while respondents were solving the Raven’s matrices, approximately 25 minutes after the interview had begun.10 Other studies relating phenotypic features to socioeconomic outcomes have relied on interviewer-based classifications (Flores and Telles 2012; Gullickson 2005; Massey and Sánchez 2010; Telles and Bailey 2013; Telles et al. 2015; Trejo and Altamirano 2016). The survey’s skin tone data are presented in Fig. 1.
Figure 1 shows the percentage of the sample in each of the 11 categories of the palette. Respondents are highly concentrated in the middle of the scale (4–6), and few individuals are at the extremes. The average skin tone is 4.94,11 with a small difference between men (5.16) and women (4.77). Given the composition of the sample, we formed five categories to facilitate analysis: white (PERLA 1–3), light brown (PERLA 4), medium brown (PERLA 5), brown (PERLA 6), and dark brown (PERLA 7–11). These groups account for 7.8 %, 24.2 %, 38.8 %, 22.3 %, and 9 % of the sample, respectively. In the regression analysis, we standardized the measure of skin tone to facilitate interpretation.
The study of intergenerational mobility focuses on the association between parents’ and adult children’s socioeconomic standing: a closer association means less mobility. We calculated intergenerational mobility using an approach similar to that of Torche (2015), with self-reported household-level information at the time of the interview (current household) and retrospective information for when the respondent was 14 years old (parental household). This information serves as a proxy for family socioeconomic status. To compare levels of socioeconomic status, we constructed two indices of economic well-being—one for the parental household and the other for the current household—using a primarily asset-based approach and principal component analysis (PCA), a common methodology in mobility research (CEEY 2013; McKenzie 2005; Torche 2015).
Family living standards for the current and parental households were evaluated with direct questions. For the parental index, dummy variables for asset ownership were used for the following: shower, washing machine, gas or electric stove, refrigerator, landline telephone, water heater, television, automobile, indoor bathroom, and domestic workers employed. We also included information on parental ethnoracial background (indigenous or not), parental education (elementary school or not), number of bedrooms per household member, and whether the household was in a rural or an urban area. For the current wealth index, we computed the PCA on dummy variables for asset ownership of the following: shower, washing machine, gas or electric stove, refrigerator, landline telephone, water heater, internet, and cable television, as well as the number of television sets, cellphones, automobiles, personal computers, complete bathrooms, and domestic workers employed. Three other household characteristics were included: number of rooms and light bulbs per household member, and a dummy variable for dirt floor. Both indices were standardized to facilitate interpretation of results.
The average age of respondents was 42.7 years, and 69.3 % were currently working. Parental educational attainment was defined as the educational level of the mother or father, whichever was highest, and respondents’ education was their self-reported level. Among parents, 67 % had elementary schooling or less, and only 6.3 % obtained a college education. Respondents had an average of 10 years of schooling, and 11.8 % finished college. We identified respondents’ ethnicity by asking whether their father or mother spoke an indigenous language. Only 8.9 % of respondents answered affirmatively.
Cognitive and Noncognitive Skills
Previous studies have strongly established that cognitive and noncognitive skills are good predictors of various economic outcomes, including employment, education, and income, which may in turn be important determinants of social mobility (Almlund et al. 2011; Heckman and Kautz 2012; Nisbett 2009; Richardson et al. 2012). The SSM-2015 included three basic measures of cognitive skill commonly used in general intelligence tests. The first was a 10-item Raven’s Progressive Matrices test, a widely used intelligence test (Jensen 1998). The respondent must complete a geometric design with a missing element, choosing from a set of possible options; the score is the number of correct answers. The second test of cognitive skill was an animal-naming task that measures the ability to solve problems using previous experience and knowledge. The respondent must name as many different animals as possible in one-half of a minute. This test has been used in other surveys, including the German Socioeconomic Panel (GSOEP), and has been proven to be highly correlated with other measures of intelligence (Schipolowski et al. 2014). Finally, a short digit-span test (five questions) was used to measure working memory. In this test, the respondent listens to a series of digits and is then asked to recall them. The number of correctly recalled digits in five different series is the measure of working memory. This test is one of the most widely used for short-term memory in clinical diagnosis and psychological and economic research (Ostrosky-Solís and Lozano 2006; Richardson 2007; Schurer et al. 2015). Our measure of general cognitive skills was constructed using a PCA for the standardized scores from these three tests, and the general cognitive score was standardized to allow comparability and interpretation.
The survey also included multiple measures of noncognitive skill. We obtained our aggregated measure of noncognitive skill using a PCA for three measures.12 The first was a 10-item locus of control scale. Locus of control is defined as an individual’s perceived control over life events. This short test is a modified version of the original Rotter Locus of Control scale (1966), which was included in the GSOEP (2006) and has been used in numerous studies (Caliendo et al. 2015; Heineck and Anger 2010; Piatek and Pinger 2010; Rammstedt 2007). In this test, respondents indicate their agreement with a statement on a scale from 1 (totally disagree) to 10 (totally agree). Five of the items represent internal control—for example, “You have to work hard to succeed.” The other five represent external control—for example, “I frequently feel that others have control over my life.” The measure of locus of control is the average of the 10 responses, with external control responses scored inversely: a higher score on the scale reflects a high internal perception of control.
The second test of noncognitive skills assessed the Big Five personality traits with 10 questions (Rammstedt and John 2007). The Big Five, the most widely used personality taxonomy (Cervone and Pervin 2013), are openness to experience (creative thinking), conscientiousness (scrupulous behavior), extraversion (outgoing behavior), agreeableness (cooperation skills), and neuroticism (emotional instability). A third test was for grit, which refers to the ability to commit to and meet long-term goals, measured here using the eight-item Short Grit Scale developed by Duckworth and Quinn (2009). Each item consists of a statement reflecting grit-related attitudes—for example, “I have difficulty maintaining my focus on projects that take more than a few months to complete.” Respondents rate each item on a scale of 1 (very much like me) to 5 (not like me at all).
Measures of Parenting Style and Stress in the Parental Home
Parenting and home environment are important factors in children’s development. Research has consistently shown the importance of positive parenting for the development of both cognitive and noncognitive abilities and the development of prosocial behavior. Children whose parents are actively involved in their education develop more positive behavior and show better performance (Avvisati et al. 2014; Castro et al. 2015), yet these factors are also related to the socioeconomic status of the household. Parents at higher socioeconomic levels foster negotiation and self-sufficiency skills in their children, while children in poorer households have less supervision, lower levels of autonomy, and fewer school-related activities. Poorer households also have higher levels of stress, which affects parental investment and children’s development (Evans et al. 2012; Nisbett 2009; Putnam 2015). Hart and Risley (1995) showed that low-income children hear far fewer words than their higher-income counterparts, and the words are of lower quality.
It is clear, then, that parenting style affects life outcomes, and it may be that skin color affects life outcomes indirectly through parenting style. Indeed, recent work has shown that racial hierarchies are found within the family. Hordge-Freeman (2015) and Rangel (2015) examined the relationship between racialization and the emotional life and educational investments of families in Brazil, and found that parents reproduced racial hierarchies by offering different treatment to children based on skin color. In general, lighter-skinned children are likely to receive greater investment in education and more signals of affection from their parents than their darker siblings.
Our analysis therefore included measures of home environment and parenting style to control for omitted variable bias between the outcome variables (schooling, earnings, and social mobility) and skin color. The social mobility survey allowed us to construct indices of parenting style and of stress in the parental home (when interviewees were 14 years of age) using data obtained through retrospective questions. For the stress measure, these data included (1) frequency of insults, shouts, and threats in the parental home; (2) frequency with which respondents felt emotionally close to other household members; and (3) frequency of fears and concerns within the household. Respondents rated each item on a scale of 1 (always) to 4 (never). For the parenting style measure, the data included (1) the closeness of respondents’ emotional relationships with their fathers and mothers, (2) the degree to which respondents’ parents understood their concerns and problems, (3) the number of school activities respondents engaged in with their parents, (4) the amount of spare time respondents spent with their fathers and mothers, and (5) the consistency of their parents’ rules. Respondents rated each item on a scale of 1 (a lot) to 4 (not at all).We followed the PCA for each measure and normalized the indices to means of 0.
Skin Color and Social Mobility
We begin this section by discussing our measurement of intergenerational mobility using a rank-rank specification (Chetty et al. 2014), which ranks the percentile of each individual relative to others in the sample based on the indices of parental and current wealth described in the preceding section. We then examine the relationship between social mobility and skin color.
Figure 2 shows the parental (panel a) and current (panel b) wealth index distribution by skin tone group (white, light brown, medium brown, brown, and dark brown). In both panels a and b, lighter skin color of a decile is associated with greater household wealth. For parental households (panel a), the percentage of persons in the white group is approximately 7 % up to the 7th decile, but in current households, a clear gradient is evident across deciles (the greater the household wealth, the higher the percentage in the white group). These findings suggest that light skin color is associated with advantages that begin in the parental home and continue through adulthood, with corresponding disadvantages for the darkest skin color. In our data, 52 % of individuals in the white category move upward, compared with only 43 % of those in the brown and dark brown groups.
We formalized the previous result using a regression framework. Figure 3 shows the resulting ordinary least squares (OLS) estimates of the relationship between the percentile wealth ranking of adults (y-axis) and the percentile ranking of their parents (x-axis) for three skin tone groups: white (PERLA 1–3), light and medium brown (PERLA 4, 5), and brown and dark brown (PERLA 6–11).13 That is, we established the relationship between current and parental wealth by skin color. The first notable result is the positive influence of wealth of origin on future attainment, implying persistence in socioeconomic status. The slopes for the white, brown, and dark brown groups are 0.36 (SE = 0.05), 0.43 (SE = 0.02), and 0.39 (SE = 0.03), respectively. Although the difference in slopes is not statistically significant, we show later that there is a great deal of heterogeneity. Second, it is clear that with the same level of parental wealth, the lighter a person’s skin color, the higher his or her current wealth. This finding implies that white individuals show higher upward mobility than their brown or dark brown counterparts. Among people who were poorest when they were 14 years old, those in the white group are, on average, currently better off than those in the brown and dark Brown groups.
The estimation in Fig. 3 does not show the heterogeneity in the relationship of current and parental wealth. However, we can see this heterogeneity with a simple quantile regression. We estimated the intergenerational association at several conditional percentiles (20th, 40th, 50th, 60th, and 80th) of individuals’ parental wealth for the three groups separately: white, light brown and medium brown, and brown and dark brown. Quantile regression coefficients are plotted in Fig. 4, which provides an assessment of the dispersion of children’s income around the central tendency at different levels of parents’ income, by skin color. When regression lines for different quantiles are plotted, the variance of the distribution of current wealth across levels of parental advantage becomes obvious. This analysis complements that of Fig. 3 by focusing on the predicted dispersion in addition to the predicted mean of individual outcomes at different levels of parental advantage. The skin tone group with wider variability around the mean can be seen as having higher mobility at that level of parental wealth. For the lowest quantile of every group, Fig. 4 describes a situation where individuals in the white group at the lower end of the parental distribution will be better off than their counterparts in the brown and dark brown groups. On the other hand, at the top of the parental distribution, dispersion is lower for those in the white group than for those in the dark brown group: those in the white group show higher upward mobility at the bottom and lower downward mobility at the top than those in the brown and dark brown groups.
Table 1 shows calculations of the social mobility matrices associated with each of the three groups considered. For each matrix, element (i, j) indicates the proportion of individuals whose current wealth falls within quintile j and whose parents’ wealth falls within quintile i. A useful way to interpret Table 1 is to recall the perfect mobility case, in which parental wealth has no influence on an individual’s current wealth. This would require the same proportion of individuals in all cells (20 %). Another benchmark is the zero-mobility scenario, in which the transition matrix would have 100 % along the main diagonal and 0 % elsewhere.
Perhaps the most striking finding is the intergenerational persistence among the wealthiest quintiles (Q4 and Q5). The proportion of individuals who started and stayed at the top of the distribution (Q4 and Q5) is almost 90 % for the white group, compared with 67 % for the brown and 57 % for the dark brown group. At the lower extreme of the distribution, individuals starting in the bottom quintile (Q1) who managed to move into the higher quintiles (Q4 and Q5) are mostly lighter-skinned: 51 % of individuals in the white group, but only 23 % and 15 % of the brown and dark brown groups, respectively.
Table 2 presents our OLS estimation of the effect of skin color on intergenerational mobility. Panel A shows the results using the standardized skin color, and panel B shows the results for skin color groups as in previous tables. Columns 1 and 2 include sex (female), age, and age squared as regressors; columns 3 and 4 additionally include indicator variables for parental ethnicity and both parents present in the household of origin as well as the variables of parenting style and stress in the household of origin. Columns 1 and 3 show calculations of the direct effect of skin color on current wealth, conditional on parental wealth. Columns 2 and 4 include an interaction between parental wealth and skin color in order to calculate differential effects on mobility. The results depict stratification in current wealth associated with skin color. Column 1 in panel A indicates that a 1 standard deviation increase in skin color on the PERLA scale, with constant sex, age, and parental wealth, translates into a lowering of the current wealth rank by 5.5 percentile ranks. The remaining columns in panel A show similar results. Social mobility is relatively low in Mexico: columns 1 and 2 show that, on average and with all else held constant, a one-unit increase in the parental wealth rank distribution increases the current wealth distribution by 0.43 units. Columns 2 and 4 do not show differences in mean mobility across skin color categories.
Panel B of Table 2 includes the results by skin color group, which support the negative color gradient hypothesis that darker skin color is associated with lesser life outcomes. Column 1 shows that the dark brown group is 20 percentile ranks below the white group in current wealth, even after parental wealth is controlled for; the remaining columns show qualitatively similar results. In other words, family and individual skills and characteristics do not explain the gap in social mobility. This result is similar to that found by Chetty et al. (2018) for blacks and whites in the United States. As in panel A, columns 2 and 4 in panel B show no differences in mobility (in the mean) across skin color. Figure 4 and Table 2 do, however, indicate heterogeneity of mobility by parental wealth.14
Why do individuals in the white group achieve better outcomes than those in the dark brown group? A conclusive answer to this question will require future research. Here we provide some possible explanations. Previous studies have shown less favorable life outcomes among darker-skinned individuals in Mexico (Flores and Telles 2012; Telles 2014; Telles and Bailey 2013; Telles and Sue 2009; Telles et al. 2015; Trejo and Altamirano 2016; Villarreal 2010). There is also compelling evidence of discrimination based on physical appearance in the Mexican labor market. Darker-skinned individuals are employed in less-skilled occupations, including domestic work, manual labor, and jobs as drivers and security guards (Arceo-Gomez and Campos-Vazquez 2014; Villarreal 2010). Undoubtedly, discrimination is a channel that leaves darker-skinned people without the same opportunities for mobility as their lighter-skinned counterparts.15
Skin Color and Life Outcomes
The main question of this study is whether skin color affects intergenerational social mobility. We completed the analysis by corroborating the effect of skin color on life outcomes described in recent studies but using additional information and introducing new variables, such as cognitive and noncognitive skills, parental wealth, and measures of stress and parenting style in the home of origin.
Figure 5 shows the relationship between skin color and life outcomes. Panel a shows the relationship between skin color and average years of schooling, and panel b shows the relationship between skin color and average hourly wage. For simplicity, the five skin tone groups described in Table 1 (PERLA 1–3, 4, 5, 6, 7–11) are used. The bars depict descriptive results: they show the respective life outcomes by skin color group. In general, the relationship between skin color and life outcomes is unambiguously negative and decreasing, evidence for what we refer to as the negative color gradient hypothesis. It is noteworthy that years of schooling decrease as skin color darkens, from an average of 11.5 years in the white category to 9 years for the dark brown group—a difference of 2.5 years between extremes. A similar trend can be seen in hourly earnings: average hourly wages range from approximately 78 MXN ($9.20 USD in purchasing power parity (PPP)) for the white group to approximately 40 MXN ($4.90 USD in PPP) the dark brown group—a 49 % difference.
Because the data set includes detailed individual information, it allowed us to control for a number of factors not previously included in the literature that could potentially bias the estimation of the effect of skin color. We carried out a regression analysis of educational attainment (years of schooling) and individual earnings (log hourly wages) on skin color (standardized score and PERLA groups) and a vector of individual productivity-enhancing characteristics.16 Coefficients of such estimations were produced for every skin color group. The analysis included the following variables as controls: gender, age, age squared, parental ethnicity, cognitive and noncognitive skills, parental wealth, parental education, parenting style in the home of origin, stress in the parental home, and an indicator of the presence of both parents while growing up. In general, the results show that with other variables held constant, darker skin color is associated with fewer years of education. Estimated coefficients shown over the bars in panel a (Fig. 5) illustrate that, on average, the white group has 1.4 more years of schooling than their darkest counterparts. However, the effect of skin color on schooling is not significant (at the 95 % confidence level) for the light brown, medium brown, and dark brown groups. That is, although the descriptive evidence does show a negative gradient, regression analysis suggests that the differences are palpable only at the extremes. People with darker skin attain significantly less education than those with lighter skin. However, adjustment for characteristics such as cognitive and noncognitive skills and parents’ education and wealth lessens the magnitude of the disparity.
These findings have also been described for other countries in Latin America. Telles et al. (2015) clustered the skin colors in the PERLA palette into three groups (light, medium, and dark) and found a negative relationship between darker skin colors and educational attainment in eight countries (Bolivia, Brazil, Colombia, Dominican Republic, Ecuador, Guatemala, Mexico, and Peru), even after controlling for ethnoracial identification. The most extreme example was Bolivia, where the white group had an average of 3.5 more years of education than the dark group. The least extreme was Brazil, where the average difference was approximately 1 year. In Mexico, we found a difference of 2.3 years. Thus, our estimate of the effect of skin color on years of schooling, controlling for a richer set of individual characteristics and family background, is smaller than that reported in other studies (Flores and Telles 2012; Telles et al. 2015; Trejo and Altamirano 2016; Villarreal 2010).
The regression analysis in panel b of Fig. 5 shows an estimation similar to that in panel a but with the log of hourly wage rather than years of education as the dependent variable. These estimations included the same controls as in panel a and an additional control for years of schooling. The estimated coefficients provide strong evidence in favor of the negative color gradient hypothesis: people in the light brown, medium brown, brown, and dark brown categories earn an average of 37 %, 38 %, 54 %, and 53 % less in hourly wages, respectively, than those in the white group—differences that are economically and statistically important. These results are consistent with a color gradient relationship found in previous studies in Mexico (Arceo-Gomez and Campos-Vazquez 2014; Trejo and Altamirano 2016) and in studies of African Americans (Keith and Herring 1991) and Latinos (Telles and Murguia 1990) in the United States, which found that dark-skinned Latinos and African Americans experienced less occupational prestige and had economic outcomes inferior to those of their lighter-skinned counterparts. These findings are also consistent with the higher wage penalties found for darker-skinned individuals: a penalty of at least 10 % for black men relative to white men in the United States (Espino and Franz 2002; Goldsmith et al. 2006; Hersch 2006; Hill 2000; Telles and Murguia 1990). Finally, our results are consistent with the extensive descriptive work presented in Bailey et al. (2014), which showed evidence of light-skinned people having the highest mean incomes in a large part of Latin America (14 of the 18 countries studied) and disadvantages increasing with darker skin color. As they noted, however, color and racial hierarchies in the Americas are a complex matter. White group advantages and indigenous disadvantages are constant across the region, whereas disadvantages attributed to dark skin color are sometimes less, depending on the context.
Our analysis controlled for different abilities in order to assess the mechanism influencing the lower hourly wages of dark-skinned individuals. The salary premium for lighter skin color in Mexico did not change greatly when we included a large set of skills and individual characteristics. Mechanisms of explicit discrimination in the labor market are a more plausible explanation (Arceo-Gomez and Campos-Vazquez 2014). Our results suggest that skin color influences the stratification system in Mexico, with large differences in wages between darker and lighter people who are otherwise similar.
The main question of this study is whether skin color affects social mobility in Mexico. The ideology of mestizaje suggests that skin color should not affect economic outcomes. However, taking together our results and the conclusions of previous studies of Mexico, we conclude that skin color is still a key variable in social inequality and intergenerational mobility.
This study contributes to the understanding of racial stratification by analyzing the effect of skin color on social mobility. We used the 2015 Survey of Social Mobility in Mexico (SSM-2015), whose general objective is to understand the underlying factors of social mobility in Mexico. We extended the analysis by corroborating existing evidence of the relationship between life outcomes (wages and educational attainment) and individual skin color but using additional information and a richer set of individual characteristics that previous studies have not been able to include: a measure of cognitive and noncognitive skills, the economic situation of the parental household, and other sociodemographic characteristics.
Our results are in line with previous studies and provide empirical evidence of profound social stratification by skin color. Mexicans with darker skin color attain significantly less education than their lighter-skinned counterparts. However, controlling for characteristics such as cognitive and noncognitive skills and parents’ education and wealth lessens the magnitude of the disparity. Those with the darkest skin color average 1.4 fewer years of education than those with the lightest skin color. This relationship also holds for hourly earnings: the lightest-skinned receive 53 % more hourly earnings than their darkest counterparts. These results are less pronounced than those reported in previous studies, but they also include a richer set of individual and family characteristics.
Our study also adds to the understanding of racial stratification by examining the relationship between intergenerational social mobility and skin color in a Latin American country. Given the design of the survey, we were able to test the effect of individuals’ skin color on their economic position, controlling for the socioeconomic position of their parental home and other determinants of economic outcomes, such as ethnicity, years of schooling, schooling of parents, and information about their home environment while they were growing up. People with lighter skin show more upward social mobility, independent of their starting wealth. Those with darker skin are 20 percentile ranks lower in the current wealth distribution than their lighter-skinned counterparts with the same level of parental wealth. Individuals with the darkest skin color also show higher downward mobility than other groups. We find that stratification by skin color persists in Mexico, and we provide evidence of its importance to life outcomes. Future research is needed to explore the importance of factors besides labor market discrimination to explain such differences.
This work was supported by the Sectorial Fund for Research on Social Development of the Mexican National Council on Science and Technology (CONACyT) and the Secretary of Social Development (Sedesol; Project No. 217909). We thank the anonymous reviewers and the editor for thoughtful comments and suggestions that substantially improved the paper. Any errors or omissions are the responsibility of the authors.
The exception is Brazil, where information on ethnoracial identification is available in the official census and incorporated into public policy (Bailey 2009; Daniel 2006; Dixon and Telles 2017; Schwartzman 2007).
Legal restrictions, for example, prohibited the members of the lower castas from occupying important government positions or living in certain neighborhoods (Meyer et al. 2013; Villarreal 2010).
Loveman (2014) provided an extensive analysis of this topic for the entire Latin American region, noting that census queries about skin color and other phenotypic features disappeared in the mid-twentieth century mainly because domestic political agendas followed international definitions of modernization and development.
In general, this research is based on two data sets: the LAPOP and the 2010 PERLA. Together, the two databases include information for 18 of the 19 countries in Latin America, omitting only Cuba. Our data set, the Survey of Social Mobility in Mexico (SSM-2015), represents an important additional resource for research on the topic.
A separate case in Latin America is Brazil, where the relationship between skin color and ethnoracial stratification is widely acknowledged, data on skin color and other phenotypic characteristics are officially collected, and the academic and political debates on racial stratification have led to policies of affirmative action (Bailey 2009; Daniel 2006; Schwartzman 2007).
In Spanish, Encuesta de Movilidad Social 2015. The survey is publicly available at http://movilidadsocial.colmex.mx/index.php/encuesta. For a detailed guide, see Campos-Vazquez (2016).
Urban was defined as a community of 100,000 inhabitants or more. By this definition, the urban population of Mexico is close to 50 % of the total. This limitation of the survey does not compromise our results. As we explain later, our results are very similar to those presented in Flores and Telles (2012), which are representative at the national level. The results associated with the variable of interest (skin color) and results with other controls are of similar magnitude and sign, as discussed in the section, Skin Color and Life Outcomes. Our sample is also larger than the one Flores and Telles used, giving us greater statistical power.
The social mobility survey consists of two separate questionnaires designed specifically for adults and teenagers. To avoid interference in answers, interviewers surveyed parents and children separately. In this article, we analyze only the adult responses.
The PERLA color palette is the result of Telles’s (2014) arduous identification and classification of the phenotypes of Latin America. More information can be found at https://perla.princeton.edu/perla-color-palette. Using this palette, several studies have analyzed the effect of skin color on economic outcomes in Latin America. For an international perspective, see Glenn (2009) for a discussion of the impact of skin color inequality independent of race in different national contexts.
One potential concern about using an interviewer-based measure is that respondents’ socioeconomic status could bias interviewer evaluations of skin color. In other words, social class may change ethnoracial perception in a “money whitening effect,” whereby wealth could drive interviewers’ classifications into whiter categories (Telles et al. 2015). The SSM sought to minimize such effects through the use of a color palette and by thoroughly training the evaluators. A further concern is that interviewers’ classifications may be influenced by factors such as their own education or gender. However, previous studies have found that neither the sex nor the educational attainment of interviewers is a significant predictor of their color ratings in Mexico (Telles et al. 2015; Villarreal 2010).
This figure is close to the value of 4.5 (SD = 1.41) reported for Mexico in Telles et al. (2015).
In the online appendix, we include results using each noncognitive variable separately instead of aggregated into a single index. The results are unchanged.
That is, we estimated yi = α0 + β × Parenti + ε, where yi is the individual’s current wealth, expressed as an index, and Parenti corresponds to the calculated parental wealth index. We estimated this regression separately for the subsamples white (PERLA 1–3), brown (PERLA 4–5), and dark brown (PERLA 6–11) and plotted the prediction from the regression.
Additional results are shown in the online appendix. We quantified the effect of skin color on mobility by estimating conditional quantile regressions of current against parental percentile rank of wealth, sex, age, and age squared. The main result of this regression is that the brown and dark brown (PERLA 6–11) groups show lower coefficients for parental wealth than the other groups. It is as if parental wealth provides less insurance against very bad outcomes for those in the dark brown group. Simply put, the dark brown group experiences higher rates of downward mobility than other groups.
However, other potential factors related to discrimination have not been explored, such as differences in personal aspirations and social identity associated with phenotypic differences (Gilliam et al. 2016; Hoff and Pandey 2014; Hordge-Freeman 2015; Jones et al. 2012; Rangel 2015; Steele 2010; World Bank 2015).
Specifically, we define five skin tone groups—white (PERLA 1–3), light brown (4), medium brown (5), brown (6), and dark brown (7–11)—and estimate , where yi is either years of schooling or log earnings for individual i, and Xi is a vector of individual characteristics. In this way, we compare life outcomes for each skin tone group relative to the lightest. We also estimate yi = α0 + δ × Skintone + β × Xi + ε, where Skintone corresponds to the standardized PERLA skin tone classification. The color scale runs from 1 to 11, with 1 the lightest color and 11 the darkest. Several specifications of this reduced-form relationship are estimated, and the results for years of schooling and log earnings are reported in the online appendix.