Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Dec 1.
Published in final edited form as: Glob Public Health. 2021 Jul 8;17(12):3670–3685. doi: 10.1080/17441692.2021.1951801

Is social-ecological risk associated with individual HIV risk beliefs and behaviours?: An analysis of Kenyan adolescents’ local communities and activity spaces

Christina N Schmidt a, Eve S Puffer b,c, Sherryl Broverman c,d, Virginia Warren b, Eric P Green b
PMCID: PMC8741821  NIHMSID: NIHMS1722286  PMID: 34236940

Abstract

The places where adolescents live, learn, and play are thought to influence behaviours and health, but we have limited tools for measuring environmental risk on a hyperlocal (e.g., neighbourhood) level. Working with 218 adolescents and their parents/guardians in rural western Kenya, we combined participatory mapping activities with satellite imagery to identify adolescent activity spaces and create a novel measure of social-ecological risks. We then examined the associations between social-ecological risk and individual HIV risk beliefs and behaviours. We found support for the conjecture that social-ecological risks may be associated with individual beliefs and behaviours. As social-ecological risk increased for a sample of Kenyan adolescents, so did their reports of riskier sex beliefs and behaviours, as well as unsupervised outings at night. This study reinforces calls for disease prevention approaches that go beyond emphasizing individual behaviour change.

Keywords: activity space, participatory mapping, social-ecological models, HIV, adolescents

Introduction

The places where young people live, learn, and play may help to shape their health outcomes—for better and worse. For instance, living near accessible neighbourhood green spaces is associated with greater physical activity (Janssen & Rosu, 2015), which is a predictor of improved cardiovascular health (Cuenca-García et al., 2014; Loprinzi et al., 2015). Conversely, proximity to tobacco stores has been linked to increased risk of tobacco use (Lipperman-Kreda et al., 2020), a predictor of health problems such as lung cancer.

Social-ecological models offer a useful framework for representing the complex connections between health and place. Psychologist Urie Bronfenbrenner formalized the first ecological systems theory in the 1970s, depicting children at the centre of several nested systems expanding outward from the child to the broader social, political, and historical context (Bronfenbrenner, 1979). McLeroy et al. (1988) built on this theory, and on work by Belsky (1980) and others, to explain how ecological systems shape individual behaviours and to identify opportunities for health promotion interventions at each level.

Ecological models like these have influenced public health research over the past 40 years, especially in the field of HIV prevention. Around the year 2000, two decades into the epidemic, the public health community began to emphasize the need for ‘structural’ (ecological) approaches to HIV prevention that targeted contextual drivers of transmission at multiple levels (Blankenship et al., 2000; Sumartojo, 2000). Researchers responded with dozens of papers that used ecological systems thinking to further conceptualize these structural factors, to identify sources of social-ecological risk, and to design prevention interventions (e.g., Parker & Aggleton, 2003; Auerbach et al., 2011; Baral et al., 2013; DiClemente & Patel, 2016; Poundstone et al., 2004; Rhodes et al., 2005).

Despite the wide adoption of ecological models in public health and HIV, however, the definition and measurement of social-ecological risk remains underdeveloped—the measurement of place-based and community-based risk in particular. With notable exceptions (Conners et al., 2016; Martinez et al., 2014; Theall et al., 2018), the most common definition of place remains subnational administrative units, such as counties and districts (Perchoux et al., 2013). In some settings, typically in high-income countries, a limited set of community data, crime statistics for instance, are available down to the block or neighbourhood level. More commonly, however, community characteristics only describe broad top-level administrative units, such as counties or states. These broad units can be home to tens or hundreds of thousands of people to millions. This renders the data of limited use for understanding community exposure to risk because top-level administrative boundaries do not capture how and where people spend time within these units (Browning & Soller, 2014; Morrison et al., 2019; Vaughan et al., 2017). To push forward our conceptualization of the interactions between health and place, we need better tools and procedures for describing and measuring characteristics of where adolescents spend the most time, often the neighbourhoods and areas closest to home. This is referred to as one’s hyperlocal community.

One approach to improving our understanding of hyperlocal communities is to measure people’s activity spaces. A person’s activity space encompasses the locations that they traverse throughout their daily routines (Schönfelder & Axhausen, 2003). There are various methods for collecting activity data and constructing activity spaces (Smith et al., 2019). For instance, Sherman et al. (2005) asked study participants to report how often they engaged in 34 possible routine activities (e.g., buy gasoline, go to church) and to report the location of each activity. The authors used this georeferenced location data to calculate a polygon (ellipse) for each participant that covered the majority of locations visited (weighted by frequency). Approaches like this require countries to have residential and retail address systems that cover even rural areas. Other common methods require participants to have access to location tracking technology. In the absence of these resources, some researchers have asked participants to hand draw community maps as a measure of activity space, but the accuracy of such an approach is uncertain (Basta et al., 2010).

We sought to overcome these limitations and expand the methods for studying activity spaces using geospatial tools that are widely available to researchers, even in low-resource settings. Building on work first reported in Green et al. (2016) where we used satellite imagery to interview adolescents in rural western Kenya about their daily activities, in this analysis we pursued the following primary aims: (a) construct individuals’ activity spaces; (b) identify and quantify ‘dangerous’ features in these spaces (i.e., risks) using data from focus groups and surveys; and (c) explore the relationship between space characteristics and previously identified psychosocial correlates of HIV risk in this sample, including gender, emotional problems, and HIV risk beliefs and behaviours. Our primary research question was, “To what extent is social-ecological risk associated with individual HIV risk beliefs and behaviours?”

The data for this analysis come from a cross-sectional study on adolescent HIV risk (Puffer et al., 2011). Preventing HIV transmission among adolescents remains a key public health goal (DiClemente & Patel, 2016; Protogerou & Johnson, 2014). Of the nearly 1 million new infections that occur in Sub-Saharan Africa each year, 46% are among youth between the ages of 15 to 24 (UNAIDS, 2020). Various psychosocial factors have been implicated in HIV transmission among adolescents. For example, gender norms are thought to shape how adolescent girls are expected to engage and interact with their communities (Blum et al., 2017; Heise et al., 2019). Anxiety, traumatic stress, depression, and emotional problems have also been associated with engagement in risky sexual behaviours (Brown et al., 2010; Hill et al., 2018; Jackson et al., 2015; Puffer et al., 2011). In Kenya specifically, researchers have identified various correlates of risky sexual activity among adolescents, including acceptance of risky behaviours, frequenting fish markets where intergenerational transactional sex often occurs, and low parental monitoring (Béné & Merten, 2008; Puffer et al., 2011).

We also report on two secondary aims in this manuscript. First, prior work using participatory mapping methods in South Africa found that girls’ social environments shrank with the onset of puberty while boys’ environments expanded (Hallman et al., 2015). We attempted to replicate this finding with our activity space data by testing the hypothesis that there would be a significant interaction between age and gender such that activity spaces get bigger with age for boys but shrink for girls. Second, a recent meta-analysis of studies of childhood and adolescent depression estimated that there is a small negative correlation between depressive symptoms and physical activity (Korczak et al., 2017). This led us to ask whether depression symptoms and emotional problems are negatively correlated with geographic mobility among adolescents.

Materials and Methods

Setting

This analysis examines previously unanalysed activity space data collected as part of a study on adolescent HIV risk. The original study took place in 2009 in Muhuru Bay, Kenya, a small, rural town on the shores of Lake Victoria. Muhuru Bay is located within the former Nyanza province, which at the time of the study had the highest HIV prevalence in the country at ~15% (NASCOP, 2014). Prior to adopting a new constitution in 2010, Muhuru Bay was a division (province > district > division) with 4 locations and 8 sublocations. This study took place in all 8 of the former sublocations (average sublocation area 5.6 square kilometres).

Participants and Procedures

We analysed three types of data: (a) surveys with adolescents and one of their caregivers; (b) participatory mapping interviews with adolescents from the survey sample; and (c) focus groups with community members. Native speakers translated all materials into Dholuo, the local language, and blind back-translated them into English for checking. Kenyan research assistants administered all surveys in-person at private areas near participants’ schools or homes (Puffer et al., 2011). Two members of the research team (one American and one Kenyan) facilitated each individual interview and focus group (Green et al., 2016); most groups opted to conduct the session in Dholuo. The research protocol was reviewed and approved by ethical review boards at Duke University and the Kenya Medical Research Institute.

Surveys

Adolescents 10 to 18 years of age were recruited from 14 local schools to participate in a survey of HIV risk behaviours and psychosocial wellbeing (Puffer et al., 2011). 353 of 1,847 eligible students were randomly selected and invited to participate, with 325 enrolling in the study and completing the survey. A caregiver (parent or guardian) for each adolescent was also recruited and surveyed (separately), resulting in data from 325 dyads. Survey measures used in the current study are described below.

Participatory Mapping Interviews

Adolescents were then invited to participate in a participatory mapping interview about their activities during a reference day in the prior week (Green et al., 2016). At the start of the interview, a facilitator helped the adolescent to locate their home on a laptop displaying high-resolution satellite imagery of the community. The adolescent then created a log of their activities from waking up until going to sleep. The location of each activity was marked as a digital waypoint on the map along with other regularly visited places. In total, 316 of the 325 adolescents in the survey sample participated, and 218 of these interviews generated sufficient data for the spatial analysis described in this paper.

Community Focus Group Discussions

As described in Green et. al, we recruited health workers, traditional male chiefs, women leaders, parents/guardians, and teachers to participate in focus group discussions and activities to understand characteristics of the local community. We conducted 15 different meetings with a total of 58 adults.

Study Constructs and Measurement

Our primary construct of interest was social-ecological risk. To help the reader understand our construction of social-ecological risk scores, we first describe how we constructed individuals’ activity spaces. Then we describe how we constructed social-ecological risk scores before describing the secondary data we accessed on adolescent HIV risk behaviours, beliefs, and emotional and behavioural wellbeing.

Adolescent Activity Spaces

We used the aspace package (version 3.2; Bui et al., 2012) in R to construct the activity space for each adolescent. Each activity space was represented by an unweighted standard deviation ellipse based on the adolescent’s georeferenced activity log. We calculated the planimetric area of each activity space in square kilometres. Panel B of Figure 1 shows an anonymized example of an adolescent’s activity space.

Figure 1.

Figure 1.

Open in a new tab

Panel A shows an example paper map created by one adult participant. Participants used red stickers to represent ‘dangerous’ places for young people to spend time (participants used the other stickers to respond to different questions not addressed in this manuscript). Panel B shows the georeferenced locations of red ‘dangerous place’ stickers extracted from all scanned paper maps, as well as a density raster of these points. Overlaid on this density raster are sublocation boundaries and an example activity space ellipse for one adolescent (moved to preserve anonymity).

Social-Ecological Risk Environments

We constructed indicators of perceived risk in the community using two different data sources: (a) risk density arising from our ‘dotmap’ focus groups and (b) caregiver-reported community dangers. We used the caregiver-reported data to assess the validity of our new ‘dotmap’ generated risk indicator.

‘Dotmap’ Focus Groups.

As described in Green et al., focus group participants not involved in the adolescent or caregiver surveys were given paper maps of the community and placed red sticker dots on the maps to indicate ‘dangerous places’ for children and adolescents to spend time. Stickers represented votes, and adults could place up to 8 red stickers anywhere on the map. Panel A of Figure 1 shows an example paper map from one adult participant. We scanned and georeferenced each paper ‘dotmap’, thereby converting each sticker into a waypoint on a digital map. Please see Green et al for additional procedural details (2016).

We then used the spatstat R package (version 1.63–3; Baddeley & Turner, 2005) to calculate kernel density estimates of the ‘dangerous places’ and converted these estimates into a raster where every pixel in the raster grid carried a density value. Panel B of Figure 1 shows the georeferenced locations of red ‘dangerous place’ stickers extracted from all scanned paper maps, as well as a density raster of these points.

To calculate the risk density for each area of interest, we overlaid sublocation and activity space polygons on the density map and summed the density values of the raster cells falling inside each polygon. The result was a quantitative social-ecological risk score for each adolescent’s activity space and each sublocation.

Individual Caregiver Surveys.

As part of the caregiver survey, we asked parents and guardians if they agreed with two statements of the form (separately about boys and girls): “There are dangerous places for [girls/boys] in this community.” Participants used a 4-point response scale from “strongly disagree” (0) to “strongly agree” (3) to answer with respect to their sublocation. We averaged each caregiver’s responses to these two questions and calculated the grand mean of perceived risk for each sublocation. We used these sublocation scores to validate the ‘dotmap’ generated risk density scores for each sublocation.

Adolescent HIV Risk and Related Psychosocial Factors

Data on adolescent HIV risk behaviours, beliefs, and emotional and behavioural wellbeing were collected through individual surveys. Adolescents and one of their caregivers completed separate 90-minute surveys administered verbally in private by a trained Kenyan enumerator (Puffer et al., 2011).

Emotional and Behavioural Wellbeing

Caregivers and adolescents completed the Strengths and Difficulties Questionnaire (SDQ, ages 11–17; Goodman, 2001). Participants responded, ‘not true’ (0), ‘somewhat true’ (1), or ‘certainly true’ (2) to prompts such as, ‘Many worries or often seems worried’/’I worry a lot’ (caregivers responded about the child, not themselves). We summed responses to the 5 emotional problems items and the 5 conduct problems items (each with a possible range of 0–10); higher scores represented more problems (α = 0.59 in the full sample, full scale; Puffer et al., 2011).

We assessed adolescents’ symptoms of depression using the 27-item Children’s Depression Inventory (CDI; Kovacs, 1992; Cluver et al., 2007). For each item, adolescents were asked to select the statement that best described their mood or behaviours over the past 2 weeks (e.g., ‘Which statement describes how you have been feeling over the past 2 weeks? Nobody really loves me, I am not sure if anybody loves me, or I am sure that somebody loves me.). We summed responses to create a scale with a possible range of 0 to 54; higher scores represented greater depression severity (α = 0.66 in the full sample; Puffer et al., 2011).

Beliefs Related to Sex and HIV

We assessed adolescents’ beliefs about sex and HIV with 16 original items that probed their acceptance of risky behaviours (e.g., ’It is ok for men to have many sexual partners; it is okay to have sex with a person so they will buy you things’; see Table A1 in the Appendix). Adolescents indicated their agreement with each belief on a 4-point scale that ranged from 0 to 3. We summed their responses to create a total score with a possible range from 0 to 48, where higher scores represented ‘riskier’ beliefs (α = 0.68 in the full sample; Puffer et al., 2011).

HIV Risk Behaviours

Adolescents indicated whether they ever engaged in vaginal or oral sex, the number of sex partners in the prior 12 months, and whether they used a condom during their last sexual encounter. Adolescents were coded as higher risk for HIV (1/0) if they reported having multiple sexual partners or not using a condom. Adolescents also indicated how often they left home at night unsupervised (never, once per week, twice per week, or more than twice per week), an indicator of low parental monitoring. We coded adolescents as 0 if they reported never leaving their house at night, otherwise as 1.

Analysis

We conducted all analyses in R (version 3.6.2; R Core Team 2016). After describing the study participants and adolescents’ activity spaces, we attempted to answer several questions with the data. The results presented in this paper should be considered exploratory rather than confirmatory since we did not register our hypotheses in advance.

Does activity space size vary by gender and age?

We used the brms package (version 2.13.5, Bürkner, 2018) with default priors to fit a Bayesian linear mixed effects model of activity space area that included fixed effects for male gender, age, and the interaction between gender and age, as well as a random effect for school (gamma response distribution with a log link function). We tested the hypothesis that there would be a significant interaction between age and gender such that activity spaces get bigger with age for boys but shrink for girls.

Is activity space size associated with depression symptoms or emotional problems?

To explore this relationship, we used the brms package with default priors to fit Bayesian linear mixed effects models of activity space area (sq. km) with fixed effects for emotional problems/depression symptoms (as measured here by the SDQ/CDI) and a random effect for school (gamma response distribution with a log link function).

Is dotmap-generated risk a valid measure of social-ecological risk?

To validate that the dotmap-generated risk metric measures youth’s risk environment, we estimated the rank correlation with the grand mean of perceived social-ecological risk constructed from caregivers’ survey responses. This analysis was only possible at the sublocation-level (not at the activity space level) as this was the smallest unit that we asked caregivers to reference in the survey when responding to the questions, “There are dangerous places for [girls/boys] in this community.” Therefore, we summed risk density at the sublocation-level (n=8) and calculated Spearman’s rank correlation coefficient. As a reminder, the average sublocation area was 5.6 square kilometres.

Is social-ecological risk associated with individual HIV risk beliefs and behaviours?

In the final analysis, we used the brms package with default priors to fit 6 Bayesian linear mixed models of 3 adolescent outcomes: leaving the house at night (1/0), being at higher risk for HIV (1/0), and ‘risky’ sexual beliefs (standardized, higher scores=riskier beliefs). Each model included a random effect for school and a different fixed effect indicator of social-ecological risk: risk density by (i) activity space or (ii) sublocation, both standardized (gaussian response distribution).

Results

Participants

Surveys and Interviews

Table 1 summarizes the characteristics of 218 adolescents and parents/guardians who completed the survey and were included in the activity space analysis. The average age of adolescents was 14.0 years, and 54.1% were female. Nearly 1 in 3 adolescents (28.9%) reported that both of their biological parents had died. The average caregiver who completed the survey about the adolescent was 42.2 years old, and most were female (75.2%), married or cohabiting (76.1%), and working for money (85.8%). Very few caregivers had completed secondary school or any higher education (6.9%).

Table 1.

Participant characteristics

Gender
Total
n = 218		 Boys
n = 100		 Girls
n = 118
Sublocation
 1 18 (8.3%) 7 (7.0%) 11 (9.3%)
 2 35 (16.1%) 17 (17.0%) 18 (15.3%)
 3 59 (27.1%) 23 (23.0%) 36 (30.5%)
 4 24 (11.0%) 14 (14.0%) 10 (8.5%)
 5 40 (18.3%) 20 (20.0%) 20 (16.9%)
 6 14 (6.4%) 7 (7.0%) 7 (5.9%)
 7 8 (3.7%) 3 (3.0%) 5 (4.2%)
 8 12 (5.5%) 7 (7.0%) 5 (4.2%)
 Missing 8 (3.7%) 2 (2.0%) 6 (5.1%)
Mean age, youth (SD) 14.0 (1.6) 14.0 (1.6) 13.9 (1.5%)
Orphan, youth 63 (28.9%) 29 (29.0%) 34 (28.8%)
Mean age, guardian (SD) 42.4 (13.4) 43.9 (11.8) 40.7 (14.5)
Female, guardian 164 (75.2%) 66 (66.0%) 98 (83.1%)
Married or cohabitation, guardian 166 (76.1%) 76 (76.0%) 90 (76.3%)
Working, guardian 187 (85.8%) 88 (88.0%) 99 (83.9%)
Completed secondary school/higher, guardian 15 (6.9%) 10 (10.0%) 5 (4.2%)
Open in a new tab

Note. See Puffer et al., 2011 for more details about the broader sample.

Focus Group Discussions

Fifty-eight adults from across the community participated in 1 of 15 different focus group discussions: 1 group of health workers (n = 5); 1 group of traditional male chiefs (n = 7); 1 group with women leaders (n = 4); and 12 groups with parents (n = 26) and teachers (n = 16) combined. Characteristics of focus group participants are described in Green et al.

Descriptive Summary of Activity Spaces

Table 2 summarizes the characteristics of adolescent activity spaces in terms of size and social-ecological risk. The average area of activity spaces was 2.7 square kilometres. 90.4% of activity space polygons contained at least 1 ‘dangerous place’ sticker, 19.8 stickers on average. The average sum of kernel density raster values contained inside activity space polygons was 0.000509 stickers per square meter. On average, sublocations are twice as large as adolescent activity spaces (5.6 sq. km vs 2.7 sq. km).

Table 2.

Average activity space areas and ecological risk scores by gender

Gender
Total
n = 218		 Boys
n = 100		 Girls
n = 118
Mean area in square kilometers (SD) 2.7 (5.8) 2.2 (3.7) 3.1 (7.1)
Mean count of ‘dangerous’ stickers (SD) 19.8 (24.1) 18.2 (19.1) 21.1 (27.7)
Any ‘dangerous stickers’ located within: Yes 197 (90.4%) 91 (91.0%) 106 (89.8%)
Mean density of stickers, per 1 sq m x 1000 0.5 (0.6) 0.5 (0.5) 0.5 (0.6)
Open in a new tab

Two percent of adolescents had activity spaces larger than 20 sq. km, and we considered removing them from the dataset. Activity spaces were intended to represent where adolescents spend the most time in daily life—their hyperlocal communities—but in these few cases such large areas could represent something else (e.g., errors in mapping, special trips to visit distant family members on the reference day). However, we concluded that it was not possible to confirm that these cases are true outliers and decided to leave them in the dataset.

Does activity space size vary by gender and age?

Figure 2 displays the marginal effects of the age (centred) by gender interaction from the model of activity space area. We see that activity space size does not differ on average by gender at the youngest ages in our sample, but it does increase in size with age. However, we do not find support for the hypothesis that girls’ social environments shrink with age. In fact, girls’ activity spaces may get larger at a faster rate than boys’ spaces as they age, but the 95% credible intervals are overlapping.

Figure 2.

Figure 2.

Open in a new tab

The marginal effects of the age (centred) by gender interaction from a Bayesian linear mixed effects model of activity space area (in sq. km).

Is activity space size associated with depression symptoms or emotional difficulties?

Overall, adolescents in this subsample had low scores on our measure of depression symptoms and low to moderate emotional difficulty scores. The average CDI score was 7.4 (SD=4.5; possible range 0–54), and the average emotional difficulties scores were 4.3 (adolescent-report; SD=2.3, possible range 0–10) and 4.1 (caregiver-report; SD=2.1, possible range 0–10).

Figure 3 displays the percent change in activity space area for every 1 standard deviation increase in emotional problems or depression. Each panel shows Markov chain Monte Carlo draws from the posterior distributions of the unadjusted and adjusted problems/depression parameters. The adjusted models control for age and gender.

Figure 3.

Figure 3.

Open in a new tab

Results of 6 separate Bayesian linear mixed effects models of activity space area (sq. km) with fixed effects for emotional problems/depression symptoms (all models) plus fixed effects for adolescent gender and age (centred) in the adjusted models. All models included a random effect for school. Plots show Markov chain Monte Carlo draws from the posterior distributions of the problems/depression parameters. The emotional problem and depression predictor variables were standardized, but outcome of activity space area was not. This means that the point estimates of the coefficients can be interpreted as the predicted change in activity space area for every 1 standard deviation increase in emotional problems or depression. CR=caregiver report about adolescents. AR=adolescent report.

Across models we see little evidence of a relationship between the size of adolescents’ activity spaces and their emotional problems or symptoms of depression. The posterior probabilities of caregiver-reported adolescent emotional problems and adolescent-reported symptoms of depression are both centred on the null, suggesting that there is no association. If there is a relationship, we can confidently rule out any large effect of practical importance. The posterior probability that the effect is less than a 10% reduction in area ranges from 77.5% to 95.0% across the adjusted models.

Is dotmap-generated risk density a valid measure of social-ecological risk?

At the sublocation level, we find that the dotmap density of ‘dangerous places’ is positively correlated with the independent survey data (rs=0.83, p=0.02). This is evidence of convergent validity.

Is social-ecological risk associated with individual HIV risk beliefs and behaviours?

Figure 4 displays the change across the three HIV outcomes for every 1 standard deviation increase in risk density measured at the activity space or sublocation levels. Each panel shows Markov chain Monte Carlo draws from the posterior distributions of the risk density parameters. We observe evidence of positive associations between individual risk and social-ecological risk at the sublocation level, but not the activity space level.

Figure 4.

Figure 4.

Open in a new tab

Results of 6 separate Bayesian linear mixed effects models of HIV risk outcomes with fixed effects for risk density of activity spaces or sublocations. All models included a random effect for school. Plots show Markov chain Monte Carlo draws from the posterior distributions of the risk density parameters. Predictors and outcomes standardized (except binary outcomes). The point estimates of the coefficients can be interpreted as the predicted change in each outcome for every 1 standard deviation increase in risk density. CR=caregiver report about adolescents. AR=adolescent report.

At the activity space level, the posterior probabilities of environmental risk density are centred around 0 for the sex beliefs outcome and around 1 for the binary outcomes of leaving the house at night and being at higher risk for HIV, suggesting that there is no association. However, the posterior distribution for the fixed effect of sublocation risk density is shifted to the right for each sex/HIV outcome, suggesting a possible increase in individual risk beliefs and behaviours as sublocation risk density increases. A 1 standard deviation increase in sublocation risk density is associated with higher odds of leaving the house unsupervised at night (OR=1.4), a 0.20 standard deviation increase in ‘risky’ beliefs about sex and HIV, and higher odds of being at higher risk for HIV (OR=1.2). The posterior probability that each effect is greater than 0 (or 1 in the case of the binary outcomes)—implying some degree of a positive association—is 93.0%, 95.9%, and 84.7%, respectively.

Discussion

In this paper we report on an analysis of data collected as part of a cross-sectional study on adolescent HIV risk in rural Kenya. We interviewed adolescents with the aid of high-resolution satellite imagery to create georeferenced activity logs that we used to construct individuals’ activity spaces. We then characterized the social-ecological risk (danger) of each activity space and the surrounding neighbourhoods using data from participatory mapping exercises and surveys of parents and guardians.

We were motivated by numerous applications of ecological models to the study of HIV prevention (e.g., Parker & Aggleton, 2003; Auerbach et al., 2011; Baral et al., 2013; DiClemente & Patel, 2016; Poundstone et al., 2004; Rhodes et al., 2005) to attempt to quantify ecological risk for adolescents at the neighbourhood level in a rural setting without existing sources of social-ecological risk data. We measured risk by digitizing risk maps that parents, teachers, and community leaders created in focus groups, described in f et al. (2016) and then we calculated the density of ‘dangerous places’ per sublocation and activity space.

We were limited in our options for validating this measure, but we did find a high correlation with independent survey data about dangerous places, an indicator of convergent validity. If our measure of ‘danger’ density is a valid representation of ecological risk—as is suggested by our analysis—then our study supports the contention of ecological models that characteristics of communities are linked to individual behaviours. We found positive associations between social-ecological risk and adolescents’ HIV-related beliefs and behaviours. As risk density increases, so do risky sex beliefs and behaviours, including sex without a condom or multiple partners. This is consistent with previous work reporting connections between place and HIV risk. As one example, in a study of men who have sex with men (Vaughan et al., 2017), activity spaces were associated with both risk and prevention behaviours, with results highlighting the need for place to be examined beyond just place of residence, as we did in this study.

We observed an association between social-ecological risk and risk beliefs/behaviour at the sublocation level, but not at the level of adolescents’ activity spaces. This could suggest that adolescents are more influenced by their broader community context compared to their immediate and routine social settings. This finding could also reflect measurement error if adolescents were reluctant to document visits to known specific areas that adults consider ‘off-limits.’ It is also possible that adolescents may live near high-risk spots and have easy access to them (e.g., live in sublocations with beaches known to be sites for transactional sex) but not visit those specific spots with high frequency. In these cases, risk exposure would be more likely to be captured at the broader sublocation level than at the level of adolescents’ hyperlocal activity spaces where they spend the most time.

In addition to investigating these questions about ecological risk, we also explored the relationship between activity space size and adolescent characteristics (demographics and emotional wellbeing). First, we sought to replicate a finding reported in Hallman et al. (2015) about the ‘shrinking world of girls at puberty’. In that study, the authors reported an interaction between age and gender with respect to the size of adolescents’ social spaces. As girls entered puberty, Hallman et al. (2015) found that their spaces shrank while boys’ spaces expanded. We did not replicate this result. In fact, our findings suggest that girls’ activity spaces may increase at a faster rate than boys’ spaces as they age, although the credible intervals are overlapping. In our particular study context, there are multiple possibilities that could drive this finding. One is that girls’ chores and home responsibilities may expand as they get older and sometimes require more movement for activities such as obtaining water or shopping. More concerning is the possibility that girls who are living in poverty may begin spending time further from home for reasons related to risky behaviours, such as seeking out opportunities to exchange sex for money in order to meet their financial and material needs (Juma et al., 2013; Zamudio-Haas et al., 2020; Puffer et al., 2011).

Aside from these potential population differences that could of course account for these discrepant findings—Hallman et al. (2015) conducted their study in two communities in South Africa, one urban one rural—it is also possible that methodological differences in the measurement of activity space are the driver. In their study, Hallman et al. (2015) asked eight groups of adolescents, segregated by grade level (grade 5 vs grades 8–9) and gender, to hand draw one reference map of the community for the group and to use markers to indicate safe and unsafe places. The research team used Google Maps to identify as many of these places as possible and created digital waypoints for each group’s map. Researchers then drew a digital rectangular bounding box—the “community space”—that encompassed each group’s waypoints and calculated the area in square meters. This resulted in four measurements of community space in each community: grade 5 girls, grade 5 boys, grades 8–9 girls, grades 8–9 boys.

In contrast, we constructed 218 individual activity spaces based on each adolescent’s georeferenced activity log. This enabled us to observe substantial variability in individual activity space size that was hidden to Hallman et al. (2015) on account of their group-based approach. We cannot know if a subsequent group data collection process in our study would have recovered the mean areas we calculated on the basis of individual activity spaces, but it seems doubtful. Given the choice between either method, starting with individual-level data may offer measurement advantages.

Finally, we also examined the relationship between activity space size and adolescents’ emotional problems and symptoms of depression. Based on the results of a recent meta-analysis of childhood physical activity and adolescent depression (Korczak et al., 2017), we hypothesized that we would observe a negative correlation between emotional problems/depression symptoms and geographic mobility. We did not. One reason for this could be our lack of a clinical sample. Most adolescents did not report high levels of depression symptoms. It could also be the case that activity space size is not a good proxy for mobility, as area does not fully capture physical activity.

Limitations

Our process of interviewing adolescents about their daily activities with the aid of high-resolution satellite imagery made it possible to create personalized and georeferenced activity logs for each participant (Green et al., 2016), but we recognize several issues that limited our measurement of activity spaces. First, we were only able to ask adolescents about their activities on a single reference day at one cross-section in the life of the community. Second, our method of capturing data on frequency of visits to each location did not produce usable data to incorporate into activity space construction as weights. Future work would benefit from a longitudinal perspective that also allows for weighting by frequency. This is trivial to accomplish with tracking hardware and software, of course, but it should also be possible with a low-tech approach like the one we adopted. Third, we were limited in our ability to validate the dotmap-generated risk density as a measure of social-ecological risk at the activity space level. We noted a strong correlation between the density of ‘dangerous places’ with independent survey data from adult caregivers, which is evidence of convergent validity at the sub-location level, but we do not have equivalent data for triangulation at the local activity space level. This is something of a ‘chicken and egg’ problem, however, because it is the lack of community and neighbourhood-level data that motivated our efforts to develop this method. Fourth, the observed internal consistency value for the SDQ score was low (α = 0.59), but the null finding reported herein is consistent with the null finding for the CDI, which measures a closely related construct. Finally, this analysis used data collected in 2009. While the methods used remain innovative and undescribed in the literature, it is of course possible that the social settings of the adolescents in our study have changed over time. Only a replication attempt could say with certainty, but it seems unlikely to us that the relationship between social-ecological risks and individual beliefs and behaviours would disappear or reverse sign as a result. In any case, our illustration of the method and analysis is still informative. Mapping technology has improved immensely over this time, but, unfortunately, the core challenges of collecting activity space data in rural, low-resource communities—the challenges our approach addresses—remain relatively unchanged.

Conclusions

We find support for the conjecture, derived from social-ecological models of HIV transmission, that community-level risks may be associated with individual beliefs and behaviours. As social-ecological risk increased for a sample of Kenyan adolescents, so did their reports of riskier sex beliefs and behaviours, as well as unsupervised outings at night. Therefore, this study reinforces calls for disease prevention approaches that go beyond emphasizing individual behaviour change.

Furthermore, this investigation, as a follow-up to Green et al. (2016), demonstrates an accessible method for quantifying localized social-ecological risk by involving community members and using free geospatial tools that are widely available to researchers everywhere, even those working in the poorest communities. As reported in our previous work, the activity log interviews only added 20 minutes to data collection, and the dotmap focus groups were easy to administer, informative, and highly rated by participants. In this manuscript we analysed the resulting data with free and open-source software. Researchers can build on this foundation to further improve our ability to measure aspects of social-ecological risk (and protection). These advances will deepen our understanding of the link between health and place, and will help program developers to design and tailor prevention and health promotion interventions.

Acknowledgements

We would like to thank Dyan Moses and Judith Andrew for their assistance carrying out the study.

Appendix

Table A1.

Survey items: Beliefs related to sex and HIV

For each of the following statements, tell me how much you agree or disagree. Your choices are: Strongly Agree, Agree, Disagree, or Strongly Disagree
It is cool to have a sexual partner who is older than you.
It is okay to have sex with a person so they will buy you things.
It is more difficult to refuse sex with someone who is older than you than to a partner who is the same age as you.
It is ok to have sex with someone even if they do not want to.
It is okay to have sex when you do not want to if your partner insists on having sex.
It is ok for men to have many sexual partners.
It is ok for women to have many sexual partners.
Condom use is a shared responsibility for both partners.
Using condoms is a sign of not trusting your partner.
Condoms can make you sick.
It is against your values for you to have sex while you are still a young person.
It is okay for people YOUR AGE to have sex.
Anal sex protects a person from HIV.
Sex is pleasurable.
Oral sex is sex.
Anal sex is sex.
Open in a new tab

Footnotes

Disclosure Statement

The authors report no conflict of interest.

References

  1. Auerbach JD, Parkhurst JO, & Cáceres CF (2011). Addressing social drivers of HIV/aids for the long-term response: Conceptual and methodological considerations. Global Public Health, 6(sup3), S293–S309. [DOI] [PubMed] [Google Scholar]
  2. Baddeley A, & Turner R (2005). spatstat: An R package for analyzing spatial point patterns. Journal of Statistical Software, 12(6), 1–42. http://www.jstatsoft.org/v12/i06/ [Google Scholar]
  3. Baral S, Logie CH, Grosso A, Wirtz AL, & Beyrer C (2013). Modified social ecological model: A tool to guide the assessment of the risks and risk contexts of hiv epidemics. BMC Public Health, 13(1), 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Basta LA, Richmond TS, & Wiebe DJ (2010). Neighbourhoods, daily activities, and measuring health risks experienced in urban environments. Social Science & Medicine, 71(11), 1943–1950. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Belsky J (1980). Child maltreatment: An ecological integration. American Psychologist, 35(4), 320. [DOI] [PubMed] [Google Scholar]
  6. Blankenship KM, Bray SJ, & Merson MH (2000). Structural interventions in public health. AIDS, 14, S11–S21. [DOI] [PubMed] [Google Scholar]
  7. Bronfenbrenner U (1979). The Ecology of Human Development. Harvard University Press. [Google Scholar]
  8. Browning CR, & Soller B (2014). Moving beyond neighbourhood: Activity spaces and ecological networks as contexts for youth development. Cityscape, 16(1), 165. [PMC free article] [PubMed] [Google Scholar]
  9. Bui R, Buliung RN, & Remmel TK (2012). Aspace: A collection of functions for estimating centrographic statistics and computational geometries for spatial point patterns. https://CRAN.R-project.org/package=aspace
  10. Bürkner P-C (2018). Advanced Bayesian multilevel modeling with the R package brms. The R Journal, 10(1), 395–411. 10.32614/RJ-2018-017 [DOI] [Google Scholar]
  11. Cuenca‐García M, Ortega FB, Ruiz JR, González‐Gross M, Labayen I, Jago R, … & HELENA Study Group. (2014). Combined influence of healthy diet and active lifestyle on cardiovascular disease risk factors in adolescents. Scandinavian Journal of Medicine & Science in Sports, 24(3), 553–562. [DOI] [PubMed] [Google Scholar]
  12. Cluver L, Gardner F, & Operario D (2007). Psychological distress amongst aids-orphaned children in urban south Africa. Journal of Child Psychology and Psychiatry, 48(8), 755–763. [DOI] [PubMed] [Google Scholar]
  13. Conners EE, West BS, Roth AM, Meckel-Parker KG, Kwan M-P, Magis-Rodriguez C, Staines-Orozco H, Clapp JD, & Brouwer KC (2016). Quantitative, qualitative and geospatial methods to characterize hiv risk environments. PLOS One, 11(5), e0155693. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. DiClemente RJ, & Patel N (2016). HIV prevention among youth: Designing effective evidence-based hiv risk-reduction programs for adolescents. In A clinical guide to pediatric hiv (pp. 273–295). Springer. [Google Scholar]
  15. Goodman R (2001). Psychometric properties of the strengths and difficulties questionnaire. Journal of the American Academy of Child & Adolescent Psychiatry, 40(11), 1337–1345. [DOI] [PubMed] [Google Scholar]
  16. Green EP, Warren VR, Broverman S, Ogwang B, & Puffer ES (2016). Participatory mapping in low-resource settings: Three novel methods used to engage Kenyan youth and other community members in community-based HIV prevention research. Global Public Health, 11(5–6), 583–599. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Hallman KK, Kenworthy NJ, Diers J, Swan N, & Devnarain B (2015). The shrinking world of girls at puberty: Violence and gender-divergent access to the public sphere among adolescents in south africa. Global Public Health, 10(3), 279–295. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Janssen I, & Rosu A (2015). Undeveloped green space and free-time physical activity in 11 to 13-year-old children. International Journal of Behavioral Nutrition and Physical Activity, 12(1), 26. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Juma M, Alaii J, Bartholomew LK, Askew I, & Van den Born B (2013). Understanding orphan and non-orphan adolescents’ sexual risks in the context of poverty: a qualitative study in Nyanza Province, Kenya. BMC International Health and Human Rights, 13(1), 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Korczak DJ, Madigan S, & Colasanto M (2017). Children’s physical activity and depression: A meta-analysis. Pediatrics, 139(4). [DOI] [PubMed] [Google Scholar]
  21. Kovacs M (1992). Children’s depression inventory: Manual. Multi-Health Systems. [Google Scholar]
  22. Lipperman-Kreda S, Finan LJ, Kowitt SD, Grube JW, Abadi M, Balassone A, & Kaner E (2020). Youth daily exposure to tobacco outlets and cigarette smoking behaviours: Does exposure within activity space matter? Addiction. [DOI] [PMC free article] [PubMed]
  23. Martinez AN, Lorvick J, & Kral AH (2014). Activity spaces among injection drug users in san francisco. International Journal of Drug Policy, 25(3), 516–524. [DOI] [PubMed] [Google Scholar]
  24. McLeroy KR, Bibeau D, Steckler A, & Glanz K (1988). An ecological perspective on health promotion programs. Health Education Quarterly, 15(4), 351–377. [DOI] [PubMed] [Google Scholar]
  25. Morrison CN, Byrnes HF, Miller BA, Kaner E, Wiehe SE, Ponicki WR, & Wiebe DJ (2019). Assessing individuals’ exposure to environmental conditions using residence-based measures, activity location–based measures, and activity path–based measures. Epidemiology, 30(2), 166–176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. National AIDS and STI Control Programme. (2014). Kenya AIDS Indicator Survey 2012: Final Report. NASCOP, Nairobi. [Google Scholar]
  27. Parker R, & Aggleton P (2003). HIV and aids-related stigma and discrimination: A conceptual framework and implications for action. Social Science & Medicine, 57(1), 13–24. [DOI] [PubMed] [Google Scholar]
  28. Perchoux C, Chaix B, Cummins S, & Kestens Y (2013). Conceptualization and measurement of environmental exposure in epidemiology: Accounting for activity space related to daily mobility. Health & Place, 21, 86–93. [DOI] [PubMed] [Google Scholar]
  29. Poundstone KE, Strathdee SA, & Celentano DD (2004). The social epidemiology of human immunodeficiency virus/acquired immunodeficiency syndrome. Epidemiologic Reviews, 26(1), 22–35. [DOI] [PubMed] [Google Scholar]
  30. Protogerou C, & Johnson BT (2014). Factors underlying the success of behavioral hiv-prevention interventions for adolescents: A meta-review. AIDS and Behavior, 18(10), 1847–1863. [DOI] [PubMed] [Google Scholar]
  31. Puffer ES, Meade CS, Drabkin AS, Broverman SA, Ogwang-Odhiambo RA, & Sikkema KJ (2011). Individual and family-level psychosocial correlates of HIV risk behavior among youth in rural Kenya. AIDS and Behavior, 15(6), 1264–1274. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. R Core Team. (2016). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
  33. Rhodes T, Singer M, Bourgois P, Friedman SR, & Strathdee SA (2005). The social structural production of hiv risk among injecting drug users. Social Science & Medicine, 61(5), 1026–1044. [DOI] [PubMed] [Google Scholar]
  34. Schönfelder S, & Axhausen KW (2003). Activity spaces: Measures of social exclusion? Transport Policy, 10(4), 273–286. [Google Scholar]
  35. Sherman JE, Spencer J, Preisser JS, Gesler WM, & Arcury TA (2005). A suite of methods for representing activity space in a healthcare accessibility study. International Journal of Health Geographics, 4(1), 24. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Smith L, Foley L, & Panter J (2019). Activity spaces in studies of the environment and physical activity: A review and synthesis of implications for causality. Health & Place, 58, 102113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Sumartojo E (2000). Structural factors in hiv prevention: Concepts, examples, and implications for research. AIDS, 14, S3–S10. [DOI] [PubMed] [Google Scholar]
  38. Theall KP, Felker-Kantor E, Wallace M, Zhang X, Morrison CN, & Wiebe DJ (2018). Considering high alcohol and violence neighbourhood context using daily diaries and gps: A pilot study among people living with hiv. Drug and Alcohol Dependence, 187, 236–241. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Vaughan AS, Kramer MR, Cooper HL, Rosenberg ES, & Sullivan PS (2017). Activity spaces of men who have sex with men: An initial exploration of geographic variation in locations of routine, potential sexual risk, and prevention behaviours. Social Science & Medicine, 175, 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Zamudio-Haas S, Auerswald C, Miller L, Amboka S, Agot I, Kadede K, … & Bukusi EA (2020). Seeking a “Sponyo”: Insights Into Motivations and Risks Around Intergenerational Transactional Sex Among Adolescent Boys and Girls in Kenya. Journal of Adolescent Health. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. UNAIDS. (2020). Seizing the moment: Tackling entrenched inequalities to end epidemics. UNAIDS: Geneva. [Google Scholar]

RESOURCES