Social network analysis is increasingly used to better understand the drivers of HIV transmission because this analytic approach moves beyond the monolith of individual risk behavior as the dominant explanatory factor of onward HIV transmission.1–3 Explanations of risk and transmission are dependent on where one is located within a network of interest and the pattern of behaviors and infections among other network members that surround a given individual.4,5 Within network analysis, there are multiple network levels that can help illuminate network risk. Most common in the field of sexually transmitted infection (STI) networks is that of the sexual network, that is, the characterization, visualization, and analysis of the direct sex connections between individuals. These have been critical to understanding patterns of STI transmission, leading scholars to observe network level risk, which has included concurrency,3,6 network position,7,8 personal network density,9 and assortative mixing.10,11 Most analyses that examine networks, whether they be social, sexual, risk, or support, focuses on the ties between individuals. However, increasingly, there has been interest in the structural features of the social environment—we call this sociostructural features—the networks and spaces that surround individuals and confers HIV risk.12 These sociostructural features that drive infection increasingly move beyond individual and person-to-person network factors and consider spaces and other institutional factors, as well. Sociostructural features could be the venues where individuals socialize or have sex,13 or the health care or prevention centers that they visit.14 Getting at this overlap between individuals within networks and the spaces that they occupy can be done by using 2-mode network analysis (the first mode is people or actors, and the second mode is places or events).15,16 Two-mode (or bipartite) network analysis is an increasingly popular analytic approach used by Fujimoto et al17 and others in health behavioral research to explicate complex sociostructural networks that identify shared venues or spaces between network members whose colocation in these spaces can provide powerful insights into both patterned risk for the individuals affiliating with these spaces and the organizational structure of these spaces.18
Within the context of STI risk networks, 2-mode network analysis helps us disentangle the complex relationship between a sex network of individuals and the environment that these networks are embedded—the sexual affiliation network.13 One can explore not only the risks or protective factors that are evident within this network but also how the patterned affiliation of network members to venues or other spaces creates the venue or space structure. Two-mode data are generally collected by asking individuals to select from an a priori list, what spaces they have been to over a set of time. The data are then displayed in a rectangular matrix form referred to as an affiliation matrix, whereby rows indicate individual respondents and columns the individual spaces they affiliate with (the entry of 0 represents no affiliation and the entry of 1 represents an affiliation). The respondent-by-respondent connections through shared space affiliation can be created by multiplying this matrix by its transpose—the transpose is a matrix on its side where the rows become columns and columns become rows. This can be done using social network programs such as UCINET,19 statnet20 in R environment, or any languages that handle matrix algebra (MATA in Stata, IML in SAS, etc). In this type of analysis, we are able to create a network of spaces created by shared respondent affiliation that allows us to characterize the venues of structural importance, for example, the spaces or venues that are most bridging or those most centrally located in the venue network. In addition, we can model how certain risks, behaviors, and protective factors might move through a social network made up of individuals through the affiliation patterns of venues that these individuals are connected to.
Fujimoto et al use 2-mode network analysis to examine male sex worker (MSW) networks in Houston, Texas, and the sociostructural networks that these sex workers are embedded within.21 The team examined venue-based affiliation networks of MSWs and the relation of network structures to risky sexual behavior (defined as unprotected anal intercourse [UAI]) and HIV status. By collecting information on the venues that MSWs nominated—but not the connections between individual MSWs—they were able to conduct several key analyses of interest (in order of complexity): (1) MSW centrality, a greater centrality represents a larger number of sex connections MSWs reported; (2) venue centrality, a greater centrality represents a larger number of MSW who nominated a given venue; (3) clustering of behaviors/statuses based on the pattern of sharing venues between MSW respondents; and (4) exposure of MSWs to behaviors, which is operationalized by the proportion of the shared venues respondents affiliated with other MSWs who engage in behavior of interest. The first 2 analyses are commonly conducted in studies of individual risk behavior and are not typically thought of as network analyses despite the existence of individuals and their ties in this data type. The third analysis is what is traditionally implemented using 2-mode network analysis and mapping distinct clusters by linking the structural pattern of respondent’s network via venue affiliation to specific characteristics or behaviors of those individuals. The final analysis was developed by the first author and others17,22 and is a network influence model applied to 2-mode affiliation network, which is called “Affiliation exposure model.” The affiliation exposure model allows us to investigate the potential impact of influence (measured by the combined information of venue affiliation with other MSWs behavior) on an individual’s own behavior. These analyses are incredibly important because little is known about MSW networks and their relation to the venues where diffusion of risk occurs.
Fujimoto et al thus were first able to characterize and visualize the sociostructural features of physical venues, in this case, bars and intersections, where MSW affiliate and coaffiliate and then to examine the exposure to behaviors via these venues. This provides us with a rich understanding of the pattern of venue affiliation and how bars and intersections are or are not related to each other through the MSWs interviewed. The information obtained is directly actionable by public health authorities. For example, an intervention may focus on prevention efforts on one bar within a cluster of intersections/bars even if it may seem more efficient to target another more popular bar (ie, higher-degree centrality), because the intervention bar may link to another group of bars/intersections through the shared behaviors across these venues. In addition, by using the affiliation exposure model, Fujimoto et al describe some counterintuitive findings. They find that MSWs who are increasingly exposed to other UAI practicing MSWs actually practice less UAI. This goes against most network theories of homophily23 or assortative mixing (like with like), which describe attributes or behaviors that cluster within groups. Fujimoto et al suggest that this may caused by risk compensation because MSWs may be likely to practice less UAI given a perception that certain environments or venues are risky. Although this is certainly possible, other alternatives should be considered in this specific context. The finding of decreasing UAI among MSWs who are in contact with MSWs who practice more UAI might also be the nature of competition within sex venues such that MSWs who are practicing more UAI are competitively eliminating other MSWs ability to practice UAI who, then by default, are having less UAI. We do not know from the data presented whether the number of sex events is included in the model and the denominator of total sex acts with or without a condom. It is also unclear whether MSWs are having sex with each other or with outside men who have sex with men (MSM) clients who were not included in the MSW only network boundary specification in this analysis.24 Cruising sites and nontransactional sex events of MSM may be consistent with the original hypothesis of the team that increasing risk behavior within a network increases the risk—which has been found to be the case among younger black MSM networks.11,25 Another limitation not highlighted was that the sample was mostly white and limits our interpretation of current network epidemiology among underrepresented minority populations most affected by HIV.26 Because causality could not be determined through the present analysis, future studies should look at the affiliation exposure model over time, which will be incredibly powerful for determining how risk behavior is propagated through venues. In the future, it may be that our analyses generated from the pattern of venues/intersections can pinpoint risk with only limited information about respondents. Increasing use of digital communication technology (cell phones and social media applications) by these populations and researchers improves data collection potential and ability to map risk in real time. Also, future studies that overlay actual person-to-person sex networks with the sexual affiliation network—that is, the venues where sex network members affiliate—will validate some of our assumptions made about exposure at venues. Thus, research could shift from these often difficult to characterize sex networks to affiliation networks. Perhaps future survey questions will move from “did you use a condom in the last 7 days?” to “which of the places on this list were you hanging out at over the past 7 days?”
1. Friedman SR, Aral S. Social networks, risk-potential networks, health, and disease. J Urban Health 2001; 78: 411–418.
2. Laumann EO, Ellingson S, Mahay J, et al. The Sexual Organization of the City. Chicago: University of Chicago, 2004.
3. Morris M, Kretzschmar M. Concurrent partnerships and the spread of HIV. AIDS 1997; 11: 641–648.
4. Laumann EO, Youm Y. Racial/ethnic group differences in the prevalence of sexually transmitted diseases in the United States: A network explanation. Sex Transm Dis 1999; 26: 250–261.
5. Rothenberg R, Hoang TDM, Muth SQ, et al. The Atlanta urban adolescent network study: A network view of STD prevalence. Sex Transm Dis 2007; 34: 525–531.
6. Kretzschmar M, Morris M. Measures of concurrency in networks and the spread of infectious disease. Math Biosci 1996; 133: 165–195.
7. Bettinger JA, Adler NE, Curriero FC, et al. Risk perceptions, condom use, and sexually transmitted diseases among adolescent females according to social network position. Sex Transm Dis 2004; 31: 575–579.
8. Fichtenberg CM, Muth SQ, Brown B, et al. Sexual network position and risk of sexually transmitted infections. Sex Transm Infect 2009; 85: 493–498.
9. Latkin C, Mandell W, Vlahov D, et al. Personal network characteristics as antecedents to needle-sharing and shooting gallery attendance. Soc Netw 1995; 17: 219–228.
10. Doherty IA, Schoenbach VJ, Adimora AA. Sexual mixing patterns and heterosexual HIV transmission among African Americans in the Southeastern United States. J AIDS 2009; 52: 114–120.
11. Schneider JA, Cornwell B, Ostrow D, et al. Network mixing and network influences most linked to HIV infection and risk behavior in the HIV epidemic among black men who have sex with men. Am J Public Health 2013; 103: e28–e36.
12. Auerbach JD, Parkhurst JO, Caceres CF. Addressing social drivers of HIV/AIDS for the long-term response: Conceptual and methodological considerations. Glob Public Health 2011; 6: S293–S309.
13. Frost SDW. Using sexual affiliation networks to describe the sexual structure of a population. Sex Transm Infect 2007; 83: I37–I42.
14. Schneider JA, Walsh T, Cornwell B, et al. HIV health center affiliation networks of black men who have sex with men: Disentangling fragmented patterns of HIV prevention service utilization. Sex Transm Dis 2012; 39: 598–604.
15. Breiger RL. Duality of persons and groups. Soc Forces 1974; 53: 181–190.
16. Feld S. The focused organization of organizational ties. Am J Sociol 1981; 86: 1015–1035.
17. Fujimoto K, Chou CP, Valente TW. The network autocorrelation model using two-mode data: Affiliation exposure and potential bias in the autocorrelation parameter. Soc Netw 2011; 33: 231–243.
18. Latapy M, Magnien C, Del Vecchio N. Basic notions for the analysis of large two-mode networks. Soc Netw 2008; 30: 31–48.
19. Borgatti S, Everett MG, Freeman LC. Ucinet 6 for Windows: Software for Social Network Analysis. Harvard: Analytic Technologies, 2002.
20. Handcock MS, Hunter DR, Butts CT, Statnet: Software Tools for the Statistical Modeling of Network Data. Available at: http://statnetproject.org
21. Fujimoto K, Williams ML, Ross MW. Venue-based affiliation networks and HIV risk-taking behavior among male sex workers. Sex Transm Dis 2013; 40: 453–458.
22. Fujimoto K, Unger JB, Valente TW. A network method of measuring affiliation-based peer influence: Assessing the influences of teammates’ smoking on adolescent smoking. Child Dev 2012; 83: 442–451.
23. Mcpherson JM, Smithlovin L. Homophily in voluntary organizations—Status distance and the composition of face-to-face groups. Am Sociol Rev 1987; 52: 370–379.
24. Laumann EO, Marsden PV, Prensky D. The boundary specification problem in network analysis. In: Freeman LC, White DR, Romney AK, eds. Research Methods in Social Network Analysis. Fairfax, VA: George Mason University Press, 1989.
25. Schneider JA, Laumann EO. Alternative explanations for negative findings in the community popular opinion leader multisite trial and recommendations for improvements of health interventions through social network analysis. J Acquir Immune Defic Syndr 2011; 56: e119–e120.
26. Hemmige V, McFadden R, Cook S, et al. HIV prevention interventions to reduce racial disparities in the United States: A systematic review. J Gen Intern Med 2012; 27: 1047–1067.