Ethnicity or race, area characteristics, and sexual partner choice among American adolescents.

A number of recent studies have documented the similarities and dissimilarities of the demographic and social characteristics of sexual partners among American adolescents and adults. (Ford, Sohn, & Lepkowsi, 2001; Harris & Ono, 2000; Laumann, Gagnon, Michael, & Michaels, 1994). The structure of sexual partnering is of importance because it is related to the likelihood of marriage and parenthood as well as the rate at which sexually transmitted infections including HIV can spread through a population (Koopman et al., 1988; Kretzchmeyer, Jager, Van Zessen, & Brouwers, 1995).

One of the most persistent empirical findings about social relationships is that people tend to initiate and maintain relationships with people of similar characteristics (Laumann et al., 1994). Several explanations have been proposed for this behavior, including geographic and social segregation, the reinforcement of a person's own identity and values, and the preference of social groups for persons of similar characteristics.

Studies of marital and cohabiting partners have generally confirmed this hypothesis (Schoen & Weinick, 1993). However, more recent studies of marriage and cohabitation among American adults have shown an increasing trend toward greater racial and ethnic intermarriage (Bersharov & Sullivan, 1996; Suro, 1999). The percent of marriages that include interracial couples had reached about 10% in the late 1990s (Bersharov & Sullivan, 1996; Suro, 1999). An analysis of 1990 U.S. census data showed that cohabitation among interracial couples ages 18 to 30 was more common than marriages among interracial couples (Harris &Ono, 2000). Black, White, Asian, and Latino men and women in this age group cohabited with people of different races more often than they married them.

Laumann et al. (1994) studied the characteristics of sexual partners of American adults in a nationally representative survey of U.S. adults conducted in 1992. This study found that while 93% of married sexual partners were of the same race, 88% of cohabiting partners and 89% of noncohabiting sexual partners were of the same race. For short-term partnerships, 91% were between persons of the same race. These data indicate that at the time of the study, longer term relationships were more likely to occur with partners of the same race.

Results from a national survey of adolescents conducted from 1994 to 1996 showed that while most partners were of the same race or ethnicity (78% overall), there was variation among adolescents according to ethnicity (Ford et al., 2001). While 87% of White adolescents and 85% of Black adolescents reported partners of the same ethnicity, 57% of Latino adolescents reported partners of different ethnicities. Differences of more than 2 years between partner ages were common in all groups. As adolescents became older, the characteristics of their partners became more heterogeneous in terms of age, school, and neighborhood. Furthermore, the less similar adolescents and their partners were, the less likely they were to use condoms or other contraceptives.

Community characteristics, including ethnic composition and region of the country, may affect people's choice of sexual partners. Recent studies of residential segregation have shown that although segregation has decreased in many metropolitan areas, many areas remain heavily segregated, especially in the East and Midwest (Glaeser & Vigdor, 2001). Harris and Ono (2000) have shown that local partner markets play a role in the selection of marriage and cohabitation partners. Propinquity should also be important for adolescents. Because adolescents may have less mobility than adults, the influence of local markets may also affect their choice of partners more strongly.

Apart from the supply of partners, communities may differ in their acceptance of interethnic or interracial dating. Research has shown evidence for community-level variation in racial climate as indicated by neighborhood preferences (Farley, Steeh, Krysan, Jackson, & Reeves, 1994; Massey & Denton, 1993) and racial attitudes (Schuman, Steeh, Bobo, & Krysan, 1997).

The aim of this research project was to examine the effects of school, neighborhood, and regional characteristics on sexual partner choice in a national sample of adolescents. First, we developed models of the effects of individual characteristics on partner choice. Following this, we examined the influence of community characteristics such as racial and ethnic composition and region of the country.

METHODS

Sample Design and Participants

Data for the study was drawn from the National Longitudinal Study of Adolescent Health (Add Health). The Add Health data was designed to assess the health status of adolescents and explore the causes of their health-related behaviors, focusing on the effects of the multiple contexts or environments (both social and physical) in which they live (Bearman, Jones, & Udry, 1997). The investigators selected a sample of secondary schools from a frame assembled by Quality Education Data, Inc. (QED). The population was stratified by region, urbanicity (urban, suburban, or rural), school type (public, private, or parochial), ethnic mix, and size. The investigators selected schools with probabilities proportional to size. More than 70% of the originally sampled high schools participated in this study. If a school refused to participate, they selected a replacement school within the same stratum. Once a high school was selected, the investigators recruited a feeder school (a school that includes 7th-grade students and sends its graduates to that high school). They selected a pair of schools in nearly all of the 80 communities where sampling occurred. Participating schools provided a roster of their students. In most cases, schools also agreed to administer an in-school questionnaire to each student during a class period. The in-school survey was completed by 90,118 students in 1994 to 1995. The investigators then selected a sample of students from school rosters to participate in inhome interviews. In 16 small schools, all students were selected for in-home interviews. In the other schools, the investigators grouped students into 12 grade and gender strata and randomly selected an average of 17 students from each within-school strata, yielding about 200 students for in-home interviews from each of 132 schools. Respondents selected through this two-stage sampling procedure constitute a nationally representative sample of adolescents. The investigators chose additional students to increase the number of students in several ethnic groups and the number of disabled students, and to create genetic samples of related individuals. Only students who were selected from the nationally representative sample were included in this study. This nationally representative sample provides data on 18,924 respondents interviewed in their homes in Wave 1. In a second wave conducted 2 years later, 13,570 Wave 1 respondents were interviewed in their homes. We included data from both the Wave 1 and the Wave 2 interviews in this study.

This study focused on the sexually active portion of the sample. During the in-home interview, information on sexual activity was obtained through two different sequences of questions. In Wave 1, 7,508 of the 18,924 respondents reported ever having sexual intercourse. In Wave II, 5,989 respondents reported sexual intercourse. Merging the two waves of data, 9,303 distinct respondents reported ever having sexual intercourse.

The unit of analysis in this paper is sexual relationships rather than individuals. We had to convert the respondent-level data file from one with a record for each individual to one with a record for each sexual partner. The 9,303 individuals who reported sexual activity in the direct question reported having 26,657 total partners.

In each wave, respondents were asked to report on relationships the last 18 months. Noting that telescoping might have occurred such that respondents may have reported in Wave 2 on relationships that began more than 18 months ago, we developed a procedure for comparing relationships reported by the same respondent in Waves 1 and 2 that examined agreement on five variables. This procedure identified 724 relationships (2.7% of sample relationships) that were likely duplicates; these relationship records were deleted from the relationship data set. Finally, we selected for analysis relationships in which the respondents reported having sexual intercourse.

Examination of the data on partnerships also revealed that although some individuals had reported never having sexual intercourse in the direct question, they later reported intercourse when asked about specific partners. From these cases, 363 additional relationships were added to the sexual relationship file. Only 20 of the reported relationships were homosexual; these were deleted from the file. In total, we analyzed 17,266 heterosexual relationships. These relationships were reported by 8,024 individuals. For operational reasons, the investigators limited reports to a maximum of three romantic and three nonromantic partners. Only a small percentage of students reported the maximum number of romantic or nonromantic partners, indicating that not many relationships were unreported.

The probability of selection of a relationship was determined by the probability of selection of the reporting student. Thus, we assigned each sexual relationship the weight of its reporting student. Relationships had the potential to be reported by either partner; however, both partners could report not all relationships. A relationship involving an individual in school and another not currently enrolled in school could only have been reported by the individual in school. That is, relationships with both partners in school had a higher probability of selection than those involving one partner in school and the other not in school. In the 16 schools where all students were interviewed, relationships reported with partners in the same school cannot be linked in the data released for use outside of the University of North Carolina.

A standard method for handling varying probabilities of selection is to weight sample reports to decrease the contribution of those who have higher probabilities of selection (Cochran, 1976; Kish, 1965). In the present investigation, the weight of the reporting student was assigned to each of his or her reported relationships. The proportion of relationships that were with partners currently in school (and thus relationships that could have been reported twice) was 0.82. Thus, 82% of relationships could have been reported more than once. We therefore adjusted the relationship weight by a factor of 1/(2 x 0.82) = 0.7098 to reduce the weight for all relationships involving two partners in school.

Procedures

Respondents were interviewed in their homes by an interviewer using a personal computer. Respondents used the computer to self-administer sensitive parts of the questionnaire, including questions on sexual behavior.

Measures

Type of partner. During the interview, respondents were asked about (a) romantic and (b) nonromantic partners. In Wave 1, these partners were designated as being from relationships that began after January 1994; in Wave 2, they were designated as being from relationships that began in the last 18 months. Romantic partners were identified by the question "Have you had a romantic relationship with anyone?" Respondents could also identify a romantic partner by reporting that they had held hands, kissed on the mouth, and/or told the partner that they loved him or her. Initially, each respondent identified up to three romantic partners and answered detailed questions about the characteristics of each partner, the timing of each relationship, and activities within each relationship. Respondents were then asked about persons with whom they had had sexual intercourse but whom they did not consider romantic partners. They were asked to identify up to three such partners. If they had held hands, kissed on the mouth, and/or told any of these partners that they loved them, they were considered romantic partners. Otherwise, partners were considered nonromantic. Respondents identified up to three nonromantic partners and answered detailed questions about the characteristics of each partner, the timing of each relationship, and activities within each relationship.

Age. We coded data on age in single years for both respondent and sexual partner. Differences between ages of the partners were coded as (a) more than two years older, (b) within 2 years, and (c) more than 2 years younger.

Race or ethnicity. We used the data on self-report of the respondent and partner race or ethnicity and coded the data as White, Black, Latino, or other. Data on multiple ethnicities or races were not included because only 2.7% of respondents reported more than one ethnicity or race in the home interview.

Grade. We coded the grade in school according to the respondent's or the partner's current grade. Differences in grade were coded as (a) in a lower grade, (b) in the same grade, (c) in a higher grade, or (d) respondent or partner not in school.

School. We coded school as the partner attending the same school as the respondent attended or not.

Neighborhood. Respondents were asked if their partners lived in the same neighborhood as they did. The word neighborhood was not defined more specifically for respondents.

We also used contextual factors.

Region. The U.S. Census Bureau's region of the country was coded as West, Midwest, South, and Northeast.

Proportion Black. This variable measured the proportion of persons living in the same census tract as the respondent who reported their race as Black in the 1990 census.

Proportion Hispanic. This variable measured the proportion of persons living in the same census tract as the respondent who reported their ethnicity as Hispanic in the 1990 census.

RESULTS

Statistical Analyses

The relationships or dyads in this study may be viewed as equal-sized but overlapping clusters of individuals. Given the weighting employed to compensate for duplicate reports, the relationships could be treated as sample elements within a stratified multistage sample design. We conducted appropriate variance estimation through SUDAAN (a computer program developed for studies with clustered samples; Shah, Bamwell, & Bieler, 1997) for means, proportions, regression coefficients, and test statistics.

The variance estimation methods assume that the samples of relationships are independent between schools. This is, strictly speaking, not correct in the Add Health sample design. A relationship could have been selected from two different schools when the partners were enrolled in different schools. The survey estimation literature and available variance estimation software do not provide methods for variance estimation in such a sample of networks drawn from a cluster sample. Employing existing variance estimation methods such as in SUDAAN to the present problem results in biased variance estimates. The lack of independence among sample clusters (in this case, schools) can be expected to increase the variability of sample estimates over the case in which clusters are selected independently. Thus, the variance estimates presented in this paper understate the true sampling variance. While 58% of the relationships reported in the Add Health survey involved individuals who were both in the same school or an individual in school and another not in school, the variance estimates and test statistics based on them in the present paper are not expected to severely understate true values.

In examining the associations between partner characteristics and demographic variables, we used Generalized Estimating Equation (GEE) models to account for the correlation among relationships reported by the same person. The GEE model (Diggle, Liang, & Zeger, 1994; Liang & Zeger, 1996) is a repeated-measures analysis of correlated outcomes and predictors. It can be applied to multiple logistic types of analyses in which the dichotomous or polytomous outcomes are predicted. This procedure is appropriate here because respondents reported on experiences with several partners. Each respondent could appear in the analysis more than once. We used the SUDAAN statistical package to generate GEE models using sampling clusters to group observations (Shah et al., 1997). Results are reported as important if the p values of test statistics were smaller than 0.05.

In addition, we calculated predicted proportions of respondents choosing a partner with the same characteristic by gender and ethnicity or race. We computed these probabilities by summing the coefficients for the relevant gender and ethnic groups. Coefficients for age groups and type of partner were weighted by the sample distribution of these variables. In the models including contextual variables, we also added coefficients for the appropriate region. Proportions were estimated for the variables Black ethnicity greater than 10% and Latino ethnicity greater than 20%. We then converted the sum of the coefficients from a logit to a proportion (Hosmer & Lameshow, 2000).

Individual Characteristics

Table 1 shows the demographic characteristics of respondents in the study who reported sexual relationships. The approximately 8,000 persons consisted of 11% age 14 or younger, 36% age 15 to 16, and 54% age 17 or older. Over half of the sample were White (62%), 19.2% were Black, 11.7% were Latino, and 7% were of other ethnicity.

Table 2 shows the distribution of partner characteristics. About 78% of partners were of the same ethnicity. While just over half of all partners were within 2 years of age of each other, 12.6% were 2 or more years younger and 32.1% were 2 or more years older. Only about half of all partners (52.6%) were in the same grade, and 18.3% were not attending school. More respondents reported partners in higher grades (18.3%) than in lower grades (7.1%). The majority of partners were from different neighborhoods (70%) and different schools (60%). Most partners were categorized as romantic (80.6%).

The relationships between respondent characteristics and five measures of similarity in partner characteristics (ethnicity or race, age, grade level, neighborhood, and school) were examined through a sequence of linear logistic models estimated under a GEE approach. We used respondent demographic variables as predictors, and after testing various two-way interactions, we included an interaction between gender and ethnicity or race in the models. Results are presented in Tables 3 and 4. Table 3 shows the logistic regression coefficients, their standard errors, and the significance level of a two-tailed t test. Table 4 shows the estimated proportion selecting a partner with the same characteristics.

For the model predicting ethnicity or race of partner (Table 3, Model 1), age, gender, ethnicity or race, and type of partner were significantly related to ethnicity or race of partner. The proportion selecting a partner of the same ethnicity or race was fairly close for males and females (.88 and .85, respectively); the proportion with a partner of the same ethnicity or race was .80 for Black males but .90 for Black females (see Table 4, Model 1). The proportions for Latino respondents were lower (.55 for Latino males and .58 for Latino females).

For the model predicting whether or not the partner's age was within 2 years of the respondent's age (Table 3, Model 2), gender and type of relationship were significant. Adolescents in romantic relationships were more likely to choose partners in the same age group (p < .01) and female respondents were more likely to choose partners outside of their age group (p < .01). Ethnicity was not significantly related to the age of the partner. For White respondents the proportions choosing a partner within the same age group were .65 for males and .44 for females (see Table 4, Model 2). For Black respondents, the proportions were .65 for males and .43 for females. For Latinos, the proportions were .62 for males and .40 for females.

Due to questionnaire design, the analysis for the grade variable included only romantic partners (see Table 3, Model 3). Compared to the youngest adolescents, the older adolescents were less likely to choose partners in the same grade (p < .01). In addition, males were more likely to choose partners in the same grade than were females (p < .01).

In the analysis of partner neighborhood (see Table 3, Model 4), only age had a significant effect. The older adolescents were less likely than the younger adolescents to choose partners in the same neighborhood (p = 0.02). The predicted proportions for race or ethnic and gender groups ranged from .29 to .34 (see Table 4, Model 4).

Gender, race or ethnicity, and type of relationship were significantly related to choosing partners from the same school (see Table 3, Model 5). Female respondents were less likely to have partners in the same school (p < .01) than were males, and romantic partners were more likely than nonromantic partners to be from the same school (p < .01). The proportions choosing a partner from the same school were .33 for White males, .26 for White females, .31 for Black males, .24 for Black females, .30 for Latino males, and .23 for Latino females (see Table 4, Model 5).

Community Characteristics

Because partner choice may be affected by the characteristics of the community, we ran a series of models that included indicators of the proportion of the population that was Black or Hispanic, the region, the poverty level of the community, and the school size (number of students) and type (public or private). In general, these variables did not have significant relationships with the dependent variables. The one exception was the model for race or ethnicity of partner, in which there were significant effects of community ethnic composition and region (see Table 5). The chi-square statistic for the difference in -2 log likelihood between the model with and the model without the contextual variables was 689.94 (p < .001). Tests of two-way interactions indicated significant interaction effects between gender and ethnicity and between region and ethnicity in this model. Differences between models with and without contextual variables were not significant for the dependent variables related to age, grade, neighborhood, and school.

Table 6 shows the estimated proportion of respondents choosing a partner of the same race or ethnicity by gender and race or ethnicity. The proportions varied by ethnicity or race, gender, and region. For White males, the estimated proportion of respondents choosing a White partner was .97 in all regions except the West, where it was .92. A slightly lower proportion of White females chose a White partner: .95 in the Northeast and West and .96 in the Midwest and South. Black males had a lower probability of choosing a Black female partner in the Northeast (0.83) than in the Midwest and South (.97 for each) and the West (.90). Latino males were least likely to choose a Latino female in the Midwest (.66) compared to the Northeast (.70), the South (.83), or the West (.90). Latino females also had the lowest probability of choosing a Latino partner in the Midwest (.69), compared to the Northeast (.73), the West (.82), and the South (.85).

DISCUSSION

This project used a national sample of adolescents to examine the association of individual and community characteristics with the demographic characteristics of sexual partners of American youth. The data indicate that the community characteristics of ethnic composition of the population and region were most strongly related to the ethnicity or race of the partner. Differences in partner's age, grade, neighborhood, and school or residence were not related to school and community characteristics. In addition, the individual characteristics of age, gender, and race or ethnicity were also related to ethnicity or race of partner.

This study has several limitations. The study relies on self-report of data on a sensitive topic. Some adolescents may be unwilling to report sexual activity in an interview and may not fully report their experience with multiple sexual partners. Additionally, reports of partner characteristics rely on the report of the respondent. In short-term relationships, individuals may not be fully aware of their partner's exact age or other characteristics.

In addition, due to concerns about confidentiality, more specific information on geographic information beyond region of the country was not available. Within each of these regions there is considerable heterogeneity in racial and ethnic composition and residential segregation.

Findings from the 1990 census data of Harris and Ono (2000) showed that among White and Black married women ages 18 to 30, 96% were married to partners of the same race, while among White and Black cohabiting couples, 93% of women were living with a partner of the same race. Among married Asian women, 68% were married to partners of the same race or ethnicity, while among cohabiting Asian women, only 42% were living with Asian men. Finally, among married Latino women, 80% were married to Hispanic men, while among cohabiting Latino women, 73% were living with Latino men (Harris &Ono, 2000). The findings from our study echo these results to some extent. The more committed relationships (the romantic ones) were more likely to be with partners of the same ethnicity or race. These data reflect the time periods of data collection and may not capture more recent trends.

Regional differences in the ethnicity or race of partners were significant even after we controlled for the ethnic composition of the community. This may reflect regional differences in the acceptability of friendships between adolescents of differing ethnic and racial groups. While the Midwest and the Northeast are the most residentially segregated areas, these were not uniformly the two areas where adolescents were least likely to choose partners of different ethnicity or race. These differences may reflect variations in interactions between ethnic and racial groups that extend beyond residential segregation.

In summary, while most adolescents choose partners with similar characteristics, this choice is related to the adolescent's age, gender, and ethnicity or race, and to the type of relationship as well as the ethnic or racial composition of the community and the geographic region. Further research should examine possible reasons for differences by region, including school and neighborhood segregation and migration and commuting patterns.

Table 1. Demographic Characteristics of Respondents
(N = 8,024)

                       N      %     se (%)

Age
  14 years or less    683   10.84   1.64
  15-16 years        2761   35.68   0.88
  17 years or more   4579   54.47   2.05

Ethnicity or race
  White              3895   61.96   3.29
  Black              1921   19.25   2.76
  Latino             1421   11.67   1.78
  Other               782    7.12   0.74

Gender
  Male               3973   50.46   0.8
  Female             4051   49.54   0.8

Note. Table includes unweighted N, weighted percent, and standard
error adjusted for clustering.

Table 2. Percent of Partners With Specific Demographic
Charateristics (N = 17,266)

Characteristics        %

Race or ethnicity
  Same               78.3
  Different          21.7

Age
  2+ years younger   12.6
  Within 2 years     55.3
  2+ years older     32.1

Grade
  2+ years lower      7.1
  Within 2 years     52.6
  2+ years highter   22.1
  Not in school      18.3

Neighborhood
  Same               30.0
  Different          70.0

School
  Same               39.9
  Different          60.1

Relationship
  Romantic           80.6
  Nonromantic        19.4

Table 3. Logistic Generalized Estimating Equation (GEE) Model
for Predictors of Partner Characteristics

                                Model 1             Model 2
                           Partner has same       Partner age
Respondent                ethnicity or race    within 2 years of
  characteristics           as respondent         respondent

                            b       se(p)        b       se(p)

Intercept                  1.75   0.16(.00)     0.36   0.11(.00)

Age
    [less than or equal
      to] 14 years         0.00                 0.00
    15-16 years            0.11   0.14(.42)    -0.04   0.10(.71)
    17+ years              0.29   0.14(.04)    -0.08   0.10(.39)

Ethnicity or race
    White                  0.00                 0.00
    Black                 -0.67   0.19(.00)    -0.04   0.10(.67)
    Latino                -1.85   0.20(.00)    -0.22   0.14(.11)
    Other                 -3.27   0.28(.00)    -0.29   0.14(.04)

Gender
    Male                   0.00                 0.00
    Female                 0.21   0.07(.00)    -0.78   0.07(0.00)

Relationship
    Romantic               0.21   0.07(.00)     0.43   0.05(.00)
    Nonromantic            0.00                 0.00

Ethnicity or race X
  Gender                             (0.00)               (0.08)
    White male             0.00                 0.00
    White female           0.00                 0.00
    Black male             0.00                 0.00
    Black female           1.21   0.23(.00)    -0.01   0.15(.96)
    Latino male            0.00                 0.00
    Latino female          0.54   0.21(0.01)    0.20   0.16(.23)
    Other male             0.00                 0.00
    Other female           0.36   0.29(0.23)    0.50   0.02(.02)

Chi-square                     2887.06               580.06
df                               10                   10
p                                 0.00                 0.00
[R.sup.2]                         0.154                0.036
N                             17240                15839

                              Model 3             Model 4
                          Partner in same   Partner lives in same
Respondent                   grade as         neighborhood as
  characteristics           respondent          respondent

                            b       se(p)       b       se(p)

Intercept                  0.88    01(.00)     -0.82   0.13(.00)

Age
    [less than or equal
      to] 14 years         0.00                 0.00
    15-16 years           -0.23    0.09(.01)   -0.10   0.10(.34)
    17+ years             -0.44    0.10(.01)   -0.27   0.11(.02)

Ethnicity or race
    White                  0.00                 0.00
    Black                  0.00    0.12(.99)    0.08   0.11(.47)
    Latino                -0.13    0.14(.35)    0.25   0.14(.08)
    Other                  0.16    0.19(.40)    0.30   0.17(.08)

Gender
    Male                   0.00                 0.00
    Female                -0.84    0.10(.00)    0.10   0.08(.21)

Relationship
    Romantic              --      --            0.07   0.06(.20)
    Nonromantic           --      --            0.00

Ethnicity or race X
  Gender                              (0.87)              (0.51)
    White male             0.00                 0.00
    White female           0.00                 0.00
    Black male             0.00                 0.00
    Black female          -0.05    0.17(.78)   -0.04   0.12(.73)
    Latino male            0.00                 0.00
    Latino female         -0.05    0.20(.82)   -0.12   0.15(.44)
    Other male             0.00                 0.00
    Other female           0.10    0.23(.66)   -0.31   -0.31(.13)

Chi-square                      641.60                 67.27
df                                9                    10
p                                 0.00                  0.00
[R.sup.2]                         0.048                 0.004
N                             13180                 17238

                              Model 5
                          Partner goes to
Respondent                same school as
  characteristics           respondent

                            b       se(p)

Intercept                 -0.40   0.13(.00)

Age
    [less than or equal
      to] 14 years         0.00
    15-16 years           -0.06   0.10(.55)
    17+ years             -0.17   0.11(.13)

Ethnicity or race
    White                  0.00
    Black                 -0.04   0.12(.74)
    Latino                -0.24   0.11(.04)
    Other                 -0.28   0.13(.03)

Gender
    Male                   0.00
    Female                -0.35   0.08(.00)

Relationship
    Romantic               0.43   0.06(.00)
    Nonromantic           00

Ethnicity or race X
  Gender                             (.07)
    White male             0.00
    White female           0.00
    Black male             0.00
    Black female          -0.22   .13(.10)
    Latino male            0.00
    Latino female         -0.19   .17(.25)
    Other male             0.00
    Other female           0.31   .21(.14)

Chi-square                     302.113
df                              10
p                                0.00
[R.sup.2]                        0.017
N                            17197

Table 4. Estimated Proportion of Adolescents Choosing a
Partner of the Same Race or Ethnicity, Age Group, Grade,
Neighborhood, and School

 Gender and     Model 1    Model 2   Model 3   Model 4
  race or        Same       Same      Same      Same     Model 5
ethnicity of    race or      age      grade    neigh-     Same
 respondent    ethnicity    group              borhood   school

White male       0.88       0.65      0.65      0.29      0.33
White female     0.85       0.47      0.44      0.31      0.26
Black male       0.80       0.65      0.65      0.30      0.31
Black female     0.90       0.46      0.43      0.32      0.24
Latino male      0.55       0.61      0.62      0.34      0.30

Table 5. Logistic Generalized Estimating Equation (GEE)
Model for Individual and Contextual Predictors of
Partner Ethnicity

                                      b       se(p)

Intercept                            3.24   0.42(.00)

                       Individual factors

Age
  [less than equal to 14 years       0.00
  15-16 years                        0.07    0.14(.58)
  17+ years                          0.23    0.13(.08)
Ethnicity or race
  White                              0.00
  Black                             -1.79    0.31(.00)
  Latino                            -2.57    0.42(.00)
  Other                             -3.97    0.54(.00)
Gender
  Male                               0.00
  Female                            -0.46    0.10(.00)
Relationship
  Romantic                           0.21    0.07(.00)
  Nonromantic                        0.00
Ethnicity or race X Gender
  White male                         0.00
  White female                       0.00
  Black male                         0.00
  Black female                       1.29    0.23(.00)
  Latino male                        0.00
  Latino female                      0.60    0.19(.00)
  Other male                         0.00
  Other female                       0.34    0.28(.22)

                       Contextual factors

Region
  West                              -0.90    0.30(.00)
  Midwest                            0.23    0.22(.31)
  South                              0.25    0.24(.29)
  Northeast                          0.00
Proportion Black
  < 0.1                             -0.68    0.25(.01)
  0.1+                               0.00
Proportion Hispanic
  < 0.2                             -0.87    0.18(.00)
  0.2+                               0.00
Ethnicity or race X Region
  White-West                         0.00
  White-Midwest                      0.00
  White-South                        0.00
  White-Northeast                    0.00
  Black-West                         0.68    0.53(.20)
  Black-Midwest                      0.77     .38(.05)
  Black-South                        0.87    0.39(.03)
  Black-Northeast                    0.00
  Latino-West                        1.45    0.48(.03)
  Latino-Midwest                    -0.42    0.47(.38)
  Latino-South                       0.53    0.47(.27)
  Latino-Northeast                   0.00
  Other-West                         2.43    0.65(.01)
  Other-Midwest                     -0.08    0.63(.89)
  Other-South                       -0.67    0.58(.25)
  Other-Northeast                    0.00
Intercept                            3.24    0.42(.00)
Chi-square                        3576.0
df                                  24
p                                    0.00
[R.sup.2]                            0.189
N                                17082

Table 6. Estimated Proportion of Adolescents Choosing a
Partner of the Same Race or Ethnicity

Gender or ethnicity
of respondent         Northeast   Midwest   South   West

White male              0.97       0.97     0.97    0.92
White female            0.95       0.96     0.96    0.95
Black male              0.83       0.93     0.94    0.90
Black female            0.92       0.97     0.97    0.95
Latino male             0.70       0.66     0.83    0.90
Latino female           0.73       0.69     0.85    0.82

REFERENCES

Bearman, J., Jones, J., & Udry, R. (1997). The national longitudinal study of adolescent health. Chapel Hill, NC: Carolina Population Center.

Bersharov, D. J., & Sullivan, T. S. (1996, July/August). One flesh. The New Democrat, pp. 19-21.

Cochran, C. (1976). Sampling techniques (3rd ed.). New York: Wiley.

Diggle, P., Liang, K.-Y., & Zeger, S. L. (1994). Analysis of longitudinal data. Oxford, OH: Clarendon Press.

Farley, R., Steeh, C., Krysan, M., Jackson, J., & Reeves, K. (1994). Stereotypes and segregation: Neighborhoods in the Detroit area. American Journal of Sociology, 100, 750-780.

Ford, K., Sohn, W., & Lepkowski, J. (2001). Characteristics of adolescents' sexual partners and their association with use of condoms and other contraceptive methods. Family Planning Perspectives, 33, 100-105, 132.

Glaeser, E. L., & Vigdor, J. L. (2001, April). Racial segregation in the 2000 census: Promising news. Washington, DC: The Brookings Institution, Center on Urban and Metropolitan Policy.

Harris, D. R., & Ono, H. (2000, April). Estimating the extent of intimate contact between the races: The role of metropolitan area factors and union type in mate selection. Paper presented at the meeting of the Population Association of America, Los Angeles, CA.

Hosmer, D. W., & Lameshow, S. (2000). Applied logistic regression (2nd ed.). New York: Wiley.

Kish, L. (1965). Survey sampling. New York: Wiley.

Koopman, J., Simon, C., Jacquez, J. A., Joseph, J., Sattenspiel, L., & Park, T. (1988). Sex partner selectiveness rates on homosexual HIV transmission rates. Journal of Acquired Immune Deficiency Syndrome, 1, 486-504.

Kretzschmer, M., Jager, J. C., Van Zessen, G., & Brouwers, H. (1995). A network modeling approach for assessing the efficiency of prevention strategies. Journal of Mathemetical Sociology, 20, 351-374.

Laumann, E. O., Gagnon, J. H., Michael, R. T., & Michaels, S. (1994). The social organization of sexuality. Chicago: University of Chicago Press.

Liang, K.-Y., & Zeger, S. (1986). Longitudinal analysis using general linear models. Biometrika, 73, 13-22.

Massey, D., & Denton, N. (1993). American apartheid: Segregation and the making of the underclass. Cambridge, MA: Harvard University Press.

Schoen, R., & Weinick, R. M. (1993). Partner choices in marriages and cohabitations. Journal of Marriage and the Family, 55, 408-414.

Schuman, H., Steeh, C., Bobo, L., & Krysan, M. (1997). Racial attitudes in America: Trends and interpretations (Rev. ed.). Cambridge, MA: Harvard University Press.

Shah, B. V., Bamwell, B. G., & Bider, G. S. (1997). SUDAAN user's manual, release 7.5. Research Triangle Park, NC: Research Triangle Institute.

Suro, R. (1999). Mixed doubles. American Demographics, 21 (11), 56-62.

Manuscript accepted January 20, 2003

The analysis of the data in this manuscript was funded by grant #HD36971 from the National Institute of Child Health and Human Development. The data was taken from the Add Health project, a program project designed by J. Richard Udry (PI) and Peter Bearman, and funded by grant #P01-HD31921 from the National Institute of Child Health and Human Development to the Carolina Population Center, University of North Carolina at Chapel Hill, with cooperative funding participation by the National Cancer Institute; the National Institute of Alcohol Use and Alcoholism; the National Institute on Deafness and Other Communication Disorders; the National Institute on Drug Abuse; the National Institute of General Medical Sciences; the National Institute of Mental Health; the National Institute of Nursing Research; the Office of AIDS Research, NIH; the Office of Behavior and Social Science Research, NIH; the Office of the Director, NIH; the Office of Research on Women's Health, NIH; the Office of Population Affairs, DHHS; the National Center for Health Statistics, Centers for Disease Control and Prevention, DHHS; the Office of Minority Health, Centers for Disease Control and Prevention, DHHS; the Office of Minority Health, Office of Public Health and Science, DHHS; the Office of the Assistant Secretary for Planning and Evaluation, DHHS; and the National Science Foundation. Persons interested in obtaining data files from the National Longitudinal Study of Adolescent Health should contact the Carolina Population Center, 123 West Franklin Street, Chapel Hill, NC 27516-3997.

Address correspondence to Kathleen Ford, Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, MI 48109-2029; e-mail: [e-mail].

Kathleen Ford, Woosung Sohn, and James M. Lepkowski

University of Michigan