Growth Trajectories of Sexual Risk Behavior in Adolescence and Young Adulthood

Abstract

Objectives. Adolescence and young adulthood (ages 18–25 years) are periods of development and change, which include experimentation with and adoption of new roles and behaviors. We investigated longitudinal trajectories of sexual risk behaviors across these time periods and how these trajectories may be different for varying demographic groups.

Methods. We developed multilevel growth models of sexual risk behavior for a predominantly African American sample (n=847) that was followed for 8 years, from adolescence to young adulthood. We investigated differences in growth parameters by race/ethnicity and gender and their interactions.

Results. The final model included linear and quadratic terms for both adolescence and young adulthood, indicating acceleration of sexual risk behaviors during adolescence and a peak and deceleration during young adulthood. African American males exhibited the highest rate of sexual risk behavior in ninth grade, yet had the slowest rate of growth. Compared with their White peers, African American males and females exhibited less sexual risk behavior during young adulthood.

Conclusions. Our results suggest that youths of different races/ethnicities and genders exhibit varying sexual risk behavior trajectories.

Adolescent sexual risk behavior may have profound health consequences,1 which can extend into later life.2 Adolescents may be susceptible to engagement in sexual risk behaviors such as infrequent use of condoms3 or having multiple partners4 because of underdeveloped decisionmaking skills.5 Young adulthood, defined here as ages 18–25 years, is a time when youths take on new roles and responsibilities and form adult identities,6^,7 which may include a relationship with a significant other and changes in sexual behavior. Sexual risk behavioral patterns adopted during adolescence and young adulthood may continue throughout adulthood. It is therefore vital to understand sexual risk trajectories during adolescence and young adulthood.

One problem with much of the research on sexual risk behavior in adolescence and young adulthood is that most studies are cross-sectional and consequently focus either on adolescence or adulthood.8 Most longitudinal studies included only 2 waves of sexual risk behavior data,9^–13 assessed sexual risk behavior at only 1 time point,3^,14^,15 did not include data that spanned both of these developmental periods,8 investigated an overall mean level of sexual risk behavior across study waves,16^,17 or conducted separate analyses for each wave of data.18 These study designs are insufficient for capturing trajectories of behavior.19 In a multiwave study of predominantly White males that spanned adolescence and young adulthood, Capaldi et al.16 suggested that their pattern of results showed the frequency of sexual intercourse increasing, the number of partners increasing then decreasing, and condom use decreasing, although the researchers did not test these suggestions statistically. Similarly, O’Donnell et al.3 suggested in their 4-wave study of urban youths that frequency of sexual intercourse increases and condom use decreases during early adolescence, but they did not provide statistical tests. Longitudinal studies that include several waves of data spanning adolescence and young adulthood and that investigate and test trajectories of sexual risk behavior are lacking.

It is further not clear how Capaldi et al.’s16 suggestions may generalize to non-White or female youths or how O’Donnell et al.’s3 results generalize to older youths. It is vital that we understand the trajectories of sexual risk behavior for different demographic groups. The consequences of sexual risk behavior, including contracting HIV/AIDS20 and gonorrhea and syphilis,21 for example, disproportionately affect African Americans compared with other groups. Also, compared with men, women are more likely to have asymptomatic sexually transmitted diseases,22 and they make up an increasingly larger share of the incidence of HIV.23 Understanding the differential trajectories of adolescent and young adult sexual risk behaviors among men and women from various races/ethnicities may help to explain these disparities.

Nationally representative studies have suggested that sexual risk behaviors among youths differ by race/ethnicity and gender. In the 2003 Youth Risk Behavior Surveillance System high school survey,24 African American students were found to be more likely to have had sexual intercourse and to report a greater number of sexual partners, yet were more likely to have used a condom during last sexual intercourse, compared with White students. In the same survey,24 boys were found to be equally as likely to have had sexual intercourse, to report a greater number of sexual partners, and to be more likely to have used a condom during last intercourse as girls. Although the Youth Risk Behavior Surveillance System and similar studies are informative in assessing trends in adolescent sexual risk behavior across time, they do not allow us to investigate individual trajectories of behavior, which provide information about the shape of change in the behavior in the same group of individuals over time. The Youth Risk Behavior Surveillance System also does not include young adulthood.

In our study, we addressed some of the shortcomings in the literature. First, we examined data over 8 years that included the period of young adulthood. Second, we employed an analytic approach—hierarchical linear modeling—that permits the study of trajectories of change over time. Finally, we investigated how trajectories of sexual risk behavior may differ by race/ethnicity and gender.

METHODS

Sample

We obtained data from adolescents from 4 public high schools in Flint, Michigan, a medium-sized, high-poverty, majority–African American city located close to Detroit. Participants were interviewed starting in 1994 every year for 4 years (corresponding to participants’ high school years), followed by a 1-year gap, and then every year for 4 additional years, for a total of 8 waves of data. Participants who dropped out of school were retained in the sample. Selection criteria included having a grade-point average of 3.0 or lower at the end of eighth grade and not being diagnosed by the school as emotionally impaired or developmentally disabled. Of the 920 who were eligible to participate, 850 (92%) agreed to participate in wave 1. This analysis included 847 participants with a total of 5203 observations. The 3 participants who were not included were missing sexual risk behavior measures for all waves. Participants provided a mean of 6.1 waves of data, with 30% providing 8, 22% providing 7, 16% providing 6, 12% providing 5, 10% providing 4, and 10% providing 3 or fewer.

Eighty percent of participants were African American, 17% were White, and 3% were biracial. Half (50%) were girls, and the mean age at wave 1 was 14.6 years (SD = 0.7). Participants reported in wave 1 that the highest education attained by their mothers was as follows: 11% grade school or some high school, 38% high school, 28% vocational school or some college, 13% college, and 3% graduate or professional school. The percentage of participants who reported never having had sexual intercourse at wave 1 was 38%, at wave 2 was 30%, at wave 3 was 27%, and at wave 4 was 17%. The percentage of participants who reported not having had sexual intercourse in the previous 12 months at wave 5 was 12%, at wave 6 was 13%, at wave 7 was 13%, and at wave 8 was 12%.

Procedure

Structured, 50- to 60-minute face-to-face interviews were conducted in school or in a community setting. Parental consent was obtained for minors. After the interview, participants completed a self-administered questionnaire about sensitive topics including sexual risk behavior.

Measures

The sexual risk behavior measure is a composite measure, calculated by taking the sum of the standardized scores (x̄ = 0, SD= 1) of 3 items. This measure is similar to a measure used in other studies of youth sexual risk behavior.16 The items are frequency of (unspecified) sexual intercourse in the preceding year (1 = 1–2 times, 2 = 3–5 times, 3 = 6–8 times, 4=9–11 times, 5=12 or more times), number of sexual partners in the preceding year, and frequency of condom use during sexual intercourse in the preceding year (1 = almost never, 2 = not very often, 3 = half of the time, 4= most of the time, 5 = always). The condom-use frequency item was reverse coded so that higher scores indicated more risk. For participants who reported never having had sexual intercourse, or not having sexual intercourse in the previous year, “1” was assigned for the sexual intercourse and condom-use frequency items and “0” for the number of partners before standardizing. Table 1 ▶ presents descriptive statistics for the items and composite scores for each wave. Overall, the skewness and kurtosis of the sexual risk behavior measure were 0.56 and 1.6, respectively. In the analyses, we used standardized scores of the composite measure.

TABLE 1—

Means (and Standard Deviations) of Sexual Risk Behavior Variables in Adolescence (Waves 1–4) and Young Adulthood (Waves 5–8): Flint, Mich, 1994–2003

Behavior Variables	Wave 1 (n = 782), Mean (SD)	Wave 2 (n = 751), Mean (SD)	Wave 3 (n = 718), Mean (SD)	Wave 4 (n = 716), Mean (SD)	Wave 5 (n = 549), Mean (SD)	Wave 6 (n = 616), Mean (SD)	Wave 7 (n = 546), Mean (SD)	Wave 8 (n = 525), Mean (SD)
Sexual intercourse frequencya	1.68 (1.80)	1.99 (1.84)	2.31 (1.96)	2.83 (2.03)	3.66 (1.86)	3.60 (1.89)	3.63 (1.90)	3.63 (1.88)
Number of sexual partners	2.10 (3.21)	1.83 (2.63)	1.96 (2.90)	1.84 (2.24)	2.03 (3.11)	1.85 (2.10)	1.81 (2.09)	1.81 (3.06)
Condom useb	1.51 (1.07)	1.54 (1.08)	1.71 (1.18)	1.96 (1.30)	2.53 (1.53)	2.48 (1.59)	2.60 (1.60)	2.48 (1.63)
Composite scorec	−0.88 (2.27)	−0.80 (1.99)	−0.48 (2.21)	−0.02 (2.27)	0.80 (2.16)	0.67 (2.01)	0.75 (2.04)	−0.02 (2.07)

Note. Study participants were interviewed starting in 1994 every year for 4 years (corresponding to participants’ high school years), followed by a 1-year gap, and then every year for 4 additional years, for a total of 8 waves of data.

^aCoded: 1 = 1–2 times, 2 = 3–5 times, 3 = 6–8 times, 4 = 9–11 times, 5 = 12 or more times.

^bCoded: 1 = always, 2 = most of the time, 3 = half of the time, 4 = not very often, 5 = almost never.

^cSum of standardized scores.

Race/ethnicity was a dummy variable signifying that the participant was White, with African American and biracial participants as the referent group. Gender was a dummy variable signifying that the participant was female. Age was a continuous measure in years, calculated from the participant’s date of birth and the date of the interview at wave 1. Participants were interviewed at approximately the same time of year at each wave.

Statistical Analyses

We developed piecewise growth models25^–28 of sexual risk behavior during adolescence (waves 1 through 4) and young adulthood (waves 5 through 8), with 2-level hierarchical linear modeling19^,26 with the HLM 5: Hierarchical Linear and Nonlinear Modeling program (Scientific Software Intl, Chicago, Ill). Our general level-1 model was as follows:

(1)

For a quadratic model, we coded a as shown in Table 2 ▶, so that Y is an individual’s value on a measure at a given time, π₀ is the average score at wave 1, π₁ is the average instantaneous linear growth at wave 1, π₂ is the average acceleration during adolescence, π₃ is the average instantaneous linear growth at wave 5, and π₄ is the average acceleration during young adulthood, with additional π values corresponding to higher-level polynomial terms. The a values indicate the data waves, and e is an error term.

TABLE 2—

Coding Scheme Used for the Piecewise Model Using Quadratic Terms: Study of Growth Trajectories of Sexual Risk Behavior in Adolescence and Young Adulthood, Flint, Mich, 1994–2003

	Adolescence				Young Adulthood
Coded Time Variables	Wave 1	Wave 2	Wave 3	Wave 4	Wave 5	Wave 6	Wave 7	Wave 8
a1t	0	1	2	3	3	3	3	3
a2t	0	1	4	9	9	9	9	9
a3t	0	0	0	0	1	2	3	4
a4t	0	0	0	0	1	4	9	16

Note. Study participants were interviewed starting in 1994 every year for 4 years (corresponding to participants’ high school years), followed by a 1-year gap, and then every year for 4 additional years, for a total of 8 waves of data.

We built the level-1 model using a forward stepwise approach. We first estimated a model with linear terms, and in 2 additional steps we added quadratic and cubic terms. In all models, we allowed the intercept and slopes to vary. We assessed which model best fit the data by comparing the χ² deviance statistics. In addition to the deviance statistics, we examined (1) whether the π terms differed from zero, using the t test, (2) whether the variance of each π term differed from zero using the χ² test, and (3) the average λ reliability for each random term.26

After the level-1 growth model was specified, we developed level-2 models with the intercept and slopes as outcomes. The general equation for these models was

(2)

where π_pi is interpreted as in equation 1, b_p0 is the average value for the intercept or slope, X_qi is the person-level variables, b_pq is the change in intercept or slope associated with the person-level variables, and r_pi is the person-level error term. Hierarchical linear modeling determined whether the b terms were different from zero, using the t test. For the models, we entered age, race/ethnicity, gender, and their interactions, to predict each p term in the level-1 model and then used a backward stepwise approach, deleting the person-level variable with the smallest nonsignificant t ratio, until all of the variables and interactions in the model were significant predictors of the respective p term.

RESULTS

Level-1 Model

The deviance statistics indicated that the model with the linear and quadratic terms specified for both adolescence and young adulthood fit the data better than did the model with only the linear terms (χ²₍₉₎ =291.4; P<.001). The cubic model did not improve the fit (χ²₍₁₃₎ =20.2; P not significant), so the final level-1 model included only the linear and quadratic terms. The model suggested that acceleration in sexual risk behavior occurred during adolescence (π₁ =0.01; P not significant and π₂ =0.04; P <.01) whereas growth decelerated during young adulthood (π₃ =0.36; P<.001 and π₄ =− 0.09; P <.001). Except for linear growth in adolescence, all coefficients differed from zero (P<.05). All random effect variances differed from zero (P<.05). The λ reliability for the intercept was 0.63, and those for the slopes ranged from 0.23 to 0.33.

Level-2 Model

The final level-2 model is presented in Table 3 ▶ and Table 4 ▶. The b parameter estimates that modeled sexual risk behavior in ninth grade (π₀) indicated that those who were older, African American or biracial, and male showed more sexual risk behavior in ninth grade than did those who were younger, White, and female. White girls did not report the lower sexual risk behavior associated with being female evident among African Americans, as indicated by the significant interaction term.

TABLE 3—

Final Piecewise Model, With Covariates, for Growth Trajectories in Sexual Risk Behavior in Adolescence and Young Adulthood: Flint, Mich, 1994–2003

Covariates	Coefficient (SE)	Average t ratio	Reliability, λ
Fixed Effects
Mean score at wave 1, π0			0.59
Base, b00	−0.06 (0.06)	−0.95
Age, b01	0.30 (0.05)	6.13***
White, b02	−0.61 (0.09)	−7.04***
Female, b03	−0.44 (0.07)	−6.02***
Female × White, b04	0.42 (0.12)	3.33***
Mean growth at wave 1, π1			0.26
Base, b10	−0.16 (0.07)	−2.49*
Age, b11	−0.05 (0.02)	−2.73**
Female, b12	0.34 (0.08)	4.11***
Mean acceleration during adolescence, π2			0.22
Base, b20	0.07 (0.02)	3.30**
White, b21	0.05 (0.01)	4.50***
Female, b22	−0.07 (0.03)	−2.65**
Mean growth at wave 5, π3			0.31
Base, b30	0.24 (0.06)	4.20***
Age, b31	−0.16 (0.05)	−2.98***
White, b32	0.38 (0.12)	3.06**
Female, b33	0.15 (0.07)	2.08*
Female × White, b34	−0.42 (0.15)	−2.79**
White × age, b35	−0.24 (0.11)	−2.19*
Mean acceleration during young adulthood, π4			.22
Base, b40	−0.06 (0.01)	−4.48***
Age, b41	0.02 (0.01)	1.92
White, b42	−0.07 (0.03)	−2.47*
Female, b43	−0.04 (0.02)	−2.20*
Female × White, b44	0.09 (0.04)	2.61*
White × age, b45	0.06 (0.03)	2.23*

*P < .05; **P < .01; ***P < .001.

TABLE 4—

Variance Components of the Final Piecewise Model, With Covariates, for Growth Trajectories in Sexual Risk Behavior in Adolescence and Young Adulthood: Flint, Mich, 1994–2003

	Variance	df	χ2
Score at wave 1, r0i	0.56	591	1469.2***
Growth at wave 1, r1i	0.35	593	766.7***
Acceleration during adolescence, r2i	0.03	593	761.6***
Growth at wave 5, r3i	0.27	590	867.7***
Acceleration during transition to adulthood, r4i	0.01	590	740.2***
Level-1 error, eti	0.36

***P < .001.

Interpretation of the b estimates that model parameters of growth and acceleration in sexual risk behavior (π₁ π₂ π₃ π₄) required that the predicted values for different groups be plotted (Figure 1 ▶). After we controlled for age, African American or biracial boys participated in the highest levels of sexual risk behavior in ninth grade, followed by African American or biracial girls, with White boys and girls participating in the lowest levels of sexual risk behavior in ninth grade. In adolescence, African American or biracial and White girls increased their sexual risk behavior faster than did their boy counterparts. All groups’ sexual risk behaviors peaked, then declined, during young adulthood, although White men and women continued to have higher levels of sexual risk behavior than did their African American or biracial counterparts.

FIGURE 1—

Growth in sexual risk behavior by race/ethnicity and gender during adolescence and young adulthood: Flint, Mich, 1994–2002.

Notes. Study participants were interviewed starting in 1994 every year for 4 years (corresponding to participants’ high school years), followed by a 1-year gap, and then every year for 4 additional years, for a total of 8 waves of data. Scores were standardized across all waves.

DISCUSSION

Our study is one of the first to describe trajectories of sexual risk behavior in a sample that is predominantly African American and spans the time from adolescence through young adulthood. The finding that sexual risk behavior increased during adolescence extends the work of Capaldi et al.16 and O’Donnell et al.3 by statistically testing changes in sexual risk behavior over time. The results of our study further suggest that overall sexual risk behaviors decreased during young adulthood, despite the frequency of sexual intercourse increasing and condom use decreasing. The decrease in sexual risk behavior may have been because of the development of better decisionmaking skills,5 which enabled young adults to better understand the consequences of sexual risk behavior, or because of the development of longer-term, monogamous partnerships.

This study further builds on Capaldi et al.’s16 and O’Donnell et al.’s3 studies by extending the results to African American males and females, White females, and older youths. The gender differences found in adolescent sexual risk behavior demonstrate the importance of longitudinal study of sexual risk behavior that incorporates several waves of data and allows for nonlinear change in behavior. In 2 cross-sectional studies,29^,30 for example, researchers formed risk groups of adolescents on the basis of frequency of sexual intercourse, number of partners, and condom-use frequency. In 1 of the studies,30 boys were found to be more likely to be in the high-risk group, whereas in the other,29 no difference in the gender distribution was found. The findings of this longitudinal study suggest that the 2 studies may simply have assessed their samples at different points along the adolescent sexual risk behavior trajectory. Newman and Zimmerman30 assessed only 10th grade students, who had a mean age of 15.6 years. However, Murphy et al.29 included students from all grades of high school; participants who had a mean age of 17.5 years. The differences in the ages of participants may explain why these studies are contradictory. In our study, we found that boys exhibited more sexual risk behavior in the early years of high school than girls, but girls overtook the boys by the end of high school. Newman and Zimmerman30 may have captured early gender differences, when boys exhibited more sexual risk behavior. By contrast, by combining all students together across age groups, Murphy et al.29 may have masked the gender differences across time.

The results of the comparison of African Americans to Whites similarly suggest that differences in sexual risk behaviors between these groups are more complicated than has been found in previous research with adolescents.13^,31^–33 We found that African Americans engaged in more sexual risk behavior in ninth grade compared with Whites, which may reflect the finding by Warren et al.33 that African Americans had their sexual initiation earlier than Whites. We found, however, that Whites had a faster rate of growth in sexual risk behavior during high school than did Africa Americans, leading to their eventual surpassing of African Americans by the time they reached young adulthood.

The finding that African Americans engaged in less sexual risk behavior during young adulthood compared with Whites is a unique contribution of this study. These results raise several issues. Although African Americans may participate in lower rates of sexual risk behavior during young adulthood, their overall risk of experiencing negative outcomes is nonetheless greater, as evidenced by the higher rates of HIV and other sexually transmitted diseases among African Americans compared with Whites.20 It is possible that an African American adolescent engaging in a particular risky behavior with another African American is at greater risk than is a White adolescent engaging in the same behavior with another White adolescent, because of the higher prevalence of HIV or other sexually transmitted diseases among African Americans compared with Whites.20^,21 Another interpretation may be that African Americans are more likely to be infected during adolescence, when their rates of sexual risk behavior are higher than their White counterparts.33 Those at younger ages, especially girls, may be physiologically more vulnerable to HIV and sexually transmitted infection, compared with older youth.34

Several limitations of the study should be noted. These results may not generalize to all populations of adolescents because of several factors. The grade point average exclusion criterion may have excluded less-risk-taking youths from our sample.35 By 12th grade, however, the sample’s average grade point average resembled a more normal distribution.36 Because the sample was drawn from a medium-sized city, it may not generalize to adolescents who live in other contexts. Our results may not generalize to youths who drop out of school before age 16, especially homeless or runaway youths. Our White participants were from a majority African American city and attended majority African American schools, and therefore they may not be representative of all White youths. We did not include sexual orientation in the study so we do not know how the results may generalize to nonhetero-sexual youths. Few participants (2.0%), however, described themselves as having had sexual intercourse with someone of the same sex.

Another limitation is that we did not specify the type of sexual intercourse in the sexual intercourse questions, so we do not know how participants may have differentially reported specific sexual behaviors. We do not know if participants reported condom use only for vaginal intercourse, for example, or if they also included oral and anal sexual intercourse when responding to the question. Some researchers have found that a majority of youths do not consider oral intercourse to be sex,37 but do consider anal intercourse to be so,37 suggesting that participants may have included anal but not oral sex in their responses. Because youths may be likely to have had oral sex but not other types of sexual intercourse,38^,39 our measure of sexual risk may have underestimated the actual sexual risks taken by youths.

Another limitation was the assumption that engaging in certain sexual risk behaviors confers risk. As with much research on sexual risk behavior,40^,41 the measure of sexual risk behavior included in this study did not consider participants’ relationship characteristics. Because participants in monogamous relationships were not differentiated from those who were not, we may have overestimated the sexual risk behaviors of the participants. This may be a particular problem during young adulthood, when some participants may have formed stable, monogamous relationships in which neither partner has a sexually transmitted disease, rendering frequency of sexual intercourse and condom use less useful as measures of sexual risk behavior. Therefore, the result concerning a decrease in sexual risk behavior during young adulthood may be an underestimation. Relationships among adolescents and young adults, however, may be particularly unstable, and adolescents’ understanding of sexuality and intimacy particularly may be undeveloped. This may present difficulties in assessing true monogamy status among adolescents and young adults. Thus, the items frequency of sexual intercourse, number of partners, and condom-use frequency likely accurately summarize the sexual risk of adolescents. In addition, few participants reported having been married at wave 8 (9.2%). We therefore likely underestimated the decrease in sexual risk for few participants.

A final limitation is that the measures were based on self-reports. Researchers have found that males tend to overreport and females tend to underreport sexual behavior.42 Because of practical and ethical limitations, however, it is difficult to avoid self-report methods when studying human sexual behavior. The sexual risk behavior items for this study were collected via a self-completed questionnaire, which tends to increase the accuracy of reports.42

These limitations notwithstanding, studies such as ours that focus on trajectories of sexual risk behavior may provide much-needed information for researchers and practitioners. Because of the developmental nature of adolescence, a snapshot of sexual risk behavior among adolescents at a particular point may be insufficient. Investigating trajectories over time allows for differences in starting points and differences in rates of change in sexual risk behavior. These differences may be vital to consider when developing interventions. An understanding of trajectories of sexual risk behavior, for example, may guide practitioners to intervene earlier with some groups than would a snapshot understanding of sexual risk behavior. Trajectories further suggest that some groups may need enhanced intervention to slow the rate of growth in sexual risk behavior among that group.