Space and Interracial Marriage: How Does the Racial Distribution of a Local Marriage Market Change the Analysis of Interracial Marriage in Brazil?

Racial Classification and Interracial Marriage in Brazil

Racial identity in Brazil is recognized as something quite different from other countries, especially because of the absence of a descent rule, which means that Brazilians consider different factors when classifying one’s race ( Degler, 1971). Another issue is the association between race and socioeconomic conditions ( Schwartzman, 2007). Politics may also influence racial self classification in Brazil ( Bailey, 2008). Another specific characteristic of the Brazilian racial system is the recognition of mixed-race people as a distinct category, which sometimes makes both the creation of specific racial boundaries and self-classification difficult. In this sense, the white/nonwhite perspective utilized in the United States, does not apply to Brazil. As noted by Ribeiro (2017) the racial classification in Brazil is much more a continuum than a set of fixed categories. While racial identity is more related to ancestry in the United States, in Brazil physical characteristics are the main indicators of race. This means that racial classification in Brazil is much more a matter of an individual’s perception of their own physical characteristics than to their family’s origins ( Bailey, 2008 ; Silva, 1987 ). It is significant to note that racial classification also depends on who is doing the reporting because important differences arise when comparing self-reporting and classification conducted by another person (i.e. Telles and Lim, 1998; Bailey, 2008). This result corroborates the argument about the low degree of groupness and how racial boundaries in Brazil are not clearly defined.

Since 1940, the Brazilian Institute of Geography and Statistics (ibge) has used four racial categories in the Census and other surveys: branco (white), pardo (brown), preto (black), and amarelo (asian). A fifth category, indígena (indigenous), was added in 1991. One of the main concerns about the official terms used by the ibge is the extent to which they correspond to informal terms largely used by the population. In 1976, the National Household Survey (pnad-Pesquisa Nacional por Amostra de Domicílio), conducted by the ibge, asked an open question about race and received more than 190 answers regarding racial categories. However, most people (95 percent) used only six terms, including three of the categories already used by the ibge: white (41.9 percent), light (cor clara) (2.5 percent), light moreno (2.8 percent), moreno (34.4 percent), brown (parda) (7.6 percent), and black (preto) (4.4 percent). Therefore, the majority of Brazilians seem to use a small number of terms, with white and moreno being the most common terms, comprising 76 percent of all answers ( Silva 1987: 71).

Although there is an understanding that racial boundaries are fluid, racial distribution in Brazil remained very stable between 1980 and 2000. During this time period, the proportion of blacks was between 5–6 percent, browns were between 39–42 percent, and whites were 54–52 percent (ibge 2004). The 2010 Census data show that whites represent 47.73 percent of the total population, 43.13 percent is brown, and 7.61 percent is black. The main change from the last Census in 2000 is the current absence of a majority racial group in Brazil. Because of that important change it is fundamental to investigate what looks like the Brazilian marriage market after 2000.

In 1998, 78 percent of all marriages were racially endogamous, with rates being higher among whites (83.2 percent) and lower among blacks (60.7 percent). When controlling for the size of the population, this relationship inverts and blacks have the highest rate of endogamy ( Petruccelli, 2001). On average, couples are 5.2 times more likely to marry someone of the same race, whereas, individuals with mixed ancestry are more likely to marry exogamously ( Heaton, 2010). Although endogamy is preferred, there is less resistance to marriage between persons of proximate color in comparison with the much higher resistance to marriage between whites and blacks ( Telles, 1993, 2004). In an interracial union, white people are ten times more likely to marry a brown spouse than a black one. In 1991, about 14.2 percent of white men had brown wives and only 1.3 percent had black spouses ( Telles, 2004).

Although there are higher levels of intermarriage in Brazil than in the United States and South Africa, marriages are far from being random. One possible explanation is the higher acceptance of browns/mulattos in society and the fact that most interracial unions are between whites and browns. Another explanation suggested by Telles (2004) is the greater exposure of whites to nonwhites in Brazil, which also helps to explain the geographical differences in interracial marriage rates. The author claims that geographical differences have little or nothing to do with variations in tolerance. Rather, propinquity or the extent of interracial interaction determines nearly all the regional variation. Telles also cautions and makes clear that exposure does not explain all the differences between Brazil and the United States.

Over time, racial endogamy has been declining, as have educational and religious endogamy as well ( Heaton, 2010; Longo, 2011; Ribeiro and Silva, 2009). Overall, endogamy was 77 percent in 1980, it declined to 73 percent in 1991, and reached 65 percent in 2000 ( Longo, 2011). But this process is not the same for all racial groups. Racial homogamy is greatest for blacks and those with mixed ancestry (brown) have much lower rates of homogamy. Homogamy is declining for whites and blacks at the same pace, but is increasing slightly for browns ( Heaton, 2010).

Intermarriage and the Local Marriage Market

Most studies about local marriage markets has focused on the analysis of sex ratios (see for instance, Freiden, 1974; Angrist, 2002; Queiroz, 2004; Posel et al., 2013). A few additional authors explored other characteristics such as racial distribution ( Harris and Ono, 2005; Telles, 1993); socioeconomic characteristics and immigrant influence ( Telles, 1993); and economic opportunity for females, male economic attractiveness, and racial differences ( Posel and Casale, 2013).

When considering Brazil, all analyses of interracial marriages using loglinear models utilized national level data (i.e., Silva, 1987; Ribeiro and Silva, 2009; Heaton, 2010; Longo, 2011). The main assumptions in this approach, are: i) race is distributed equally across the country or ii) the probability of marrying someone that lives close or further away is the same ( Harris and Ono, 2005). Another assumption is that race classification is the same in the whole country. However, the racial classification of people can vary across the country, and in this sense, racial self and the perception of others is influenced by exposure and local social norms. A fourth factor that may alter across the country is the geographical difference in the acceptability of having a nonwhite in one’s own family. Telles (2004) cites Degler’s work who discussed a series of studies showing that there are important geographical differences in this regard. In a research project conducted among college students, the acceptability rate of a brown or a black person in one’s own family ranges from 14 to 24 percent in Recife, and from 10 to 14 percent in São Paulo. Amongst a sample of white high school students in Rio de Janeiro the range was between 37–43 percent. In Florianópolis, the range was 23–28 percent among middle class people.

For the year 2000 in Brazil, there was a concentration of whites in the South, where 83.6 percent of the total population was white. Nonwhites were concentrated in the North of the country: 4.97 percent of the population was black and 63.97 percent was brown. The proportion of whites in mesoregions during this period ranged from 11.07 percent (Norte Amazonense) to 93.86 percent (Vale do Itajaí). The distribution of black people ranged from 1.47 percent (Vale do Itajaí) to 20.24 percent (Metropolitana de Salvador); and the range for browns was from 3.19 percent (Sul Catarinense) to 72.25 percent (Nordeste Paranaense). Curiously, the mesoregion with the highest proportion of browns was in the South, followed by other mesoregions in the North (Norte do Amapá, Baixo Amazonas, and Marajó) that were composed by about 70 percent of brown people. This difference in racial distribution influences exposure, which consequently influences opportunities and preferences within the marriage market. The group size influences the possibility of finding a partner within or outside of one’s racial group. Exposure also influences an individual’s preference by defining what is more socially accepted in his/her social circle. Another element of exposure is that it establishes the presence of other interracial couples in the individual’s life.

Therefore, the consideration of local marriage markets is important when studying intermarriage (see for instance, Freiden, 1974; Angrist, 2002; Queiroz, 2004; Posel et al., 2013). As Kalmijn (1998) states, there are three causes for intermarriage: 1) the preferences for certain characteristics in a spouse; 2) the interference of “third parties” in the selection process, and 3) the constraints of the marriage market in which candidates are searching for a spouse. Hence, marriage patterns are a result of preferences and opportunities. The characteristics of the geographical area considered as the marriage market, including the racial distribution, are perceived as constraints or as facilitators of intermarriage. In this article, I also argue that the local marriage market may influence the first intermarriage cause noted by Kalmijn. In other words, the amount of exposure to other racial groups experienced by an individual may influence his/her preference in terms of a spouse’s characteristics. This occurs because an individual’s decision also depends on social expectations and what is considered “normal” (in the sense of what is the norm) among one’s social network and family. Moreover, how people are racially perceived of by others also depends on local characteristics.

The most common explanation for lower rates of intermarriage than homogamy (i.e. racial, religious, educational) is based on Weber’s (1978) concept of group closure. The understanding is that intermarriage is an indicator of social distance. However, without considering demography (group size) it is not possible to fully understand intermarriage as a social boundary ( Silva,1987). Thus, it is important to consider the conditions of local marriage markets, including the racial distribution, because part of the rate can be explained as a lack of opportunity and not necessarily the result of preferences.

The consideration of local characteristics is also important for helping to elucidate the main reason for intermarriage. Kalmijn (1998) indicates that intermarriage rates in smaller areas might be due more to preferences than opportunities. Therefore, this article sheds light on to what extent previous national rates of intermarriage in Brazil are related to opportunities, while answering the question: how would homogamy rates be affected if the racial distribution in different areas was considered in the analysis? In other words, how much of the drop in the odds ratio is due to a different racial distribution. In this sense, the intermarriage rates that cannot be explained by racial distribution may be justified by other local characteristics that are not being controlled for. These local characteristics might include individual preferences, opportunities in a smaller area, or more specific marriage markets, like the school or workplace environment, for example.

For Brazil, Telles (1993) used metropolitan areas as the local marriage market and found that racial composition explains most (but not all) of the geographical variation in interracial marriage, because the likelihood of out-marriage depends partly on the extent of exposure to the outside group. His work was inspired by the debate about regional differences in marriage rates between Frazier (1942) and Degler (1986). These authors concluded that the marriage rate differences reflected profound regional differences in racial tolerance across the country. Telles’ work included different variables to explain the proportion of marriages between two racial categories, in order to question whether regional differences are due to differences in racial tolerance or are related to variations in racial composition. He found that despite controls for racial composition, white out-marriage to blacks in the South of Brazil is significantly lower than in other regions, and thus the white/black social divide is greatest in the South. Similarly, brown and black social relations in the marriage market, are more distant in the Northeast and frontier regions (Amazonia, Acre, the Federal District, Goiás, Mato Grosso, Mato Grosso do Sul, Pará, and Rondônia). In addition, Telles highlights the importance of other variables, such as the level of industrialization, as well as others like the sex ratio, which are not included in the analysis. One notable message in his work is related to racial asymmetry. Telles concludes that there are small regional differences in the white tolerance level of browns, but the differences in white racial tolerance of blacks is more relevant.

The main way in which this article differs from Telles (1993) work is the estimation of the homogamy-heterogamy rates and the analysis of potential differences among the local marriage markets. His work only utilized one model for all markets. Moreover, our concept of the local marriage market differs. Where he used metropolitan areas, I am using mesoregions. He justified the use of metropolitan areas as the representation of where people from different racial groups are mostly likely to come into contact with each other. However, one important point worth noting is that only considering metropolitan areas results in excluding areas with a concentration of one specific racial group from the analysis. Therefore, my study provides a more complete picture of the importance of considering the local racial composition. Moreover, my analysis tests the hypothesis of whether what is not explained by racial distribution is the same or is different among the local marriage markets. This refers to examining whether values, preferences, racial tolerance, and other factors influence intermarriage in the same way throughout the country.

Finally, it is relevant to highlight that discussing local marriage markets is difficult because there is no common agreement on the definition of a marriage market. Spouses can meet one another in their neighborhood, but also outside of their community. However, studies have shown that people are more likely to find a spouse in their local context ( Blau, Blum, and Schwartz, 1982; Lewis and Oppenheimer 2000); even people who often travel ( Harris and Ono, 2005). The other problem is that it is often difficult to determine where people meet because many databases do not have data on marriage history. Different studies have used varying concepts of what constitutes a local marriage market, including schools ( Blossfeld and Timm, 2003; Mare, 1991), workplaces (i.e., Svarer, 2007), and small geographic areas (i.e., Blau et al., 1982; Lewis, 2000; Harris and Ono, 2005). In this sense, another important reason for considering the geographical factor when analyzing intermarriage, is the possible effect of neighborhood or school related peer groups (Holmlund, 2006). However, the effect of these groups is not evaluated in my work. Moreover, it is also significant to mention that residential segregation, which is an indicator of how racial groups relate to each other, reflects racial tolerance and in some extension racial economic inequality. Therefore, identifying areas with more nonwhite people may indicate a lower tolerance for this racial group. Although a lack of exposure decreases the likelihood of whites out-marrying, the ultimate cause of this phenomenon is not only a result of exposure or the composition of the population, but is directly related to racial segregation.

Data

The data are from the Brazilian Census from the Integrated Public Use Microdata Series (ipums), for the years 1991 and 2000. I linked the head of the family with his/her spouse and considered each one’s race and the mesoregion where they lived at the time of the Census. I used family as an identifier instead of household, because a household can consist of multiple families. Married and cohabiting couples were considered together, because no significant differences when the models were done separated in preliminary analysis. The total number of mesoregions is 137 and they all correspond to the same areas during the two periods. For this analysis I am considering black, brown, and white people and the six possible combinations among them: white/white, brown/brown, black/black, brown/black, white/brown, and white/black couples.

In order to minimize the effect of migration after marriage, I limited my sample to couples in which both women and men are between twenty and thirty-four years of age. Spouses in this age range are more likely to be in their first marriage and less likely to have moved away from where they met. This selection also decrease the potential effect of characteristics associated with second or higher order marriages (See Mare, 1991; Kalmijn, 1994; Qian, 1997). Mare (1991) noted that the educational pattern of a higher order marriage is different from that of a first marriage. However, Blackwell and Lichter (2004) did not find substantive differences in terms of racial assortative mating.

I am using mesoregion as an indicator of the local marriage market. Mesoregions are combinations of municipalities. They are geographically stable throughout time, with minor exceptions, and divide the entire country into relatively coherent geographic units beneath the state level ( Minnesota Population Center, 2011). Internally, they are socially and economically homogenous. They have transportation linking the cities within each mesoregion and do not cross state boundaries.

I am using three of ibge’s official racial categories: white, black, and brown. The racial classification is self declared, therefore, the categories reflect whether people consider themselves to be branco (white), preto (black), or pardo (brown). The brown category usually includes a larger range of skin color variation, including moreno and its variants. Interracial marriages are the combination of whites and browns, whites and blacks, and browns and blacks. I do consider blacks and browns as separate groups due to the fact that they have different probabilities of marrying whites. These probabilities even differ by sex. Although these two groups are much more alike in socioeconomic terms, in the marriage market they have different statuses. Finally, I excluded the Indian and Asian racial groups because they represent a small part of the population. In some regions there is no Indian or Asian population, which would compromise the use of the racial distribution of a local marriage market.

Loglinear Approach

I used loglinear models in order to analyze the possible association between the local marriage market (mesoregions) and a spouse’s racial distribution, following the previous work conducted by Harris and Ono (2005). Loglinear models can be compared to ordinary least squares regressions, in the sense that the models estimate a parameter for the row variable, the column variable, and interaction effects between these variables for each cell in the table. The row and the column effects are considered to be main effects and the interaction effects are considered to be joint effects that describe how the main effects are related to each other. The objective is to best describe the association between the variables in a more parsimonious way ( Goodman 1970, 1972). The main difference is that in loglinear models, a specification of dependent and independent variables does not necessarily exist, and all variables in the model are categorical. The outcome is the frequency of each cell. When it is possible to differentiate independent and dependent variables, the loglinear model can be converted into a logit or linear logistic response model ( Fienberg, 2007)

The models I present follow the work done by Harris and Ono (2005). The first model is the independence model. This model states that there is no association among the variables (wives’ race, husbands’ race and mesoregions). The main objective is to establish a baseline to which the adjustment of the other model can be compared, so if the models with the interaction terms better describe the data the assumption of independence is not corroborated. Considering F as the fitted frequency for each cell, i=1 or 2 refers to husband’s race; j=1 or 2 refers to wife’s race and k=1 to 137 refers to mesoregions and ß^M indicates the market’s parameters, ß^H and ß^W are husband’s race and wife’s race parameters. The equation of the independence model is as follow:

$\textcolor{red}{}$ $\ln \left(\frac{F_{ijk}}{F_{i{j}^{\prime }k}}\right)={\beta }_{0jk}+{\beta }_{1jk}^H{\chi }_i+{\beta }_{2ik}^W{\chi }_j+{\beta }_{3ij}^M{\chi }_k$

In the second model, an interaction term between husband’s and wife’s race is included, represented by ß^HM . This term means that the spouse’s race is associated to the individual’s own race. For instance, the interaction between black/black might be higher than an interaction between black/white because a black person is more likely to marry a black spouse than a white spouse. Because there is no interaction with the mesoregions, this model reflects the usual assumption of independence between the local marriage market characteristics and spouses’ characteristics, or a similarity of all local marriage markets. The model is represented by Equation 2:

$\ln \left(\frac{F_{ijk}}{F_{i{j}^{\text{'}}k}}\right)={\beta }_{0jk}+{\beta }_{1jk}^H{\chi }_i+{\beta }_{2ik}^W{\chi }_j+{\beta }_{3ij}^M{\chi }_k+{\beta }_{4k}^{HW}{\theta }_{ij}$

In the third model, two interaction terms are added. One term is added between husband’s race and mesoregion and one between wife’s race and local marriage market, represented by ß^HM and ß^WM , respectively. These two terms reflect eliminating the assumption of combining all marriage markets, meaning that there is an interaction between the racial distribution of a local marriage market and the person someone chooses to marry (or his/her spouse’s race). The inclusion of these two terms also assumes that the marriage markets are different and that this difference is important to be considered. The hypothesis being tested is whether it is necessary or unnecessary to consider local marriage market racial distribution for calculating homogamy-heterogamy rates. In other words, I seek to determine if model 3 is actually better than model 2.

$\ln \left(\frac{F_{ijk}}{F_{i{j}^{\prime }k}}\right)={\beta }_{0jk}+{\beta }_{1jk}^H{\chi }_i+{\beta }_{2ik}^W{\chi }_j+{\beta }_{3ij}^M{\chi }_k+{\beta }_{4k}^{HW}{\theta }_{ij}+{\beta }_{5j}^{HM}{\theta }_{ik}+{\beta }_{6i}^{WM}{\theta }_{jk}$

In sum, the models represent the main concern of this article: the interaction between the race of a husband and wife (equation 2) and how race is affected when an interaction term is added between the husband’s race and mesoregion and between the wife’s race and mesoregion (equation 3). Contemplating an interaction is the same as thinking about non independence. The main objective of using interaction terms is to analyze whether or not the variables, in this case a couple’s race and mesoregion, have or do not have a joint distribution. In other words, the objective is to determine whether the racial pairing depends on the local racial distribution.

The coefficient of the loglinear model is the natural logarithm of the ratio of the observed value and its predicted value, called the log odds ratio. A positive coefficient means that there are more unions in the specific table cell compared to how many there would be in cases where unions were randomly assigned. A negative coefficient means that there are fewer unions than if the spouse’s characteristics were independent. In other words, a positive coefficient indicates that the probability of marrying a brown or black person is greater than a marriage with a white spouse (reference category). If the coefficient refers to an interaction term and is positive it shows that the married between the racial groups, for instance, a black wife and a brown husband, or a brown wife and a brown husband, are more likely than the marriage between a white wife and a white husband. The coefficients can be expressed by the odds ratio e ^ß or as percentages 1-e ^ß. Therefore, the odds ratio allows us to interpret each logit as the odds of marrying a black or a brown spouse as opposed to a white spouse (reference category). The white category is the baseline, which means that this category’s coefficients are equal to zero.

Using the interaction terms between a husband’s and a wife’s race, we can calculate the homogamy-heterogamy rates. The homogamy-heterogamy rates are odds ratios. This means that the rates indicate whether intraracial union is more or less likely than intramarriage. An odds ratio is the ratio of the odds of an outcome occurring within a group compared to another group. In this case, I seek to determine the odds of an intraracial union happening in two specific racial groups compared to the chance that an interracial union between these two racial groups will occur. Because white is the omitted category in the analysis, all loglinear coefficients for whites will be equal to ‘‘0,’’ and the exponentiated coefficients will be equal to ‘‘1.’’ As a result, homogamy–heterogamy odds ratios for white/black and white/brown marriages can be expressed as

$\exp {\beta }_{h=2=,w=2}^{WH}$ , $\exp {\beta }_{h=3=,w=3}^{WH}$

Odds ratios larger than 1 indicate a greater likelihood of intraracial marriage as opposed to interracial marriage. The equation for calculating the homogamy-heterogamy rates is below and follows the course established by Powers and Xie (2000).

$\frac{\left[\exp \left({\beta }_{i=1=,j=1}^{WH}\right)*\exp \left({\beta }_{i=2,j=2}^{WH}\right)\right]}{\left[\exp \left({\beta }_{i=1=,j=2}^{WH}\right)*\exp \left({\beta }_{i=2,j=1}^{WH}\right)\right]}$

One example that demonstrates this equation is a coefficient of black/brown equal to 1.7, a coefficient of brown/black equal to 1.99, a coefficient of black/black equal to 4.76, and a coefficient of brown/brown equal to 2.57. The odds ratio (or homogamy-heterogamy rates) is

$\frac{\exp (4.76)*\exp (2.75)}{\exp (1.7)*\exp (1.99)}=38.32$

This result indicates that intraracial unions among blacks and among browns (black/black or brown/brown marriages) are 38.32 more likely to happen than interracial unions (browns marrying blacks).

In order to analyze the model adjustment or judge to what extent model 3 is better than model 2, I report the Likelihood-Ratio Chi-Square Statistic (L²), the Pearson Chi-Square Statistic (X²), and the Bayesian Information Criterion (bic) for each model. The first two measures follow the Chi-Square distribution and in order to choose the better model it is necessary to consider the difference in degrees of freedom from the compared models and judge this difference against the Chi-Square distribution. When analyzing the bic, a more negative bic indicates a better model ( Raftery, 1995). A better model means that it is the most appropriate model given the data. In other words, it is the model that best describes the data.

Descriptive Analysis

The final sample represents 454,616 couples, in which both spouses were between 20 and 34 years of age in 1991, and 492,957 couples, who were also between this same age range, in 2000. Most of these couples were in racially endogamous unions: 75.43 percent in 1991 and 68.89 percent in 2000. White-white couples made-up the majority of these unions, representing 43.85 percent of all couples in 1991 and 39.97 percent of all couples in 2000. Among interracial couples, the white/brown union represented the majority (Table 1).

An important regional difference ³ emerges in terms of racial distribution. Table 2 shows that in the South a little more than 80 percent of the population was white in 1991 and in 2000, whereas in the North, where there are fewer whites, about 70 percent were brown in 1991 and 64.82 percent were brown in 2000. The region with a more balanced distribution between whites and nonwhites is the Central area of Brazil. There are a few more white wives than white husbands, which points to a gender asymmetry in terms of more white women marrying nonwhite spouses than white men.

Table 3 presents the analysis of the distribution of interracial couples among the mesoregions. Because there are more than one hundred regions, the table shows only the maximum, the minimum and the average proportion, so the mesoregions are identified in the text. In Vale do Itajai (Santa Catarina), only 4.30 percent of unions are interracial in 1991. On the other hand, in Central Potiguar (Rio Grande do Norte) 35.54 percent of the couples are mixed race in 1991. It is worth highlighting that Vale do Itajai is in the South, where most of the population is white, and Central Potiguar is in the Northeast region, where there is a concentration of blacks and browns in Brazil. In 2000, the lowest percentage of interracial unions was in Sul Catarinense (6.63 percent) and the highest was in Baixadas (40.88 percent),. which is in Southeast region, where there are also more browns and blacks than in the South.

The white/black couples range from zero (Norte do Amapá) to 4.93 percent (Centro Fulminenses) in 1991, the average proportion for this period was 1.7 percent; and it ranged from 1.35 percent (Sertão Sergipano) to 7.42 percent (Baixadas) in 2000, with an average of 7.42 percent. When comparing unions between whites and browns one can observe that in 1991 the minimum was 2.94 percent (Sul Catarinense) and the maximum was 32.74 percent (Pantanais Sul Mato Grossense). These proportions are much higher than for black/white couples. The percentage of black/black couples in 1991 ranged from zero (Norte do Amapá -North) to 8.08 percent (Centro-Fluminense ) and in 2000, the variation was from 0.51 percent (Centro Oriental Paranaense) to 11.33 percent (Metropolitana de Salvador). In contrast, for white/white couples in 1991, the minimum was 2.45 percent in Norte Amazonense and the maximum was 94.11 percent in Vale do Itajaí, located in the South of Brazil. In 2000, the minimum was 4.76 percent (Norte Amapá) and the maximum was 94.11 percent (also in Vale do Itajaí). The increase of white/black couples was much higher than any other mixed race couples.

The changes over time (1991–2000) were not the same for all groups. Considering the average proportion, same race couples decreased from 74.96 percent in 1991 to 68.83 percent in 2000. Brown/brown and white/white couples decreased to 14.81 percent and 3.33 percent respectively. However, the average proportion of black/black couples increased to 21.4 percent. The proportions of all mixed race couples increased-white/black (107.38 percent); followed by black/brown (33.22 percent); and white/brown (only 16.63 percent).

These descriptive results confirm the importance of considering geographical differences when analyzing interracial marriage in Brazil. In the following section, I discuss the results of the loglinear models. It is worth highlighting that where there is a concentration of whites, the interracial marriage rates are lower. Both, racial tolerance and population composition or exposure, can be the explanation, however it is clear that local racial distribution play an important role, a good example is how small is the white/black union in the North area, but is also the place with lower rates of black/black marriages.

Loglinear models

Table 4 shows the adjustment measures for the three models in order to analyze which one better adjusts to the data. The main conclusion is that the third model (that includes two interaction terms: one between a wife’s race and mesoregion, and the other between a husband’s race and mesoregion) has a better fit in both 1991 and 2000. This result indicates that a spouse’s race and the local marriage market are associated, which reassures the importance of considering the local marriage market racial distribution when analyzing the partner choice process. Considering the difference in the degrees of freedom (1088-544=544) and the critical chi-squared test (599.37 for alpha=0.05) the L² and the X² are statistically significant. This means that the third model is more effective than the second model. Moreover, the bic is negative for the third equation. As a result of the fact that the coefficients of model three are significant and statistically different from the second model, ⁴ the statement can be made that local racial characteristics influence the level of intermarriage. This information indicates that the assumption that there is only one national marriage market, which is made in equation 2, is not correct, but I will present the results of both models (2 and 3) for comparison.

In table 5, I present the coefficients (log odds ratios) for models 2 and 3 for both periods. The coefficients are the increase in the log of the predicted frequency, taking, for example, the result for black/black couples of model 2 in 2000, 3.393 it means that a black/black couple about 3.4 times more likely to happen than a white/white couple (reference). The odds ratio. representing the homogamy-heterogamy rates, at the bottom of the table for each year, indicate whether a union between two racial groups, black and brown for instance, is more likely to be homogamous, meaning between black and black or brown and brown, or heterogamous. I also included the difference in the odds ratio for models 2 and 3 in order to better understand how the inclusion of the interaction terms changes the interpretation of the homogamy-heterogamy rates.

By analyzing this table, I find that in 1991 unions involving whites and blacks are nearly nine times more likely to be homogamous than marriages among whites and browns (116.98 versus 13.105). When the local racial distribution is considered this difference is 11.5 times more likely (85.89 versus 7.45). In 2000, whites and blacks were about four times more likely to be homogamous and 4.6 times more likely when considering the spouse’s racial distribution. These results confirm that whites prefer to marry a brown spouse as opposed to a black spouse, although blacks and brown are socially and economically closer. Browns seem to be in an intermediary position within the marriage market. Moreover, endogamy is more pronounced at the end of the color spectrum, among blacks and among whites ( Berquó, 1987; Longo, 2011; Silva, 1987; Telles, 2004).

Changes over time are also important. Comparing both periods, it is clear that the differences are smaller in 2000, except for white/black couples that have a little higher difference. This result can be explained by more homogeneity of the population across regions or a decrease in the importance of local marriage market characteristics over time, for black/brown and white/brown couples. Despite the decrease over time, the rates between model 2 and model 3 are still very different, meaning that local racial distribution is still important for explaining intermarriage in Brazil. The results also confirm the descriptive analysis by showing that, between 1991 and 2000, the proportion of black/white couples increased much more than the proportion of brown/white couples. Moreover, this increase seems to be due to changes in cultural norms because the difference from equation 2 to equation 3 was not drastic between1991 and 2000, like in the case of white/brown couples. It may raise the hypothesis of influence of racial declaration across Census. There is actually an increase in the difference for black/white homogamy-heterogamy rates, meaning that the differences over time depend more on racial distribution.

The overall results point to the importance of local marriage markets but since homogamy is more common than heterogamy, especially when considering blacks, social distance also plays an important role in Brazil. The changes in 2000 show that local racial distribution has decreased importance while racial tolerance may have increased over time, or that the differences may also be an effect of racial classification changes, as analyzed by Miranda (2015). Furthermore, these results strongly suggest that intermarriages rates have been overestimated in past research because of the lack of consideration of racial composition in the local marriage market. However, it is worth noting that the adoption of a different concept of the marriage market would likely change the results, which does not mean that the substantive conclusions would be altered. In sum, the differences in homogamy-heterogamy rates between models 2 and 3 are all negative (see Graph 1), which specifies that after taking into account the local racial distribution of husbands and wives, the odds ratios of marrying homogamously compared to marrying heterogamously are lower.

The main importance of these results is the fact that racial distribution which is an indicator of opportunities and constraints in the marriage market reduces in 15.3 to 43 percent (depending on the year and spouses’ race). Therefore, although it does not fully explain the rates it is an important explicative factor to be considered. It is also worth noting that other local characteristics need to be added in order to better understand interracial marriage in Brazil, moreover the differences among the couples show that different factors, such as tolerance, influence differently the racial pairings, also the racial distribution is not random in the country being also affected by cultural and social factors. Changes over time demonstrate a decrease of importance of local racial distribution and higher racial tolerance and also may indicate differences in racial classification. These highlight the importance of the documentation of more recent data.

Local Marriage Market Differences

The main results discussed previously showed that overall the inclusion of the interaction terms between spouses’ race and mesoregions has a statistically significant impact on how to interpret interracial marriage rates in Brazil. In addition, it is important to investigate if including these interaction terms is significant for all mesoregions or only for a select group of them. In order to answer this question, I calculated the homogamy-heterogamy rates for each mesoregion and tested whether each result was statistically significant different from zero. Figure 1 shows the distribution of these rates. Values around zero indicate that there is no significant difference from model 2 and model 3 in a specific local marriage market. If this result occurs, the effect of the racial distribution is not important in that specific area. There are few mesoregions where the homogamy-heterogamy rates are not statistically different from zero. In 1991 these areas include: Norte do Amapá (North region) for all three types of couples; Litoral Sul Paulista (Southeast Region) and Sudeste Paranaense (South region) for brown/black couples; and Norte Amazonense (North region) for white/black couples; note that most of these mesoregions are in the North region, where there is a concentration of browns. For 2000, the areas where the homogamy-heterogamy rates are not statistically significant are: Norte Amazonense for black/brown couples; and Sul de Roraima for black/brown and white/black couples; and both mesoregions in the North region. The main message from this result is that racial distribution is not an important factor to explain interracial marriage in some areas, especially some mesoregions in the North of the country. Knowing that the rates are not random the explanation in these areas may be more a factor of local cultural and historical values. These cases are exceptions and not the rules. Therefore, the main message that can be derived from figure 1 is that, with very few exceptions, the consideration of the local distribution of spouses’ race is statistically significant for each mesoregion.

Figure 2 presents the same information of figure 1. the homogamy-heterogamy rates for each racial combination and each time period, in form of maps. Lighter colors indicate lower rates and darker colors specify higher values. It is important to remember that homogamy-heterogamy rates are odds ratios which indicate whether intraracial marriages are more or less likely than interracial unions. The main result is that clusters emerge from the data. The South region is a good example of this because in this area rates tend to be higher than in the rest of the country for all three racial combinations during both periods. The South region is where whites are concentrated. Another area that stands out is the North region, where there are also higher rates of concentration. Another interesting result from the map is the fact that brown/white couples are the ones with lower odds ratio in the whole country. The main message from figure 2 is that marriages between whites and browns are better accepted in Brazil, although in the South region seems to have some resistance to it, not explained by the racial composition of the population. Overall the ratios are lower between 1991 and 2000, confirming the previous results that there is an increase of interracial marriage over time.

In order to assess whether the differences among the mesoregions are statistically significant. I tested whether the rate for each mesoregion is different from the average level. This test is very important because if all of the areas have the same difference, the results would suggest that although the homogamy-heterogamy rates, when considering the local racial distribution, differ from zero, they are the same throughout the country. Rates that do not differ from each other, suggest that other possible explanations, besides local racial distribution, are the same everywhere. On the other hand, if there are statistically significant differences among the mesoregions, the explanations and preferences within the marriage market are also different between the local marriage markets, since the model considers the racial composition. This indicates that an investigation about what these differences are would be worth for future research. Figure 3 presents the maps with results for values statistically significant below the average, equal to the average, and above the average. The averages are specific for each type of couple for each year. If the value is below the average, this means that there are more mixed unions than the average. Values above the average level indicate that there are fewer interracial unions than the average.

The main results show that most areas have rates equal to the average. However, there are some important clusters, especially in the South region. These clusters refer to white/brown and white/black couples in 1991 and 2000, and black/brown couples in 1991. The values that are below the average are more dispersed throughout the country, with some evidence of clustering in the Central and Northeast regions for white/brown couples during both periods. Although some important differences arise from the analysis, it is worth noting that the values of the homogamy-heterogamy rates are close to each other. The lowest values are approximately two and the highest are about five. In some cases, such as for white/brown couples in 2000, the range is even smaller 1.35 to 3.5. Therefore, although the differences are statistically significant they may not necessarily be large.

The results point to important issues when analyzing intermarriage. First, the importance of considering the racial composition of local marriage markets, this is very consistent though the whole analysis of this article. It is worth highlighting that the differences in the homogamy-heterogamy rates are all lower 50%. in most cases the explanation is lower than 30%. meaning that most of the difference is not capture by population racial composition. The case with higher explanation is the white/brown couple in both periods, and the lowest explanation is among the brown/black unions Second, from figures 1 and 2 the South region stands out, as well as part of the North region. In these areas other variables may be even more important than population racial composition for explaining interracial unions’ rates, such as racial tolerance. Telles (1993) also found more differences in these regions, highlighting the lower black/white marriages and brown/black relations in the Northeast and frontier regions; part of the explanation is the industrialization level, although it does not explain the higher social distance between whites and blacks in the South region. And third, the fact that unions between whites and browns are more accepted in all parts of Brazil, which may indicate a common cultural behavior towards this kind of relationship, moreover, this is the main symbol of intermarriage in Brazil.