The Impact of the African Youth Alliance Program on the Sexual Behavior of Young People in Uganda Author(s): Ali Mehryar Karim, Timothy Williams, Leslie Patykewich, Disha Ali, Charlotte E. Colvin, Jessica Posner and Gideon Rutaremwa Source: Studies in Family Planning, Vol. 40, No. 4 (Dec., 2009), pp. 289-306 Published by: Population Council Stable URL: http://www.jstor.org/stable/25593973 Accessed: 17-12-2015 17:11 UTC Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Population Council and Wiley are collaborating with JSTOR to digitize, preserve and extend access to Studies in Family Planning. http://www.jstor.org This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions The Alliance Youth of the African on the Sexual of Behavior Program in Young Uganda People Impact Ali Mehryar Karim, Timothy Williams, Leslie Patykewich, Disha Charlotte E. Colvin, Jessica Posner, and Gideon Rutaremwa Ali, This study evaluates the impact of theAfrican Youth Alliance (AYA) program on the sexual behavior ofyoung people aged 17-22 inUganda. Between 2000 and 2005, thecomprehensivemulticomponent AYA program implemented behavior-change communication and youth-friendly clinical services, and it coordinated policy and advocacy. The program provided institutional capacity building and established coordination mechanisms between agencies that implementedprogramsfor young people. The analysis offindings from both a self-reported exposure design and a static group comparison a on sexual behavior among young females but not design indicated thatAYA had positive impact were at least 13 percentage points more likely to report males. among young AYA-exposed girls a more likely to report that theyhad having used condom at last sex, at least 10 percentage points current at their least 10 used condoms with partner, percentage points more likely to consistently have used contraceptives at last sex, and 13 percentage points more likely to have had fewer sex partners during thepast 12 months, comparedwith girls who were not exposed to theAYA program. Scaling up theAYA program inUganda could, therefore,be expected to improve significantly the sexual and reproductive health of young women. Effective strategies for promoting safer sexual behaviors among boys and young men must be identified,however. (Studies in Family Planning 2009;40[4]: 289-306) are especially vulnerable to the adverse of unsafe sexual relations, including un consequences intended pregnancy and sexually transmitted infections Young people such as HIV/AIDS (Bearinger et al. 2007). Promoting safe sexual behavior among young people is, therefore, consid ered essential to curbing such adverse reproductive health outcomes (Bearinger et al. 2007; Tylee et al. 2007; UNAIDS and WHO 2007). The importance of protecting the health of young people and encouraging and promoting safe sex ual behavior is particularly great in sub-Saharan Africa. Ali Mehryar Karim is SeniorMonitoring & Evaluation and Research Technical Advisor, The Last Ten Kilometers Project, and Disha Ali isResearcher and Evaluation Technical Advisor, John Snow, Inc., Post Office Box 13898, Addis Ababa, Ethiopia. TimothyWilliams and Leslie Patykewich are Senior Technical Advisors, and Jessica Posner isMonitoring & Evaluation Technical Advisor, John Snow, Inc., Rosslyn, VA. Charlotte E. Colvin is TB/HIV Technical Officer, PATH, Washington, DC. Gideon Rutaremwa isHead, Department ofPopulation Studies,Makerere University, Kampala, Uganda. E-mail: akarim@jsi.com. in sub-Saharan Africa one-third of the population is in the 15-24 age group (PRB 2006), and about half of all new HIV infections occur within that age group (Monasch and Mahy 2006). About Recent research consistently suggests that compre programs are more effective hensive multicomponent than narrowly focused programs for increasing safe sexu al and reproductive health (SRH) behavior among young 2000; Focus people (Kirby 1997 and 2001; Senderowitz on Young Adults 2001; et al. Karim 2002; 2003; Magnani Karim et al. 2003; WHO et al. 2006). The development of programs targeting the young is based multicomponent on the notion that the and sexual develop psychosocial ment of adolescents takes place under the influence of overlapping contexts, or ecological systems, within which they live. These contexts include the nuclear family, ex tended family, peer groups, neighborhood, community, institutions such as school or workplace (Brooks and Gunn et al. 1993; Duncan et al. 2001). Comprehensive mul combine strategies to address ticomponent programs simultaneously myriad individual and contextual factors that influence safe sexual behavior. For example, inmul ticomponent programs, interventions such as life-skills Volume 40 Number This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions 4 December 2009 289 are aimed at improving the negotiation and practice of safe sex, and youth-friendly services at clin ics are designed to augment contextual factors such as peer relationships, social ex parental communication, pectations, and institutional support that directly or in education directly influence the sexual behavior of young people. Moreover, multicomponent programs provide informa tion through multiple channels, including schools, insti and tutions, peers, the mass media, reaching different target groups and reinforcing positive messages. They increase the likelihood that people will recall and act on the information they provide (Piotrow et al. 1997; Jato et al. 1999; Kim etal. 2001). con to a growing body of knowledge Responding SRH for programs young people, the Bill & Me cerning linda Gates provided funding for the Afri (AYA) program in 2000. AYA was Foundation can Youth Alliance a comprehensive multicomponent prevention program to and encourage promote designed healthy SRH be among people aged 10-24 in four sub-Saharan countries: Botswana, and Uganda. Ghana, Tanzania, was a The project partnership managed by the United Nations Population Fund (UNFPA), Pathfinder Interna haviors in tional, and the Program forAppropriate Technology Health (PATH), drawing upon their expertise and expe rience in implementing SRH program components for in developing countries. At the country young people AYA collaborated with level, public and private organi zations to implement its six components, which included to the evaluation, two types of study designs with dif ferent sets of assumptions?the self-reported exposure design and the static group comparison design?were implemented simultaneously to draw conclusions about the impact of AYA on the desired outcomes. This study describes the analysis conducted inUganda to illustrate the impact of the comprehensive AYA program on SRH behaviors of the young and its potential implications for inUganda, in other countries inAfrica, such programs and elsewhere. HIV and Preventive Epidemic inUganda Measures Uganda faces a generalized HIV/AIDS epidemic inwhich the predominant mode of transmission is heterosexual contact. The epidemic peaked in Uganda in the early in 1990s; HIV seroprevalence among pregnant women urban areas was estimated cent. HIV to be as high as 25 to 30 per to recent has since declined seroprevalence levels of 10 percent forurban adult populations, with the national average currently estimated at 6 percent (MOH [Uganda] and ORC Macro 2006). Much of the decline has been attributed to behavioral changes, specifically to de lay in sexual initiation and reduction in number of part ners (Asiimwe-Okiror da AIDS Commission et al. 1997; Kilian et al. 1999; Ugan 2003 and 2005). High mortality among those infected in the early stages of the epidemic is an important influence on the overall thought to have had innovative behavior-change communication (BCC) pro clinical services; integration of grams; youth-friendly decline building among in-country implementing partners; and establishment of coordination mechanisms among imple ly increased since then (Mukuria et al. 2005). in response to the The change in sexual behavior HIV epidemic inUganda is attributed to the leadership SFxH interventions with livelihood-skills training; coordi nation of local and national policy and advocacy activities for SRH of the young; provision of institutional capacity menting agencies. impact of the comprehensive program on SRH knowledge, The multicomponent and perceptions, behaviors of young people was assessed inGhana, Tan zania, and Uganda. (Botswana was not included in the assessment because of budget and timing constraints.) As a result of the complexity of the partnership among AYA in imple international and local organizations theAYA strategy, the geographical coverage of multiple menting the comprehensive AYA program did not correspond the expected program coverage during adequately the baseline survey. Therefore, appropriate preinterven tion data were not available for the evaluation, and the with team had to choose from post-test-only study In order tomake best use of the postintervention evaluation designs. data available 290 and to apply rigorous analytic approaches in prevalence (Wawer et al. 1997; Low-Beer 2002). Some research suggests that condoms did not play a large role in the initial decline, although condom use has steadi in promoting a national campaign of the government to promote primary and secondary sexual abstinence, mutual fidelity between married or cohabitating part ners, and condom use. This "ABC" strategy was further expanded prevention to include voluntary of mother-to-child counseling transmission, and testing, antiretrovi ral therapy, care, and support (MOH [Uganda] and ORC Macro 2006). The national policy environment for SRH programs foryoung people is an enabling one. The government has on a number of policies for developed youth that focus health (particularly sexual and reproductive health and HIV/ AIDS), gender, and education (K2-Consult Uganda Limited 2001). Several programs and organizations sup port the government's priorities in this area and imple ment various components of youth-service Studies inFamilyPlanning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions interventions and (African Medical and Research Foundation-Uganda Secretariat Unresolved AIDS Commission 2001). Uganda issues remain, however, and require attention. Currently, HIV seroprevalence among those aged 15-24 is estimated at 3 percent; young women are estimated to be nearly four times more likely than young men to be infected (MOH exposure [Uganda] and ORC Macro 2006). Mass-media and other aspects of modernization have reportedly re formany laxed social controls over sexual behaviors (Bohmer and Kirumira 1997). Surveys young Ugandans are less informed show that out-of-school adolescents about sexual matters, have fewer resources available to them, and are more likely to engage in risky sex than their schoolgoing counterparts (Bohmer and Kirumira 2000; et al. 2004). Young women whose sex part Ndyanabangi ners are older than they are often feel powerless to insist on condom use (Luke 2005). Most young Ugandans know but many do not perceive themselves about HIV/AIDS, to be at personal risk of becoming infected (Hulton et al. 2000). Even those with substantial knowledge ofHIV and sexually transmitted diseases report that they sometimes engage in risky sex (Sekirime et al. 2001). menting partners and government officials at all levels. The institutional capacity-building component aimed to and technical ability organizational strengthen partners' to sustain SRH programming among young people by providing them with general and intensive technical as sistance. The youth-friendly services component aimed to increase the use of high-quality SRH services by estab lishing youth-friendly health facilities, extending out reach service, establishing peer providers of services, and institutionalizing an appropriate SRH service curriculum in theMinistry ofHealth in-service training (see Tylee et al. 2007 for details). The behavior-change communica tion (BCC) component, including life-planning skills and "enter-education" (educational entertainment activities designed to reach young people), is intended to increase skills, norms, and positive attitudes toward knowledge, of safer sexual practices through in-school and adoption activities. The AYA component designed to integrate SRH programs with vocational training was not implemented because of local priorities (for further out-of-school details see AYA 2005a, 2005b, 2005c, and 2007; Daniels to 20 of 2007). Between 2001 and 2005, AYA expanded some 69 districts. of the of coverage Uganda's Although activities, such as BCC using mass media, was the coverage of all five components of the nationwide, program was limited to eight districts. the AYA The African Youth Alliance Program in Uganda program inUganda has com and scaled up many of the existing SRH inter plemented ventions foryoung people thatwere implemented by the donors and stake government and by nongovernment The African Youth Alliance thus expanding their scope and coverage. Ac inUganda partnered with AYA secretariat the cordingly, holders, 35 local implementing partners and agencies (public and and private), and with community-based organizations to various the implement components religious groups of its program. Each partner worked on implementing a specific component while other AYA partners. coordinating its activities with Five of the six components of theAYA program were The policy and advocacy co implemented in Uganda. ordination component of AYA aimed at promoting an enabling national and local environment for SRH pro gramming, which included mass-media campaigns and and stake members, reaching young people, community holders through networking activities, workshops, and student essay competitions activities. The coordination nent was and debates, among other and dissemination compo aimed at establishing a mechanism within the government structure thatwould ensure integration of AYA program components by sharing partners' work and and plans by networking collaborating among imple Framework Conceptual Evaluation for the Figure 1 illustrates the conceptual framework motivat ing this evaluation. The framework is based on the theory adapted by AYA that the sexual and psychosocial devel of adolescents takes place under the influence opment of individual tors are and contextual expected to influence factors. The contextual SRH behavior and fac its an tecedents (knowledge, attitudes, and self-efficacy) that are presumed to act as precursors to behavior change. The framework indicates thatAYA interventions affect ed these antecedents directly by interacting with young people indirectly by influencing the context within which they live, as well as by enhancing established SRH or programs for the young. The antecedents are assumed to influence behavioral outcomes such as abstinence, reduc tion of the number of sex partners, and condom or other modern contraceptive use. Finally, improved SRH behav ior should logically contribute to improved health con ditions among the young. This framework is consistent with the health belief model (Rosenstock 1974; Janz and Becker 1984), the social cognition model (Bandura 1986), and other models of health behavior. Volume 40 This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions Number 4 December 2009 291 Figure 1 Conceptual frameworkforthe effectof theAfricanYouth Alliance (AYA) program on the sexual behavior of young people_ Adolescent sexual and reproductive health (ASRH) Contextual program interventions factors / Individual/\ / \ / commun.ty / / \ Peers/partners \ \ - Fam.ly/ household / Behavioral Antecedents \Age at firstsex r/ \ Knowledge Number of \ sexpartners Attitudes -^ / Self-efficacy Modern \ Two variants of post-test-only evaluation designs were implemented for this study to assess the impact of AYA on the sexual behavior of young Ugandans: the self reported exposure design and the static group compari son design. The findings from the two post-test-only for the designs were synthesized to draw conclusions assessment. The self-reported exposure AYA-impact extent towhich safe SRH behavior measured the design in the improved among young people who participated AYA program, compared with the behavior of those who were not exposed to it.The static group comparison de measured the extent towhich safe SRH behavior im sign proved among young people who were living in the areas where the AYA program was implemented, compared with the outcome foryoung people in comparison areas. threats to the validity of the self-reported exposure design need to be mentioned. First, the report ing of exposure toAYA is nonrandom and, therefore, is Two major -^ use I Condom because encouraged resulting in an overestimation This study uses a simultaneous to address nonrandom endogeneity healthy sexual behavior, of the program's impact. (SEM) equations model or selection bias associated with reporting of program exposure. Second, ob a measure of self-reported exposure is com valid taining 292 Studies in Family \- Incidence of / \ coercedsex/ exposure toAYA and implementation of this evaluation (AYA field activities ended inNovember 2005, whereas the data collection for this evaluation was conducted in 2006). These challenges inmeasuring respondents' recall were of AYA-specific met by the meticulous exposure design of the data-collection detail below). instruments (described in The results of the static group comparison design are included here because this design is based upon a dif ferent set of assumptions from those of the self-reported exposure design. The intervention (or treatment) areas are the eight districts where all components of theAYA were implemented. The static group comparison design on the assumption that all of the young people in the intervention areas had been exposed to the program; nomic similarities between that promote I \ abortions/ "branding" quently, young people and other community members were not likely to recall the association of SRH activities with AYA. The validity of the self-reported exposure measure is further complicated by the time lag between is likely to influence respondents' personal motivation the practice of healthy behavior, including safe sex. Some respondents may have been likely to self-select into pro like AYA pregnancies Unsafe local capacity, which promoted the implementing partners to use their own of AYA-supported interventions. Conse works grams I AYA not a true independent variable. Itmay be endogenous in terms of the expected sexual behavior outcomes; that is, a third factor (unmeasured or unobserved by the survey) may have influenced the reporting of exposure toAYA and the reporting of sexual behavior outcomes simultane ously, leading to biased impact estimates. For example, ,\ / ? \ / Reduced: \ ' HIWSTI rateS / *Unwanted \\ / use / contraceptive plicated Methodology Long-term ASRH outcomes outcomes therefore, poor coverage of the program would potential the impact ofAYA. Because of regional lyunderestimate ethnic (including language) variation,1 comparison areas were selected purposively to ensure ethnic and socioeco the two groups. Consequent data the from ly, comparison areas could be contaminated because of the areas' proximity to each other. threat to the validity of the static group design is nonrandom program placement? The major comparison that is, program placement unobserved factors?which in an unknown having been determined by could bias impact estimates direction. The program-impact estimates from the static group comparison would Planning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions be biased if the program areas is associated with the level of safe sexual behavior of the youthful population. selection of AYA Such a possibility exists because the comprehensive AYA program was intended to target areas with few or no SRH foryoung people and where appropriate local if implementing partners were available. Accordingly, AYA expanded to all areas where the level of safe sexual programs behavior among young people is low owing to the lack of youth-friendly SRH services, the source of informa tion for the comparison group would most likely be those areas where the sexual behavior among young people is safe. The program impact for such areas to be negative or low even where ithad comparatively would appear had a positive if the appropriate only in areas with impact. Alternatively, implementing agencies were available longstanding SRH youth programs, the preintervention levels of sexual behaviors among young people would be higher in the intervention areas than the comparison areas, resulting in false estimates or overestimates of pro gram effect. In order to reduce bias due to nonrandom program placement, nique was the propensity score matching tech used. Data The sampling strategy for this study was devised tomeet requirements for the static group comparison and self reported exposure designs and to allow the analysis to be stratified by sex. Six districts were selected to serve as the areas for the eight intervention districts. The comparison ethnic (including language) and socioeconomic compo sition of Bushenyi district was considered similar to the Kabarole and Kasese enyi district was Similarly, Hoima intervention districts; thus, Bush selected as a comparison area forboth. district was area for theKyenjojo was as selected the selected as the comparison intervention district; Kamuli district area comparison for the Iganga and intervention districts; Bugiri was selected as the Mayuge intervention district; and comparison area for theMbale Kumi was selected as the comparison area for the Sironko intervention district. No districts were trict.Because the Rubaga socioeconomically comparable found for the Kampala intervention dis the integrated AYA division of Kampala, sion of Kampala program was limited to however, another divi was selected as the comparison area for the Nakama division. The sample size Rubaga: namely, for the intervention areas was determined purposively to be 1.5 times the sample size of the comparison areas in order to obtain an adequate number of self-reports of ex posure toAYA. Institutional review board approval for the study was obtained from theUganda National Coun cil of Science and Technology. to 17-22 study limited the target population were unmarried or who had been married year-olds who within the past two years, thereby reducing the sample The size requirement to half of what would have been re impact had been evaluated according quired ifAYA's to the three age groups that were targeted by the AYA interventions decision married The (10-14-, 15-19-, and 20-24-year-olds). to limit the sample size to unmarried or recently is justified because the young 17-22-year-olds and unmarried population reached by AYA during 2000 to 2005 had aged, and some may have been married dur ing the time interval between their exposure toAYA and this evaluation. The 17-22 age group is also considered to be considered as a single sufficiently homogeneous age group. Two-stage cluster sampling was employed to obtain the required sample size. In the first stage, 86 enumera tion areas (EAs) from the intervention areas and 57 EAs from the comparison areas were randomly selected with to the size of the EA. In the probability proportionate second stage, ten households with eligible males and ten households with eligible females were systematically and randomly selected from the chosen EAs. All eligible young people from the selected households were inter viewed using a structured questionnaire translated into the local language. A household questionnaire was also completed ifan eligible young person from the household was interviewed. All interviewers obtained informed consent from the heads of household to conduct the interviews, from the respondent to conduct the individual interviews, and from parents of 17-year old respondents. for all questionnaires Respondents household were interviewed by same-sex interviewers. The survey took place between April and June 2006. Interviews were completed with 1,548 males (633 from comparison areas and 995 from intervention areas), 1,628 females (615 from comparison areas 2,732 household Measurements The and 933 from intervention areas), and respondents. and Statistical Analyses five sexual behavioral outcomes among for the young people impact evaluation of theAYA: delay of sexual initiation (defined as reporting initiation of sex after 16 years of age); fewer sex partners following are considered (defined as reporting having fewer than two partners in the past 12months); increase in condom use (defined as reporting condom use during last sex); consistent condom use (defined as reporting always using a condom with the current sexual partner); and increased use of contracep tives (defined as reporting use of a modern method of contraception during last sex). Volume 40 Number 4 December 2009 293 This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions As discussed sure toAYA was above, measuring self-reported expo To increase the validity of complicated. than those areas with comprehensive consequence of all categorizing AYA. The possible respon comparison-area the self-reported exposure variable, a precise inventory of AYA-supported activities was developed tomake the dents as not exposed toAYA will be underestimated. by implementing partners but funded only or mainly by AYA (such as mass-media programs and youth-friendly services), as well as activities (such as peer education and funded by AYA and other donors. The enter-education) of interest), the following simple probit model can esti mate the unbiased program effect from the self-reported exposure design: as specific as possible. questionnaires Exposure ques tionswere constructed to capture all activities carried out questionnaire contained questions about 11AYA-specific activities: radio programs, newsletters, peer education, youth-friendly services, youth or Red Cross camps, life in school, and the five enter-education skills planning programs (for details, see JSI2007, page 56). Some "false interventions" (for example, incorrectly named radio shows) were included to further test for validity. Any or fully funded by AYA) (partly activities AYA-specific in the intervention areas were considered to be within the overall AYA strategy that built local capacity and coordinated linkage between all local SRH programs for young people. The assumption was made that three or more channels of communication behavior-change were required for theAYA program to have had an impact on behavior; thus, respondents in the intervention areas who could recall three ormore AYA-specific activities were de fined as having been exposed to the integrated AYA pro gram or as having had high exposure toAYA. Interven tion-area respondents who reported exposure to one or two AYA-specific activities were classified as having had some exposure to AYA. All respondents living in com areas and those parison respondents in intervention areas who did not report exposure to any AYA-specific activi tieswere classified as not having been exposed toAYA. comparison of the self-reported exposure com of this evaluation was between those respondents ponent to three or more AYA activities and those classi exposed The main fied as "not exposed." Comparisons of program outcomes no exposure and some exposure and between some exposure and high exposure (used forobserving the between dose-response relationship) were not performed, largely because of a lack of statistical power to detect such partial effects due to incremental program exposure. areas may have in the comparison Young people to one or more AYA-specific activities been exposed ser as or mass-media (such campaigns youth-friendly were funded Neverthe that other vices) by agencies. less, all comparison-area respondents were classified as not exposed to AYA because the SRH interventions for young people in those areas were not integrated as were AYA interventions the comprehensive multicomponent and, therefore, were 294 Studies in Family considered less effectively exposed is that the impact of AYA In the absence of selection bias or endogeneity (that is, if the reporting of exposure toAYA is not confounded by unmeasured variables that also influence the outcome (i) Vi.r^i^y^^^y where y2 is the propensity of the outcome of interest (safe sexual behavior) for individual i from cluster;; xt is a vec tor of exogenous explanatory variables (that is, respon dent's such characteristics, as age, marital status, educa tion, religion, and geographical region); J3is the vector of coefficients estimating the effect of x2 on y2; y2 is an indi cator variable defining exposure toAYA; 7 is the effect of on u is the error term that includes the variance y2 y2; and of y3 unexplained fac by x2 and y1 (that is, unobserved tors such as exposure to other SRH programs that also assumes that i is independent influence y2). The equation and that there are no unobserved confounders (that is, that u is not correlated with x2 or We know that / is y2). not independent but is conditional upon being within j (cluster), which implicates variance estimates, and that u as may be correlated with y2 if the latter is endogenous, suspected. Taylor series linearization is used to correct for the nonindependence of i?that is, the cluster-survey effect design (StataCorp 2005). Ify2 remains endogenous after controlling forxv however, equation (1) gives biased and inconsistent 7(impact of AYA on y2).2 The simultaneous equations model (SEM) allows us to testwhether y2 is endogenous in equation (1). If y2 is in equation (1), u would be correlated with endogenous error term the (e) of equation (2) that predicts exposure to comprehensive multicomponent AYA: = + + E, (2) y2ij x1jp x2ja x2 is an instrumental variable (IV)3 or a vector of IVs, 0.05) when it is added to equation (1). Hutchinson tisticalmethods and Wheeler (2006) discussed several sta that are available to control forbias due to program placement in the static group com parison design. Many studies have used single-equation multiple regression to approximate a randomized clinical nonrandom trial (a quasi-experiment) when program placement was nonrandom by controlling for individual-level charac teristics and other observed confounders (for example, see Kincaid et al. 1993 and 1996; Van Rossem and Meek ers 2000). Recently, the propensity score matching (PSM) and Rubin (1983) has technique described by Rosenbaum gained popularity as an alternative technique to single forprogram-impact evalu equation regression methods ation. It compares outcomes between exposed individuals who are to the program and other individuals who are not likelihood of being ex exposed posed is similar (Babalola and Vonrasek 2005; Yanovitsky et al. 2005; Kincaid and Do 2006). Application of PSM for to the program but whose two or more nonequivalent group comparisons has also been applied (Kaplan 1999; Dehejia and Wahba 2002; Guo et al. 2006). One of themost efficientmethods for account ing for nonrandom program placements in communities has been described The method models by Angeles and his colleagues (1998). employs integrated simultaneous equations with multilevel (or random-effects) regression to account for bias due to nonrandom program models in order to demonstrate the impact of the placement and of timing availability family planning programs in communities gram impact, themajor assumption of the PSM method or at least is that confounders are known and measured are measured. Moreover, highly correlated with those that the static group comparison design assumes that the lev els of sexual behaviors that existed before the AYA pro gram was implemented were similar among with similar background characteristics. The major advantage of PSM over normal regression is that it is not sensitive to variable specifications and is free of parametric distribution assumptions (Conniffe et al. 2000; Imbens 2004; Moffitt 2004). The advantage of PSM over conventional one-to-one matching is that the latter is limited by the difficulty in finding an appropri ate match between intervention and comparison groups when the number of covariate patterns of the confound ing variables increases. Instead ofmatching according to the covariate patterns of the confounders, the PSM meth od suggests summarizing preintervention characteristics of each subject into a single-index variable (that is, the propensity score) for the purpose of selecting appropriate matches (Hutchinson and Wheeler 2006). The propensity score of program placement single-equation list of predetermined known confounders as the program-placement by using a that employs a factors and all is firstobtained probit regression model sociodemographic predictors. The goodness-of-fit of is considered ad probit model test when the for equate balancing property is satisfied. The test for balancing property (using Stata's "pscore" procedure) ensures that the distributions of the predic tors of the program-placement probit model were not > 0.01) different between the intervention significantly (p and comparison-group subjects with similar propensity scores. If the balancing property is not satisfied, the pro scores are re-estimated pensity using a different specifica tion of the predictors until balance is achieved (fordetails, see Becker and Ichino 2002). Nevertheless, the choice of predictors in the program-placement probit models re the same as in the SEM models, in large part. The propensity score is used tomodel intention to treat effect mained (ITTE). ITTE is a measure of the weighted ferences in the outcome variables between and comparison-group matched on individual-level fertilityoutcomes. This not for the static group com methodology adapted of this parison design study, however, because the de tailed measurements of the timing and determinants of individuals subjects across pairs or blocks or strata. average dif intervention propensity-score was AYA program placements in a community were not avail able. Because the PSM design has been gaining popular ity,we used themethod to analyze AYA program impact from the static group comparison design. However, like the single-equation regression models used to assess pro Results The characteristics of the study respondents according to intervention and comparison areas are presented inTable 1. The proportion of respondents aged 17-18 (48 percent) is almost three times higher than the proportion of re spondents aged 21-22, largely because those who have Volume 40 Number 4 December 2009 295 This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions Table 1 Percentage distributionof survey respondents, by selected background characteristics, according to sex and to interventionand comparison areas, Uganda, 2006_ Female Characteristic Intervention Male Intervention Comparison (933) (N) (615) (995) Total Comparison . (633) (3,176) Age group 17-18 49.6 45.340.8 47.9 32.0 19-20 35.4 35.037.6 34.7 13.421-22 15.0 54.6 19.8 21.6 17.3 25.5 29.2 27.3 Able to read newspaper or letter 48.1 Easily With28.3 difficulty Not at all 23.6 50.0 49.147.3 39.8} 26.8 1* 33.3; 24.5 21.625.4 In-school status Never attended school 5.4 7.8 or Out of school forfiveyears Left school during past five more 4.6 5.1 5.5 17.015.1 14.5 12.8 14.8 28.833.5 34.1 29.7 31.1 years 48.8 Currently in school 43.6 52.9 46.3 48.6 20.1 Currently married 26.2 8.9 15.8** 16.9 Religion Catholic 34.5 35.4 35.9 38.3Anglican Pentecostal 38.1 39.3 7.8 \ 35.7 37.4 41.4 I 39.2 13.2 5.2 10.1 >* 8.5 Muslim 15.8 9.615.6 Other 3.6 Attends church every day 10.5 since Living inthe area more Traveled forone month or last 14.9 8.8 birth 8.9 I 13.1 2.2/ 3.5 3.7 4.0 11.2 11.7 65.1 68.5 83.2 79.1 73.7 29.6 year Ever worked forwage/salary 21.9 29.8 33.8 33.8 31.8 28.6* 39.6 36.2 31.2 Residence 11.9 Kampala Other urban 13.9 74.2 Household-asset 11.5 11.4 11.1 11.5 7.213.4 Rural 81.3 75.383.9 5.1 10.7 77.8 quintile 21.8 Poorest 33.8 15.3 Poorer 15.9 15.4 Middle 15.1 Richer 24.5 17.2 22.2 22.9 Richest 17.9 21.3 \ 27.4 39.0 14.1 16.3 15.5 19.0 19.0 >** 17.2 12.5 20.0 21.2 15.5/ 19.9 Education of household head 12.4 None 17.019.3 16.4 50.9 Primary 22.8 Secondary 49.1 13.8 Higher 17.2 17.2 48.7 16.0 52.0 50.1 21.9 17.4 20.4 11.4 12.4 13.5 Listens to radio weekly_84^3_806_89^_91?_86.8 *Significantat p < 0.05; **p< 0.01. formore than two years are likely to be in the older age group and to be excluded from the sample because they were not the principal target of the AYA been married program. About three-fourths of the respondents have been living in the same place of residence since birth, giving them the opportunity to be exposed to local youth programs at different stages of their lives. Most (80 per cent) of the respondents are either currently in school or school during the five years preceding the survey, giving them the opportunity to be exposed to in school life-skills planning programs. The literacy level of have attended the study population (75 percent able to read) is the same as the national average for that age group. The distribu 296 Studies in Family tion of respondents by religion is 39 percent Anglican, 9 percent Pen 36 percent Catholic, 13 percent Muslim, 4 These and other. tecostal, percent proportions are also not very different from the national levels of 35 percent, 42 percent, 11 percent, 8 percent, and 3 percent for those religions, respectively (as reported inUBOS and Macro International 2007). The urban-rural residence status of the study respondents is slightly different from the na tional average: 78 percent of the study respondents are from rural areas, which is lower than the national aver age of 83 percent. The crude differences of the respondents between in background characteristics the two study arms, stratified Planning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions sex and adjusted only for the cluster-survey design ef by fect,are obtained using design-based Pearson's statistics (using the two-way tabulation function of survey data implemented by Stata). Among female respondents, the characteristics that are significantly differ background ent between the intervention and comparison areas are forwages. The female ability to read and ever worked are more likely to areas respondents in the intervention be able to read (76 percent in the intervention areas ver sus 67 percent in the comparison areas) but less likely to have worked forwages (22 percent in the intervention areas versus 29 percent in the comparison areas). Among are signifi respondents, the characteristics that and different between intervention the comparison cantly areas are marital status, religion, and household-asset themale quintile. Compared with those in the comparison area, males in the intervention area are less likely to be mar ried (16 and 9 percent, respectively), less likely to be Pen sified as having high exposure to AYA. High exposure toAYA was found to be at similar levels among females in the intervention areas (36 percent and 38 and males percent, respectively). Only 13 percent (242) of respon to dents in the intervention areas report no exposure AYA activities. Our analysis of self-reported exposure to theAYA program excludes the 459 female and 516 male as respondents who are categorized having low exposure toAYA, comparing only the no-exposure group and the high-exposure group along the outcomes of interest. (Be low, the high-exposure category is referred to simply as to AYA.) exposed The magnitude of exposure toAYA activities among those categorized as exposed toAYA is shown in Table all of those who were exposed toAYA report at least one enter-education program (94 per attending cent among females and 93 percent among males). Enter 3. Almost tecostal (10 and 5 percent, respectively), more likely to be (9 and 16 percent, respectively), and more likely to be in the poorest household-asset quintile (39 and 21 Muslim percent, respectively).4 The observed differences in the characteristics of the respondents between background the intervention and comparison areas could also be as sociated with safer sexual behaviors. Therefore, the influ ence of background variables was subsequently adjusted in the PSM technique and the SEM to isolate the effect of theAYA are classified as having low exposure toAYA, whereas 22 percent (338) and 23 percent (373), respectively, are clas program. Exposure toAYA The patterns of reporting exposure toAYA indicate that the program reached a wide range of young people and that a significant portion were reached through multiple channels. For the self-reported exposure design, as indi cated above, all respondents in the comparison areas were classified as not exposed toAYA (see Table 2). According ly, of the 1,548 females and 1,628 males in the two study arms, more than half (51 percent of females [797] and 55 percent ofmales [889]) report having at least some degree of exposure toAYA. Thirty percent of the females (459) and 32 percent (516) of themales across the two settings activities combined funwith the promotion of SRH messages. Five kinds of enter-education ac were tivities presented in different parts of the country: education AYA's drama groups, debates, puppet shows, sporting events, and youth clubs. The enter-education and the other AYA to young people on sev interventions provided messages en SRH topics: pregnancy prevention and reproductive condom use to prevent sexually transmitted in fections, including HIV/ AIDS; having sex with only one partner (practicing fidelity); postponing sex or practicing health; abstinence; knowing about HIV/AIDS; knowing about transmitted and (venereal) diseases; sexually undergoing voluntary counseling and testing forHIV/ AIDS. After the enter-education activities, the most fre mentioned is quently activity reading AYA-supported newsletters such as Straight Talk or Young Talk (81 per cent among females and 80 percent among males). Ex radio programs is the third posure to AYA-supported most commonly reported component of the program towhich respondents were exposed (75 percent among both females and males). AYA aired at least eight radio programs?Voice ofTooro, Twogele Kaati, Ngalaletap Manta, Erikania, Tusheeshuure, KhuKhanikha Lubuula, Einer Eitene, Table 2 Percentage distributionof survey respondents, by exposure to theAYA program, according to sex and to intervention and comparison areas, Uganda, 2006_ Female Intervention Exposure None to AYA Percent (n) (136) Low 14^6 49.2 High 36.2 (338) Total_100.0 Male Total Comparison (459) (933) Percent (n) 100.0 0.0(0) 0.0 100.0 (0) (615) Percent 48^5 (615) 29.7 21.8 100.0 Intervention Total Comparison (n) Percent (7510 106 (106) 100.0 (459) 51.9 (516) 0.0 (338) 37.5 (373) 0.0 (1,548)_100.0 (n) Percent (995)_100.0 (n) (633) 31.7 (0) Percent 45^4 (n) (739) (0) (516) 22.9 (373) (633) 100.0 (1,628) Volume 40 Number 4 December 2009 297 This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions Percentage of female and male survey respondents self-reportingexposure toAYA activities, by typeof activity, Uganda, 2006_ Table 3 AYA intervention Female = 338) (n Listened toAYA radio program Read AYA newsletter 80.8 75.1 Male (n = 373) percent among males). The expected effect ofAYA on the behavior "always use condom with current partner" is 75.4 80.2 Visited youth-friendlyclinic 17.6 21.2 28.5 27.3 Peer-educator interaction only for females, however. Thirty-five percent the of females who were exposed toAYA report that they observed 20.2 22.6 Life-skills planning program (in-school) Life-skills planning program (out-of-school) Engaged inat least one enter-education activity 12.1 10.1 93.8 92.5 Debate Puppet show 59.6 5.5* Sport Club 63.1 10.9 40.0 44.1 34.5 33.9 Drama_1^2_12.2 *Significantat p < 0.05. Note: The sample size forthisanalysis is limitedto those who reported exposure to three or more AYA-specific activities in the interventionareas. and Twogele Lwatu?that provided information to young seven SRH topics. people about the Smaller but still substantial proportions of respon dents report being exposed to AYA activities that com messages through more interpersonal mecha nisms and for longer durations (such as interactions with peer educators), an approach expected to have a greater retention. Among those exposed to potential formessage municated interactions with peer educators are reported by 29 percent of females and 27 percent ofmales; exposure to AYA, youth-friendly clinics is reported by 18 percent of females and 21 percent ofmales; exposure to in-school life-skills programs is reported by 23 percent of females and 20 and exposure to out-of-school life-skills programs is reported by 10 percent of females percent ofmales; planning and 12 percent of males. Other than one enter-education toAYA activities is program (the puppet show), exposure not significantly different between females and males. Crude Analysis of Effects ofAYA The effects of theAYA behavior?observed tials in respondents' program on young people's sexual as unadjusted and crude differen sexual behaviors according to pro gram exposure by respondent's sex?are shown inTable 4. The crude analyses of the self-reported exposure de sign, adjusting only for the cluster-survey design effect, indicate that the AYA program has an impact on young sexual behaviors, as expected. The analysis in people's dicates that condom use at last sex is significantly higher among those who reported that they were exposed to (54 percent for both sexes), compared with those who reported that they were not exposed (27 percent among females and 42 percent among males). Similarly, AYA 298 Studies contraceptive use during last sex is also higher among those who were exposed toAYA (59 percent among fe males and 61 percent among males), compared with those who were not exposed (34 percent among females and 46 always use a condom with their current partner, which is about 20 percentage points higher than the proportion of those who were not exposed toAYA. No other effects among young people are in the crude analysis of the self-reported expo sure design. The observed crude relationships between sexual behavior exposure to AYA and young people's of AYA on sexual behaviors observed may be biased and inconsistent, however, because of the characteristics differentials in respondents' background between the exposure categories or due to selection bias (that is, endogeneity), or both. Therefore, the unbiased effect of AYA on young people's sexual behavior is ana lyzed below using SEM. in Table 4, ad The crude static group comparisons for indicate that AYA the survey design, justing only had a significant impact on females but not on males. The analysis indicates that, as expected, the young fe males in the intervention areas are more likely to report using a condom at last sex (44 percent versus 27 percent in the comparison areas), always using a condom with their current partner (27 percent versus 15 percent in the a contraceptive at last sex comparison areas), and using (50 percent versus 35 percent in the comparison areas). in the intervention areas are less Female respondents sex partners during the 12 likely, however, to have fewer months preceding the survey (79 percent) than are those areas (87 percent), which in the comparison indicates either thatAYA promoted having more sex partners or that the program targeted the geographical areas charac terized by more risky sexual behaviors. Similarly, male were less likely to respondents in the intervention areas 16 initiation until sexual (75 percent) than were age delay areas in the comparison those (80 percent). This finding also indicates that either AYA promoted early sexual initiation among males or that the program targeted the areas characterized by more risky sexual geographical behaviors and, therefore, had no or very little impact on delaying sexual initiation among males. The crude analy sis also indicates thatAYA had no influence on condom or contraceptive the differentials use among young males. in sexual behaviors Nevertheless, between the inter vention and comparison areas in the crude analysis could be biased, reflecting the differentials in the background characteristics of the respondents that are also associated in Family Planning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions Unadjusted differentials inthe percentage of survey respondents engaging inselected sexual behaviors, by sex, according to study design and whether respondent was exposed to theAYA program, Uganda, 2006_ Table 4 Self-reported exposure Male Female NoT Sexual Exposed to AYA behaviors exposed to AYA NoT Exposed to AYA exposed to AYA Full sample Sexually (337) (n) sex Ever had 53.4 (751) (372) 52.3 57.8 55.2 (739) (393) (215) (408) initiated sample (n) (180) Delay insexual initiation3 Fewer sex partnersb 79.1 77.8 77.1 78.8 82.6 84.2 71.6 75.0 Condom use at last sex 53.9 26.8** 54.2 42.3* Always use condom Contraceptive use at last sex 34.5 14.3** 39.0 31.5 33.9** 60.5 45.6** 58.9 Intervention Full sample Com parison size sex Ever had Sexually parison Intervention (n) (933) 52.3 (615) (995) (633) 52.0 57.1 55.9 (320) (568) (354) initiated sample Delay insexual initiation8 Fewer sex partners6 Condom use at last sex (n) (487) Always use condom Contraceptive use at last sex 76.7 78.4 74.6 79.4 86.7** 71.6 76.4 44.4 26.9** 46.5 43.3 27.4 15.0** 33.4 32.6 49.5 34.8** 53.0 45.3 79.8* *Significantat p < 0.05; **p< 0.01. a b occurred afteragel 6. Respondent reported thatsexual initiation Respondent reported having had fewer than two partners during the past 12 months. Therefore, to control for the influence of such background differentials between the with their sexual behavior. intervention and comparison areas, the PSM technique applied for the static group comparison analysis. is SEM and PSM Analyses of Effects ofAYA impact of AYA on young people's sexual as gleaned from the self-reported exposure de a is assessed simultaneous sign using equations model. The model adjusts for the influence of both observed and The unbiased behavior unobserved determinants of the outcome of interest to isolate the unbiased impact of program exposure. Table 5 illustrates the steps of the simultaneous equations (or bi variate probit) model used to assess the effect ofAYA on the propensity among females to have fewer sex partners. The of exposure to AYA and the likelihood fewer sex partners?that is, equations (2) and estimated. The identification of simultaneously likelihood of having having tal variable criterion, "listens to radio weekly" does not 1, Table 5). predict having fewer sex partners (see Model The choice of the instrument is based on the assumption that young people who listen regularly to radio are more likely than those who do not to listen toAYA radio cam paigns and are, therefore, more likely to seek access to available AYA services. The statistically significant en dogeneity test of the biprobit with one degree of freedom (5.61), shown in Table 5, indicates that the unexplained variance of the "exposure toAYA" variable is correlated variance of the "fewer sex part with the unexplained ners" variable. Therefore, the likelihood of having fewer partners among females and the propensity of exposure served) variable Male Com- exposure equation but not in the equation for . to the instrumen fewer sex partners According toAYA Static group comparison Female in theAYA (1)?are the biprobit model is achieved by including the significant < (p 0.05) instrumental variable "listens to radio weekly" are jointly determined by a common (and unob factor. In other words, the "exposure to AYA" is endogenous in the equation that predicts hav the biprobit model ing fewer sex partners. Accordingly, is appropriate forobtaining the unbiased effect ofAYA. If the endogeneity test of the biprobit model were not signif icant,Model 2 in Table 5 would have been themost effi cient estimator for obtaining the effect ofAYA. However, the impact of theAYA program on the propensity of hav ing fewer than two partners during the past 12months, shown inModel 2 of Table 5, is biased and inconsistent as a result of endogeneity. The reversal of the direction of the relationship between exposure toAYA and having fewer = sex partners from insignificant 0.872) and negative (p < in the simple probit to significant and positive (p 0.001) in the biprobit model suggests the presence of selective program targeting or selective program participation. The AYA program may have targeted young females who are likely to have two or more sex partners, or perhaps young females who are likely to have more than one sex more partner systematically program, or both. exposed Because of themselves to the AYA the uncorrected selection issue, the impact ofAYA is obscured and a negative rela tionship is observed in the single equation probit model the biprobit analysis (Model 2 of Table 5). Nevertheless, corrected for the program targeting or self-selection or both and demonstrated that exposure toAYA is associ ated with greater likelihood of having among 17-22-year-old females. fewer sex partners The predicted probabilities from the biprobit model of sex partners among young females having fewer exposed toAYA and among those not exposed toAYA are given inTable 6. The probability of having fewer sex partners is 17 percentage points higher among the females exposed to with those females who were not exposed. AYA, compared In other words, the effectofAYA on having fewer sex part ners is 17 percentage points in the expected direction. Volume 40 Number This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions 4 December 2009 299 Table 5 Analysis of the impactof theAYA program on the likelihoodof having fewersex partners among female respondents, = according to the self-reported-exposure design (n 565), Uganda, 2006_ Simple Bivariate probit model Fewer sex partners" Variable Exposed Coefficient toAYA 0.292 to AYA Exposed SE 1.269** having Coefficient Model SE probit models that predict fewer sex partners8 1 Model 2 Coefficient - SE - Coefficient SE -0.027 0.159 Age group 0.000 0.000 17-18(r) 19-20 -0.204 0.143 21-22 -0.028 0.196 -0.085 0.183 0.166 0.387 0.000 0.099 0.000 0.149 -0.195 0.160 -0.212 0.154 0.175 -0.127 0.206 -0.125 0.198 -0.361* 0.163 -0.013 0.172 -0.020 0.170 0.220 -0.936** 0.189 -0.027 0.195 -0.059 0.183 -0.138 0.314 0.510 0.411 -0.049 0.329 -0.062 0.292 -0.108 0.297 0.714 0.414 0.060 0.297 0.078 0.289 -0.247 0.368 1.291** 0.423 0.234 0.356 0.246 0.325 0.203 0.398* 0.195 0.385* 0.176 Able to read newspaper or letter 0.000 Easily (r) With difficulty Not at all 0.000 0.000 0.000 In-school status Never attended school (r) Out of school forfiveyears or more 0.000 Left school during past fiveyears Currently inschool Currentlymarried 0.000 0.402* Religion Catholic (r) 0.175 -0.233 0.000 0.000 0.000 0.000 0.000 0.000 Anglican Pentecostal -0.106 0.146 -0.154 0.146 -0.183 0.158 -0.175 -0.155 0.214 -0.643* 0.265 -0.422 0.226 -0.431 0.230 Muslim -0.348 0.181 0.063 0.215 -0.394* 0.184 -0.386 0.206 0.495 0.481 0.056 0.238 Other 0.161 0.289 0.422 0.108 0.313 0.491 0.505 Attends church every day 0.116 0.215 -0.085 0.216 0.068 0.237 Living inthe area since birth Traveled forone month or more last year 0.015 0.146 -0.064 0.160 -0.009 0.156 -0.008 0.149 0.151 0.127 -0.103 0.128 0.127 0.135 0.137 0.155 Ever worked forwage/salary 0.036 0.122 -0.145 0.161 -0.004 0.145 0.004 0.152 Residence 0.000 0.000 0.000 0.000 Kampala (r) Other urban -0.583 0.299 0.922 0.526 -0.277 0.274 -0.300 0.334 Rural -0.369 0.299 0.539 0.389 -0.208 0.321 -0.221 0.315 Household-asset quintile Poorest (r) Poorer -0.482** 0.177 0.486* 0.202 -0.414* 0.190 -0.407* 0.191 Middle -0.209 0.209 0.606* 0.236 0.048 0.214 0.059 0.227 Richer -0.599** 0.176 0.732** 0.196 -0.393 0.203 -0.382 0.205 Richest -0.778** 0.277 0.884** 0.275 -0.547 0.311 -0.518 0.297 -0.029 0.181 -0.007 0.194 0.246 - 0.222 0.000 0.000 0.000 0.000 Education of household head 0.000 None(r) Primary Secondary or higher Listens to radioweekly 0.000 0.000 0.000 -0.017 0.154 0.080 0.195 0.276 0.182 -0.141 - 0.614** 0.221 0.222 0.191 0.180 0.200 0.171 Constant 1.043 -2.350** 0.542 0.664 1.237* 0.572 ** 351.90 (50) 29.46 (25) goodness-of-fit (degrees of freedom) Wald's test of rho= 0 (degrees of (1)* freedom)_5.61 ? = Variable not = included inthemodel. (r) Reference category. *Significantat p < 0.05; **p< 0.01. a Respondent reported having had fewer than two sex partners during the past 12 months. Note: The sample size forthis analysis excludes the intervention-area respondents who were exposed to one or twoAYA-specific activities. 1.393** Wald's two equations of the biprobit model in Table 5 also identify the determinants ofAYA program participa tion and partner reduction among the females. Females The who can read in school, who newsletters very are Catholic are Pentecostal), who listen to radio weekly 300 Studies who are currently (as compared with those who are not in the poorest quintile, and who ing been exposed easily, toAYA. are more likely to report hav The respondents' characteris 0.521 28.12 (25) tics that determine exposure to AYA could plausibly be sex linked to safer behaviors; therefore, the background characteristic variables are included in the likelihood of the "fewer sexual partners" equation to account for the females who were exposed and those who were not. The determinants of hav observed differences between toAYA ing comparatively fewer sex partners among females are being married, being Catholic (compared with being Mus in Family Planning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions Percentage of survey respondents, by predicted probabilities of theirreportingselected sexual behaviors, according to sex and exposure toand effectofAYA, (self reportedexposure design), Uganda, 2006_ Table 7 Outcome Outcome Impactestimates fromthe static group comparison design using propensity score matching analysis, by sex, Uganda, 2006_ (n) Table 6 Not variable (n) exposed AYA effect Exposed Com parison ITTE SE Female Female Delayed sexual initiation3 Had fewer sex partners15 Delayed sexual initiation3 Had fewer sex partners6 (918) (600) -0.023 0.024 (480) (313) -0.051 0.031 13.2** Used condom at last sex (478) (314) 0.139** 0.040 11.5** Always uses condom Used contraceptive at last sex (475) (312) 0.098** 0.035 (483) (314) 0.101** 0.042 Delayed sexual initiation3 Had fewer sex partners15 (965) (611) -0.061** 0.024 (557) (345) -0.025 0.038 Used condom at last sex (557) (348) -0.047 0.041 Always uses condom Used contraceptive at last sex (556) (347) -0.060 0.040 (553) (350) -0.018 0.039 (1,066) 79.3 75.5 (565) 8.3 25.4 17.1** Used condom at last sex (566) 31.0 44.2 Always uses condom Used contraceptive at last sex (562) 16.5 28.0 -3.8 (569) 37.8 50.3 12.4** Male Male (1,074) 79.3 76.0 -3.3 (608) 75.8 70.0 -5.8 Used condom at last sex (616) 46.3 46.8 0.5 Always uses condom Used contraceptive at last sex (613) 34.6 33.4 -1.2 (616) 49.4 52.9 Delayed sexual initiation3 Had fewer sex partnersb 3.5 **Significantat p < 0.01. 3 b Respondent reported that sexual initiationoccurred after age 16. Respondent sex two 12 had fewer than months. Because partners during past reported having of endogeneity, the simulation of the predicted probabilities were derived fromthe bivariate probitmodel. Note: The denominators of these analyses exclude the intervention-area respon dents who were exposed to one or twoAYA-specific activities. lim), and being in the poorest household-asset quintile (compared to belonging to the richest quintile). Similar to the analysis in Table 5, a series of probit models Intervention variable are estimated to assess the other four outcomes the impact of AYA on The effects of AYA considered. from themost efficient model for each of the outcomes are shown inTable 6. Endogeneity was not detected with of the model other estimates. Therefore, theAYA ef any fects for the other outcomes are estimated from single equation probit models.5 The analyses indicate thatAYA among significantly influenced safe sexual behaviors females for all variables except delay of sexual initiation. females, the effects of AYA on condom use at Among last sex, always using condoms with the current partner, and contraceptive use during last sex are, respectively, 13 percentage points, 12 percentage points, and 12 percent age points, all in the expected direction. Nevertheless, the significant impact of AYA among themale respondents observed in the unadjusted analysis of the self-reported in the adjusted analysis. exposure design disappeared Results of the propensity score matching analysis of the static group comparison design are provided in 7. The PSM adjusts only for the observed con founders of the outcome of interest to isolate the impact Table of program exposure. For themost part, the PSM analy sis of the static group comparison design corroborated the results from the adjusted analysis of the self-report ed exposure design. The analysis confirmed that AYA promoted condom use, consistent use of condoms, and contraceptive use among females, with the intention to treat effects (ITTEs) being 14 percentage points, 10 per = Intentionto treateffect. **Significantat p < 0.01. ITTE 3Respondent reported that sexual initiationoccurred after age 16. bRespondent reported having had fewer than two sex partners during past 12 months. centage points, and 10 percentage points, respectively. The impact ofAYA on partner reduction among females observed in the self-reported exposure design was not seen in the PSM analysis of the static group comparison design, however. This finding is not surprising because the PSM models did not correct for the endogeneity that was present. Interestingly, when endogeneity is not ac counted for,both the single-equation probit model from the self-reported exposure design and the PSM analy sis of the static group comparison design indicate that AYA's impact on reduction of the number of sex partners, although not statistically significant, was negative; that is, exposure toAYA was associated with having a higher number of sex partners. Nevertheless, the presence of en exposure toAYA and number of sex partners in the simultaneous equations model of the self reported exposure design suggests thatAYA exposure is selective for young females who are more likely to have sex partners. In the presence of multiple endogeneity, dogeneity between the simultaneous effect ofAYA equations model shows the unbiased on reduction of the number of sex partners among young females. Consistent with the self-reported exposure design, the static group comparison design revealed no impact on delaying sexual initiation among females. The PSM analysis of the static group comparison, which also shows consistency with the adjusted analysis of the self reported exposure design, did not detect among males a significant impact ofAYA on partner reduction, condom use during last sex, consistent condom use with current partner, or contraceptive use during last sex. Contrary to expectation, the PSM analysis detected among males a negative impact of AYA on delaying sexual initiation: males in the intervention areas were 6 percentage points Volume 40 Number 4 December 2009 301 This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions more likely than those in the comparison areas to report earlier sexual initiation. Although the relationship be tween exposure toAYA and delaying sexual initiation in the self-reported exposure design is not significant (see Table 6), the direction of the relationship is similar to that from the static group comparison design; that is,AYA is associated with earlier sexual initiation. Therefore, find on of the of AYA sexual behavior outcomes ings impact are males consistent between the two study de among signs. ITTE from the PSM analysis is smaller but is con sistent with the unadjusted analysis of the static group comparison design. Lastly, the advantage of the application of the PSM technique over single-equation regression for analyzing the static group comparison design is not demonstrated. Analogous single-equation probit models accounting for the cluster sample design and respondents' back were characteristics also estimated for static the ground group-comparison the impact of AYA design (not shown). The findings of on the outcomes of interest from the probit estimates were similar to those from the PSM mod els?which is consistent with the conclusion derived by Hutchinson and Wheeler (2006). Discussion The impact of the comprehensive African Youth Alliance program on the SRH knowledge, perceptions, and behav iors of young people was assessed in Ghana, Tanzania, and Uganda using the self-reported exposure design and the static group comparison design. Rigorous analytic were to the postintervention data applied techniques available. In general, the findings concerning the impact of AYA on SRH behavioral outcomes in the three coun tries did not differ greatly6 (see Williams et al. 2007 for and Williams his colleagues was details). The analysis by complex but offered limited technical methodologically details. To illustrate the impact of the comprehensive program on the SRH behaviors of young people AYA its broader and implications for youth SRH programs, in Uganda was revised the impact analysis conducted for this study.7 Given of the AYA program, certain required to determine were assumptions were young people exposed contextual which the complexity to AYA for the self reported exposure design. The respondents who were classified as exposed to AYA were exposed not just to three or more AYA activities, but also to other, less tan strategies such as coordination of policy and institutional advocacy, capacity building, and partner communication. and ship coordination gible AYA 302 Studies Coverage of theAYA activities is impressive; only 13 percent of the intervention-area respondents were unable to recall exposure to a single AYA activity, whereas more than one-third of the intervention-area respondents re ported exposure to the comprehensive multicomponent AYA program. Among AYA activities, exposure was higher for the more readily accessible and less lengthy programs. For example, more respondents reported ex posure to programs provided on the airwaves, through in their distribution, or as events organized communities than to other aspects of the program. The reported exposure tomore-structured programs, such as national the life-planning programs or those that young people would have to seek out, such as youth-friendly clinics, was noticeably lower. The validity of the self-reported exposure measure in question, mainly because of the systematic and the random errors associated with recall. The error in recall was is systematic when recall is jointly associated with both the exposure and the outcome of interest. The recall error is random when it is not associated with the outcome of interest. Nevertheless, after controlling for endogeneity (unmeasured confounders such as self-selection), the re lationship between exposure toAYA and the outcomes of interest in the expected direction among the females indicates that the self-reported exposure toAYA is valid, at least for females; that is, the existing bias related to re call error is, for themost part, nondifferential. Moreover, the validity of the self-reported exposure design results is confirmed by the results from the static group com parison design (when endogeneity is absent), forwhich defining exposure did not rely on recall. The static group comparison design holds other as that are not required by the self-reported sumptions exposure design. The main assumptions are (1) that all intervention-area respondents are exposed toAYA and (2) that the levels of sexual behaviors before AYA were similar among respondents with similar background con characteristics (in other words, that no unobserved founders are present). The first assumption of the static group comparison design can be considered adequate from the high coverage of the AYA activities observed from the self-reported exposure design. The validity of the second assumption of the static group comparison de sign can also be considered adequate when the findings of the two study designs confirm each other, which was mainly the case. of the key AYA goals was to reduce the preva lence of HIV and other sexually transmitted infections and of unplanned among sexually active pregnancy One young people by providing information about and access to condoms and other modern contraceptives through in Family Planning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions BCC programs, peer educators, and youth-friendly ser vices. The findings from both study designs indicate that AYA substantially increased condom use, consistency of condom use, and contraceptive use among female re spondents but not among male respondents. In the presence of endogeneity, the simultaneous source were of unbiased effect the only equations models of AYA on for this study. Thus, in the case of AYA's of sex partners, of the number reduction we may effect con had a significant positive impact among females, although thiswas not corroborated by data from the static group comparison design. Conversely, AYA clude thatAYA did not achieve a significant impact on partner reduction among males; neither study design detected an impact of the program on this variable among males. or understated Perhaps males overstated sure toAYA status or their behaviors, which their expo could have translated into a null impact for the self-reported exposure analysis. The static group comparison analysis, which does not rely on recalling exposure to SRH messages, also found a null impact, however, strengthening the argument that abortions, and dropping out of school (Focus on Young Adults 2001; Kim et al. 2001; Bearinger et al. 2007; UNFPA focus on females may the program's 2007). Although have seemed tomake sense inmost environments, and although it clearly benefited young female respondents, future SRH programming foryoung people should be de as well. Positive female SRH signed to reach young males outcomes ultimately depend on positive male behaviors, and future programs must be creative in undertaking in terventions that differ from?but complement?interven tions for females. They must tailor messages specifically tomale needs (Peacock and Levack 2004; Pulerwitz et al. 2006; Barker etal. 2007). AYA's to delay young people goals included empowering sexual initiation and practice abstinence, goals cas sought especially for the youngest individuals and in es of coercion. and Macro UBOS International Although (2007) found that early sexual initiation among young adults has been declining inUganda, neither of the two an AYA contribu study designs described here detected tion toward this goal. AYA's interventions in these areas AYA had littleor no impact on male sexual behavior. Interestingly, the levels of condom and contraceptive use among males not exposed toAYA were similar to the were are exposed to AYA, which tionmale condom and contraceptive among males. Although SRH programs foryoung males may unintentionally have encouraged sexual activity as a result of the increased availability of condoms, a per ceived decreased risk of HIV transmission, or both (see levels for those behaviors among females who indicates that preinterven observed use were generally use females. than (This finding is sup among greater 2006, which also shows substantially ported by UDHS use among young males than females.) condom higher Females exposed toAYA likely "caught up" with their male peers during the period ofAYA; the remaining un exposed females are likely the ones lagging behind and still in need of attention from large-scale SRH programs for the young. Males may also have better access to infor mation through themass media, schools, and community events, and theymay obtain most of their SRH informa tion (and determine their behaviors) from sources other than AYA-type programs. Finally, other possible reasons for the lack of program impact on male behavior could include issues with theAYA program itself. For example, messages may have been more focused toward a female audience, or themessages directed toward males may not have been as well received as those directed AYA's toward females. Much of AYA's effort to address the is sue of gender norms was focused on gender equity from the perspective of improving young females' status. So cial norms of Africa assign a lower status to fe males than tomales, and as a result young females face adverse treatment, including forced or coerced sex and unwanted inmuch advances consequences from "sugar daddies," and negative including unwanted pregnancies, unsafe clearly less successful than the program's efforts thatwere focused on condom and contraceptive use. the static group comparison design Paradoxically, found a negative impact ofAYA on early sexual initiation Gray et al. 2003), the finding is only weakly supported by this study. The self-reported exposure design did not confirm the negative impact of AYA on early sexual ini tiation among males, even in the absence of endogeneity, indicating that the effect ofAYA on delayed sexual initia tion among males in the static group comparison design ismost likely spurious. Therefore, the synthesis of the two study designs suggests that AYA had no effect on delaying sexual initiation among males or females. Conclusions The objective of this study was to determine whether ex African posure to the comprehensive multicomponent Youth Alliance program promoted safe sexual and re productive health behaviors among young people aged 17-22. In light of the limitations of post-test-only study designs, the impact of AYA was determined by synthe sizing the findings from a self-reported exposure design with a static group comparison design. The results show that exposure to AYA led to a substantially higher pro pensity among young people to use condoms and con Volume 40 Number 4 December 2009 303 This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions traceptives, and, among females but not among males, to have fewer sex partners. The coverage of the AYA pro gram in the intervention areas was similar for female and 7 scaled-up, African Medical use, use, contraceptive and partner Activities had the greatest impact on SRH behavior, but it suggests that integrated programs can attain successful outcomes. Additional research beyond the current evaluation could also illuminate the best approaches for future SRH In particular, qualitative programs foryoung Ugandans. research is needed to determine themost acceptable and effective approaches for improving safe SRH behaviors males. research should investigate such top This among ics as sexual initiation, abstinence, how best to influence males and reduction, partner is Baganda, of Uganda The ethnic composition kole, 9.5 percent; Basoga, 8.4 percent; Bakiga, 16.9 percent; Banya 6.9 percent; Iteso, 6.4 4.7 percent; Bagisu, 4.6 percent; Langi, 6.1 percent; Acholi, 4.2 percent; Bunyoro, 2.7 percent; Lugbara, cent (UBOS and Macro International 2007). percent; 2 Valid conclusions known; no valid of inconsistent 3 4 "other," 29.6 per the direction of the bias is may be drawn when in the presence conclusion can be drawn, however, effect. in theory, is a predictor of exposure, variable, not for the outcome of interest (Bollen et al. 1995). An instrumental but forTable 1 tests for statisti chi-square test conducted across whole categories for each char cally significant differences tests but acteristic, (not shown) confirm the statistical chi-square 5 All of themodel 6 As estimates some degree in Tanzania, (Williamsetal. Studies noted in the discus from the first author. on selected SRH behavioral out observed mainly among females; of impact was also observed among males 2007). to improve PATH, Uganda." Uganda: 29 January 2008. -. ''Bringing youth and adults to in sexual and reproductive health Accessed . 2005a. (AYA). adolescent 2005b. "Partnerships: technical paper on theAYA experience." 29 January 2008. Accessed A . -. 2005c. "Reaching out-of-school with sexual youth in Uganda and reproductive health information and services." Uganda: PATH. Accessed 29 2008. / / . January -. 2007. "Improving health, improving lives: End of programme report of theAfrican Youth Alliance." New York: UNFPA. . Angeles, David Gustavo, 1998. "Purposive Guilkey, and Thomas Mroz. and the estimation of family planning program Journal of theAmerican Statistical Association program placement effects in Tanzania." 884-899. 93(443): Alex A. Opio, and Michel George ual behaviour and decline inHIV Elizabeth Joshua Musinguzi, in sex Carael. 1997. "Change infection among young pregnant women AIDS 11(14): Godwil, Tembo, in urban Uganda." and Claudia 1,757-1,763. ide "Communication, ation and contraceptive use in Burkina Faso: An application of the score matching method." Journal ofFamily Planning and propensity 207-212. Health Care 31(3): Reproductive Babalola, Stella Bandura, Albert. Vonrasek. 2005. of Thought and Action: A Social Cliffs, NJ: Prentice-Hall. 1986. Social Foundations Cognitive Theory. Englewood Nascimento. 2007. "En and Marcos Barker, Gary, Christine Ricardo, men and boys in inequity in health: gaging changing gender-based from programme interventions." Geneva: WHO. Evidence Bearinger, 2007. E. Sieving, Jane Ferguson, and Vinit Sharma. on the sexual and reproductive health perspectives Linda H., Renee "Global (9,568): Becker, Sascha O. ment Patterns, prevention, and potential." Lancet 369 1,220-1,231. and Andrea effects based on of average treat Stata Journal 2(4): 358-377. Ichino. 2002. "Estimation propensity scores." to reproductive Kirumira. 2000. "Access for control services: Participatory research with adolescents Pacific Institute forWomen's Health. of STDs." Los Angeles: Bohmer, Lisa and Edward health are available inUganda, the impact of AYA inGhana and Tanzania was comes differences Youth Alliance of adolescents: The Pearson's significance of thewithin-category sion of results shown in the table. 304 and African Madraa, Notes 1 and HIV/AIDS Asiimwe-Okiror, to adopt safe sex practices. and Uganda AIDS Foundation-Uganda 2001. Inventory ofAgencies with HIV/AIDS Interventions inUganda: A Review ofActors, Secretariat. and Constraints Relating to theHIV/AIDS in Challenge Uganda, . gether essary for achievement of an optimal impact, nor which single component or which combination of components to and his col Interventions, Achievements reduction among females) would be the goal. This study did not test whether all five integrated program components are nec are similar complex analysis in addition to the simultaneous and Research Commission lar effects. In theory, for expansion of a program to other sites, replication of those interventions associated with the greatest positive impact (for example, those focused condom here by Williams References as that ofAYA can be effective in proach such improving certain SRH behavioral outcomes, and that expanding such programs to other sites inUganda could have simi on presented equations the PSM technique to analyze the self-reported implemented conclusions from three analytic exposure design and synthesized in not The models did for literacy. that control techniques. study ap multicomponent analysis model, AYA achieved impressive positive effects on sev eral sexual and reproductive health outcomes among the inUganda. The results suggest young female population comprehensive, in themore (2007), which, leagues respondents and sufficient to observe population level changes in sexual behavior. a results of the revised those found male that The Bollen, Kenneth K. Guilkey, and Thomas A. Mroz. 1995. "Bi Tests and and endogenous variables: explanatory use an to the demand for contraceptive application Demography 32(1): 111-131. A., David nary outcomes solutions with in Tunisia." in Family Planning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions and Nao Pamela Kato Klebanov, Jeanne, Greg J.Duncan, 1993. "Do neighborhoods influence child and adoles cent development?" American Journal of Sociology 99(2): 353-395. Ali Mehryar. 2003. Individual-, Family-, and Neighborhood-level on Adolescent in Kwazulu-Natal, South Africa: A Effects Childbearing Karim, Brooks-Gunn, mi Sealand. Multilevel 2007. "Improving Ugochi. of the African Youth Alliance Daniels, health, and Statistics 84(1): Greg J., Johanne Boisjoly, "Sibling, peer, neighbor, and schoolmate of the importance of context for adolescent correlates 2001. as indicators sibility among H. Greene, William 2003. Econometrics Analysis. River, NJ: Prentice Hall. Saddle Guo, AIDS 17(13): Upper Kincaid, Louise 2006. Jennifer Wheeler. the impact of family planning Evidence from Tanzania and Nepal." grams: ning 37(3): 169-186. for evaluating Imbens, Guido ment W. 2004. effects under "Advanced methods communication pro Studies in Family Plan estimation of average treat "Nonparametric review." A Review Economics and exogeneity: of Statistics 86(1): 4-29. Janz, Nancy K. and Marshall A decade Jato,Miriam later." Health N., Calista H. Becker. Education Simbakalia, 1984. "The health belief model: and Behavior 11(1): JoanM. Tarasevich, 1-47. David N. Awa sum, Clement N.B. Kihinga, and Edith Ngirwamungu. 1999. "The impact ofmultimedia family planning promotion on the contracep tive behavior of women in Tanzania." International Family Planning Perspectives 25(2): 60-67. jsi.com /Managed tion_uganda.pdf /Docs /Publications >. Accessed /Evaluation /aya_evalua 29 January 2008. -. (PATH), Program International. 1999. "An extension of the propensity score Kaplan, David. adjustment method for the analysis of group differences inMIMIC models." Multivariate Behavioral Research 34(4): 467^92. International Family Planning 2002. "HIV-1 incidence Lancet 360(9,347): Uganda." Perspec and prevalence in trends 1,788-1,789. Luke, Nancy. 2005. "Confronting the 'sugar daddy' stereotype: Age and economic asymmetries and in urban Kenya." risky sexual behavior International Family Planning Perspectives 31(1): 6-14. Robert, Ali Karim, Lisa Weiss, Kate Bond, Musonda Lemba, Magnani, and Gwendolyn 2002. "Reproductive health risk and pro Morgan. tective factors among youth in Lusaka, Zambia." Journal ofAdoles centHealth 30(1): 76-86. of Health and ORC Macro. 2006. Uganda (MOH) [Uganda] Sero-Behavioural Survey, 2004-2005. Calverton, MD: Min and ORC Macro. istry of Health HIV/AIDS Moffitt, Robert. 2004. "Introduction rics of matching." Review to the symposium on the economet and Statistics 86(1): 1-3. of Economics Roeland and Mary Mahy. 2006. "Young people: The center of In in HIV/AIDS epidemic." Preventing Young People: A Sys tematic Review of theEvidence from Countries. Eds. David Developing Monasch, Technology Alice and Research. Daniel. Low-Beer, Ross, Population and Pathfinder Sage Sandra Payne Merritt, Liza Nickerson, P. Paulo Marcos de and Bernadette Castro, Buffington, de Castro. 1996. "Impact of a mass media pro vasectomy grams No. for Appropriate CA: Oaks, 2001. Emerging Answers: Research Findings on Programs toReduce Sexual Risk-Taking and Teen Pregnancy. DC: National Washington, to Prevent Teen Campaign Pregnancy, Task Force on Effective Pro Nations Fund, in the Phil 2): 69-90. 1997. No Easy Answers: Research Findings on Programs Kirby, Douglas. toReduce Teen Pregnancy. to DC: National Campaign Washington, Prevent Teen Pregnancy. Limited. 2001. "A rapid fact Uganda finding study (situation sexual and reproductive health inUganda: analysis) on adolescent Final report." Kampala: African Youth Alliance Initiative, United K2-Consult and Everett M. Rogers. Thousand in Brazil." campaign tives 22(4): 169-175. Ministry of the African Youth Alliance John Snow, Inc. (JSI). 2007. "Evaluation inUganda: Program Impact on sexual and reproductive health be havior among young people." Washington, DC: JSI. . John Snow, 29 Janu ary 2008. World origins of the health belief mod 2(4): 328-335. Senderowitz, StataCorp. . 2008. ior among Jerome Tamale, John C. Lule, and Fred Kigongo, 2001. attitude, and practice about Wabwire-Mangen. "Knowledge, among university students in Kam sexually transmitted diseases African Health Epidemic Update: December and WHO. Tim, Stephanie Mullen, Ali Karim, and Jessica Posner. 2007. of theAfrican Youth Alliance Program inGhana, Tanza on sexual and and nia, Uganda: Impact reproductive health behav Sekirime, Wilberforce pala." and World (UNAIDS) 2007. AIDS "Evaluation 2006. Rosenstock, Organization 2007. Geneva: UNAIDS on HIV/AIDS (WHO). Williams, Julie, Gary Barker, Mardo Segundo, and Marcos Nacimento. more norms and behaviors "Promoting gender-equitable among young men as an HIV/AIDS prevention strategy." HORI DC: Population Council. ZONS Final Report. Washington, Pulerwitz, Paul Programme inHIV-1 Reference Bureau (PRB). 2006. The World's Youth 2006 Data Population 4 January 2009. Accessed Sheet. . score in observational the propensity Biometrika 70(1): 41-55. Nations Health 173-188. Piotrow, Phyllis Tilson, D. Lawrence Kincaid, JoseD. Rimon II, and Ward Rinehart. 1997. Health Communications: Lessons from Family Planning Rosenbaum, United and Family Health Health Organization, Council, Population a better future for youth: Learning International. 2006. "Building and evidence." Africa Regional Forum on Youth from experience DC: World Washington, and Family Health Health Council, Population Organization, International. Itzhak, Elaine Yanovitzky, Tanzania. 6-9 June, Dar-es-Salaam, and HIV, Health Reproductive and Robert Hornik. Zanutto, effects of public mating score propensity methodology." 28(2): 209-220. causal health 2005. "Esti education Evaluation using campaigns and Program Planning Station, TX: StataCorp Tylee, Andre, Dagmar M. Haller, Tanya Graham, Rachel Churchill, and services: How Lena A. Sanci. 2007. "Youth-friendly primary-care more needs to be done?" Lancet 369(9,572): are we doing and what 1,565-1,573. Uganda AIDS Commission. Impact, 2003: Kampala, 29 January 2008. -. Epidemic: Prevalence and Accessed . 2003. The HIV/AIDS 2005. National HIV/AIDS Atlas, . January 2008. Uganda Bureau of Statistics (UBOS) Uganda Demographic and Health International. and Macro United for action 306 International. MD: 2007. UBOS framework 2007. "UNFPA Fund (UNFPA). Population on adolescents doors with young and youth: Opening 4 keys." New York: UNFPA. Nations people: and Macro Survey 2006. Calverton, 29 Acknowledgments This study was funded by the Bill & Melinda Gates Founda tion and implemented by theResearch and Training Institute of John Snow, Inc., Arlington, and VA, the Makerere Univer sity Institute of Statistics and Applied Economics, Kampala. We acknowledge Elizabeth Aisu, JohnsonKagugube, Emanuel Sekatawa, Francis Tumukwasibwe, Godwin Turyasingura, and Benita Rumanzi for implementing the fieldactivities inUganda. We would also like to acknowledge Gwyn Ainsworth, Abeja Apunyo, queline Robert Carolyn Darroch, Magnani, Boyce, Douglas Michael Aisha Kirby, Camara, Brian McQuestion, Ugochi Kironde, Fatuma Daniels, Anne Jac LaFond, Mrisho, Lisa Mueller, and Susan Rich for their contributions to the design of the study and interpretation Studies inFamilyPlanning This content downloaded from 130.240.43.43 on Thu, 17 Dec 2015 17:11:09 UTC All use subject to JSTOR Terms and Conditions of the results.