Abstract
Purpose
Evaluate whether adolescent women who received economic benefits from their boyfriends were more likely never to use condoms.
Methods
Data is from a longitudinal HIV prevention intervention study with 715 African-American adolescent women in urban Atlanta surveyed at baseline, 6 months, and 12 months. The primary outcome was never using condoms in the past 14 and 60 days at 6 and 12 months. The primary predictor was having a boyfriend as primary spending money source at baseline. Analysis minimized confounding using propensity weighting to balance respondents on 81 variables.
Results
A boyfriend was primary spending money source for 24% of respondents, who did not differ in neighborhood or family context but had lower education, more abuse history, riskier sex, and more sexually transmitted infections. After propensity score weighting, no statistically significant differences for 81 evaluated covariates remained, including age distributions. Women whose boyfriend was their primary spending money source were 50% more likely never to use condoms at 6 and 12 months and less likely to respond to the intervention at 12 months. Women whose boyfriend had been their primary spending money source but found another spending money source were more likely to start using condoms than women who continued. Women whose boyfriends owned cars were more likely never to use condoms.
Conclusions
Receiving spending money from a boyfriend is common among adolescent women in populations targeted by pregnancy and STI prevention interventions, and may undermine interventions’ effectiveness. Clinicians and reproductive health interventions need to address females’ economic circumstances.
Implications and Contribution
Disadvantaged teenage women who receive spending money from their boyfriends may not explicitly trade unsafe sex for money but are nonetheless more likely to have unsafe sex. Safe sex interventions and clinicians must consider economic factors that may interfere with adolescents’ practice of safe sex, particularly during an economic recession.
Introduction
Condoms are widely promoted for contraception and sexually transmitted infection (STI) prevention, but only 62% of US adolescents report having used condoms at last intercourse (1). Interventions to increase condom use improve individuals’ motivation, knowledge, and self-efficacy to use condoms but have limited effectiveness on condom use (2). Women with less relationship bargaining power, and hence limited ability to insist on safe sex, are particularly at risk of condom non-use (3). Relationship partners’ dyadic power appears to be influenced by the resources they contribute (4), both money and large capital investments such as cars. We hypothesize that adolescent females whose boyfriend is their primary source of spending money or whose boyfriend owns a car are more likely never to use condoms.
Some disadvantaged women receive material benefits from their committed relationship partners and may thus be at risk for low dyadic power. Explicitly transactional sex — sex traded directly for money or drugs — is a known risk factor for risky sex (5). Few studies in US populations have investigated long-standing relationships where the economic dimension is implicit rather than explicit (6). Evidence from southern Africa suggests that economic benefits occur in relationships of both disadvantaged and non-disadvantaged women, including university students (7), and that some women may not use condoms and birth control to avoid jeopardizing their relationships’ economic benefits (8–11). Some African women receiving economic benefits from their relationships describe coercion in sexual relationships as normal (12). Other women claim to avoid partner coercion and describe economic benefits as empowering economic opportunities (13, 14). Conversely, African women who are economically empowered relative to their partners have greater bargaining power (15), higher condom use (16, 17), and lower HIV risk (18, 19). Relationship power differentials have been identified as a factor in HIV spread in Africa (6, 20), and several interventions in Africa have succeeded in increasing stigma for intergenerational and power-disparate sexual relationships (21). These issues have not been studied extensively in the US. Based on the research from southern Africa, we hypothesize that disadvantaged adolescent women whose boyfriends are their primary source of spending money are more likely never to use condoms than women who receive spending money from other sources.
This research tests hypotheses that teen women who receive spending money from their boyfriends are more likely never to use condoms and to not respond to an effective safe-sex intervention. Among women whose initial primary source of spending money was their boyfriend, we also test a hypothesis that women who change to another primary source of spending money are more likely to start using condoms than if their boyfriend continues to be their primary source of spending money. A car is a large investment, particularly for urban adolescents, so this research also tests the hypothesis that female adolescents whose boyfriends own a car are more likely never to use condoms. This research is unique for using propensity score weighting to minimize confounding.
Methods
Theoretical models
This study is guided by the theories of money and commodities by Lee Rainwater and Michael Walzer who hypothesized that full membership in modern industrial societies requires deploying money for consumer goods beyond the level of subsistence (22, 23). Relationship partners receive decision-making precedence proportional to the resources that they bring to their relationships, according to theories of sociology of the family (4). Together, these theories predict that boyfriends who own cars or give spending money to their girlfriends will have greater power within relationships, and some boyfriends will use that power to promote condom non-use.
Data
We evaluated these hypotheses using data from the study of Horizons, an HIV prevention intervention for African-American females in urban Atlanta. From March 2002 through August 2004, the study enrolled low SES African-American adolescent women ages 15–21 at 3 clinics: a publicly funded STI clinic, a teen clinic based in a large public hospital, and a family planning clinic. Respondents answered surveys at baseline (n=715), 6 months (wave 2, n=607), and 12 months (wave 3, n=605) (24). Unmarried African-American females were eligible to participate if they were sexually active in the past 60 days and neither pregnant nor attempting pregnancy: 847 participants were eligible, of whom 84% agreed to participate. At baseline, 83.6% of respondents reported having a boyfriend, as did 76.8% at 6 months and 72.4% at 12 months. At baseline, 4 of the 715 participants had children whom they lived with. At each of the three waves, a 40-minute interview was administered via audio computer-assisted subject interviewing, and participants were tested for chlamydia, gonorrhea, and trichomoniasis using nucleic-acid-based metrics. Participants were paid $50 upon completion of each survey. Emory University’s Institutional Review Board approved the study protocol prior to implementation.
Measures
Condom non-use represents no attempt to protect against STIs and, for the 85% of the sample who do not use hormonal contraceptives, pregnancy. This study’s primary outcomes were having not used condoms in the past 14 days and past 60 days, at waves 2 and 3. Condom non-use was measured with 3 questions for the past 14 and 60 days: whether the respondent had sex in the last 14 (60) days, how many times, and “Out of the X times you’ve had vaginal sex, in the past 14 (60) days, how many times did you use a condom?” Respondents who reported having had vaginal sex in the past 14 (60) days but having used condoms 0 times were coded as having not used condoms in the past 14 (60) days.
Women’s primary spending money source is the answer to the question “Where do you get most of your spending money?” with possible answers: job, allowance from parents, TANF/public assistance, boyfriend, school/loans/work study, friends, other family, or other. The survey did not define the term “spending money.” Boyfriend and respondent car ownership were assessed with yes/no questions: “Does your boyfriend have a car?” and “Do you have a car?”
The survey defined a boyfriend as a committed sexual relationship partner and casual partners as non-committed sexual partners. We compared women who responded that their boyfriend was their primary source of spending money with women who selected any of the other sources of spending money.
Data Analysis
Data analysis used Stata SE version 11.
Propensity score weighting motivation
Matched sampling is a nonparametric method for reducing confounding by comparing a set of respondents with a characteristic with similar respondents who lack that characteristic. The respondents with the characteristic are analogous to the treatment group in a randomized experiment, and the respondents without the characteristic are analogous to the control group.
Women whose boyfriends are their primary source of spending money may differ from women with other spending money sources, potentially confounding any associations between condom non-use and receiving spending money. Regression analysis attempts to reduce confounding but relies on assumptions that are rarely satisfied, so confounding remains after regression analysis (25, 26). Several matched sampling methods could minimize confounding. This study uses propensity score weighting because it preserves sample size and thus power better than other matched sampling methods (27). After weighting, the only observable difference between groups will be the source of spending money.
Formulation of propensity score models
The predictors in the propensity model are potential confounders of the relationship between spending money source and never using condoms at 6 and 12 months (28), including never using condoms at baseline.
A second propensity score model was formulated among the 173 women whose boyfriends were their primary spending money source at baseline, to predict which women would discontinue having their boyfriend as their primary spending money source at 6 months.
A third propensity model was formulated to predict which women are likely to have at 6 months a boyfriend who owns a car, and likewise predicting propensity to have a boyfriend at 12 months who owns a car.
Potential confounders
Factors are considered potential confounders in this study if they satisfy two criteria: first, they differ between women receiving spending money from boyfriends, the “treatment” group (n=173), and women receiving spending money from another source, the “control” group (n=542); second, these factors either (a) are associated with condom non-use in past research or (b) predict condom non-use according to the National Institute of Mental Health integrated health behavior model (29). We describe potential confounders, how they may confound the relationship between spending money source and condom use, and why the two groups must be balanced on these factors.
Irrespective of parent socioeconomic status, adolescent women with greater human and financial capital (measured by last grade completed in school, employment status, and high school graduation) may be less likely to receive money from boyfriends and more likely to use condoms.
Women with greater potential relationship inequality may be more likely to receive money from boyfriends and less likely to use condoms (3). Related potential confounders include baseline condom non-use and history of abuse or forced sex; greater potential relationship inequalities, such as greater age difference with boyfriend; more permissive sexual norms, such as more lifetime partners; and risk-taking propensity indicated by cigarette and marijuana use.
Women in more committed, monogamous relationships (measured by number of months in relationship, concurrent casual partners, prediction of relationship’s future, and whether the relationship has broken up recently) may be more likely to receive money and less likely to use condoms due to greater trust.
We verified that the propensity weighting method balanced the groups on boyfriend-specific factors including boyfriend age, but boyfriend-specific factors were not used in the propensity score regression model because 16% of respondents lacked boyfriends; age would have been missing for these respondents with no logical value to impute.
Propensity score weighting procedure
Matched sampling methods, including propensity score weighting, limit comparisons to the “region of common support,” the set of individuals with overlapping estimated propensity scores because individuals in one group without a counterpart in the other group are not pertinent for comparisons of the two groups (25, 26, 30). Limiting the analysis to the region of common support excluded 20 control respondents who were predicted less likely to receive money from their boyfriends than respondents within the treatment group. These 20 control respondents had estimated propensity scores lower than the minimum estimated propensity score among the treated. One treated subject was predicted to be more likely to receive money from her boyfriend than any subjects within the control group, so was likewise excluded. The 172 remaining respondents whose boyfriends were their primary sources of spending money were given a weight of 1. The remaining 522 control respondents were given a weight of p^/(1 – p^), where p^ is the estimated propensity score (31, 32). A parallel process was used to implement the other 3 propensity weighting models.
Balance assessment
The goal of propensity weighting is to eliminate significant differences between treatment and control groups so that all statistical comparisons are insignificant (p > 0.05.) Distributions of each variable were compared before and after propensity score weighting using the Pearson chi-squared test. The cumulative distributions of age, boyfriend age, and age difference with boyfriend were compared with the chi-squared test to ensure that the groups did not differ in distribution (Table 1).
Table 1.
Background characteristics at baseline, comparing women whose boyfriend is her primary source of spending money at baseline with women whose spending money at baseline comes from another source, before and after propensity score weighting. The goal of propensity score weighting is to make all differences between the two groups have p > 0.05. Factors are sorted within category in order of p-value prior to matching. Age differences may affect power between women and their boyfriends, so ages are compared both in cumulative distribution and in mean.
| All factors measured at baseline
| ||||||
|---|---|---|---|---|---|---|
| Before weighting | After weighting | |||||
| Source of money | Source of money | |||||
| Boyfriend (n=173) | Other (n=542) | p | Boyfriend (n=172) | Other (n=522) | p | |
| Assigned intervention | 52.0 | 47.6 | 52.3 | 49.9 | ||
| Personal background | ||||||
| Ever physically abused | 56.6 | 44.1 | ** | 56.4 | 60.8 | |
| Ever forced to have vaginal sex | 31.2 | 22.7 | * | 31.2 | 25.3 | |
| Ever forced sex | 31.2 | 22.7 | * | 31.4 | 27.3 | |
| Ever emotionally abused | 65.9 | 59.6 | 65.7 | 66.0 | ||
| Current age (years) | 17.8 | 17.8 | 17.8 | 17.9 | ||
| Self-esteem (0 to 1) | 0.77 | 0.78 | 0.77 | 0.76 | + | |
| Neighborhood quality | 0.79 | 0.81 | 0.79 | 0.79 | ||
| Stress | 0.40 | 0.40 | 0.40 | 0.40 | ||
| Coping | 0.34 | 0.36 | 0.34 | 0.36 | ||
| Residence | ||||||
| Live with boyfriend | 17.1 | 6.7 | **** | 17.2 | 17.9 | |
| Live alone | 4.0 | 10.1 | * | 4.1 | 8.2 | |
| Live with roommate/friends | 2.9 | 6.6 | + | 2.9 | 4.4 | |
| Live with parent | 59.0 | 62.0 | 58.7 | 53.6 | ||
| Live with older family | 75.7 | 75.8 | 75.6 | 68.3 | ||
| Live with children | 0.6 | 0.7 | 0.6 | 1.0 | ||
| STIs | ||||||
| Trich, positive test | 17.3 | 11.4 | * | 16.9 | 18.6 | |
| Prior STIs | 51.4 | 41.7 | * | 51.7 | 42.0 | + |
| Gonorrhea, positive test | 7.5 | 4.1 | + | 7.6 | 11.5 | |
| Chlamydia, positive test | 22.0 | 16.2 | + | 22.1 | 22.6 | |
| Bacterial vaginosis | 12.1 | 10.9 | 12.2 | 11.9 | ||
| Education | ||||||
| High school drop-out | 28.3 | 17.0 | *** | 27.9 | 30.8 | |
| Plan graduate high school | 53.2 | 67.2 | *** | 53.4 | 50.2 | |
| Enrolled post-secondary school | 9.2 | 17.7 | ** | 9.3 | 8.1 | |
| High school graduate | 9.3 | 17.9 | ** | 9.4 | 8.1 | |
| High school student | 45.7 | 50.9 | 45.9 | 44.7 | ||
| Age - last grade completed (years) | 7.1 | 7.1 | 7.1 | 7.2 | ||
| Personal money | ||||||
| Have a job | 17.3 | 33.2 | **** | 17.4 | 19.1 | |
| Have car | 6.4 | 15.5 | ** | 6.4 | 6.8 | |
| Past year TANF or SSI | 12.1 | 19.2 | * | 12.2 | 12.8 | |
| Past year any public assistance | 57.8 | 50.4 | + | 57.6 | 55.8 | |
| Past year food stamps | 43.9 | 38.0 | 43.6 | 41.4 | ||
| Eligible for free STI care | 66.5 | 62.2 | 66.3 | 64.4 | ||
| No. hours work per week (n=210) | 25.9 | 25.8 | 25.9 | 28.3 | ||
| Hourly pay (dollars) (n=210) | 7.00 | 6.84 | 7.00 | 6.90 | ||
| Total earnings (dollars) (n=210) | 189 | 179 | 189 | 198 | ||
| Boyfriend (n=596) | ||||||
| Have a boyfriend | 94.8 | 80.1 | **** | 94.8 | 94.5 | |
| Boyfriend makes more money | 70.1 | 51.4 | **** | 69.9 | 70.9 | |
| Boyfriend age | 22.0 | 20.3 | **** | 22.0 | 22.0 | |
| Age difference (years) | 4.3 | 2.6 | **** | 4.3 | 4.1 | |
| Boyfriend has job | 78.0 | 63.6 | *** | 77.9 | 77.3 | |
| Boyfriend emotional abuse | 19.5 | 11.5 | ** | 19.0 | 22.2 | |
| Time to sex (weeks) | 10.0 | 12.3 | * | 10.1 | 11.6 | |
| Boyfriend physical abuse | 11.0 | 6.0 | * | 11.0 | 13.0 | |
| Boyfriend has car | 56.1 | 47.2 | * | 56.4 | 48.2 | |
| Boyfriend any concurrent partners | 29.9 | 26.0 | 29.4 | 31.9 | ||
| Casual partner | 28.7 | 25.1 | 28.2 | 29.2 | ||
| Boyfriend has casual partner | 28.7 | 25.1 | 28.2 | 29.2 | ||
| Boyfriend relationship quality | ||||||
| See marrying current boyfriend (0 to 1) | 0.71 | 0.65 | ** | 0.71 | 0.66 | |
| Months with boyfriend | 17.4 | 14.4 | ** | 17.3 | 16.4 | |
| Future of relationship (0 to 1) | 0.67 | 0.63 | 0.67 | 0.66 | ||
| Partner communication (0 to 1) | 0.82 | 0.82 | 0.83 | 0.82 | ||
| Partner communication frequency (0 to 1) | 0.46 | 0.42 | 0.46 | 0.45 | ||
| Recently broke up with boyfriend | 40.9 | 43.3 | 41.1 | 45.1 | ||
| Sexual risks | ||||||
| No. times sex, last 60d | 15.5 | 10.7 | **** | 15.5 | 14.9 | |
| Ever douche | 72.3 | 56.8 | **** | 72.1 | 65.4 | |
| Sex with high partner, last 60d (num. days) | 4.3 | 2.2 | **** | 4.3 | 4.1 | |
| Sex with recently released incarcerated | 26.0 | 14.2 | **** | 26.2 | 22.4 | |
| Sex while high/drunk, last 60d (no. days) | 3.4 | 1.7 | **** | 3.4 | 5.6 | |
| Douche, past 60 d | 2.0 | 0.9 | ** | 2.0 | 1.4 | |
| No. times sex, last 14 d | 4.9 | 3.7 | ** | 4.9 | 3.9 | |
| On oral contraception | 9.2 | 16.6 | * | 9.3 | 9.2 | |
| Ever HIV test | 86.7 | 79.0 | * | 87.2 | 85.0 | |
| Lifetime no. partners | 10.9 | 7.9 | + | 11.0 | 9.6 | |
| Ever oral sex | 63.0 | 55.9 | + | 62.8 | 60.9 | |
| Partners past 60 days | 1.5 | 1.6 | 1.5 | 1.9 | ||
| Sexual adventurism (0 to 1) | 0.33 | 0.31 | 0.32 | 0.31 | + | |
| Lack of condom self-efficacy (0 to 1) | 0.16 | 0.18 | 0.16 | 0.16 | ||
| Condom effectiveness (0 to 1) | 0.48 | 0.51 | 0.48 | 0.49 | ||
| Refusal efficacy (0 to 1) | 0.70 | 0.70 | 0.70 | 0.70 | ||
| Age first sex | 14.4 | 14.6 | 14.4 | 14.4 | ||
| Pleasure from sex (0 to 1) | 0.85 | 0.83 | 0.85 | 0.85 | ||
| Ever sex with 2 men | 6.9 | 4.3 | 7.0 | 7.4 | ||
| Ever anal sex | 28.9 | 23.0 | 28.5 | 27.4 | ||
| Substance use | ||||||
| Marijuana, no. days/60 | 14.6 | 7.0 | **** | 14.4 | 16.1 | |
| Any marijuana past 60 d | 56.1 | 41.3 | *** | 55.8 | 56.7 | |
| Smoke | 31.8 | 20.3 | ** | 31.4 | 37.8 | |
| Ever marijuana | 84.4 | 77.3 | * | 84.3 | 82.5 | |
| Any alcohol past 60 days | 63.0 | 52.4 | * | 62.8 | 61.5 | |
| Alcohol past 60 days | 4.6 | 3.6 | + | 4.5 | 5.2 | |
| Usual no. alcoholic drinks | 1.2 | 1.0 | * | 1.2 | 1.2 | |
| Age | ||||||
| Age ≤15 | 9.8 | 10.9 | 9.9 | 11.6 | ||
| Age ≤ 16 | 23.7 | 25.7 | 23.8 | 25.6 | ||
| Age ≤ 17 | 45.1 | 45.6 | 45.4 | 42.3 | ||
| Age ≤ 18 | 65.3 | 65.1 | 65.1 | 62.0 | ||
| Age ≤ 19 | 85.0 | 81.7 | 84.9 | 82.6 | ||
| Age ≤ 20 | 94.8 | 92.1 | 94.8 | 93.0 | ||
| Boyfriend age | ||||||
| Boyfriend age ≤ 15 | 1.2 | 1.4 | 1.2 | 0.01 | ||
| Boyfriend age ≤ 16 | 4.9 | 7.8 | 4.9 | 4.4 | ||
| Boyfriend age ≤ 17 | 11.6 | 20.1 | * | 11.7 | 12.7 | |
| Boyfriend age ≤ 18 | 25.0 | 37.6 | ** | 25.2 | 29.8 | |
| Boyfriend age ≤ 19 | 37.8 | 51.6 | ** | 38.0 | 38.7 | |
| Boyfriend age ≤ 20 | 47.0 | 60.8 | ** | 47.2 | 44.8 | |
| Boyfriend age ≤ 21 | 57.3 | 71.9 | *** | 57.1 | 57.9 | |
| Boyfriend age ≤ 22 | 62.2 | 80.2 | **** | 62.0 | 69.0 | |
| Boyfriend age ≤ 23 | 70.1 | 83.9 | *** | 69.9 | 72.8 | |
| Boyfriend age ≤ 24 | 76.8 | 87.8 | *** | 76.7 | 78.1 | |
| Boyfriend age ≤ 25 | 83.5 | 91.9 | ** | 83.4 | 82.3 | |
| Boyfriend age ≤ 26 | 87.8 | 94.5 | ** | 87.7 | 85.1 | |
| Boyfriend age ≤ 27 | 90.2 | 95.2 | * | 90.2 | 87.4 | |
| Boyfriend age ≤ 28 | 93.3 | 97.0 | * | 93.3 | 91.7 | |
| Boyfriend age ≤ 29 | 95.1 | 97.2 | 95.1 | 91.8 | ||
| Boyfriend age ≤ 30 | 95.7 | 97.7 | 95.7 | 92.6 | ||
| Age difference with boyfriend | ||||||
| Boyfriend is ≤ 0 years older | 12.8 | 10.2 | 10.3 | 8.1 | ||
| Boyfriend is ≤ 1 years older | 26.5 | 33.2 | + | 26.7 | 27.6 | |
| Boyfriend is ≤ 2 years older | 38.1 | 51.9 | ** | 38.4 | 42.8 | |
| Boyfriend is ≤ 3 years older | 49.7 | 69.6 | **** | 50.0 | 57.3 | |
| Boyfriend is ≤ 4 years older | 64.0 | 78.3 | *** | 63.7 | 64.3 | |
| Boyfriend is ≤ 5 years older | 74.2 | 84.0 | ** | 74.0 | 72.2 | |
| Boyfriend is ≤ 6 years older | 81.6 | 87.5 | + | 81.5 | 75.5 | |
| Boyfriend is ≤ 7 years older | 85.0 | 91.6 | * | 84.9 | 82.0 | |
| Boyfriend is ≤ 8 years older | 86.4 | 93.5 | ** | 86.3 | 84.9 | |
| Boyfriend is ≤ 9 years older | 89.8 | 96.5 | ** | 89.7 | 89.6 | |
| Boyfriend is ≤ 10 years older | 93.2 | 96.7 | + | 93.2 | 89.6 | |
p ≤ 0.1,
p ≤ 0.05,
p ≤ 0.01,
p ≤ 0.001,
p ≤ 0.0001
Regressions to predict never using condoms
The primary outcomes are condom non-use in the past 14 and 60 days, measured at waves 2 and 3, a total of 4 measures of condom non-use. The 3 predictors of interest were having a boyfriend as primary spending money source, changing from a boyfriend to another primary spending money source, and having a boyfriend who owns a car. We performed 12 propensity-weighted Poisson regressions: 4 outcome measures of condom non-use and 3 predictors of interest (Figure 1). To avoid spurious statistical significance due to multiple comparisons, the logistic regression model was formulated in one of the 12 eventual regressions: with outcome condom non-use in the past 14 days measured at wave 2, and predictor having a boyfriend as primary spending money source measured at baseline. The control variables chosen for this regression model were used in the remaining 11 regressions.
Figure 1.
Adjusted relative risk of not using condoms in the past 14 and 60 days as a function of boyfriend resources, giving the average treatment effect on the treated. The boyfriend resources were: (1) having a boyfriend as primary spending money source at baseline versus having a different primary spending money source at baseline (n=694), (2) among women whose boyfriend was their primary spending money source at baseline (n=173), changing to another primary spending money source at wave 2 versus keeping boyfriend as primary spending money source (n=173), (3) having a boyfriend who owns a car versus not having a boyfriend who owns a car (n=596 at wave 2, n=509 at wave 3); (4) among women who own cars, having a boyfriend who owns a car versus not having a boyfriend who owns a car (n=87 at wave 2, n=89 at wave 3). Adjusted relative risks were computed from the coefficients in propensity-weighted Poisson regressions. The propensity weights (1) estimated the likelihood of having a boyfriend as primary spending money source at baseline; (2) estimated the likelihood of discontinuing spending money at wave 2; and (3) (4) estimated the likelihood of having a boyfriend who owns a car. All regressions adjusted for relationship length, concurrent partners, oral contraceptive use, and intervention status. Regressions (3) and (4) also adjusted for having a boyfriend as primary spending money source.
To avoid artificially low standard errors, all regressions were clustered by the three clinics from which women were sampled.
The regressions included baseline condom non-use as a predictor. The regressions with outcome wave 3 condom non-use were repeated using wave 2 condom non-use instead of baseline condom non-use, and the results were not substantially different.
The coefficient for the predictor of interest in each propensity-weighted regression represents the estimated average treatment effect on the treated (31). We estimated relative risks using a Poisson working model because estimators from logistic regression are inconsistent when the outcome (here, condom non-use) is not rare; Poisson regression yields consistent and unbiased estimates (33–35).
Results
Receiving money from a boyfriend
At baseline, 24% of respondents said that their boyfriend was their primary source of spending money. Women whose boyfriends were their primary source of spending money were 10 percentage points more likely not to use condoms in the past 14 and 60 days than respondents with non-boyfriend sources of spending money: 31% versus 21% in the past 14 days (Fisher’s exact p=0.005) and 25% versus 15% in the past 60 days (Fisher’s exact p=0.004). This bivariate association could be confounded. The analysis uses propensity-weighted regressions to minimize confounding.
Compared with women with other primary spending money sources, women whose boyfriend was their primary source of spending money had greater drug and alcohol use, were more likely to test positive for STIs, had experienced more abuse and forced vaginal sex, and had lower economic and educational attainments than other women in the sample (Table 1). They were more likely to report physical and emotional abuse by their boyfriends in the last 60 days; had a larger age gap with their boyfriends; had sex with their current boyfriend more quickly after first meeting; had more frequent and riskier sex, including sex under the influence of drugs or alcohol, with a partner under the influence, and with recently released incarcerated men; and had lower oral contraceptive use (Table 1). Educationally, they were less likely to have graduated high school, plan to graduate high school, or be enrolled in school, but not more likely to be behind in school; less likely to work or have a car, but employed women did not differ in hours worked or hourly wage. They were not more likely to be in households receiving public assistance.
Achieving covariate balance with propensity score weighting
The propensity score model for having the boyfriend as a primary source of spending money included 11 variables significant at the 0.05 level (Table 2).
Table 2.
Logistic regression model to estimate propensity scores with the dichotomous outcome that primary source of spending money is boyfriend (n=715). All variables are from wave 1.
| Odds ratio | 95% CI | p | |
|---|---|---|---|
| Boyfriend makes more money | 2.40 | (1.60, 3.63) | **** |
| Family receives TANF/SSI | 0.24 | (0.13, 0.46) | **** |
| No. times used marijuana past 60 days | 1.02 | (1.01, 1.03) | **** |
| Work for pay | 0.41 | (0.25, 0.68) | *** |
| Months with boyfriend | 1.05 | (1.01, 1.09) | ** |
| Months with boyfriend, squared | 0.999 | (0.999, 1.000) | + |
| Have a car | 0.36 | (0.17, 0.79) | ** |
| Live with boyfriend | 2.33 | (1.23, 4.41) | * |
| Future of relationship, scale | 1.07 | (1.01, 1.13) | * |
| Plan to graduate hs or already did | 0.62 | (0.41, 0.95) | * |
| Positive test for CT, GC, trich | 1.58 | (1.03, 2.41) | * |
| Family receives any public assistance | 1.59 | (1.04, 2.41) | * |
| Ever physically abused | 1.45 | (0.97, 2.16) | + |
| Birth control pill | 0.58 | (0.31, 1.10) | |
| Broke up in the past 6 months | 0.71 | (0.47, 1.08) | |
| Never use condoms in past 60 days | 1.45 | (0.91, 2.29) | |
|
|
|||
p ≤ 0.1,
p ≤ 0.05,
p ≤ 0.01,
p ≤ 0.001,
p ≤ 0.0001
Outcome and all covariates at wave 1. P is from z-test.
The goal of matching is to make the groups comparable, and having all evaluated differences insignificant implies that the groups do not differ on average on any of the 81 evaluated variables. After propensity score weighting, all 81 evaluated differences were insignificant (p > 0.05) (Table 1). Cumulative age distributions also did not differ for respondent age, boyfriend’s age, or age difference with boyfriend (Table 1).
Predicting condom non-use in the propensity-weighted sample
At baseline, 27% (21%) of all respondents reported not using condoms in the past 14 days (60 days). Women whose boyfriends were their primary source of spending money were more likely never to use condoms than if they received money from another source, adjusting for 11 covariates and weighted by propensity to receive money from a boyfriend (Figure 1, Table 3.) Respondents randomly assigned to the Horizons intervention had lower rates of condom non-use at 6 months (both among all respondents and respondents receiving money from a boyfriend); and at 12 months among all respondents, but not among respondents receiving money from a boyfriend (Table 3, Table 4).
Table 3.
Results from 4 propensity-weighted regressions with outcome having not used condoms in the past 14 and 60 days during vaginal sex, at waves 2 and 3 (n=695). Adjusted relative risks are approximated with a Poisson working model.
|
|
||||
|---|---|---|---|---|
| Outcome at wave 2: Never use a condom in the past… | ||||
| 14 days | 60 days | |||
| Relative Risk, 95% CI | p | Relative risk, 95% CI | p | |
| Money from boyfriend, wave 2 | 1.64 (1.20, 2.22) | ** | 1.49 (1.02, 2.18) | * |
| Never use condom, wave 1 | 1.81 (1.34, 2.44) | **** | 2.05 (1.43, 2.94) | **** |
| Relationship length, wave 2 | 1.04 (1.01, 1.08) | * | 1.05 (1.00, 1.09) | * |
| Relationship length squared, wave 2 | 0.999 (0.998, 1.000) | * | 0.999 (0.999, 1.000) | + |
| Same boyfriend, wave 2 | 2.13 (1.28, 3.56) | ** | 1.99 (1.04, 3.81) | * |
| Respondent has casual partner, wave 2 | 1.36 (0.45, 4.13) | 0.52 (0.11, 2.38) | ||
| Boyfriend has casual partner, wave 2 | 0.67 (0.20, 2.25) | 1.38 (0.27, 6.97) | ||
| Respondent takes oral contraception, wave 2 | 1.05 (0.60, 1.83) | 0.99 (0.58, 1.71) | ||
| Intervention | 0.74 (0.54, 1.01) | + | 0.54 (0.36, 0.79) | ** |
|
|
||||
| Outcome at wave 3: Never use a condom in the past… | ||||
| 14 days | 60 days | |||
| Relative Risk, 95% CI | p | Relative risk, 95% CI | p | |
| Money from boyfriend, wave 3 | 1.44 (0.98, 2.11) | + | 1.49 (1.00, 2.22) | * |
| Never use condom, wave 1 | 1.67 (1.16, 2.40) | ** | 2.09 (1.38, 3.16) | **** |
| Relationship length, wave 3 | 1.04 (1.01, 1.07) | * | 1.04 (1.01, 1.07) | ** |
| Relationship length squared, wave 3 | 0.999 (0.999, 1.000) | + | 0.999 (0.999, 1.000) | + |
| Same boyfriend, wave 3 | 0.87 (0.53, 1.40) | 0.87 (0.51, 1.46) | ||
| Respondent has casual partner, wave 3 | 0.64 (0.27, 1.54) | 0.71 (0.25, 2.00) | ||
| Boyfriend has casual partner, wave 3 | 1.28 (0.48, 3.44) | 1.18 (0.37, 3.76) | ||
| Respondent takes oral contraception, wave 3 | 1.08 (0.59, 1.99) | 1.14 (0.65, 2.01) | ||
| Intervention | 0.56 (0.38, 0.82) | ** | 0.54 (0.36, 0.82) | ** |
|
|
||||
p ≤ 0.1,
p ≤ 0.05,
p ≤ 0.01,
p ≤ 0.001,
p ≤ 0.0001
Regression results (n=695), inverse propensity score weighted. Relationship length is in months, both linear and quadratic terms. P is from z-test.
Table 4.
Effectiveness of Horizons intervention among women whose boyfriends are their primary sources of spending money. Results from 4 propensity-weighted regressions with outcome having not used condoms in the past 14 and 60 days during vaginal sex, at wave 2 (n=141) and wave 3 (n=127). Adjusted relative risks are approximated with a Poisson working model.
|
|
||||
|---|---|---|---|---|
| Outcome at wave 2: Never use a condom in the past… | ||||
| 14 days | 60 days | |||
| Relative Risk, 95% CI | p | Relative risk, 95% CI | p | |
| Horizons intervention | 0.63 (0.43, 0.94) | * | 0.38 (0.21, 0.68) | *** |
| Never use condom, wave 1 | 1.50 (1.05, 2.14) | * | 1.70 (1.05, 2.77) | * |
| Relationship length, wave 2 | 1.00 (0.98, 1.01) | 1.00 (0.98, 1.02) | ||
| Same boyfriend, wave 2 | 1.48 (0.94, 2.33) | + | 1.79 (0.91, 3.52) | + |
| Respondent has casual partner, wave 2 | 0.68 (0.36, 1.30) | 0.66 (0.28, 1.59) | ||
| Respondent takes oral contraception, wave 2 | 1.50 (0.82, 2.74) | 1.42 (0.64, 3.13) | ||
|
|
||||
| Outcome at wave 3: Never use a condom in the past… | ||||
| 14 days | 60 days | |||
| Relative Risk, 95% CI | p | Relative risk, 95% CI | p | |
| Horizons intervention | 0.87 (0.53, 1.42) | 0.82 (0.49, 1.38) | ||
| Never use condom, wave 1 | 1.87 (1.16, 3.00) | ** | 2.08 (1.25, 3.47) | ** |
| Relationship length, wave 3 | 1.01 (0.99, 1.02) | 1.01 (1.00, 1.03) | * | |
| Same boyfriend, wave 3 | 0.74 (0.43, 1.27) | 0.71 (0.39, 1.31) | ||
| Respondent has casual partner, wave 3 | 0.44 (0.21, 0.92) | * | 0.40 (0.19, 0.84) | * |
| Respondent takes oral contraception, wave 3 | 1.11 (0.43, 2.85) | 1.51 (0.71, 3.25) | ||
|
|
||||
p ≤ 0.1,
p ≤ 0.05,
p ≤ 0.01,
p ≤ 0.001,
p ≤ 0.0001
Relationship length is in months. P is from z-test. To compensate for loss of power due to restriction to the subsample of women whose boyfriends are their primary spending money sources, these regressions do not include two colinear factors present in previous regressions: quadratic term for relationship length and the dichotomous factor boyfriend has casual partner. All respondents who reported that their boyfriends had casual partners reported having casual partners themselves.
Among the 173 women who received most of their spending money from boyfriends at baseline, women who changed to a non-boyfriend spending source were less likely never to use condoms at wave 2, but not wave 3, after adjusting for baseline condom use and weighting by propensity to discontinue money (Figure 1).
Boyfriend’s car ownership
At all three waves, about 40% of all respondents say that they have a boyfriend who owns a car. After propensity score weighting, no differences were observed between respondents with car-owning boyfriends and respondents with boyfriends without cars or without boyfriends (not shown). Respondents with car-owning boyfriends were 30–50% more likely never to use condoms than if they did not have boyfriends with cars (Figure 1). Restricting to respondents who owned cars at wave 2 (n=87) and wave 3 (n=89), the results were mixed: car-owning respondents with car-owning boyfriends were twice as likely never to use condoms at wave 3 than car-owning respondents without car-owning boyfriends, but no difference was observed at wave 2 (Figure 1).
Discussion
Adolescent women whose boyfriend is their primary source of spending money may not explicitly exchange risky sex for money, but their relationships may be implicitly transactional. Women who received most of their spending money from boyfriends were more likely not to use condoms than if they had received spending money from another source, but women who discontinued receiving most of their spending money from their boyfriend were more likely to start using condoms.
Women whose boyfriends owned cars were also more likely never to use condoms than if their boyfriends did not own cars. For women who owned cars, this difference was attenuated, which is consistent with theory of resource contribution: if both partners own cars, a boyfriend’s car does not represent a resource disparity unless it is higher quality.
These differences are consistent with sociology of family theories that state that partners’ decision-making precedence is proportional to the economic resources that they contribute to the relationship (4). Resource disparities such as spending money and car ownership may change relationship power balance and alter teen women’s safe sex choices.
Qualitative research has found that in relationships with power imbalances, women may not use condoms due to partner coercion (37, 38), to avoid losing spending money (7), to deepen relationships, either intangibly through trust implicit in condom non-use (38), or tangibly through pregnancy if they consciously or unconsciously interpret resources as signaling ability to support children (39).
Condom non-use is likely not attributable to relationship closeness or other relationship factors. Propensity weighting balanced the groups on baseline relationship factors, and wave 2 and 3 relationship factors did not predict condom use in propensity-weighted regressions.
Age disparities are recognized as a source of inequality in adolescents’ committed relationships. This work implies that boyfriend-girlfriend dyads with resource disparities but no age difference are more likely to have unsafe sex. Propensity weighting allowed us to compare groups with similar age distributions.
Women who received spending money from their boyfriends were not from more impoverished backgrounds than women who did not, nor did they have lower self-efficacy or condom self-efficacy. The spending money may be used for consumer goods for social standing, as in Africa (7, 13). While not essential for survival, these consumer goods may be perceived as necessary for full membership in the U.S. consumerist society (22, 23).
Receiving money from a boyfriend could undermine interventions that attempt to improve women’s ability to negotiate safe sex. In the full sample, the Horizons intervention reduced condom non-use. Among women whose boyfriends were their primary spending money source, Horizons was effective at 6 months but not 12 months. For women receiving money from boyfriends, the failure to persist at 12 months suggests that interventions must address economic factors to have long-term impact.
Limitations
The propensity weighting method may be sensitive to model formulation, but this propensity weighting achieved the required covariate balance, so additional optimization likely would not have improved balance (31).
These results were found among HIV prevention program participants and may not generalize to all adolescents.
The study excluded women trying to become pregnant, but participants may have concealed their fertility preferences to enter the study; condom non-use may have been intentional rather than due to boyfriend coercion.
Implications for policy and practice
Clinicians may be more effective in their safe sex counseling if they consider economic factors. Clinicians can ask adolescents about their primary sources of spending money, remind adolescents of the tangible and intangible resources that they bring to a relationship, provide guidance about when to decline offered gifts, and guide patients to obtain social goods outside of romantic relationships.
Interventions can encourage communities to monitor and speak against unequal relationships, as has been done in Uganda with a social media campaign against age discrepant relationships with exchange of significant economic resources (21). Safe sex interventions seem to be less effective among young women whose primary spending money sources are their boyfriends. Safe sex programs that empower participants economically may be more effective. Randomized experiments with conditional and unconditional cash transfer programs in Africa suggest that providing money to young women decreases sexual risk-taking (40).
Conclusions
Safe-sex interventions may be undermined if adolescent women receive spending money from their boyfriends. Clinicians and safe sex interventions should consider that adolescents who receive spending money from their boyfriends may be at risk for coercive relationships or unsafe sex.
Acknowledgments
The authors thank Jeanne Brooks-Gunn, Jonathan Ellen, Charles Horn, Susan Kegeles, Carl Latkin, Willi McFarland, Clea McNeely, Katherine Newman, James Rosenbaum, Dan Scharfstein, Daniel Liam Singer, Freya Sonenstein, Donna Strobino, Elizabeth Stuart, Mark Tisdahl, and Laurie Zabin for helpful conversations, as well as the participants at seminars at the Johns Hopkins Biostatistics Department’s Causal Inference Working Group, Planned Parenthood of California, and the University of Massachusetts Statistics Department.
This research was funded by T-32 AI050056 from the Sexually Transmitted Diseases Division of the Centers for Disease Control and Prevention (Dr. Zenilman.) The data collection was funded by R01 MH061210 from the Center for Mental Health Research on AIDS, National Institute of Mental Health, Bethesda, Maryland (Drs DiClemente and Wingood).
Abbreviations
- STI
sexually transmitted infection
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Eaton Danice K, Kann Laura, Kinchen Steve, Shanklin Shari, Ross James, Hawkins Joseph, Harris William A, Lowry Richard, McManus Tim, Chyen David, Lim Connie, Brener Nancy D, Wechsler Howell. Youth Risk Behavior Surveillance — United States, 2007. Mortality and Morbidity Weekly Report. 2008 Jun 6;57(SS04):1–131. [PubMed] [Google Scholar]
- 2.Kirby Douglas B. National campaign to prevent teen pregnancy. Washington DC: 2007. Emerging Answers 2007: Research findings on programs to reduce teen pregnancy and sexually transmitted diseases. [Google Scholar]
- 3.Billy John OG, Grady William R, Sill Morgan E. Sexual risk-taking among adult dating couples in the United States. Perspectives on sexual and reproductive health. 2009;41(2):74–83. doi: 10.1363/4107409. [DOI] [PubMed] [Google Scholar]
- 4.Collins Randall. Sociology of marriage and the family: Gender, love, and property. Nelson-Hall; Chicago: 1985. [Google Scholar]
- 5.Bobashev GV, Zule WA, Osilla KC, Kline TL, Wechsberg WM. Transactional sex among men and women in the south at high risk for HIV and other STIs. J Urban Health. 2009;86(Suppl 1):32–47. doi: 10.1007/s11524-009-9368-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Kuate-Defo Barthelemy. Young people’s relationships with sugar daddies and sugar mummies: what do we know and what do we need to know? African Journal of Reproductive Health. 2004;8(2):13–37. [PubMed] [Google Scholar]
- 7.Masvawure T. ’I just need to be flashy on campus’: Female students and transactional sex at a university in Zimbabwe. Cult Health Sex. 2010 Jan;12(1) doi: 10.1080/13691050903471441. [DOI] [PubMed] [Google Scholar]
- 8.Leclerc-Madlala S. Age-disparate and intergenerational sex in southern Africa: the dynamics of hypervulnerability. AIDS. 2008;(Suppl 4):S17–S25. doi: 10.1097/01.aids.0000341774.86500.53. [DOI] [PubMed] [Google Scholar]
- 9.Silberschmidt M, Rasch V. Adolescent girls, illegal abortions and ”sugar-daddies” in Dar es Salaam: vulnerable victims and active social agents. Social Science and Medicine. 2001;52(12):1815–26. doi: 10.1016/s0277-9536(00)00299-9. [DOI] [PubMed] [Google Scholar]
- 10.Longfield K, Glick A, Waithaka M, Berman J. Relationships between older men and younger women: implications for STIs/HIV in Kenya. Studies in Family Planning. 2004;35(2):125–34. doi: 10.1111/j.1728-4465.2004.00014.x. [DOI] [PubMed] [Google Scholar]
- 11.Weiser SD, Leiter K, Bangsberg DR, Butler LM, Percy de Korte F, Hlanze Z, Phaladze N, Iacopino V, Heisler M. Food insufficiency is associated with high-risk sexual behavior among women in Botswana and Swaziland. PLoS Medicine. 2007;4(10):1589–97. doi: 10.1371/journal.pmed.0040260. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Wagman J, Baumgartner JN, Waszak Geary C, Nakyanjo N, Ddaaki WG, Serwadda D, Gray Ron, Nalugoda FK, Wawer Maria J. Experiences of sexual coercion among adolescent women: qualitative findings from Rakai district, Uganda. J Interpersonal Violence. 2009;24(12):2073–95. doi: 10.1177/0886260508327707. [DOI] [PubMed] [Google Scholar]
- 13.Nkosana J, Rosenthal D. The dynamics of intergenerational sexual relationships: the experience of schoolgirls in Botswana. Sex Health. 2007 Sep;4(3):181–7. doi: 10.1071/sh06070. [DOI] [PubMed] [Google Scholar]
- 14.Hawkins K, Price N, Mussa F. Milking the cow: young women’s construction of identity and risk in age-disparate transactional sexual relationships in Maputo, Mozambique. Global Public Health. 2009;4(2):169–82. doi: 10.1080/17441690701589813. [DOI] [PubMed] [Google Scholar]
- 15.Pollak Robert A. Technical Report. National Bureau of Economic Research; Apr, 2005. Bargaining power in marriage: Earnings, wage rates and household production; p. 11239. [Google Scholar]
- 16.Wakasugi N, Nishiyama A, Randrianarivo A, Sacko A, Siwale M, Kankasa C. The role of women’s empowerment as a factor facilitating condom use meta-analysis: The results from African countries. International Conference on AIDS; July 11–16 2004. [Google Scholar]
- 17.Randrianarivo Armand, Razafinanefa Maminirina, Rasolomaharo Monique, Nishiyama Ayako, Saito Tomohiro, Wakasugi Naomi. Child death and women’s own earnings are associated with condom use in Madagascar. Tropical Medicine and Health. 2004;32(1):27. [Google Scholar]
- 18.Kim J, Pronyk P, Barnett T, Watts C. Exploring the role of economic empowerment in HIV prevention. AIDS. 2008 Dec;(Suppl 4):S57–S71. doi: 10.1097/01.aids.0000341777.78876.40. [DOI] [PubMed] [Google Scholar]
- 19.Jukes Matthew, Simmons Stephanie, Bundy Donald. Education and vulnerability: the role of schools in protecting young women and girls from HIV in southern Africa. AIDS. 2008 Dec;(Supplement 4):S41–S56. doi: 10.1097/01.aids.0000341776.71253.04. [DOI] [PubMed] [Google Scholar]
- 20.Luke N. Confronting the ’sugar daddy’ stereotype: age and economic asymmetries and risky sexual behavior in urban Kenya. International Family Planning Perspectives. 2005;31(1):6–14. doi: 10.1363/3100605. [DOI] [PubMed] [Google Scholar]
- 21.Fiscian VS, Obeng EK, Goldstein K, Shea JA, Turner BJ. Adapting a multifaceted U.S. HIV prevention education program for girls in Ghana. AIDS Educ Prev. 2009;21(1):67–79. doi: 10.1521/aeap.2009.21.1.67. [DOI] [PubMed] [Google Scholar]
- 22.Walzer Michael. Spheres of justice: a defense of pluralism and equality. Basic Books; 1983. [Google Scholar]
- 23.Rainwater Lee. What Money Buys: Inequality and the Social Meaning of Income. John Wiley and Sons; New York: 1974. [Google Scholar]
- 24.Rose E, Diclemente RJ, Wingood GM, Sales JM, Latham TP, Crosby RA, Zenilman J, Melendez J, Hardin J. The validity of teens’ and young adults’ self-reported condom use. Arch Pedatri Adolesc Med. 2009 Jan;163(1):61–64. doi: 10.1001/archpediatrics.2008.509. [DOI] [PubMed] [Google Scholar]
- 25.Rubin DB. Estimating causal effects from large datasets using propensity scores. Annals of Internal Medicine. 1997 Oct;127(8Pt2):757–763. doi: 10.7326/0003-4819-127-8_part_2-199710151-00064. [DOI] [PubMed] [Google Scholar]
- 26.Rubin DB. The design versus the analysis of observational studies for causal effects: parallels with the design of randomized trials. Statistics in Medicine. 2007 Jan;26(1):20–36. doi: 10.1002/sim.2739. [DOI] [PubMed] [Google Scholar]
- 27.Austin PC. The relative ability of different propensity score methods to balance measured covariates between treated and untreated subjects in observational studies. Medical Decision Making. 2009 Nov-Dec;29(6):661–77. doi: 10.1177/0272989X09341755. [DOI] [PubMed] [Google Scholar]
- 28.Austin PC, Grootendorst P, Anderson GM. A comparison of the ability of different propensity score models to balance measured variables between treated and untreated subjects: a Monte Carlo study. Statistics in Medicine. 2007 Feb;26(4):734–753. doi: 10.1002/sim.2580. [DOI] [PubMed] [Google Scholar]
- 29.Jaccard J, Dodge T, Dittus PJ. Parent-adolescent communication about sex and birth control: a conceptual framework. New directions for child and adolescent development. 2002;97:9–41. doi: 10.1002/cd.48. [DOI] [PubMed] [Google Scholar]
- 30.Morgan SL, Winship C. Counterfactuals and causal inference: Methods and principles for social research. Cambridge University Press; New York: 2007. [Google Scholar]
- 31.Gelman Andrew, Hill Jennifer. Data Analysis using regression and multilevel/hierarchical models. Cambridge University Press; New York: 2007. [Google Scholar]
- 32.Rubin Donald B. Using Propensity Scores to Help Design Observational Studies: Application to the Tobacco Litigation. Health Services and Outcomes Research Methodology. 2009;2(3–4):169–88. [Google Scholar]
- 33.Cummings Peter. The Relative Merits of Risk Ratios and Odds Ratios. Archives of Pediatric and Adolescent Medicine. 2009;163(5):438–45. doi: 10.1001/archpediatrics.2009.31. [DOI] [PubMed] [Google Scholar]
- 34.Lumley Thomas, Kronmal Richard, Ma Shuangge. University of Washington Biostatistics Working Paper Series. 293. 2006. Relative Risk Regression in Medical Research: Models, Contrasts, Estimators, and Algorithms. [Google Scholar]
- 35.McNutt Louise-Anne, Wu Chuntao, Xue Xiaonan, Hafner Jean Paul. Estimating the Relative Risk in Cohort Studies and Clinical Trials of Common Outcomes. American Journal of Epidemiology. 2003;157(10):940–3. doi: 10.1093/aje/kwg074. [DOI] [PubMed] [Google Scholar]
- 36.Pallitto Christina, Campbell Jacquelyn C, O’Campo Patricia. Is intimate partner violence associated with unintended pregnancy? A review of the literature. Trauma Violence Abuse. 2005;6(3):217–35. doi: 10.1177/1524838005277441. [DOI] [PubMed] [Google Scholar]
- 37.Miller Elizabeth, Decker Michelle R, McCauley Heather L, Tancredi Daniel J, Levenson Rebecca R, Waldman Jeffrey, Schoenwald Phyllis, Silverman Jay G. Pregnancy coercion, intimate partner violence and unintended pregnancy. Contraception. 2010;81(4):316–22. doi: 10.1016/j.contraception.2009.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Michelle Corbett A, Dickson-Gomez Julia, Hilario Helena, Weeks Margaret R. A little thing called love: Condom use in high-risk primary heterosexual relationships. Perspectives on Sexual and Reproductive Health. 2009 Dec;41(4):218–24. doi: 10.1363/4121809. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Geary DC, Vigil J, Byrd-Craven J. Evolution of human mate choice. Journal of Sex Research. 2004 Feb;41(1):27–42. doi: 10.1080/00224490409552211. [DOI] [PubMed] [Google Scholar]
- 40.Baird Sarah, McIntosh Craig, Özler Berk. Cash or condition? Evidence from a randomized cash transfer program. Quarterly Journal of Economics. 2011 [Google Scholar]