Mobilizing homeless youth for HIV prevention: a social network analysis of the acceptability of a face-to-face and online social networking intervention

Abstract

The objective of the study is to use social network analysis to examine the acceptability of a youth-led, hybrid face-to-face and online social networking HIV prevention program for homeless youth.Seven peer leaders (PLs) engaged face-to-face homeless youth (F2F) in the creation of digital media projects (e.g. You Tube videos). PL and F2F recruited online youth (OY) to participate in MySpace and Facebook communities where digital media was disseminated and discussed. The resulting social networks were assessed with respect to size, growth, density, relative centrality of positions and homophily of ties. Seven PL, 53 F2F and 103 OY created two large networks. After the first 50 F2F youth participated, online networks entered a rapid growth phase. OY were among the most central youth in these networks. Younger aged persons and females were disproportionately connected to like youth. The program appears highly acceptable to homeless youth. Social network analysis revealed which PL were the most critical to the program and which types of participants (younger youth and females) may require additional outreach efforts in the future.

Introduction

There is a growing interest in online and cell phone-based interventions [1–7], an extension of face-to-face social network-based interventions, which evidence suggests are effective HIV prevention interventions [8–17]. As such, a great deal of interest has developed around creating explicit conceptual models to explain the effectiveness of these efforts in reducing HIV-related risk-taking behaviors [18–20]. We provide evidence from a pilot study, examining the acceptability of a new hybrid face-to-face and online social networking intervention for homeless youth. This study focuses on an analysis of social network data to understand the acceptability of recruitment, assessment and participation of homeless youth in a pilot test of this new HIV prevention program model.

Runaway and homeless youth are at great risk for HIV, with prevalence in urban settings ranging from 2 to 11.5% [21–23]. Risk-taking peers encountered in street life have consistently been shown to be a significant influence on HIV risk-taking [24–29]. Specifically, homeless youth who are engaged with other high-risk street-based peers are prone to increased levels of substance use/abuse, sexual risk-taking (including survival sex), criminal activity and violence [24–29]. While the limited numbers of HIV prevention interventions for homeless youth are shown to be effective, they require substantial resources, making widespread dissemination challenging. As such, new approaches like social network-based and technology-utilizing interventions are needed [30].

Homeless youth are often difficult to engage in face-to-face HIV prevention interventions over time because of their transient lifestyles and unstable housing situations [30]. Social networking technologies (e.g. Facebook) may help to overcome these structural barriers, assuming homeless youth use social media at rates which approach that of housed adolescents [4, 31–35]. Recent work suggests that they do [36–38]. Data on homeless youth collected from a drop-in center in Los Angeles, indicate that 85% of homeless youth use the Internet at least once per week and more than 80% use social networking websites like Facebook and MySpace. Social networking websites may be a viable new technology for mounting peer-led HIV prevention efforts for this unstable group of high-risk youth. The acceptability of such a modality has yet to be tested.

Measuring networks and intervention acceptability

The social networks which are created in the context of social network-based interventions (whether face-to-face or online) are critical to the intervention effort. There is a difference between the structure of social network ties and the diffusion of behaviors along those network ties. Diffusion, however, is impossible without an antecedent social network. Measuring the structure of these networks does not inform behavior change per se but rather it informs researchers about the explicit pathways along which behavior change can occur. Although there has been much conceptual work on how social network structures affect diffusion processes in general [39–44], this work has yet to be explored explicitly in the context of social network-based HIV prevention models [8–17].

Network size and the rate of growth are of primary interest to intervention efforts. All social network-based HIV prevention interventions are based on, at least in part, the Diffusion of Innovations theory [45]. The classic mid-twentieth century empirical studies of diffusion, which inform the theory, almost always tracked the growth and adoption of interventions [45]. Quite literally, the size and rate of adoption were the outcomes of interest. Measuring and quantifying these growth curves in the context of HIV prevention interventions have been extremely rare.

Network density measures the number of ties in a network as a proportion of the total possible number of ties given the size of a network; density is important to the diffusion process—information and adoption flow faster in denser networks [39–44]. An important note, density and size are not mathematically independent. As the size of a network increases, the density of ties tends to decrease [46].

Centrality of positions is another critical concept in diffusion models [39–43]. In general, centrality refers to the status, prestige or ‘importance’ of a particular actor, or set of actors, in a network. There are more than a dozen metrics for assessing positional centrality. The three most commonly discussed include (i) degree (the number of ties a given person has to other persons in the network), (ii) betweenness (the amount to which a particular person lies on the shortest path between two others in the network) and (iii) eigenvector centrality (centrality as a function of direct ties and the ties-of-ties in the larger network [47]). Centrality measures allow understanding of the extent to which varied types of actors are ‘important’ to the network.

Social network theory posits that similarity breeds connection, a simple notion commonly referred to as homophily [48]. Networks tend to be constructed of people who share sociodemographic and behavioral characteristics. In general, people are more likely to share network ties with other persons of similar race/ethnicity, age, religion, education, occupation and gender [48]. This propensity to form homophilous ties has implications for the kind of information persons have access to, the attitudes they form, their behavioral norms and the quality of their interactions. Homophily informs intervention design by elucidating the sociodemographic and behavioral characteristics most salient to the established network by measuring the extent to which the proportion of homophilous ties within a network exceeds what would be expected by chance [49]. For example, if ties to others are disproportionately constructed of like-aged others, then we can interpret that to mean that age was a salient social dimension in the formation of the group. Analyses such as this can inform intervention development by uncovering salient dimensions and characteristics of persons in the network who may be more difficult to reach and who require additional outreach efforts.

Summary

Because creating social networking interventions with online components is of great interest to HIV prevention science [1–7] and because the resulting networks are quantifiable, this study focuses on examining pilot data on the acceptability of a new network-based intervention model. Recent work by Leon et al. [50] suggest that pilot data should not be used to conduct hypothesis testing about outcomes but rather efforts should be focused on collecting data that can guide in the design and implementation of larger scale efficacy studies. Following their lead, these pilot study data focus on issues of the acceptability of recruitment to, assessment of and participation in this pilot program. The social network analysis of participation, in particular, we believe sheds light on needs for the future.

Materials and methods

Intervention acceptability pilot

Three different categories of participants were included in the pilot study: peer leaders (PL) (n = 7), face-to-face youth (F2F) (n = 53) and online youth (OY) (n = 103). The pilot ‘Have You Heard’ was named by the youth who participated in this study. The model used trained PLs to engage face-to-face peers (F2F) in the creation of digital media (e.g. You Tube videos) that promoted HIV prevention. Both the PL and F2F participants invited ‘friends’ from their MySpace and Facebook networks to join the profile/group pages on MySpace and Facebook. The process of creating and disseminating the digital media provided youth a platform for informal discussions about the importance of condom use and regular HIV testing.

The pilot intervention used social identity theory [51, 52] to guide the augmentation of pro-social identities among homeless youth, as has been done in other HIV prevention settings (i.e. self-help in eliminating life-threatening diseases, Self-Help in Eliminating Life-Threatening Diseases; [15, 16, 53]). The intervention included psychosocial training and skill building for PL [54], as is typical of most effective HIV prevention interventions [55]. For PL and F2F, engagement in the creation of youth-conceptualized and youth-produced digital media was based on theories of community mobilization and empowerment via participatory community theater models [56]. For all participants, dissemination of online media and the accompanying HIV prevention dialog utilized the Diffusion of Innovations [45], which has successfully been employed in face-to-face HIV prevention strategies [8, 57, 58].

Recruitment and participants

Peer leaders

All PL were recruited at a single community-based drop-in agency serving homeless youth aged 13–25 years through referrals from agency staff. All PL were screened for interest and consented privately.

Face-to-face

All F2F were invited by PL during agency drop-in hours to participate in workshops to be held in a separate room at the agency. F2F were consented privately by a member of the research team; informed assent was obtained for minors (a waiver of parental permission was obtained from the university Institutional Review Board).

Online youth

No eligibility criteria were used for OY. PL and F2F asked their MySpace and Facebook friends to send ‘friend requests’ to the program's profiles on MySpace and Facebook. Moreover, PL logged into the profiles and sent ‘friend requests’ directly to persons who were their current MySpace and Facebook friends. Prominently featured on the profile pages was language informing potential OY that the profile was part of a research project and participation was voluntary and would be recorded and monitored.

Assessment

All participants were asked to complete a 39-question self-administered survey. After being consented, PL and F2F were asked to complete a paper and pencil version of the survey. All seven PL completed the survey (100%) and 26 of the 53 F2F (49%) completed a baseline survey. All OY were sent a link to an online version of the survey; only 2 of the 103 (1.9%) completed the survey. No additional incentives were provided for completing the survey. As we will discuss in the Discussion section, we wanted to determine the acceptability of the pilot's programming and did not want to bias participation toward youth who would agree to participate just to receive compensation for completing the assessments.

Participation

Peer leaders

PL participated in 1 week of PL training, 1 week of website development and 9 weeks of peer engagement and prevention message dissemination. All training and workshop time were co-facilitated by at least one trained intervention facilitator and one of two designated agency staff members who partnered with the program. PL training was adapted from the SHIELD program for HIV prevention outreach work with injection drug users [53]. Topics included (i) assertive communication skills, (ii) HIV prevention information, (iii) how to craft norm-changing messages, (iv) outreach techniques such as ‘lead ins’, (v) how to deliver messages online that are engaging and not considered as spam, (vi) brainstorming about online media (e.g. You Tube) to engage peers in discussions of HIV prevention and testing and (vii) online presentation of self and potential ramifications (e.g. employment) of one's web presence and online activities.

Following their training, PL participated in 1 week of semi-structured time where they created the initial profile pages on MySpace and Facebook and conceptualized HIV prevention digital media projects that would be used to engage F2F in the program. For the final 9 weeks of the program, PL were required to participate in a minimum of 5 hours each week (maximum of 10 hours). Time was evenly split between PL-only work time and outreach where PL recruited F2F to work on the creation of HIV prevention digital media (e.g. You Tube videos) and where informal norm-changing conversations about condom use and HIV testing were delivered by PL to F2F. PL were compensated $8 for every hour of training or participation ($80 maximum per week) and provided a weekly bus pass ($17 value).

Face-to-face

F2F were invited to 17 workshops by the PL. During 14 workshops (82% of the workshops), F2F participated in the creation of HIV prevention digital media, conceptualized by the PL (e.g. You Tube videos promoting safeer sex or HIV testing). F2F were allowed Internet access during nine (53%) workshops, during which time they could ‘friend’ the group profile, look at completed media projects, disseminate media projects and encourage their online peers to become OY. All F2F were provided with refreshments during their participation in workshops. Attendance data were recorded.

Online youth

Because of the interactive nature of online social networking sites, OY were more than merely an audience for the ‘Have You Heard’ produced media, they were also active participants in the online community, commenting and posting ideas, media and messages related to HIV prevention.

Network data sources included (i) Daily ‘screen grabs’ of the Facebook and MySpace profile pages, recording the cumulative number of friends to date and the identity of those friends (plotted over time in Fig. 1). (ii) Publically accessible information from friends profiles on age, gender and ‘common friends’. Two matrices of social network data were created from the information pulled from youth who ‘friended’ the Facebook and MySpace profiles. Using the common friends functionality, a list of each common friends was collected. This created a complete set of mutual ties among the participants, independently for the Facebook and MySpace profiles.

Fig. 1.

Growth curves of the Have You Heard intervention, homeless youth and peers, Hollywood, CA, 2009. Note: Left-hand graph depicts all study participants; right-hand graph depicts MySpace and Facebook growth independently including PL, F2F and OY in the total counts for Facebook and MySpace network sizes.

These data were entered into NetDraw 2.090 and the spring embedder routine was used to generate the network visualizations presented in Fig. 2. Spring embedding is based on the idea that two actors may be thought of as pushing or pulling each other and two points located close together represent actors who have a pull on each other, while distant actors push one another apart. The algorithm seeks a global optimum where there is the least stress on the ‘springs’ connecting actors to one another [59].

Fig. 2.

Resulting MySpace and Facebook networks of the Have You Heard intervention, homeless youth and peers, Hollywood, CA, 2009. Note: Left-hand graph depicts the MySpace network and right-hand graph depicts the Facebook network, circles indicate participants, lines indicate online social media ‘friend’ ties, white = OY, gray = F2F and black = PL.

Using NetDraw, general descriptive properties of the networks were calculated separately for each network, including size and density (number of ties in the network divided by the number of possible ties in the network). Position-specific measures of centrality were also calculated for each network separately: degree, betweenness and eigen centrality [47]. In addition, two homophily measures were created for each person in both networks: percent of ties of the same gender (0 meaning, no same gendered ties; 50 meaning, half of the ties were male and half female and 100 meaning, all ties are the same gender) and percent of ties that are within plus or minus one year of the person's age [48]. Because of high-refusal rates in the baseline survey among F2F and OY, homophily analyses were limited to an examination of age and gender (data publicly available on Facebook and MySpace profiles).

Results

Table I displays descriptive statistics from the baseline assessments. Most youth reported having been tested for HIV within the prior 6 months (71% PL and 74% F2F). Very few reported ‘always’ using condoms when they had sex in the prior 90 days (28% of PL and 7% of F2F). Most youth at the time of the survey were staying in shelters, transitional living programs or sleeping on the streets.

Table I.

Descriptive statistics of participants (n = 163), Los Angeles, CA, 2009

	Participant characteristics
	PLs		Face-to-face youth		OY
	n	%	n	%	n	%
Total youth	7	100.0	53	100.0	103	100
Gender
Male	4	57.1	33	62.3	55	53.4
Female	3	42.9	20	37.7	48	46.6
Race
African–American	4	57.1	7	25.9
Latino	0	0.0	7	25.9
White	2	28.6	5	18.5
American–Indian	1	14.3	0	0.0
Mixed race	0	0.0	7	8.0
Sexual orientation
Homosexual	4	57.1	3	11.1
Bisexual	0	0.0	7	25.9
Heterosexual	3	42.9	17	63.0
Living situation
Streets	0	0.0	9	34.6
Shelter	2	28.6	3	11.5
Transitional living program	2	28.6	4	15.4
‘Couch surfing’ with friends or relatives	1	14.3	5	19.2
Other (e.g. motel, boyfriend)	2	28.6	5	19.2
HIV-related risk
HIV test past 6 months	5	71.4	20	74.1
‘Always’ used condoms past 90 days	2	28.6	2	7.4
	Mean	SD	Mean	SD	Mean	SD
Age	22.8	1.8	21	2.3	20.9	1.9
Hours participated	92.6	8.5
Workshops attended			2.5	1.9

Hours participated does not include data from the 1 PL lost due to incarceration. Age and gender of OY collected from social media profiles.

Figure 1 displays the recruitment over time. After the fourth week of intervention (at this time approximately 50 F2F had participated), there was a rapid increase in online participants, especially on MySpace. Seven PL began the training. One PL was arrested after he completed training but before F2F and OY recruitment began; he spent the remainder of the intervention incarcerated. The remaining six PL successfully recruited 53 F2F participants and collectively these 59 PL and F2F recruited an additional 103 youth. As data from Table II show, most OY were MySpace users rather than Facebook, but this difference did not emerge until half way through the program. Given the rapid dissemination of Facebook, since 2009, we would anticipate lower MySpace and higher Facebook participation in future versions of this study.

Table II.

Descriptive statistics of participants (n = 163) and networks (n = 2), Los Angeles, CA, 2009

	Network characteristics
	MySpace		Facebook
Network properties
Size	111		36
Overall density	0.1292		0.1841
Participant type	n	%	n	%
PL	5	4.5	6	16.7
F2F	30	26.8	3	8.3
OY	76	68.8	27	75.0
Gender
Male	57	51.4	22	61.1
Female	54	48.6	14	39.9
	Mean	SD	Mean	SD
Age	21.0	2.1	23.6	6.9

Participation can be assessed by individual participation, group-level productivity and network engagement. Excluding the incarcerated youth, on average, PL participated in 93 (86%) of a possible 108 hours of intervention programming. Twenty-seven (51%) of the 53 F2F attended two or more workshops. By the end of the 11 weeks, participants created 24 You Tube videos promoting condom use and HIV testing (some humorous and some serious), viewed a total of 1212 times (as of 11 November 2010), three digital comic books depicting fictional stories of HIV risk and two ‘photo essays’ promoting condoms and safe sex.

Table II presents the overall descriptive statistics for the two online networks. The MySpace network was much larger than the Facebook network, although the Facebook network was denser. Density is a function of size; larger networks are typically less dense [46]. Overall, 30 (57%) of the 53 F2F youth friended the program's profiles. Three F2F who friended the Facebook profile also friended the MySpace profile. Based on results from a bivariate logistic regression, youth who attended more workshops were significantly more likely to ‘friend’ the program profiles (odds ratio = 2.78; 95% confidence interval = 1.34–5.80; chi-square test = 7.47, P < 0.01).

The network visualizations presented in Fig. 2, along with the position-specific analyses presented in Table III, explain much of this growth pattern. One PL was an active Facebook user and refused to create a MySpace account because of perceived past online harassment on MySpace. Five of the six PL were active MySpace users who created Facebook accounts for the purpose of the program. The PL in the center of the Facebook network was largely responsible for the creation of this network. He had the highest degree centrality (C_D = 9) and betweenness centrality (C_B = 281.9) for the entire network. After a month of extending this network, his efforts reached a plateau. It is important to note, however, that when one compares the PL and F2F participants as a group to the OY participants, there was no significant difference overall in either of these two measures of centrality. Eigen centrality was significantly lower for the PL and F2F youth as a whole.

Table III.

T-test for variation in network properties by participant type (n = 163), Los Angeles, CA, 2009

Network properties	Participant type				t-statistics
	PL and F2F		OY
	Mean	SD	Mean	SD
MySpace network
Degree centrality	4.66	4.97	8.90	6.31	3.83***
Between centrality	68.80	118.90	83.06	165.40	0.52
Eigen centrality	0.04	0.05	0.08	0.06	3.79***
Percent homophily by age	29.87	32.22	40.48	27.87	1.58
Percent homophily by gender	50.03	36.22	52.89	26.46	0.39
Facebook network
Degree centrality	3.11	2.52	3.48	2.62	0.38
Between centrality	39.65	91.65	24.15	44.52	−0.49
Eigen centrality	0.03	0.05	0.12	0.15	2.75**
Percent homophily by age	44.05	30.70	51.11	48.77	0.39
Percent homophily by gender	38.89	23.34	59.42	38.81	1.81

***P < 0.001 and **P < 0.01.

The MySpace network was a more collective endeavor and its growth never hit a similar plateau. Two of the PL held relatively central positions in the network (C_D = 8, C_B = 389.2 and C_D = 10, C_B = 33.6, respectively). Four F2F, however, had higher degree centrality scores (C_D ranging from 13 to 21) and one F2F youth had a higher betweenness score (C_B = 431.9). The most central actor in the MySpace network was an OY who had 32 ties and a betweenness score of 1165.9. OY were significantly more central, as compared with F2F and PL (Table III).

Table III also presents the results of the homophily analyses. Neither gender nor age homophily was associated with OY relative to F2F and PL participation. Two additional homophily analyses were run, examining the association between gender and age on homophilous associations. Age was significantly negatively associated with the percent of homophilous age ties in the MySpace network (r = −0.26, P < 0.01) and the Facebook network (r = −0.48, P < 0.05). Hence, younger participants had more similar-aged network ties. In the MySpace network, more females had same gender ties (61 versus 43%; t-test = 3.31, P < 0.01).

Discussion

Pilot study results provide critical information, which can guide in the design and implementation of larger scale efficacy studies [50]. In the present study, our data provide guidance in issues of recruitment, assessment and participation in particular. First, recruitment was a success. In just 9 weeks of outreach, a group of seven youth were able to engage 53 face-to-face peers in the creation of digital media and the dissemination of these media projects to online networks. These 60 youth were able to recruit an additional 103 youth to participate in the program as friends of the online program.

The social network analysis revealed important intervention dynamics that would have been impossible to observe otherwise. Specifically, a ‘critical mass’ of 50 F2F was needed before the OY network began to ‘take off’. While 50 may not be the requisite number in all settings, for a drop-in center, which sees on average more than 400 unique clients in a month, this was the number of participants required.

The centrality analyses were also illuminating. Three of the PL held central positions in the final networks, whereas the others were relatively unimportant, as reflected by their different degree, eigenvector and betweenness centrality scores. PL studies, especially pilot studies, rarely include information about which PLs held central positions in the networks, which emerge in the process of the intervention. The three PL who held relatively central positions all had larger online networks at the outset of the study. In future, recruiting more PLs with large online networks is important if online dissemination is one of the goals of the project.

The homophily analysis revealed two important issues: younger youth are disproportionately connected to same aged peers and females are disproportionately connected to other females. We would like to suggest that the homophily analysis in particular may be an important new tool for testing the acceptability of peer-based prevention programs. This analysis suggests for the future feasibility phase, efforts will need to be made to reach younger youth and female youth who tend to be located in networks dominated by others like themselves. Different populations may uncover different demographic or even behavioral dimensions along which persons cluster, which may help direct the subsequent phases of studies for those populations.

The assessment data revealed two important pieces of information. First, PL and F2F youth who completed the assessments were indeed at risk for HIV, with only 4 of the 31 total youth surveyed indicating consistent condom use. Second, the process of collecting the assessment data will need to be modified in the future. Ninety-eight percent of OY and 51% of the F2F failed to complete an assessment, likely because they received no compensation for completing the assessment. We did not provide an incentive for completing the assessment because we were primarily concerned with understanding the acceptability of the new intervention programming to homeless youth. It has been well established in many prior studies that homeless youth will participate in survey research if they are compensated financially [21–30]. We did not want youth to participate in the program simply to receive financial compensation for completing a baseline survey. Rather, we wanted to know if PL could effectively recruit F2F to engage in making digital media about HIV prevention. The only compensation provided F2F youth were refreshments (e.g. a can of soda, a bag of chips). This decision gives us greater confidence in the acceptability of the program among homeless youth and yields two basic conclusions and insight for the future: (i) homeless youth can recruit their peers into an HIV prevention media creation intervention without relying on large incentives. (ii) Programming engagement does not translate into engagement around assessment. Future iterations of this program must compensate homeless youth participants for their time if assessment data are to be collected, which will allow one to test for program effectiveness in the future.

There were two important ‘lessons learned’ from this pilot which do not come across simply in these data. First, the online networks served as a great tool for participant retention. One of the greatest challenges facing prevention program implementation for homeless youth is getting youth to attend a large number of intervention sessions over time because of the transience induced by their living situations. The online component of this intervention provided a means by which youth maintained their connection to the program over time and felt comfortable returning after missed sessions because their online presence meant they never ‘left’ the group.

Second, connecting to pro-social peers was a critical component to the success of this intervention, both face-to-face and online. In the context of the self-directed and produced digital media creation, youth were able to form a community of other pro-social youth where they could freely discuss sexual health. Moreover, recent research suggests that youth who connect out of street life back to family and positive home-based peers will reduce their risk-taking [36–38]. Some youth leveraged their work in the program as an opportunity to reconnect with positive network ties by sharing online media they had created. There were two instances where youth reported back to the intervention team that they had shown their work to their family or friends from home. This sharing was the first point of reconnection to those positive home-based networks in several months.

As with any pilot study, this study has several limitations. First, the social network analysis assesses the structure along which diffusion can occur but not the actual diffusion of behavior change. The network formation demonstrates the acceptability of the intervention but not its effectiveness. Second, we have no comparison group. Without the preliminary acceptability data, we could not estimate the necessary size or composition of such a group or more importantly what activity/activities should youth engage in as the focus of a comparison network. Furthermore, we were unable to collect data on non-response coverage, an inherent difficulty of conducting online social network research. The low response rate to the baseline survey limited the breadth of homophily analyses available to us. Last, this intervention was piloted at one agency serving homeless youth, and so these results are not generalizable to settings with less Internet access or a different demographic profile of clients.

Despite these limitations, this study provides some important steps forward in HIV prevention work. First, a peer-led program, which focused on creating HIV prevention-focused digital media and disseminating that media to online networks, was highly acceptable and would appear to warrant further development. Second, to our knowledge, this is the most explicit social network analysis of an HIV prevention network to be examined within the context of intervention program development. Since social network data are so easy to collect in the context of social networking websites, we highly recommend that other researchers we are developing new online prevention efforts collect and analyze data as we did here. Without this analysis, we would not have known which PLs were most central in recruitment efforts. In addition, the homophily analyses point to network spaces which require additional outreach efforts. We are utilizing these insights in the development of the feasibility trial for this program and believe that others would benefit from this strategy as well.

Funding

National Institute of Mental Health (K01 MH080605).

Conflict of interest statement

None declared.