Neighbourhood Determinants of Early Sexual Debut and Multiple Sexual Partnerships: A cross-sectional Analysis of Adolescents in Rwanda, Ghana and South Africa.

Background: The relationship between neighbourhood characteristics and youth involvement in risky sexual behaviour such as early sexual debut and multiple sexual partnerships is well established in the literature. However, there are very few empirical studies using Demographic and Health Surveys to unpack the nature of this relationship in Africa. This study aims to identify the neighbourhood characteristics inuencing young people's engagement in risky sexual behaviour in sub-Saharan Africa. Methods: Univariate, bivariate, and multivariate analyses on young people aged 15 to 24 years were conducted using the most recent Demographic and Health Survey Data (DHS 2014-2016) from Ghana, Rwanda, and South Africa to investigate the relationship between neighbourhood characteristics and youth risky sexual behaviour. Results: Individual and, in particular, neighbourhood characteristics (community poverty, community occupation, community media access, and community education) were found to be substantially linked with youth risky sexual behaviour. Conclusion: To lower the incidence of risky sexual behaviour in the community, programs aimed at appropriate policy options must be intensied. Adopting the implications of these ndings is critical for a developmental approach aimed at reaching Africa’s long-term development goal of eliminating STIs among young people.

have documented an association between neighbourhood characteristics and risky sexual behaviour [25,26]. For example, in a crosssectional study in SSA examining the in uence of poverty and sexual behaviour, the result revealed that not being poor delays sexual debut in females and poor female adolescents are vulnerable to infectious diseases [27]. Similarly, in another study in SSA using a crosssectional design document age, urban residence, education attainment and media exposure at the individual level resulted in the likelihood of risky sexual behaviour [28]. However, very few studies recommended that further research exploring neighbourhood level characteristics need to be done in SSA [29,30]. Durkelman and colleagues stressed the in uence of lack of food in the community on the likelihood of engaging in multiple sexual partnerships among young females. Another study by Liu and colleagues found a link between neighbourhood characteristics and the risk of youth engaging in multiple sexual partnerships. The study emphasized how cultural norms and values in neighbourhoods have a detrimental impact on risky sexual behaviour [31]. As a result, it is critical to lessen the burden of sickness that plagues individuals and families in SSA, particularly among young people, by looking beyond individual variables.
The social disorganization frameworks were chosen for this study because they demonstrate that high-risk sexual activity is a complex phenomenon that involves the interaction of individual, community, and societal components [32]. Furthermore, individual and community self-regulatory capacities are eroded by age, education, employment status, neighbourhood poverty, volatility, and family upheaval. The theory of social disorganisation is used to explain how individual and community factors such as poverty, family disruptions, residential instability, and race/ethnic disparities in uence the reproductive health of people living in certain neighbourhoods, including youth themselves, directly or indirectly [33]. Therefore, this paper seeks to examine the relationships between neighbourhood factors and sexual behaviour among youth. It hypothesizes that neighbourhood factors such as community poverty, community media access, community education and community occupation are associated with adolescents' sexual behaviour in SSA.

Methods
This study used data from the Demographic and Health Survey (DHS) from Ghana, Rwanda and South Africa, which was a survey conducted nationally within ve years. For this study, a total of 14,751 youth aged 15 to 24 years comprising 9,271 females and 5,480 males were used.
Inclusion criteria for the countries selected.
The availability of data on an adolescent's sexual risk, as well as consistency in sexual-risk behaviour in connection to neighbourhood variables, were the criteria for including the DHS data for the selected countries in this study

Outcome Variables
The study made use of two outcome/dependent sexual risk behaviour variables of multiple sexual behaviour and age at rst sex. For multiple sexual partnerships, respondents were asked if they have had more than one sexual partner in the 12 months preceding the survey and risky sex was de ned as any sexual intercourse with a non-marital or non-cohabiting partner in the last 12 months before the survey.
Their responses were then grouped into 1 = yes if they have more than one sexual partner, and 0 = No if they responded otherwise. For the age at rst sexual intercourse, respondents were asked the age they had their rst sex. Their responses were then grouped into <18 years: <24 years and No sex between 15 and 24 years.

Measurements of variables 1. Individual-level variables
The following individual-level explanatory variables were included as control variables: age of the respondent at the time of the interview (15-19 or 20-24): education attainment (Primary/less education, secondary or higher): employment status (not working or working): household size (1-4, 5-6 or 7+) members.

Neighbourhood-level variables
The neighbourhood-level variables included in the study were (1) neighbourhood poverty: percentage of households in the poorest quintile of the wealth index [34]. (2) Place of residence was de ned as either rural or urban, as administratively de ned by each of the countries surveyed. (3) Community education was de ned as the level of educational attainment in the household, categorized as (i) low (ii) high. (4) Community occupation categorized as (i) Not working (ii) professional (iii) Sales/unspeci ed (iv) agriculture (v) manual. (5) Community media access was de ned as those who have access to radio in the community which was categorized as (i) low (ii) high. To make analyses and interpretations simpler and more meaningful, some variables were regrouped from their original categories in the dataset.

Data Analysis
Data were weighted using the weight weighting factors of the demographic health surveys. The relationship between each of the outcome variables of risky sexual behaviour and independent variables were assessed using descriptive, bivariate and multivariate analysis. Multilevel logistic regression analysis was employed to adjust to interrelationships between covariance. Three models were tted for each of the outcome variables: empty models (model 1) individual (model 2) and neighbourhoods (model 3). Multi-collinearity was assessed among the variables. And there was no multi-collinearity. STATA 14 Software was used for the analysis.
The study utilized secondary datasets from Ghana, Rwanda and South Africa with all identi er information removed. The survey was approved by the Ethics Committee of the Micro at Calverton in the USA and by the National Ethics Committee of each of the countries surveyed. All study participants gave informed consent before participation and all information was collected con dentially. The datasets were downloaded from the DHS website and are free to be used by researchers for further analysis.

Background characteristics
The descriptive analysis of the study population by the background characteristics were presented in Table 1 below. In the three countries surveyed, the proportion of respondents for both the male and females aged 15 to 19 years ranges from 65.7% (Ghana) to 64% (Rwanda). Only South Africa had less than 60% of those between age 15 and 19 years for both males and females. Rwanda had 39.3% of the male adolescent, 38% male adolescents were from Ghana. South Africa recorded the highest proportion of the males and females without any employment at the time of the survey at 91.4% and 83.7% respectively, while the least is Rwanda at 43% females and 34.6% for the males. With regards to education attainments, respondents who had secondary education attainment were the majority, 87.9% for females and 82.4% males (South Africa), 73.5% females and 72.4% males (Ghana) and 44.4% females and 38.6% males (Rwanda). In terms of household size, South Africa recorded the highest proportion of those from a household made up of 1 to 4 members among the males at 45.7%, 37.5% (Ghana) and 34.7% (Rwanda).
Additionally, Rwanda recorded the highest proportion of male and female respondents who are living in a neighbourhood with a low poverty level at 73.8% males and 69.6% females. Ghana and South Africa had more than 40% of their male and female youth living in a neighbourhood with high poverty levels. In terms of community media accessibility, those with no access ranges from 34.6% females (South Africa), and 18.7% females (Ghana) to 10.5% females (Rwanda). With regards to the community occupation, more than 50% of males and females did not have any job from all the countries surveyed. Apart from South Africa which recorded the highest proportion of those living in a community with low education for the males and females at more than 70%, the rest of the countries had a little above 50% of those living in a community with low education. More so, apart from South Africa which recorded the highest proportion of young people who resided in the urban areas, the rest of the surveyed countries had the highest proportion residing in the rural areas, ranging from 73.5% males and 71.2% females (Rwanda), and 54.3% males and 49.3% females (Ghana) to 53.1% males (South Africa). In all, there was substantial representation across gender in all the countries surveyed. The prevalence of adolescent risky sexual behaviour In the three countries surveyed as shown in gure 1 below, the rate of adolescents who reported to have engaged in multiple sexual partners ranges from 12% females and 13% males (Rwanda), 38% females and 30% males (Ghana), to 56% female 62% males (South Africa) as shown in Table 2 below. Those that started their rst sexual encounter between ages 18 and 24 years old were highest among South Africa male adolescents at 22%, while the least is Rwandan female adolescents at 11%. In all the countries surveyed, those who have not had sex before their 24th birthday ranges from 75% Females (Rwanda), 49% females (Ghana), and 37% females (South Africa).
Association with adolescent risky sexual behaviour and employment status was evident in all the three countries surveyed. The proportion of those without any employment to have been exposed to risky sexual behaviour ranges from 63% females and 45% males (Ghana), 43% females and 35% males (Rwanda), to 91% females and 84% males (South Africa).
In all the three countries surveyed as shown in gure 2, it was found that more than 50% of adolescent males and females with secondary education attainments had engaged in risky sexual behaviour compared to the 22% and 44% with a primary or fewer education attainments. Association of adolescent risky sexual behaviour with the neighbourhood poverty level ranges from 70% females and 74% males (Rwanda) to more than 50% for both genders in South Africa and Ghana. A similar pattern was observed for community media access as more than 60% of adolescent males and females with regular access to media in Rwanda have engaged in risky sexual behaviour. While 50% males and 47% females (South Africa) to 49% males and 69% females (Ghana) with regular access to media have engaged in risky sexual behaviour. For those in a high education community, the highest proportion was Rwanda with 46% females and 41% males. In all the countries surveyed, there was evidence of community characteristics in uencing adolescents to have engaged in risky sexual behaviours.

Adolescent risky sexual behaviour predictors
The association between background characteristics and multiple sexual partnerships is presented in Table   The association between background characteristics of male adolescents and multiple sexual partnerships is presented in  The association between background characteristics of female adolescents and age at rst sex is presented in sex. However, there was no association with age at rst sex with those from a household of 5 to 6 members in Ghana. There was an association with age at rst sex in Rwanda and South Africa among those from a household size of 5 to 6 members. The association between background characteristics of male adolescents and age at rst sex is presented in Africa (AOR = 6.30, 95% CI: 4.37-9.10). Also, male adolescents with employment in Ghana and Rwanda were 85% and 32% more likely to be associated with age at rst sex unlike those in South Africa with 84% unlikelihood to be associated with age at rst sex than those without any employment. Education was found to be associated with age at rst sex among male adolescents in South Africa and Ghana at 2.48 and 1.52 odds among those with secondary education attainment, only those in Rwanda showed no association at an age at rst sex.
There was no association with age at rst sex with household size in all the three countries surveyed among male adolescents.   Table 6 presents the results of the multilevel logistics regression analysis of the risk of engaging in multiple sexual partnerships in all the countries surveyed among female adolescents. The results indicate that the test statistics was signi cant, thus it shows evidence that the between-neighbourhoods variance is non-zero, as was revealed on the variance partition coe cient (VPC) values. Therefore, the total variance in the risk of engaging in multiple sexual partners among female adolescents as a result of differences between neighbourhoods ranged from 5% (Ghana) to 10% (Rwanda) and 5% (South Africa). It was discovered that the variations decreased after controlling for individual (model 2) and neighbourhoods (model 3). More so, there were signi cant compositional effects as shown in Table 6. For instance, female adolescents living in high neighbourhood poverty signi cantly increased the odds of multiple sexual partnerships by 54% (South Africa) and 3% (Rwanda). There were 99% lower odds of engaging in multiple sexual partnerships among female adolescents living in high neighbourhood poverty in Ghana compared to those living in low neighbourhoods. Additionally, the results revealed that there was 14% and 1% (Ghana) for those without regular access to community media to have multiple sexual partners compared to those with no access. However, South Africa and Rwanda revealed lower odds of engaging in having multiple sexual partners among female adolescents with regular access to the community media.
Furthermore, the results showed strong signi cant effects with community occupation. For instance, female adolescents in agriculture increased the odds of having multiple sexual partners by 16% (South Africa), 3% (Rwanda) and 45% (Ghana). Again, community education increased the odds of engaging in multiple sexual partners among female adolescents by 53% (South Africa), to 4% (Ghana). Meanwhile, the higher the community education in Rwanda, the lower the odds of engaging in multiple sexual partners among female adolescents.
The results from the three countries indicated that female adolescents living in rural areas had a higher level of exposure to engaging in multiple sexual partners in South Africa and Ghana, although the result was not signi cant. However, Rwanda had signi cantly lower odds of engaging in multiple sexual partners among female adolescents residing in rural areas compared to those residing in urban areas. The results presented in Table 7 reveals the risk of engaging in multiple sexual partnerships among male adolescents in all the countries surveyed. The results indicate that the test statistics was signi cant, hence, it indicates evidence that the between-neighbourhoods variance is non-zero, as was revealed on the variance partition coe cient (VPC) values. Therefore, the total variance in the risk of engaging in multiple sexual partners among male adolescents as a result of differences that existed between neighbourhoods ranged from 15% (Ghana) to 11% (Rwanda) and 14% (South Africa). It was discovered that the variations decreased after controlling for individual (model 2) and neighbourhoods (model 3) characteristics. More so, there were signi cant compositional effects as observed in Table 7 below. For example, male adolescents living in high neighbourhood poverty signi cantly increased the odds of multiple sexual partnerships by 45% (South Africa), to 55% (Rwanda) and 45% (Ghana) compared to those living in low neighbourhoods. Additionally, the results revealed community media access signi cantly increased the odds of multiple sexual partnerships among male adolescents in Ghana and Rwanda among those with regular access. Also, there was a 63% (Ghana), to 48% (Rwanda) and 33% increase in the odds among male adolescents without regular access to community media, though not signi cant. Furthermore, the results showed strong signi cant effects with community occupation. For instance, male adolescents in the category of professionals signi cantly increased the odds of having multiple sexual partners by 60% (South Africa), 23% (Rwanda) and 97% (Ghana).
Again, community education signi cantly increased the odds of engaging in multiple sexual partners among male adolescents by 23% (South Africa), to 20% (Ghana). Meanwhile, the higher the community education in Rwanda, the lower the odds of engaging in multiple sexual partners among male adolescents. The results from the three countries indicated that male adolescents living in rural areas had a higher level of exposure to engaging in multiple sexual partners in South Africa, Rwanda and Ghana, although the result was not signi cant in South Africa. Therefore, the total variance in age at rst sex among female adolescents as a result of differences that existed between neighbourhoods ranged from 15% (Ghana) to 11% (Rwanda) and 14% (South Africa). After controlling for individual (model 2) and neighbourhoods (model 3) characteristics, the variations decreased.
The results indicate that there were signi cant compositional effects as shown in the results displayed in Table 8 below. For instance, adolescent females living in high neighbourhood poverty in South Africa signi cantly increased the age of rst sex at 81%. In addition, there was an increased odds of 8% in Ghana, though not signi cant, compared to female adolescents in Rwanda with a lower odds of age at rst sex at 98%.
Among all the countries, results revealed that community media access signi cantly reduced the odds of age at rst sex, except in Ghana where female adolescents had an increased likelihood of age at rst sex among those without regular access to community media, though not signi cant. Furthermore, the results showed strong signi cant effects with community occupation. For instance, female adolescents in all the categories of occupations signi cantly increased the odds of age at rst sex. Again, community education signi cantly increased the odds of age at rst sex among female adolescents by 50% (South Africa), to 46% (Ghana). Meanwhile, the higher the community education in Rwanda, the lower the odds of age at rst sex among female adolescents. The results from the three countries indicated that female adolescents living in rural areas had a higher level of exposure to age at rst sex in South Africa, Rwanda and Ghana, although the results were not signi cant in South Africa. The results of age at rst sex in all the countries surveyed among male adolescents are presented in Table 9 below. The test of statistics was signi cant, as the result of variance partition coe cient (VPC) values indicated that the between-neighbourhoods variance is non-zero.
Therefore, the total variance in age at rst sex among male adolescents as a result of differences that existed between neighbourhoods ranged from 14% (Ghana) to 12% (Rwanda) and 17% (South Africa). In addition, after controlling for individual (model 2) and neighbourhoods (model 3) characteristics, the variations decreased.
The results further revealed that there were signi cant compositional effects. For example, adolescent males living in high neighbourhood poverty in South Africa signi cantly increased the age of rst sex by 26%. The results indicate that there was no association in age at rst sex in Ghana and Rwanda among females in high neighbourhood poverty compared to those in low neighbourhood poverty. In all the countries, results revealed that community media access signi cantly increased the odds of age at rst sex among those that regularly and not regularly accessed community media, except in Ghana among male adolescents without regular access to community media.
Furthermore, the results showed strong signi cant effects with community occupation. For instance, male adolescents in all the categories of occupation signi cantly increased the odds of age at rst sex. Again, community education signi cantly increased the odds of age at rst sex among male adolescents by 50% (South Africa), to 46% (Ghana). Meanwhile, the higher the community education in Rwanda, the higher the odds of age at rst sex among male adolescents, although not signi cant. The results from the three countries indicated that male adolescents living in rural areas had a higher level of exposure to age at rst sex in South Africa and Ghana, although the results were insigni cant in South Africa compared to those in Rwanda with 66% signi cant lower odds of age at rst sex.

Discussions
The purpose of this study was to examine the neighbourhood factors that encourage young people to engage in risky sexual behaviour in three African countries that cut across West, East and Southern African regions. Our analysis of nationally weighted and representative survey data of the three countries found that a high proportion of male and female adolescents that were exposed to the sexual risk behaviours were those below the age of 18 years. Reports of early sexual debut and multiple sexual partnerships have exposed this young productive age cohort to the risk of sexually transmitted infections, including HIV/AIDS. We found that male adolescents were more likely to report the age at which they start having sexual intercourse and multiple sexual partnerships than their female counterparts. These ndings are in line with previous observations that revealed that male adolescents tend to indulge in risky sexual behaviours, [35,36]. most especially their sexual risk practices [37,38]. However, our ndings established various pathways through which neighbourhood factors in uence sexual risk-taking among male and female adolescents in all the countries surveyed. The ndings have great implications for STIs prevention in SSA, particularly among young people whose activities might not be monitored as a result of the prevalence of social disorganisation factors in their communities. This is because there is the tendency that the current sexuality education programs, particularly at the community level, are insu cient in addressing variables that predispose young people to engage in sexual risk-taking.
Thus, attitudinal change towards positive sexual behaviour becomes problematic.
Unlike previous studies [39,40,41], our ndings reveal that adolescent sexual risk-taking could be compounded as a result of an association between community occupations, poverty and media access. This association may be explained by the possibility that adolescents from a community with high-level occupation employment (e.g., Doctors, Lawyers, Engineers etc.) and residing in a lowpoverty neighbourhood, despite having better media access (e.g., family radio, television, internet, etc.) without a better household control over media use, tend to engage in risky sexual behaviour. This nding was expected considering the proportion of young people found in the neighbourhood with adequate media access without proper control either from their parents or guardians. As a result, while awareness of sexual and reproductive health safety and risk reduction measures such as early sexual debut and multiple sexual partnerships without a condom are high, this knowledge is not well utilized among youths, due to limited control over media content in the community.
Regarding community education, there was a strong relationship between community education and adolescent risky sexual behaviour. In Ghana and South Africa, community education increased the likelihood of engaging in risky sexual behaviour, which is consistent with a study conducted in Ghana and Kenya [42,43], which found that community literacy level in uences adolescents to engage in risky sexual behaviour. This may be linked to the unique community's cultural and religious practices con icting with the global goals and policies on adolescents' sexual behaviour. This nding exposed a gap in the transmission of knowledge in a disorganized neighbourhood with high literate individuals. While this explanation remains open for further interrogation, one implication of the ndings is the effects of globalization on the traditional method of learning (i.e., within the family) to a dependency on the conventional style of learning (formal education). Moreover, coming from a larger-than-average-sized household reduced the likelihood of engaging in risky sexual behaviour, according to the ndings. This result was expected given the proportion of young adolescents living in a non-poor neighbourhood. As a result, parents' household status may act as a protective factor for young people engaging in risky sexual behaviour, such as multiple sexual behaviours [44,14], as well as improve their self-esteem, self-con dence, and self-autonomy in sexual relationships [45].
Regarding the place of residence, the ndings reveal an association with risky sexual behaviours in all the countries surveyed. For example, urban neighbourhoods have a higher prevalence of adolescents engaging in risky sexual behaviour. Despite government and nongovernmental groups' efforts to raise awareness about risky sexual activity, this study indicated that neighbourhood characteristics contribute to a high prevalence of risky sexual conduct among adolescents. As a result, this research backs up earlier ndings that preventive and "sensitization to risky sex" programs do not ensure a change in behaviour patterns [46,47,48]. Future research should investigate how limited media availability and poverty are linked to adolescent risky sexual behaviour in communities with a high number of literate parents since community media access and education were found to have a lower correlation with risky sexual behaviour among young people in the study.

Limitation
The current study was limited to never-married young individuals in Ghana, Rwanda, and South Africa, so there are some limits to this research. This restriction on never-married young individuals may have a high risk of bias/discordance when it comes to neighbourhood measures impacting early sexual debut and multiple sexual engagements. Despite these limitations, the ndings are critical for more strategic policies and programs in preventing sexually transmitted infections, including HIV, and monitoring intervention programs to control the epidemic's spread, particularly among never-married young people in the three countries studied.

Conclusion
Following previous research, this study found a signi cant incidence of neighbourhood factors on risky sexual behaviour among adolescent males and females in all the countries analyzed, independent of their neighbourhood. Further research combining qualitative investigations is highly recommended to gain a deeper knowledge of the neighbourhood's predictor elements of higher risk sexual behaviour among adolescent males and females in SSA. Policymakers should create programs that help young people grasp the seriousness of the risks connected with early sexual debut and multiple sexual partnerships to make better decisions about their sexual lives.

Policy implications
The ndings of this study on the in uence of neighbourhood determinants of risky sexual behaviour (early sexual debut and multiple sexual partnerships) have several policy implications for future HIV/AIDS prevention among never-married young people in Ghana, Rwanda and South Africa. The ndings would inform and help policymakers to consider appropriate policy options to reduce the prevalence of high-risk sexual behaviour, especially at the community level in sub-Sahara Africa. Additionally, in line with (12)