|Home | About | Journals | Submit | Contact Us | Français|
To estimate the long term (36 month) effects of Project iMPPACS, a multi-site randomized control trial of mass media and small group intervention for African-American adolescents.
Six waves of longitudinal data were collected on program participants. Seemingly unrelated regressions at each wave estimate the effects of three types of mass media messages (“thematic mediators”) on condom use intention and self-reported unprotected vaginal sex events.
The mediators of behavior change that were introduced during the media intervention were sustained to the follow-up assessments at least 18 months after the intervention ended.
Project iMPPACS demonstrates that mass media influence delivered over an extended period, when adolescents were beginning to learn patterns of behavior associated with sex, persisted after the media program ended.
Project iMPPACS (an acronym for the sites of the project: in Macon, Philadelphia, Providence, Atlanta, Columbia, and Syracuse) was a longitudinal intervention for African American youth designed to evaluate the effect of community-wide mass media campaigns to increase condom use and reduce sexual risk-taking associated with HIV and other sexually transmitted infections (STIs). The focus on African American adolescents was justified because they have higher prevalence and incidence of HIV infection than other ethnic groups 1. Although small group prevention programs are effective in enhancing teen’s prevention skills and motivating teens to lower their risk for HIV and other STIs, the impact of such programs, in the absence of booster sessions, diminishes with time 2–4.
Although long-term effects of mass media interventions have been observed for smoking5, less is known about the effects of media interventions directed to adolescents regarding sexual behaviors. Zimmerman et al. tested a brief safer-sex media intervention directed to young adults that only produced short-term effects 6. Project iMPPACS was designed to determine whether a media intervention could influence the trajectory of sexual risk taking during mid to late adolescence. Our previous reports have shown that iMPPACS was effective in producing changes in risk-reducing beliefs and reducing unprotected sex while the mass media campaign was running7. However, these studies did not examine what happened subsequent to the termination of the media exposure. In this report, we examine the effects of the media program on both mediating beliefs and behavior.
The original design of Project iMPPACS was a 2 (sexual risk reduction or a general health promotion intervention) by 2 (media present or media absent) by 5 (time: at recruitment, 3, 6, 12, and 18 months post-recruitment) randomized controlled trial implemented in two northern cities (Providence, RI and Syracuse, NY) and two southern cities (Columbia, SC and Macon, GA). The media program was randomly assigned to one city in each region. Because little is known about the long-term effects of safer-sex media interventions with youth, a subsequent data collection point (between December 2009 and December 2010) 36 months after the intervention began was added to determine the longevity of the intervention.
The iMPPACS team recruited African American adolescents (ages 14–17) in cohorts of 25–30 youth for random assignment to one of two interventions: Focus On Youth, the small-group treatment condition 8 or Promoting Health Among Teens, the control status condition 9. Participants were eligible if they were African American and between 14–17 years old at recruitment. The project recruited 1657 respondents between August 2006 and January 2008 through a variety of channels anchored in our collaboration with community-based organizations including Boys and Girls Clubs and community centers that provided recreational, social, and educational services for young people. We recruited 21% of participants directly in those centers and 29% from participant referral. However, we also recruited youth using street outreach (9%), respondent-driven sampling (15%), and referral from adults in the community (14%). Only 25 (1.5%) of the eligible adolescents refused or were unable to participate in the study. An attempt was made to oversample adolescent girls who were expected to have higher rates of STIs. The recruitment procedures in the 4 participating cities produced equivalent experimental groups at baseline in terms of unprotected sex, lifetime vaginal sex, gender, and age.10
Once recruited, adolescents provided informed written assent (and parents provided consent). Then adolescents completed an audio computer-assisted self-interview to assess their sexual attitudes, beliefs, condoms use self-efficacy, and sexual behaviors. Youth in the media cities were exposed to the media intervention for at least 18 months after their recruitment and participation in the small-group intervention. Television and radio messages were developed through a comprehensive qualitative-quantitative process 11. The media campaign used a total of nine 30-second television spots and twelve 60-second radio clips. The media messages were paid for by the project (e.g., did not run during “donated time” slots when the audience is sparse) and were placed on channels/radio shows and during hours that were popular with African American adolescents 12. The media campaign ended in June of 2009. Additional information on the implementation details of Project iMPPACS can be found elsewhere 12, 13.
Published results for iMPPACS document the effects of STI testing on behavior and/or the combination of intervention exposure and STI positive status 12, 14 as well as short term media intervention effects for STI negative respondents15. Here we examine media effects for adolescents who were not positive for an STI at the baseline test or at any time during the 18-month follow-up period using the complete 36 month data set. We focus on three of the mass media themes directly related to condom use, which we label as “thematic mediators” because it is through changes in causal mediating variables that behavioral change is produced16, 17. The media program was designed to reduce belief that a steady partner is a safer partner (the Selection mediator), that consistent condom use has negative consequences such as reducing pleasure (the Pleasure mediator), and that respondent-partner negotiations about consistent condom use result in negative outcomes (the Negotiation mediator) 12.
Two of the thematic mediators are constructed from items in the Condom Attitude Scale18. The Selection mediator is composed of: “A condom is not necessary if you are pretty sure the other person doesn’t have a sexually transmitted disease”, “A condom is not necessary if you know your partners”, and “A condom is not necessary when you and your partner agree not to have sex with anyone else”. All these items were coded from 1 = Strongly Disagree to 6 = Strongly Agree, so that high values of the index represent a belief that safer sex partners can be identified (at recruitment, the polychoric α was .85, M = 2.03, SD = 1.25). The Pleasure mediator was constructed from five items from the same scale: “Condoms take away the feeling a guy has during sex”, “Condoms are messy”, “Condoms make sex hurt for a girl”, “Condoms take away the pleasure of sex”, “Using a condom takes ‘the wonder’ out of sex”. All these items were coded from 1 = Strongly Disagree to 6 = Strongly Agree, so that high values of the index represent a belief that condom use increases negative expectancies (at recruitment, the polychoric α was .82, M = 2.62, SD = 1.13). The Negotiation mediator was based on seven items derived from results of an earlier study 19. All items used the stem: “If you talked to a potential sex partner about using condoms” and the outcomes were “he/she would respect you more”, “he/she would threaten to leave you”, “he/she would feel more affection for you”, “he/she would swear at you or call you ugly names”, “he/she would hit, punch, or kick you”, “he/she would threaten to break up with you”, and “he/she would feel safer”. The items were coded on a 1 to 6 metric with 1 = “Very Unlikely” and 6 = “Very Likely” and the positive outcome items were reversed, so this scale reflects negative outcomes of condom negotiation (at recruitment, the polychoric α was .85, M = 2.26, SD = 1.06).
From a behavioral theory perspective, these three themes are attitudinal beliefs (i.e., outcome expectations) about using condoms 20. Thus, we also include in our model as an additional mediating variable, the intention to use condoms. This is the average of two different items: “ If I have vaginal sex in the next 3 months, I am willing to use a condom every time” and “ If I have vaginal sex in the next 3 months, I intend to use a condom every time” both coded on a 1–6 metric with 1 = “Strongly Disagree” and 6 = “Strongly Agree” (at recruitment, M = 5.33, SD = 1.29, the polychoric correlation between these two measures at recruitment was .90). Our expectation was that the media program’s effect on behavior (lower unprotected sex events) would be mediated most proximally by intention and that intention would be affected by the thematic mediators 21 .
The behavioral outcome was the (log) number of condom unprotected vaginal sex events (i.e., penile-vaginal sex without a condom) during the follow-up period. At recruitment, the mean was .58 (SD = .80, range from 0 to 4.37). While this measure naturally is only available for “sexually active” participants at each wave of data collection, we do include all the respondents who reported having vaginal sex anytime during data collection in the analysis and control for sexual activity at each wave in the analysis.
Longitudinal data present many analysis alternatives. Growth curves are attractive, but the functional form of time needs to be accurately identified to yield valid results 22. Cross-sectional analyses of the same model repeated at each time period is less complex, but overstate the actual degrees of freedom of the data similar to step-wise regression 23. In contrast, seemingly unrelated regression 24 has neither of these two disadvantages: it estimates results for all waves of the data simultaneously and uses longitudinal data without any assumption of the functional form of time, so this is our preferred approach.
The generic model of the iMPPACS analysis is shown in Figure 1. The first set of parameters (A) shows the treatment effects of media exposure and the small group intervention on the three thematic mediators (for clarity, all the thematic mediators are enclosed in a single box, although strict graphing conventions would require one box for each mediator). Media exposure should decrease the level of all the mediators (after controlling for the effects of the small group intervention) because the mediators as scaled should be positively related to unprotected vaginal sex. The next set of parameters (B) is the effect of each thematic mediator on contemporaneous intention. Again, because of their scaling, all the mediators should be negatively correlated with behavioral intention. The C parameters reflect the correlation between prospective intention and the behavior in the corresponding follow-up data collection period after controlling for the respondent reporting having vaginal sex. This vaginal sex control adjusts for selection bias in the behavioral outcome that is only measured for respondents who report vaginal sex at each data collection wave 25. All three parameters are important because the intervention effects on the mediators are transformed into behavioral change via the model’s indirect process. In other words, to explain the intervention’s effect on behavior, the A, B, and C parameters all need to be non-zero.
To estimate the regression analysis we used Mplus 26 because it uses advanced maximum likelihood estimation methods that eliminate the bias of list wise deletion of observations when missing values are present 27 and also enables estimation of the serial correlation between error terms, an inevitable feature of repeated measure designs. We also weight the data using a constructed propensity score predicting being in the Media intervention sites: this helps to reduce bias due to preexisting differences between the media cities 28.
At recruitment, there were 1139 STI negative sexually active participants divided into four experimental categories (24.14% in non-media/control group condition, 26.16% in media/control group condition, 23.53% in non-media/ intervention group condition, and 26.16% in media/intervention group condition). Fifty-five percent of the participants were females because iMPPACS oversampled females, but the distribution of gender did not differ across experimental conditions (Chi2 (3) = .67, p = 0.88). Fewer females than males reported lifetime vaginal sex at recruitment (males: 67%, female 49%, Chi2 (1) = 35.75, p < .05, OR = .48, CI = .38-−.61). Follow-up rates were high: 78% of the respondents completed all 5 follow-up ACASI surveys and 12.6% only missed one assessment.
Table 1 shows the correlations between the thematic mediators and intention at the recruitment data collection and their correlations with behavior at the first (e.g., 3 month) follow-up period. To be effective mediators for behavioral change, all three mediators should be contemporaneously correlated with intention, and they are in the correct direction. Intention, in turn, has the largest correlation with behavior in the correct direction (i.e., intention to use condoms should be negatively related to prospective reports of unprotected vaginal sex). The mediators are also negatively correlated with unprotected sex.
Average unprotected sex events by media exposure and wave of data collection are shown in Figure 1. Unprotected vaginal sex increases with wave of the study, although it appears that the increases are smaller for respondents in media cities compared to no media cities. The regression analysis presented next connects media exposure to changes in the thematic mediators as well as the mediators to intention and (indirectly) to behavior.
Table 2 has the regression results for all mediators and time period. There are significant reductions in the Selection mediator due to Media exposure for all periods except the last, for the Pleasure mediators for all periods except for 6 months, and for the Negotiation mediator at 6 and 18 months after recruitment. We also find consistent negative effects between the mediator and intention for all time periods for the Selection and Negotiation mediators. As a check on model misspecification 29, we added to the model five direct effects of the Media treatment exposure on behavior: a theoretically unmediated effect that should reflect any source of treatment effect not captured by our measures. All of these parameters were small and failed to attain statistical significance. We also repeated the analysis with a direct effect of the Media treatment exposure on intention: four of these five estimates were non-significant.
The results support the major hypothesis motivating the study: effects on mediators of behavior change that were introduced during the media intervention were sustained to the follow-up assessments at least 18 months after the intervention ended. Thus, the study shows that mass media influence delivered over an extended period, when adolescents were beginning to learn patterns of behavior associated with sex, persisted after the media program ended. Media exposure was able to produce differences in respondent levels of the thematic mediators although the media effect was the strongest for the Selection and Pleasure mediators. All three mediators were (negatively) associated with intention, although Selection and the Negotiation performed the best. The intention/behavior correlation (the C parameter) was always significant even when the temporal lag was as much as 18 months (i.e., between intention measured at 18 months and behavior measured at 36 months, the average number of days between the end of the media campaign and the 36 month assessment was 306 days).
The Selection mediator performed well at all stages of the meditational process. This result is consistent with past research using survey, experimental, and qualitative research 30–32 that demonstrates that adolescents (and adults) use informal rules (e.g., “heuristics”) in an attempt to choose safe sex partners. The media intervention emphasized that one cannot tell how many partners someone has had in the past and therefore that one should always use a condom when having sex. Young people (as well as adults) are particularly susceptible to discontinuing condom use after initiating sex with a new partner for as short a period as a few weeks 33, 34. Nevertheless, even if such partnerships are monogamous, they will not protect one from transmission of STIs if a partner was infected by a previous partner. Project iMPPACS is the first example to our knowledge of a media intervention that attempted to change this problematic practice.
Media effects on the Pleasure mediator were apparent in five of six data periods, but Pleasure had the weakest association with condom use intention. Nonetheless, the Pleasure mediator was an important target in the media intervention because young people often believe that condoms will interfere with sexual enjoyment. Although it was not strongly associated with intention to use condoms apart from the other mediators, the Pleasure mediator may have played a role in supporting the other belief changes that were encouraged by the media program. Youth could be more motivated to use condoms if they are not under the impression that condoms are used by people who don’t care about their negative effects. Indeed, the media program emphasized that condoms actually enhance pleasure because they reduce worry about contracting a STI.
The Negotiation mediator showed the fewest treatment effects but was consistently associated with condom use intention. This mediator is particularly important in the context of a media intervention because it involved expectations about how sexual partners would respond to a request to use a condom. Youth in the media cities could more confidently expect their potential partners to respond favorably to such requests because everyone in the city was likely to have been exposed to the media program. This process of normative change may explain why the Negotiation mediator took some time to show effects of the media program. Youth could more confidently believe that their peers and partners would respond favorably to condom-use requests as the media program continued. They may have been less sure of this possibility in the early stages of the media program.
We restricted the present study to vaginal sex because we did not have a measure of intention for anal or oral sex. Hence, we cannot say that the intervention had long-term effects for those engaging in these forms of sex. Finally, our results are based on self-reports of sexual behavior. However, we used ACASI for all our assessments, and this was likely to decrease reporting bias 35. Hence, the results are likely to accurately reflect the rates of unprotected sex occurring in this population.
The findings from project iMPPACS strongly support the use of media interventions directed to adolescents as a way to produce long-lasting effects on sexual risk behavior. The media program not only increased condom use among the highest risk youth within the first 8 months of its introduction 12 but it also reduced the trajectory of unsafe sex among the broader youth audience as the program continued15 . Finally, it maintained the effects of an STI screening intervention carried out face-to-face in the community 14. Thus, mass media provide an effective way to enhance the durability of interventions carried out on the ground while they independently change sexual norms and behaviors among youth on their own.
This study was conducted through the iMPPACS network supported by the National Institutes of Mental Health (Pim Brouwers, Project Officer) at the following sites and local contributors: Columbia, SC (MH66802; Robert Valois [PI], Naomi Farber, Andre Walker); Macon, GA (MH66807; Ralph DiClemente [PI], Gina Wingood, Laura Salazar, Rachel Joseph, Delia Lang); Philadelphia PA (MH66809; Daniel Romer [PI], Sharon Sznitman, Bonita Stanton, Michael Hennessy, Susan Lee, Eian More, Ivan Juzang, and Thierry Fortune); Providence, RI (MH66785; Larry Brown [PI], Christie Rizzo, NanettaPayne); and Syracuse, NY (MH66794; Peter Vanable [PI], Michael Carey, Rebecca Bostwick).The content of this article is solely the responsibility of the authors and do not necessarily represent the official views of the NIMH.