Abstract
CONTEXT:
Heterosexual adolescents are less likely to use contraceptives when their partners are of different backgrounds or social circles. It is unclear why risky behaviors are more common within these relationships, and characteristics associated with school-discordant partners—that is, partners who do not attend the adolescent’s school—are particularly understudied.
METHODS:
Characteristics of romantic relationships and contraceptive use were examined among 1,043 female and 961 male respondents in Waves 1 (1994–1995) and 2 (1996) of the National Longitudinal Study of Adolescent to Adult Health. Generalized linear regression and structural equation models identified individual- and relationship-level characteristics that were associated with school-discordant relationships and contraceptive use.
RESULTS:
For females, the likelihood of ever having used a condom with a partner was reduced when that partner had not attended her school at the start of the relationship (odds ratio, 0.5) or when he was two or more years older (0.6). Notably, however, having a school- or age-discordant partner was not associated with ever having used hormonal or dual methods for females, or with any contraceptive measures for males. For females, lower emotional intimacy with school-discordant partners was positively associated with ever having used condoms with such partners (1.2).
CONCLUSIONS:
Characteristics of school-discordant relationships are correlated with condom use in varying and sometimes countervailing ways. Future research examining the mechanisms linking school discordance to decreased condom use, particularly for females, would deepen understanding of the emotional and psychosocial processes involved in adolescents’ sexual and contraceptive behavior.
Adolescence is a uniquely risky period in the life course for sexual and reproductive health. U.S. adolescents have high rates of unwanted pregnancy and STDs, in large part because of low and inconsistent use of contraceptives.1,2 An important feature of adolescents’ inconsistent use is the substantial variation in contraceptive behavior from one relationship to the next.3,4 This variation highlights that sexual and contraceptive choices are made within the context of relationships. A common finding in the growing literature on relationship-level influences is that heterosexual adolescents are less likely to use condoms or other contraceptives when they are in discordant relationships, that is, when their partners come from different backgrounds or social circles. Further, adolescents in relationships with partners who do not attend their school, compared with those whose partners are in the same school, have a reduced likelihood of using contraceptives4 and an increased likelihood of having had a self-reported STD.5
Previous research has shown that decisions about whether to have sex and use contraceptives are shaped by the characteristics of one’s partner and the quality of the relationship with that partner.3–9 However, little research has examined the extent to which individual-level behaviors and attitudes that are associated with less safe sex may also be correlated with the formation of relationships with school-discordant partners, which could partially account for the apparent riskiness of these relationships. Limited attention to how individual-level risk factors are associated with discordant relationships also leaves unanswered questions about which adolescents are at the highest risk for relationship discordance. Another important issue that has received little attention is which relationship characteristics (e.g., intimacy, conflict) may play intermediary roles in shaping contraceptive use. By using generalized structural equation models and longitudinal data from the first two waves of the National Longitudinal Study of Adolescent to Adult Health (Add Health), this study sheds light on these interconnections for school-discordant relationships.
Background
During adolescence, many young people attempt to become more independent from their families and explore alternative roles and identities.10 Dating people of different backgrounds and social groups may be an important part of a normal developmental process; indeed, romantic relationships that cross social or demographic boundaries appear to be more common in adolescence than in other periods of the life course.4,11,12 However, discordant relationships, particularly those crossing these boundaries, still warrant critical examination, as there is substantial evidence that they are comparatively riskier for sexual and reproductive health. Several studies have found that dating older partners is associated with lower contraceptive use; this association may be driven mostly by adolescent women dating men four or more years older.4,13–15 Evidence regarding race composition and contraceptive use is more mixed; some studies have shown that different-race couples use condoms and contraceptives less consistently than others, while several studies have found that these couples use condoms and other contraceptives more consistently or no differently than same-race couples.4,16,17 Notably, a handful of studies have shown that when adolescents date partners outside their immediate social environment—for example, outside their neighborhood or school—they engage in riskier sexual behaviors.4,5,13 We focused our analysis on dating relationships that cross school boundaries, since this form of discordance is very common, characterizing more than half of adolescent relationships reported in nationally representative data.4
Previous studies on school discordance and adolescent sexual behavior relied primarily on adolescents’ reports of their dating relationships collected in the first two waves of Add Heath. Using relationships reported by female adolescents in this survey, Koon-Magnin and colleagues13 found that dating a school-discordant partner was associated with having had sexual intercourse with the partner. In analyzing Add Health relationships reported by both male and female adolescents in which intercourse had occurred, Ford et al.4 showed that the likelihood of using contraceptives was reduced by approximately 30% if the partner was school-discordant. Ford and Lepkowski5 further found that the likelihood of reporting ever having had an STD was 37% higher if a respondent had dated a school-discordant partner. Regression analyses in these two studies included controls for individual- and dyadic-level demographic variables (e.g., respondent age and race and partnership discordance on age and race) and were stratified by whether the relationship was romantic or casual. Our analysis complements and extends this prior work, particularly that of Ford and colleagues,4 in three ways: First, we stratify all estimates by the respondent’s gender; second, we consider how prior individual-level attitude and behavioral risk factors are jointly associated with school discordance and contraceptive use; and third, we investigate which school-discordant relationship characteristics may be correlated with lower contraceptive use.
We hypothesize that correlations between partner discordance and reduced contraceptive use reflect two processes. First, selecting partners may be an important way that adolescents express their developing attitudes and identities, and adolescents who are more willing to take sexual risks, regardless of partners’ influences, may be more likely to select discordant partners. Second, given adolescents’ young ages and relative inexperience with intimate relationships, the dynamics of relationships and contraceptive behavior may be particularly susceptible to partners’ influences.
According to the health belief model and related models, such as the information-motivation-behavioral skills model, contraceptive use will be more likely if individuals perceive that getting an STD or having an unintended pregnancy is a likely outcome, if they perceive these outcomes as being very negative, and if they have adequate knowledge and confidence (i.e., self-efficacy) in using contraceptives.18,19 An important question is whether these individual-level risk factors also predict discordant relationship formation. For instance, perceiving lower susceptibility to an STD or a pregnancy may be associated with overconfidence or tolerance of unfamiliarity and risk, which should increase the likelihood of meeting and forming romantic relationships outside of one’s school. Alternatively, being more knowledgeable about and skilled with contraceptives may reflect “worldliness” or maturity; this could also make one more likely to venture outside of one’s school and typical social network when forming romantic relationships.
Closeness and conflict are two key relationship dimensions that may shape contraceptive behavior.7–9,20 While studies have often accounted for whether partners differ on demographic variables, they have not explored how these relationship characteristics vary with partner discordance and whether they help explain lower contraceptive use with discordant partners. Since discordance typically implies lower similarity in contexts, experiences and frames of reference,21 we expect that school-discordant relationships will, on average, have lower intimacy—both emotional and social—and greater potential for conflict. Whereas prior literature has most often focused on emotional closeness between partners, we extend the notion of closeness to include social closeness or embeddedness—for instance, shared friends or familiarity with a partner’s family—which is particularly relevant when examining discordance across a social context like school.22
Theory and prior evidence suggest that relationship closeness has the potential to both increase and decrease contraceptive use. On one hand, the contraceptive behavior communication model suggests that closer partners are more likely to use contraceptives because closeness makes it easier to communicate about and plan sexual behavior.8 Some research has shown that adolescents and young adults are more likely to use contraceptives when they have known their partner for a longer time prior to dating and have shared more intimate activities prior to first sex.3,17,20 On the other hand, a contrasting body of evidence suggests that, as relationships become more serious and committed, contraceptive use becomes less consistent because an individual is more willing to periodically forgo use when the other partner is perceived as more important and trustworthy.8,9 This tendency is particularly pronounced for condom use, which is often perceived as being contradictory to romantic ideals of love and trust.23,24 Conflict in relationships, ranging from relatively mild forms (e.g., arguments) to more extreme forms (e.g., physical violence), has generally been found to be correlated with less consistent contraceptive use,8,25 although this may be more pronounced for females than for males.9
METHODS
Data
Data for this analysis come from the first two waves of Add Health, a nationally representative, school-based survey that began in 1994–1995. Within each sampled school, all students in attendance were given short questionnaires. A random sample of those students completing the short in-school questionnaire were then asked to take a more extensive in-home questionnaire; those who did so were asked to complete a second survey in 1996.26
The relationships analyzed in this study are based on data collected in Wave 2, in which respondents were asked about up to six “special romantic” relationships that had occurred in the past 18 months. Add Health also asked respondents to list additional sexual relationships that are often more casual than “special romantic” ones. However, we did not include these relationships because Add Health asked limited questions about them, making it impossible to assess characteristics like intimacy and conflict. This exclusion is unlikely to be consequential, given that Ford et al.4 found that negative associations between school discordance and condom use in Add Health data were limited primarily to the “special romantic” relationships.
Because this analysis is concerned with contraceptive use in heterosexual relationships, we limited our sample to relationships in which the respondent reported having sexual intercourse with an opposite-sex partner. Questions about risk perceptions and self-efficacy regarding birth control were asked only of respondents aged 15 or older. Therefore, we excluded 1,007 (15%) of the 6,790 “special romantic” sexual relationships reported in Wave 2 because respondents were younger than 15 at Wave 1. To adjust our estimates for respondents’ attitudes and risk perceptions measured prior to relationship formation, we further limited our sample to relationships that were reported to have started after the Wave 1 interview:* This led to the exclusion of 2,393 (35%) of the 6,790 “special romantic” sexual relationships reported in Wave 2 because the reported start date was prior to the Wave 1 interview date. This means that our analytic sample contains a disproportionate share of current and relatively new relationships. For instance, the average duration for included relationships was three months, whereas for excluded relationships it was 13 months. Shorter relationships provide fewer opportunities for using contraceptives, so it is not surprising that the included relationships had slightly lower rates of having involved certain contraceptives. For example, in 77% of included relationships, but in 84% of excluded ones, respondents reported having ever used a condom with the partner. Similar patterns apply to ever-use of hormonal and dual methods. While these exclusion criteria reduce the generalizability of our results, they are critical for identifying individual-level risk factors that occurred before the relationship was formed, and therefore strengthen our findings’ internal validity.
The foregoing exclusion criteria resulted in an analytic sample of 2,004 relationship observations, of which 961 were reported by males and 1,043 by females. Most respondents in this analysis (62%) reported only one relationship over the past 18 months; however, 29% reported two relationships and 9% reported more than two. Using Stata 15, we employed the Huber-White sandwich estimator to adjust standard errors for relationship clustering within respondents and for respondents within sampled schools, and applied population weights to all estimates. Because we conducted secondary analysis of existing data, our study was exempted from institutional review board approval.
Measures
• Dependent variables.
For each reported relationship, respondents were asked whether they had ever used any of a list of contraceptive methods with the partner. Respondents who reported more than one method were then asked whether they had ever used more than one method at the same time. From these questions, we constructed three dichotomous indicators for ever having used a condom, ever having used hormonal methods and ever having used dual methods with a given partner. Respondents who reported any contraceptive use were also asked whether they had “used some method of birth control every time [they] and [partner] had sexual intercourse.” We used this question to construct a fourth dichotomous indicator for always having used birth control with the partner.
• School discordance and sociodemographic controls.
School discordance was coded as a dichotomous variable indicating whether the relationship partner had attended the respondent’s school at the start of the relationship. Because the question specifically asked about the start of the relationship, it should eliminate cases that became discordant only when the partner later graduated or dropped out. To adjust for relationship-level demographics, we included dichotomous variables for whether the partner was two or more years older* or from a different racial or ethnic group than the respondent. To adjust for individual-level demographics, we included a continuous variable for the respondent’s age at the start of the relationship and dummy variables for whether the respondent was black, Hispanic or of other racial or ethnic background (white was the reference category).
• Respondents’ prior attitudes and behavior.
We identified attitudes and behavior that may be associated with school discordance and contraceptive use.3,27,28 All of these variables were measured at Wave 1 and coded so that higher values should correlate with lower contraceptive use.
Impulsivity is a dichotomous variable indicating whether the respondent agreed or strongly agreed with the statement, “When making decisions, you usually go with your ‘gut feeling’ without thinking too much about the consequences of each alternative.” Self-efficacy regarding contraceptive use is a latent variable based on confirmatory factor analysis specified as part of generalized structural equation modeling. When presented with a six-category scale that ranged from “very sure” to “very unsure,” respondents were asked how sure they were that they could plan ahead to have some form of birth control available, use birth control once they were highly aroused and resist intercourse if their partner did not want to use birth control. These variables measure respondents’ ability to use contraceptives and roughly align with the “skills” portion of the information-motivation-behavioral skills model.18
Attitudes about pregnancy, AIDS and STD protection formed a second latent variable based on confirmatory factor analysis. Presented with a five-category scale that ranged from “strongly agree” to “strongly disagree,” respondents were asked to evaluate four statements: getting (someone) pregnant at this time in your life is one of the worst things that could happen; it wouldn’t be all that bad if you got (someone) pregnant at this time in your life; if you got the AIDS virus you would suffer a great deal; and it would be a big hassle to do all things necessary to protect yourself from an STD. These variables measure respondents’ attitudes about the costs and benefits of contraceptive use, and roughly align with the “motivation” portion of the information-motivation-behavioral skills model.18
Respondents were then asked to gauge, using a five-category scale, their chances of contracting HIV if having unprotected sex for a month and of becoming pregnant (or getting someone pregnant) if having unprotected sex just once. We coded these as separate dichotomous indicators of whether the respondent replied “no chance” or “some chance, but probably not.” Respondents were also administered an abridged version of the Peabody picture vocabulary test to gauge verbal ability and scholastic aptitude; this continuous variable was reverse-coded, with higher values indicating lower scores. Finally, we also included a measure of respondents’ total number of sexual partners measured at Wave 1.
• Relationship characteristics.
Respondents answered a series of questions about each reported romantic relationship. In addition to a continuous variable that measured relationship duration in months, we included a number of dichotomous measures to indicate whether respondents answered “yes” to questions about the following activities or conditions: partner and respondent said they loved each other; both thought or spoke of themselves as a couple; either gave the other a gift; both talked about STDs or contraception; most or all of the respondents’ friends knew the partner; the respondent had met the partner’s parents (these last two measures assess “social embeddedness”); the respondent knew the partner before the relationship began; and there was verbal or physical conflict in the relationship (name calling, insults, swearing, threats of violence, pushing or shoving, or throwing things that could cause harm).
Analysis
We evaluated differences between school-concordant and school-discordant relationships by comparing means and percentages for contraceptive use, respondent attributes and relationship characteristics, separately by respondents’ gender. Next, we employed generalized linear regression models and assessed school discordance, again stratified by gender, first presenting unadjusted estimates and then controlling for relationship- and individual-level characteristics. For outcomes for which school discordance was statistically significant at p<.05 after adjusting for controls, we conducted two generalized structural equation models (GSEMs). In the first, associations between respondents’ Wave 1 attitudes and behaviors and having a school-discordant partner, as well as having ever used a condom with that partner, were identified. The second model examined associations between relationship characteristics and both school discordance and condom use.
GSEMs are similar to standard linear structural equation models (SEMs), but allow for the fitting of generalized linear models. Structural equation modeling is a useful strategy for exploring direct and indirect relationships. It allows for the inclusion of measurement models, such as confirmatory factor analysis, that integrate latent, underlying concepts using observed measures. While standard linear SEMs require that observed variables be jointly normally distributed, GSEMs relax this assumption and treat observed exogenous variables as given and produce estimates conditional on their values. The relaxation of this assumption allows us to model dichotomous outcomes for contraceptive use.
In our first GSEM, pathways from respondents’ attitudes and behaviors to school discordance and condom use indicate which respondent characteristics are jointly associated with these measures. In addition, a pathway from school discordance to condom use reflects an independent association test, net of respondents’ prior behavior and attitudes. In this model, relationship-level characteristics (i.e., race and age discordance) are treated as being associated with condom use but not school discordance, because they could plausibly impact such use, but cannot reasonably be thought to cause a relationship to be school-discordant.
In our second GSEM, pathways are specified from school discordance to relationship characteristics, and from characteristics to condom use. For characteristics that are significantly correlated with both school discordance and condom use, multiplying the corresponding pathway parameters yields an estimate of the indirect association. In this GSEM, all associations are adjusted for individual-level sociodemographic characteristics and prior attitudes and behaviors; estimates associated with the pathways leading to condom use are further adjusted for the full set of relationship-level covariates. Of particular interest in this model is the direct pathway from school discordance to condom use, which captures the association that remains after adjusting for several respondent and relationship risk factors that may confound associations between discordance and condom use.
In both GSEMs, contraceptive self-efficacy and attitudes at Wave 1 were modeled with confirmatory factor analysis using ordered logit link functions. The only continuously measured endogenous variable in the analysis is relationship duration (used in the second model), which is modeled using an identity link function. All other endogenous variables were measured dichotomously and therefore modeled with logit link functions. We relied on the Bayesian information criterion to assess model goodness of fit. We did not use goodness-of-fit SEM statistics that require the full joint-normality assumption, which is not consistent with GSEMs that model dichotomous outcomes. We estimated all our models using Stata 15.
Because school-discordant partners are also more likely to be age-discordant, and because having older male partners has been associated with reduced contraceptive use for female adolescents,4,13–15 we conducted supplementary analyses using a school and age discordance measure that distinguished among three groups: school-discordant partners who were two or more years older, school-discordant partners who were younger or not more than a year older, and school-concordant partners. The number of school-discordant partners who were younger or not more than a year older was relatively modest, making it difficult to replicate the GSEMs with this specification. However, we ran supplemental generalized linear models with this alternative specification.
RESULTS
Female adolescents reported a higher proportion of school-discordant partners than did males (69% vs. 52%). Such partners were more likely than school-concordant partners to be older than female (66% vs. 21%) and male respondents (16% vs. 7%—Table 1). Moreover, respondents were slightly older when relationships with school-discordant partners began than they were when relationships with school-concordant partners began. Females reporting school-discordant partners had higher scores on several attitude and behavior measures; however, the confidence intervals overlap between the two groups. By contrast, male respondents had a less consistent pattern for the attitude and behavior scores, but again confidence intervals overlapped. Regarding relationship characteristics, school-discordant relationships had lower scores for nearly all measures capturing intimacy and social embeddedness, although many confidence intervals overlapped.
TABLE 1.
Percentages and means (and 95% confidence intervals) for selected measures of respondents’ relationships, by gender and school concordance, National Longitudinal Study of Adolescent to Adult Health, 1994–1995 and 1996
| Measure | Females | Males | ||
|---|---|---|---|---|
| School-concordant (N=323) |
School-discordant (N=720) |
School-concordant (N=458) |
School-discordant (N=503) |
|
| Contraceptive use | ||||
| Ever used condom | 84.0 (78.4–89.7) | 67.8 (63.0–72.6) | 82.8 (77.3–88.3) | 77.8 (71.3–84.3) |
| Ever used hormonal method | 31.2 (23.6–38.7) | 25.4 (20.6–30.2) | 21.1 (15.9–26.3) | 17.9 (13.8–21.9) |
| Ever used dual methods | 36.3 (29.3–43.2) | 29.3 (25.6–33.1) | 28.3 (21.9–34.6) | 25.9 (20.3–31.5) |
| Always used birth control | 62.3 (54.1–70.6) | 49.5 (44.8–54.1) | 64.1 (58.4–69.8) | 57.2 (50.7–63.8) |
| Sociodemographic | ||||
| Older partner | 20.6 (13.9–27.4) | 65.5 (60.9–70.0) | 7.0 (3.6–10.5) | 16.3 (11.6–21.0) |
| Different-race partner | 15.9 (11.2–20.7) | 18.6 (14.6–22.5) | 18.3 (3.6–10.5) | 20.6 (15.1–26.1) |
| Black | 17.9 (9.8–25.9) | 18.6 (12.7–24.6) | 22.0 (13.4–30.6) | 17.8 (10.2–25.4) |
| Hispanic | 7.4 (3.1–11.6) | 10.2 (5.6–14.9) | 13.4 (7.3–19.6) | 11.9 (6.6–17.3) |
| Other | 5.1 (2.0–8.2) | 2.7 (0.9–4.5) | 2.3 (0.7–3.9) | 7.0 (2.5–11.5) |
| Mean age at start of relationship (range, 15–21) | 16.9 (16.8–17.1) | 17.2 (17.1–17.4) | 17.2 (17.0–17.3) | 17.5 (17.3–17.6) |
| Prior attitudes and behavior | ||||
| Impulsivity | 34.1 (26.9–41.3) | 39.6 (34.4–44.9) | 48.8 (43.7–53.8) | 54.2 (47.7–60.7) |
| Mean contraceptive self-efficacy (range, 1–6) | ||||
| Could plan to have birth control | 1.6 (1.4–1.8) | 1.7 (1.6–1.9) | 1.7 (1.5–1.8) | 1.7 (1.6–1.8) |
| Could use birth control if highly aroused | 2.0 (1.8–2.2) | 2.0 (1.8–2.2) | 2.1 (2.0–2.2) | 2.2 (2.0–2.4) |
| Could resist sex if partner refused | 1.7 (1.5–2.0) | 1.7 (1.6–1.9) | 2.3 (2.2–2.5) | 2.3 (2.1–2.4) |
| Mean attitude scores (range, 1–5) | ||||
| Pregnancy worst | 1.6 (1.5–1.7) | 1.8 (1.7–1.9) | 1.6 (1.5–1.8) | 1.8 (1.6–2.0) |
| Pregnancy not so bad | 1.8 (1.6–1.9) | 1.9 (1.8–2.1) | 1.8 (1.6–1.9) | 1.9 (1.7–2.0) |
| Suffer with AIDS | 1.6 (1.5–1.7) | 1.7 (1.6–1.8) | 1.6 (1.5–1.8) | 1.5 (1.4–1.6) |
| Protection hassle | 1.9 (1.7–2.2) | 2.0 (1.9–2.1) | 2.2 (2.0–2.4) | 2.2 (2.0–2.4) |
| Risk perception | ||||
| HIV risk | 16.2 (7.9–24.4) | 15.5 (11.6–19.5) | 26.2 (19.6–32.8) | 31.2 (25.9–36.5) |
| Pregnancy risk | 9.0 (4.7–13.3) | 15.5 (11.1–19.9) | 23.2 (17.7–28.7) | 21.1 (15.2–27.0) |
| Mean picture vocabulary test score (range, 21– 139) | 50.0 (48.0–51.9) | 53.4 (51.6–55.1) | 49.2 (47.0–51.5) | 50.6 (48.7–52.4) |
| Mean no. of sex partners (range, 0–200) | 2.0 (1.5–2.5) | 2.8 (2.4–3.3) | 3.8 (2.7–5.0) | 6.2 (4.4–8.0) |
| Relationship | ||||
| Mean duration (range, 0–76 mos.) | 3.8 (3.2–4.4) | 3.3 (2.9–3.8) | 3.1 (2.6–3.5) | 3.1 (2.5–3.8) |
| Said they loved each other | 74.8 (68.5–81.2) | 63.5 (59.5–67.5) | 66.1 (59.7–72.6) | 61.1 (54.5–67.6) |
| Thought of themselves as a couple | 86.7 (82.2–91.2) | 83.6 (79.9–87.3) | 77.1 (72.0–82.1) | 79.3 (74.7–83.9) |
| Gave a gift | 70.6 (64.1–77.0) | 55.6 (49.9–61.3) | 60.1 (53.5–66.8) | 51.3 (42.9–59.6) |
| Discussed STDs or contraception | 65.5 (59.4–71.5) | 58.7 (53.1–64.3) | 51.1 (44.6–57.6) | 47.2 (39.3–55.0) |
| Friends knew partner | 79.2 (74.6–83.8) | 44.9 (38.5–51.2) | 65.9 (58.8–73.0) | 39.8 (33.4–46.3) |
| Met partner’s parents | 78.3 (72.9–83.8) | 66.2 (61.1–71.2) | 75.2 (70.2–80.2) | 68.6 (62.9–4.3) |
| Knew partner before relationship | 98.6 (97.1–100.2) | 93.2 (90.6–95.8) | 95.7 (92.9–98.5) | 90.2 (85.9–94.4) |
| Conflict in relationship | 25.6 (19.8–31.4) | 27.2 (21.7–32.7) | 29.0 (23.3–34.8) | 27.1 (22.6–31.6) |
Female adolescents’ likelihood of ever having used a condom was reduced by 60% when the partner was school-discordant; this association was attenuated to a 49% reduction in odds after adjusting for sociodemographic characteristics (Table 2). Having a school-discordant partner was not associated with ever having used hormonal or dual methods. However, the odds of always having used some type of birth control were reduced by 41% when the partner was school-discordant, but this association lost significance after adjusting for sociodemographic characteristics. Having an older partner was negatively associated with condom use for females (odds ratio, 0.6), but was not associated with any other contraceptive use measures. Among males, having a school-discordant partner was not associated with any contraceptive outcome (Table 3).
TABLE 2.
Odds ratios (and 95% confidence intervals) from generalized linear regression analysis assessing associations between contraceptive use and whether a female was in a school-discordant relationship
| Measure | Ever used condom | Ever used hormonal method |
Ever used dual methods | Always used birth control | ||||
|---|---|---|---|---|---|---|---|---|
| School-discordant partner | 0.40 (0.26–0.62)*** | 0.51 (0.31–0.84)** | 0.75 (0.50–1.13) | 0.68 (0.43–1.08) | 0.73 (0.50–1.06)† | 0.67 (0.44–1.02)† | 0.59 (0.41–0.85)** | 0.68 (0.45–1.03)† |
| Older partner | 0.58 (0.37–0.90)* | 1.23 (0.80–1.89) | 1.14 (0.77–1.69) | 0.76 (0.52–1.12) | ||||
| Different-race partner | 1.14 (0.68–1.92) | 0.63 (0.38–1.05)† | 0.72 (0.42–1.24) | 0.98 (0.61–1.58) | ||||
| Race/ethnicity | ||||||||
| Black | 1.15 (0.66–2.02) | 0.62 (0.38–1.03)† | 0.79 (0.50–1.25) | 0.94 (0.61–1.44) | ||||
| Hispanic | 0.37 (0.21 0.66)** | 0.50 (0.25–1.00)† | 0.55 (0.29–1.07)† | 0.63 (0.35–1.13) | ||||
| Other | 0.45 (0.18–1.12)† | 0.97 (0.35–2.71) | 0.88 (0.33–2.31) | 0.72 (0.32–1.62) | ||||
| Age at start of relationship | 0.95 (0.77–1.17) | 1.11 (0.91–1.35) | 1.14 (0.95–1.37) | 0.95 (0.78–1.14) | ||||
| Constant | 5.25 (3.63–7.61)*** | 14.87 (0.42–522.97) | 0.45 (0.32–0.64)*** | 0.09 (0.00–2.53) | 0.57 (0.41–0.78)** | 0.07 (0.00–1.54) | 1.65 (1.20–2.27)** | 4.77 (0.18–125.36) |
| N | 1,043 | 1,043 | 1,043 | 1,043 | 1,043 | 1,043 | 1,043 | 1,043 |
| BIC | 1,465,991 | 1,422,219 | 1,504,615 | 1,477,525 | 1,599,694 | 1,580,792 | 1,761,211 | 1,749,596 |
p<.05.
p<.01.
p<.001.
p<.10. Notes: Because of how Stata uses Ns to calculate the BIC, the application of population weights generates relatively large figures in this analysis. However, in all cases, the estimates decrease in size when controls for sociodemographic variables are added, implying a better model fit. BIC=Bayesian information criterion.
TABLE 3.
Odds ratios (and 95% confidence intervals) from generalized linear regression analysis assessing associations between contraceptive use and whether a male was in a school-discordant relationship
| Measure | Ever used condom | Ever used hormonal method | Ever used dual methods | Always used birth control | ||||
|---|---|---|---|---|---|---|---|---|
| School-discordant partner | 0.73 (0.45–1.16) | 0.84 (0.53–1.35) | 0.81 (0.55–1.21) | 0.82 (0.55–1.23) | 0.89 (0.61–1.29) | 0.91 (0.62–1.34) | 0.75 (0.52–1.07) | 0.80 (0.55–1.15) |
| Older partner | 0.59 (0.30–1.17) | 0.82 (0.41–1.65) | 0.92 (0.47–1.80) | 0.97 (0.52–1.79) | ||||
| Different-race partner | 0.87 (0.51–1.46) | 1.03 (0.59–1.79) | 0.81 (0.48–1.37) | 0.91 (0.58–1.45) | ||||
| Race/ethnicity | ||||||||
| Black | 0.83 (0.46–1.50) | 0.79 (0.48–1.33) | 0.69 (0.43–1.09) | 0.83 (0.53–1.31) | ||||
| Hispanic | 0.56 (0.32–1.00)† | 0.70 (0.37–1.33) | 0.39 (0.21–0.73)** | 0.73 (0.43–1.24) | ||||
| Other | 0.37 (0.12–1.10)† | 0.58 (0.18–1.94) | 0.32 (0.11– 0.95)* | 0.54 (0.22–1.31) | ||||
| Age at start of relationship | 0.86 (0.69–1.07) | 1.05 (0.86–1.28) | 0.99 (0.82–1.20) | 0.88 (0.74–1.04) | ||||
| Constant | 4.81 (3.46–6.68)*** | 75.26 (1.72–3,285.84)* | 0.27 (0.20–0.36)*** | 0.13 (0.00–3.91) | 0.39 (0.30–0.51)*** | 0.61 (0.02–16.19) | 1.79 (1.38–2.31)*** | 19.04 (0.97–373.62)† |
| N | 961 | 961 | 961 | 961 | 961 | 961 | 961 | 961 |
| BIC | 1,263,422 | 1,227,423 | 1,245,785 | 1,239,775 | 1,480,479 | 1,446,315 | 1,699,403 | 1,682,611 |
p<.05.
p<.01.
p<.001.
p<.10. Notes: Because of how Stata uses Ns to calculate the BIC, the application of population weights generates relatively large figures in this analysis. However, in all cases, the estimates decrease in size when controls for sociodemographic variables are added, implying a better model fit. BIC=Bayesian information criterion.
To assess whether any of these associations may have been partially a result of women who have more individual-level barriers to condom use disproportionately selecting school-discordant partners, we conducted our first GSEM (Figure 1; complete results appear in Appendix Table 1—Supporting Information). Focusing first on the pathways pointing to school discordance, we found that, for female respondents, having a lower picture vocabulary test score was positively associated with reporting a school-discordant partner (odds ratio, 1.02). Having more lifetime sex partners and perceiving a lower risk of pregnancy with unprotected sex were marginally associated with school discordance. Regarding the pathways leading to condom use, perceiving a lower chance of pregnancy with unprotected sex and having a lower score on the contraceptive self-efficacy scale were negatively associated with ever having used condoms (0.5 and 0.8, respectively). None of the attitude or behavior variables was jointly associated with both school discordance and condom use. Hence, it is not surprising that the association between school discordance and condom use was robust to the rich set of controls included in the GSEM. After adjusting for the full set of Wave 1 controls, the odds of ever having used a condom with a partner decreased by 48% when the partner was school-discordant; this is nearly identical to the 49% reduction in odds reported in Table 2, which included only sociodemographic controls.
FIGURE 1.
Odds ratios from generalized structural equation models assessing associations between female respondents’ attitudes and behaviors and whether they had a school-discordant partner and had ever used condoms
*p<.05. **p<.01. †p<.10. Notes: Significant pathways are in bold. All estimates are adjusted for individual-level sociodemographic characteristics and attitudes and behaviors; estimates listed on the pathways leading to condom use are further adjusted for racial-ethnic and age discordance in the relationship.
Our second GSEM explored which relationship characteristics correlated with both school discordance and condom use for female respondents (Figure 2; complete results appear in Appendix Table 2 —Supporting Information). Respondents with school-discordant partners were less likely to report that they and their partner said they loved each other or exchanged a gift (odds ratios, 0.3 and 0.2, respectively). While females were not less likely to talk with school-discordant partners about contraception or STDs, school-discordant relationships were, unsurprisingly, less socially embedded: A school-discordant partner was less likely to know most of the respondent’s friends and to have known the respondent when the relationship began (0.2 for each), and a respondent was less likely to have met the parents of a school-discordant partner (0.3).
FIGURE 2.
Odds ratios from generalized structural equation models assessing associations between female respondents’ relationship characteristics and whether they had a school-discordant partner and had ever used condoms
*p<.05. **p<.01. ***p<.001. Notes: Significant pathways are in bold. All estimates are adjusted for individual-level sociodemographic characteristics and attitudes and behaviors; estimates listed on the pathways leading to condom use are further adjusted for the full set of relationship-level covariates.
Three of the relationship characteristics that were sensitive to school discordance also correlated with condom use: saying they loved each other (odds ratio, 0.9), meeting the partner’s parents (1.1) and knowing the partner before the relationship began (1.3). For each of these, we tested whether the product of the pathways was significant. For knowing the partner before the relationship began, the product was negative in direction (a negative coefficient was multiplied by a positive coefficient), but marginally significant (odds ratio, 0.67; p<.10). The product of the pathways for having met the partner’s parents was also negative in direction, but not significant. In contrast, the product of the pathways for having said they loved each other was positive in direction (two negative coefficients multiplied; odds ratio, 1.24; p<.05). After accounting for all the relationship- and individual-level measures, the association between having a school-discordant partner and condom use was substantially reduced in magnitude, reflecting a 7% decrease in the likelihood of ever having used a condom with the partner.
In the supplementary analysis, we found that the likelihood of ever having used condoms or of always having used birth control was significantly lower for female respondents when their partner was both school-discordant and two or more years older. However, when the partner was school-discordant but younger or of similar age, the negative associations were smaller in magnitude and only marginally significant. This alternative school and age discordance measure was not associated with hormonal or dual method outcomes for females, and was nonsignificant for all outcomes for males.
DISCUSSION
Our analysis found that having a school-discordant partner was negatively associated with ever having used condoms for young women, but not young men. This gender difference reflects a new finding, as prior work in this area was not stratified by gender.4 Other contraceptive use measures—ever having used hormonal methods, ever having used dual methods or always having used some type of birth control—were not associated with school discordance for females or males after adjusting for sociodemographic controls.
One of the goals of this analysis was to evaluate how a wide range of individual-level risk factors for unsafe sex, measured prior to relationship formation, correlate with school discordance and may confound associations with ever having used condoms. With regard to these factors, the most notable finding was that female respondents with lower picture vocabulary scores were more likely to have a school-discordant partner. Importantly, however, key condom-use corollaries, such as self-efficacy, were not associated with school discordance, challenging our hypothesis that lower condom use in discordant relationships may partially reflect the selection of individuals with more risk factors into discordant relationships. Indeed, the associations between discordance and ever having used condoms were robust after adjusting for prior behaviors and attitudes, suggesting that, for young women, the risks associated with having a school-discordant partner cannot be explained away by individual-level selection.
In evaluating which relationship attributes vary with both school discordance and condom use, we refrain from offering a causal interpretation because Add Health does not include all the variables that could be associated with condom use. We wanted to spark additional discussion and research by identifying relationship attributes that may play an intervening role in associations between partner attributes and sexual behaviors. On one hand, it is noteworthy that we found no evidence that conflict in a relationship or communication about STDs or contraception varies with school discordance, making it unlikely that these relationship features play an important role. The nonsignificant finding for conflict seems particularly surprising, given that several out-of-school relationship features (e.g., friends not knowing the partner or not meeting the partner’s parents) suggest that females with school-discordant partners are at risk of social isolation, which could increase conflict and risk of intimate partner violence. On the other hand, we found modest evidence that being less likely to know school-discordant partners before the relationship began was correlated with a reduced likelihood of ever using condoms. In contrast, being less likely to make statements about loving a school-discordant partner was positively associated with condom use.
Given the intricacy of relationship dynamics, it is not surprising that relationship characteristics correlated with discordance sometimes have complex and countervailing associations with contraceptive behavior. Our finding that statements of love were negatively correlated with condom use confirms an extensive literature showing that condom use is less frequent in more intimate relationships.23,24 Furthermore, while the overall association between school discordance and ever having used condoms was negative for young women, there were discordant relationship features—namely, less communication about love—that were linked to safer sex.
Although age discordance was not our primary analytic focus, it is noteworthy that, for female adolescents, having a partner two or more years older was negatively associated with condom use. However, age discordance was not associated with condom use for male respondents, nor were the other contraceptive use measures significant for either gender. Age and school discordance are difficult attributes to separate, because partners who are four or more years older are virtually always school-discordant partners, which introduces collinearity between these attributes when distinguishing greater age differences. The findings from our supplemental analysis imply that female respondents’ likelihood of ever having used condoms or of consistent use of any method was lower primarily when school-discordant partners were also older.
Limitations
This analysis benefited from the rich longitudinal data available from Add Health, but there are data limitations. First, we limited our analysis to “special romantic” relationships with reported start dates after the respondent’s Wave 1 interview. This was necessary to properly adjust for prior risk factors and likely strengthened our findings’ internal validity, but it also means that our findings may primarily reflect relatively recent and ongoing relationships, and may not apply to relationships that were formed at earlier points in adolescents’ dating careers. Our decision to exclude nonromantic relationships is justified because information on more casual relationships was limited, and because prior work has shown that associations between school discordance and condom use were nonsignificant in such relationships.4 However, this exclusion means that our findings apply primarily to sexual behaviors within adolescents’ romantic relationships and may not extend to more casual encounters.
A second limitation is that, while the contraceptive use variables in Add Health are more detailed than those in many other surveys, they still have important drawbacks. Most notably, the measure of whether a respondent “ever” used a given method with a partner does not address whether the method was used consistently, and the measure of whether a respondent always used “some type of birth control” does not address the specific method used. Additionally, the outcome measures lacked potentially important information regarding the context of contraceptive decision making—for instance, the respondent’s perception of the partner’s STD status.
Finally, Waves 1 and 2 of Add Health date to the mid-1990s. While these remain the richest nationally representative data available for these questions, contextual factors in recent decades (e.g., increased social media use and access to cell phones) should be kept in mind when considering interventions, and other studies may want to attempt to replicate some of our results. While our main findings—gender differences and the robustness of relationship influences, for example—seem unlikely to be highly dependent on these factors, the more suggestive, marginally significant finding about not knowing school-discordant partners when a relationship begins may be less relevant for current teenagers, given their greater access to social media.
Conclusions
Most interventions to prevent STDs and unintended pregnancy adopt a knowledge-based approach and teach individuals about the risks associated with unprotected sex. In recent years, several researchers have argued that these approaches should be supplemented by interventions that attend to the emotional and psychosocial processes involved in sexual and contraceptive behavior, and should recognize that these processes occur within the contexts of particular relationships.4,5,8,29 Our findings support this argument for heterosexual adolescent females by showing that having a partner who does not go to one’s school or who is more than two years older is associated with reduced odds of ever using condoms with that partner. Future research should explore why these associations emerge for young women in discordant relationships, but not for young men. Although we found that selected relationship characteristics (e.g., intimacy or social embeddedness) may differ between discordant and concordant pairs, our understanding of these aspects remains preliminary. Future work using richer relationship measures or qualitative techniques could deepen our understanding of how discordance interacts with risk factors in adolescent relationships.
Supplementary Material
Acknowledgments
This work was supported by grant 1K01HD065868–01 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health. We thank Han Liu for research assistance. This research was also supported by grant 16RT-0169 from the California Tobacco-Related Disease Research Program of the University of California. This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry, Peter S. Bearman and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Information on how to obtain the Add Health data files is available on the Add Health website (http://www.cpc.unc.edu/addhealth). No direct support was received from grant P01-HD31921 for this analysis.
Footnotes
Whereas previous authors working with these relationship-level data had to remove repeated observations when respondents reported the same partner in both waves, we did not encounter this issue since we used only relationships reported in the second wave.
Given that difficulties in negotiating condom use should be more pronounced with larger age differences, we tested the age discordance measure by applying more refined categories for older partners. However, these schemes introduced collinearity because once a partner is four or more years older, he or she is virtually always an out-of-school partner. Distinguishing partners who are two or more years older was the best option for maintaining variation in school discordance.
REFERENCES
- 1.Darroch JE, Singh S and Frost JJ, Differences in teenage pregnancy rates among five developed countries: the roles of sexual activity and contraceptive use, Family Planning Perspectives, 2001, 33(6):244–250 & 281, doi: 10.2307/3030191. [DOI] [PubMed] [Google Scholar]
- 2.Hamilton BE and Ventura SJ, Birth rates for U.S. teenagers reach historic lows for all age and ethnic groups, NCHS Data Brief, 2012, No. 89, https://www.cdc.gov/nchs/products/databriefs/db89.htm. [PubMed] [Google Scholar]
- 3.Manlove J, Ryan S and Franzetta K, Contraceptive use patterns across teens’ sexual relationships: the role of relationships, partners, and sexual histories, Demography, 2007, 44(3):603–621, doi: 10.1353/dem.2007.0031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ford K, Sohn W and Lepkowski J, Characteristics of adolescents’ sexual partners and their association with use of condoms and other contraceptive methods, Family Planning Perspectives, 2001, 33(3):100–105 & 132, doi: 10.2307/2673765. [DOI] [PubMed] [Google Scholar]
- 5.Ford K and Lepkowski JM, Characteristics of sexual partners and STD infection among American adolescents, International Journal of STD & AIDS, 2004, 15(4):260–265, doi: 10.1258/095646204773557802. [DOI] [PubMed] [Google Scholar]
- 6.Kenny DA, Kashy DA and Cook WL, One with Many Designs: Dyadic Data Analysis, New York: Guilford Press, 2006. [Google Scholar]
- 7.Manlove J et al. , Relationship types and contraceptive use within young adult dating relationships, Perspectives on Sexual and Reproductive Health, 2014, 46(1):41–50, doi: 10.1363/46e0514. [DOI] [PubMed] [Google Scholar]
- 8.Manlove J et al. , Relationship characteristics and contraceptive use among young adults, Perspectives on Sexual and Reproductive Health, 2011, 43(2):119–128, doi: 10.1363/4311911. [DOI] [PubMed] [Google Scholar]
- 9.Manning WD et al. , Relationship dynamics and consistency of condom use among adolescents, Perspectives on Sexual and Reproductive Health, 2009, 41(3):181–190, doi: 10.1363/4118109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Karney BR et al. , Adolescent Romantic Relationships as Precursors of Healthy Adult Marriages: A Review of Theory, Research and Programs, Santa Monica, CA: RAND Corporation, 2007, http://www.rand.org/pubs/technical_reports/TR488.html. [Google Scholar]
- 11.Ford K, Sohn W and Lepkowski J, American adolescents: sexual mixing patterns, bridge partners, and concurrency, Sexually Transmitted Diseases, 2002, 29(1):13–19, doi: 10.1097/00007435-200201000-00003. [DOI] [PubMed] [Google Scholar]
- 12.Joyner K and Kao G, Interracial relationships and the transition to adulthood, American Sociological Review, 2005, 70(4):563–581, doi: 10.1177/000312240507000402. [DOI] [Google Scholar]
- 13.Koon-Magnin S, Kreager DA and Ruback RB, Partner age differences, educational contexts and adolescent female sexual activity, Perspectives on Sexual and Reproductive Health, 2010, 42(3):206–213, doi: 10.1363/4220610. [DOI] [PubMed] [Google Scholar]
- 14.Staras SAS et al. , The influence of sexual partner on condom use among urban adolescents, Journal of Adolescent Health, 2013, 53(6):742–748, doi: 10.1016/j.jadohealth.2013.06.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Volpe EM et al. , What’s age got to do with it? Partner age difference, power, intimate partner violence, and sexual risk in urban adolescents, Journal of Interpersonal Violence, 2013, 28(10):2068–2087, doi: 10.1177/0886260512471082. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Ku L, Sonenstein FL and Pleck JH, The dynamics of young men’s condom use during and across relationships, Family Planning Perspectives, 1994, 26(6):246–251, doi: 10.2307/2135889. [DOI] [PubMed] [Google Scholar]
- 17.Manlove J, Ryan S and Franzetta K, Adolescent sexual relationship, contraceptive consistency, and pregnancy prevention approaches, in: Booth A and Crouter AC, eds., Romance and Sex in Adolescence and Emerging Adulthood: Risks and Opportunities, Mahwah, NJ: Lawrence Erlbaum, 2006, pp. 181–212. [Google Scholar]
- 18.Fisher JD and Fisher WA, The information-motivation-behavioral skills model, in: Ralph A et al. , eds., Emerging Theories in Health Promotion and Practices and Research: Strategies for Improving Public Health, San Francisco: Jossey-Bass, 2002, pp. 40–60. [Google Scholar]
- 19.Strecher VJ and Rosenstock IM, The health belief model, in: Baum A et al. , eds., Cambridge Handbook of Psychology, Health, and Medicine, Cambridge, UK: Cambridge University Press, 1997, pp. 113–116. [Google Scholar]
- 20.Kusunoki Y and Upchurch DM, Contraceptive method choice among youth in the United States: the importance of relationship context, Demography, 2011, 48(4):1451–1472, doi: 10.1007/s13524-011-0061-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.McPherson M, Smith-Lovin L and Cook JM, Birds of a feather: homophily in social networks, Annual Review of Sociology, 2001, 27:415–444, doi: 10.1146/annurev.soc.27.1.415. [DOI] [Google Scholar]
- 22.Aalsma MC et al. , Family and friend closeness to adolescent sexual partners in relationship to condom use, Journal of Adolescent Health, 2006, 38(3):173–178, doi: 10.1016/j.jadohealth.2004.10.019. [DOI] [PubMed] [Google Scholar]
- 23.Afifi WA, Harming the ones we love: relational attachment and perceived consequences as predictors of safe-sex behavior, Journal of Sex Research, 1999, 36(2):198–206, doi: 10.1080/00224499909551985. [DOI] [Google Scholar]
- 24.Corbett AM et al. , A little thing called love: condom use in high-risk primary heterosexual relationships, Perspectives on Sexual and Reproductive Health, 2009, 41(4):218–224, 10.1363/4121809. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Manlove J, Ryan S and Franzetta K, Contraceptive use and consistency in U.S. teenagers’ most recent sexual relationships, Perspectives on Sexual and Reproductive Health, 2004, 36(6):265–275, doi: 10.1363/3626504. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Harris KM, The Add Health: Design and Accomplishments, Chapel Hill: University of North Carolina, Carolina Population Center, 2013, https://www.cpc.unc.edu/projects/addhealth/documentation/guides/DesignPaperWave_IIV.pdf. [Google Scholar]
- 27.Sheeran P, Abraham C and Orbell S, Psychosocial correlates of heterosexual condom use: a meta-analysis, Psychological Bulletin, 1999, 125(1):90–132, doi: 10.1037/0033-2909.125.1.90. [DOI] [PubMed] [Google Scholar]
- 28.Ryan S et al. , Adolescents’ discussions about contraception or STDs with partners before first sex, Perspectives on Sexual and Reproductive Health, 2007, 39(3):149–157, doi: 10.1363/3914907. [DOI] [PubMed] [Google Scholar]
- 29.Manning WD et al. , Young adult dating relationships and the management of sexual risk, Population Research and Policy Review, 2012, 31(2):165–185, doi: 10.1007/s11113-011-9226-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.