Socially Connected Neighborhoods and the Spread of Sexually Transmitted Infections

Abstract

Sexually transmitted infections (STIs) not only continue to increase at record levels in the United States but also exhibit unequal spatial patterning across urban populations and neighborhoods. Past research on the effects of residential and nearby neighborhoods for STI proliferation has largely ignored the role of socially connected contexts, even though neighborhoods are routinely linked by people moving across space for work and other social activities. We showcase how commuting and public transit networks contribute to the social spillover of STIs in Chicago. Examining data on all employee-employer location links recorded yearly by the Census Bureau for more than a decade, we assess network spillover effects of local community STI rates on interconnected communities. Spatial and network autoregressive models show that exposure to STIs in geographically proximate and socially proximate communities contributes to increases in local STI levels, even net of socioeconomic and demographic factors and prior STIs. These findings suggest, socially connected communities influence each other’s infection rates through social spillover effects.

Sexually transmitted infections (STIs) are occurring at unprecedented levels in the United States. For example, the Centers for Disease Control and Prevention (CDC) reports that in 2018 chlamydia reached the highest number of reported cases ever (1.8 million cases in 2018; a 3% increase since 2017 and 19% increase since 2014) (CDC, 2019)¹. STIs are an important population health concern due to the ways in which they can reflect and reproduce inequalities. The highest concentration of cases occur for young adults, racial and ethnic minorities, and the low income (Adimora & Schoenbach, 2005, 2013; Harling et al., 2014; Thomas & Thomas, 1999), and cases are disproportionately clustered in urban areas (Adimora & Schoenbach, 2005; De et al., 2004; Potterat et al., 1985; Risley et al., 2007). Multiple factors contribute to this link such as inequalities in healthcare access (Nelson, 2002), racially segregated sexual networks (Adimora & Schoenbach, 2005; Laumann & Youm, 1999; Liljeros et al., 2003), and residential location (Dembo et al., 2009).

STIs contribute to population health disparities as untreated infections can lead to life-threatening complications such as cervical cancer through a contraction of HPV (Trottier & Franco, 2006) and other permanent health problems (Newton & McCabe, 2008; Tolnay, 1989; Trottier & Franco, 2006; Ward & Rönn, 2010). Research estimates that the lifetime direct medical cost of STIs is several billion dollars, with a large portion spent on HIV infections (Owusu-Edusei et al., 2013). Further, high rates of STIs may have important implications for at risk populations’ fertility levels (Bongaarts, 1978; Tolnay, 1989).

Understanding the effects of social, demographic, and ecological factors on these infection patterns and their contribution to individuals’ risk of contraction is crucial to decrease STI prevalence (Adimora & Schoenbach, 2013; Frieden, 2010; Grassly et al., 2001). Valuable past research has begun to account for the importance of meso-level factors in the shaping of STI patterns, in the form of social networks (Bearman et al., 2004; Kohler et al., 2007; Merli et al., 2015), schools (Jiskrova & Vazsonyi, 2019), and residential neighborhoods (Jennings et al., 2014; Jennings et al., 2012). In particular, social and sexual networks are highly influential in the transmission of STIs (De et al., 2004; Liljeros et al., 2003; Moody, 2002) as they can shape sexual partnerships (Bearman et al., 2004; Kretzschmar & Morris, 1996), perceptions of sexual risk (Kohler et al., 2007; Morris et al., 1995), and contraceptive use (Behrman et al., 2002; Kohler, 1997; Valente et al., 1997).

However, recent work in the neighborhood effects literature highlights the importance of looking beyond residential neighborhoods² and instead toward activity spaces - destination places of routine population mobility - to better understand the dynamics of selection effects, social interactions, and other contextual factors that shape health outcomes (Browning et al., 2017; Browning et al., 2004; Levy et al., 2020; Matthews & Yang, 2013). The way people connect neighborhoods through their everyday mobility may contribute to the spatially stratified pattern of STIs by influencing factors such as dating markets, norms, and attitudes surrounding sexual risk, and access to medical resources (Crosby & Holtgrave, 2006; Cubbin et al., 2005; Jennings et al., 2014; Singer et al., 2006; White et al., 2017). Although sexual behavior would not be generally observable in these networks, we build our study on past research that finds social relationships and social contexts can influence, without direct observation, sexual beliefs and behavior (Browning et al., 2004; Jiskrova & Vazsonyi, 2019; Upchurch et al., 1999), perceptions of sexual risk (Jennings et al., 2012; Kohler et al., 2007; Morris et al., 1995), and contraceptive use (Behrman et al., 2002; Kohler, 1997; Valente et al., 1997). These processes can occur through multiple pathways such as social learning, diffusion, or role modeling of behaviors (Ali et al., 2011; Kohler, 1997; Upchurch et al., 1999). For example, friends might discuss their use of contraception or contraction of a STI in their social network, which might encourage individuals to adopt an effective method to prevent STIs (Kohler, 1997; Morris et al., 1995). Using this foundation, we extend these ideas onto our analysis of community networks and their implications for STIs.

Our study aims to examine the significance of public transit networks and workers’ commute networks for the shaping and maintenance of STI patterns across neighborhoods in the urban setting of Chicago. We seek to answer whether the areas that neighborhoods are socially tied to are more important for their STI rate than the ones they are geographically nearby. One important theoretical line of debate in the neighborhood effects research has been in conceptually defining neighborhoods and understanding the spatial interaction mechanisms within areas of larger geographic scales. Although some studies focus on theoretical mechanisms relevant for small scale influence, such as census blocks or tracts, others highlight the importance of mechanisms relevant for wider contexts such as counties or metro areas, including neighborhood effects from socially connected neighborhoods (Sampson, 2012). This study proposes that wider scale effects may be expected because of population mobility flows and the inherent social interactions and exposures that form because of routine activities and mobility patterns such as commuting.

Historically, neighborhood effects research has highlighted the problem of residential segregation and the concentration of social disadvantage and vulnerability. However, recent work by Hall et al. (2019) highlights important differences between nighttime segregation (residential) and daytime segregation (based on commuters’ work location). Our approach builds on and extends this work by focusing on the connectivity across social space, above and beyond spatial segregation. Our study thus highlights the importance of looking beyond residential and geographically proximate neighborhoods to understand neighborhood effects by considering the effects from socially connected neighborhoods.

We build on past research by examining the structural networks that spread infectious diseases across Chicago. We adopt a dynamic spatial regression approach to assess how STIs spillover from geographically contiguous neighborhoods and commuting and public transit networks. By examining commuter mobility and public transit networks, the current study draws on existing approaches focused on spillovers among geographically proximate places and advances the literature further by assessing for the first time, to our knowledge, two key structural mechanisms potentially underlying spillover effects, in this case social spillovers, that feed into and amplify the differential clustering of health problems across urban space.

Literature Review

A Social Epidemiology Approach to STIs

A social epidemiology approach to health pushes the focus from the individual onto the larger social context individuals are encased in to better understand health patterns. These environments shape health outcomes through multiple forces such as norms, social control, and opportunities (Berkman & Kawachi, 2014). Such contextual factors result in an amalgamation of constraints and incentives for an individual’s health-related behaviors. Although the importance of a social epidemiologist perspective to health is uncontested by social scientists, STI research continues to primarily focus on the importance of individual-level factors for prevention such as gender (Burstein et al., 1998), sexual partnerships (Kelley et al., 2003), race (Laumann & Youm, 1999), condom use (Chatterjee et al., 2006), and socioeconomic status (Adimora et al., 2006; Harling et al., 2014; Thomas & Thomas, 1999).

However, these infections transmit through social interactions and are best studied interdependently through their interpersonal and environmental exchanges. Although individuals may contract an STI directly from their sexual partners, understanding the social structures within which people choose partners and in which they or their partners become infected allows for better identification of the social and ecological forces that shape and perpetuate disparate STI patterns. Additionally, STIs are intricately linked to socioeconomic status. STIs are only one of many adverse health conditions associated with socioeconomic status, which contributes to it being a fundamental cause of disease (Phelan et al., 2010). The understanding and modification of individual behaviors and risks needs to account for ecological inequalities stemming from socioeconomic status, which will continue to replicate even if individual-level determinants are addressed (Phelan et al., 2010). These factors highlight the necessity for research to move the focus beyond individual-level factors and behaviors to understand the disproportionate spread of STIs in urban environments.

Residential Neighborhood Effects on STIs

A crucial step in recognizing the ecological and social drivers of STIs is understanding how residential locations shape individual risk of STI contraction. Like other health behaviors (Arcaya, Tucker-Seeley, et al., 2016; Cubbin et al., 2005; Sampson, 2003), residential neighborhoods play an influential role in shaping and maintaining patterns of STIs and risky sexual behaviors (Brahmbhatt et al., 2014; Cubbin et al., 2005; Ellen et al., 2004; Jennings et al., 2014, 2010). One pathway these environments influence STIs is through attitudes and norms surrounding sexual risk and behavior that in turn shape individuals’ beliefs and behaviors (Cubbin et al., 2005; Jennings et al., 2014; Singer et al., 2006). These norms might transmit through the social interactions people have with their neighbors and other individuals in these contexts. Social relationships may also better inform individuals’ health decisions (Hernandez et al., 2019) and safer sex practices (Crosby & Holtgrave, 2006), as people can learn about health practices and information through their social network. Additionally, sexual behaviors and STI patterns might be influenced by community-level factors, such as social cohesion. Social cohesion means individuals have people within their residential community they can turn to for social support. Research finds that lower social cohesion is associated with higher rates of STIs (Ellen et al., 2004) and higher levels are positively linked to condom use (Kerrigan et al., 2006). Residential neighborhoods may also influence STI rates by providing medical resources to their residents such as clinics where free STI testing and/or condoms are provided.

Beyond Residential Neighborhoods

Even though residential neighborhoods are important contributors to STIs, they are not the only social environments that shape individual-level outcomes. Individuals are not static but often live dynamic lives within their social environments. A social epidemiology approach to STIs pushes past residential barriers and considers the places people inhabit everyday beyond their residential neighborhood. For one, neighborhoods are not isolated islands; their spatially contiguous neighborhoods often exert health relevant spillover effects (Baller et al., 2001). But moreover, most individuals spend enormous amounts of time in areas outside their home which might make these communities even more influential regarding individual health outcomes than places of residency. STI outbreaks tend to occur in concentrated clusters in urban environments (De et al., 2004; Potterat et al., 1985; Risley et al., 2007). However, past interventions only targeting these highly infected areas have not been successful (Rothenberg et al., 2005). Additionally, some outbreaks occur in random areas quite distant from highly infected areas (De et al., 2004). It may be that these interventions cannot address the spillover occurring from spatially proximate neighborhoods or spillover from distant neighborhoods socially connected by population mobility flows. Focusing only on residential areas ignores the many meaningful connections that people make in other social spaces (Small & Adler, 2019).

The social spaces individuals visit during their routine activities are inhabited by people who may have different attitudes surrounding sexual behaviors than those of an individual’s residential neighborhood (Cubbin et al., 2005; Jennings et al., 2014; Jiskrova & Vazsonyi, 2019; Singer et al., 2006). Socially connected communities might shape patterns in dating, which in turn shape and maintain STI patterns. People are more likely to have social ties to (Small & Adler, 2019) and sexual relationships with (Adimora & Schoenbach, 2005; Zenilman et al., 1999) those who are spatially proximate or socially near them. People encounter others through social organizations and routine activities which may expose them to potential sexual partners who live in different communities. Laumann et al. (2004) show that the distribution of sexual partnership ties in Chicago are sometime spread far and wide. Highly mobile people may serve as sexual links, termed “bridging,” that connect one neighborhood, and sexual network, to the other, thus increasing the risk of STI spread (Aral, 2000; Cassels et al., 2017). In sum, we hypothesize that inter-neighborhood networks are structural drivers of STI patterns and spread across urban space.

Inter-Neighborhood Commuting Ties

We focus specifically on the importance of inter-neighborhood networks based on commuting. Work environments are particularly important in people’s lives as they are the second, after residential areas, most frequently inhabited activity spaces by individuals (Kahneman et al., 2004). Inter-neighborhood commutes may shape STI patterns through social spillover or selection. Social spillover can occur when infected workers serve as bridges by introducing STI risk from their work neighborhoods into their residential neighborhoods via sexual partnerships (Aral, 2000; Cassels et al., 2017; Morris et al., 1996). Infected neighborhoods with many commuters may increase the possibility of long-distance transmission and STI outbreaks. Additionally, the places where people work, and their surroundings are important environments where daily social interactions occur. In these interactions it is possible that individuals are exposed to beliefs and norms surrounding health and sexual risk that might reaffirm or change their current views on STIs. Selection may also play a role in why some neighborhoods may be connected to each other through their commuters. Workers may be selecting into communities that are like their residential neighborhoods. Individuals select into the social spaces they inhabit which can help perpetuate systems of inequality (Arcaya et al., 2016; Sampson & Sharkey, 2008; van Ham et al., 2018). If individuals self-select into work environments where there are similar rates of STIs as their neighborhood of residency, these rates may persist over time. These individuals would be exposed to infected individuals to partner with and be exposed to the same norms of sexual risk that are in their own home community. Though our data preclude us from examining selection effects, we can investigate how communities look similar in an outcome of interest by examining autocorrelation. We account for social spillovers in addition to social autocorrelation between home and work communities to understand how commuting ties influence STI rates.

Inter-Neighborhood Public Transit Ties

Public transit ties between communities may also shape STI patterns as individuals use public transit to conduct their routine activities. Inter-neighborhood public transit ties are less malleable than inter-neighborhood commuting ties. Building or removing public transportation connections is an economic and political endeavor which takes time and resources (Farmer, 2011). Though the literature demonstrates that advantaged communities are often adept at influencing such processes (Karner & Niemeier, 2013; Sanchez, 2008), we believe that public transit ties may influence STI patterns through processes of social spillover more than selection. Communities directly linked together by a public transportation line, i.e., a bus route or a train line, are bridged by individuals using public transportation during their routine activity patterns. Individuals can easily visit communities connected through public transportation lines to conduct their shopping, routine medical visits, and attend social outings. These communities are the locations in which individuals may meet potential sexual partners during social outings and may provide them access to resources such as health clinics that are not present in their own neighborhoods. Though we cannot study the routine activity spaces of all individuals in Chicago, we propose that inter-neighborhood public transit ties may be a feasible and reasonable way to assess how communities connected through their residents’ routine activity spaces influence STI patterns. We account for social spillovers in addition to social autocorrelation between communities sharing the same public transit lines to understand how Chicago’s public transit infrastructure influences STI rates.

Methods

Study Setting

We situate our study in the urban environment of Chicago. As in many U.S. cities, STIs in Chicago have steadily increased in recent years, with chlamydia being the most pervasive (IDPH, 2017). Illinois had the 9th highest rate of chlamydia and 16th highest rate of gonorrhea of U.S. states in 2018 (CDC, 2019). These higher rates in STIs are due to the urban environment of Chicago that heavily weights the state’s STI statistics. Chicago ranked as the second highest city for total STI cases in 2018, preceded by Los Angeles, CA and followed by Houston, TX (CDC, 2019). Like other large cities, residents of Chicago have higher infection rates than those that reside in other areas of the state (IDPH, 2017). Though Chicago is a highly segregated environment, Sampson (2012) demonstrates that the processes shaping spatial inequalities in Chicago are not unique. Additionally, Chicago, like other cities, has a large population that commutes for work. Importantly, public transportation in a city like Chicago is used not only by individuals of lower socioeconomic status, but also by more affluent city residents (Farmer, 2011). Even though our study focuses on Chicago as a case study, the community networks we examine will likely operate similarly in other urban environments.

Data

We use multiple data sources to assess our research question. We configure the inter-neighborhood commuting network of Chicago using data from the Longitudinal Employer Household Dynamics (LEHD) Origin-Destination Employment Statistics (LODES). The LEHD, sponsored by the U.S. Census, collects information on the location of employers and employees based on unemployment insurance forms. To protect the confidentiality of individuals, it aggregates these data to commuting flow statistics within and between communities (Abowd et al., 2005). The commuting flow statistics allows researchers to examine how the geographic locations of both employers and employees are connected. From these data, we create an inter-neighborhood commuting network of Chicago’s 77 community areas³. Public transit data on Chicago Transit Authority (CTA) bus stops, rapid transit system stations (elevated “L”), and commuter rail (Metra) stations are from the City of Chicago’s data portal. We geocode the stations’ geographic coordinates to identify their CTA location and define the links between any two communities based on whether they share CTA bus routes, rapid transit lines, or Metra rail commuter lines.

We obtain socio-demographic indicators of these communities using data from the Decennial Census and American Community Survey (ACS). Finally, data from the City of Chicago’s Data Portal provide us information on the prevalence of STIs among different Chicago neighborhoods. These data are provided by the Surveillance, Epidemiology and Research Section, Sexually Transmitted Infection/HIV/AIDS Division of the Chicago Department of Public Health (CDPH).

Measures

STI Prevalence

We combine multiple measures to assess the prevalence of STIs in each area. The CDHP tracks the yearly number of lab-confirmed cases of chlamydia and gonorrhea among males and females aged 15 to 44. The City of Chicago’s Data Portal provides the incident rates per 100,000 people for each of Chicago’s 77 community areas. There are four incident rates provided: chlamydia incident rate among females aged 15 to 44; chlamydia incident rate among males aged 15 to 44; gonorrhea incident rate among females aged 15 to 44; and gonorrhea incident rate among males aged 15 to 44⁴. We combine these four items into a single standardized measure representing each community’s STI prevalence. This measure is created for each year in our study, i.e., 2002 to 2014⁵. In addition to serving as our dependent variable, we include a time-lagged variable that accounts for previous STI rates, referred to as “Prior STI Rate”. The inclusion of this time-lagged variable allows us to conservatively control for processes of selection and homophily that shape these connections at a prior time. Our reliance on lab-confirmed cases of STIs might underestimate community rates as many cases are unrecorded due to a lack of regular testing for STIs in clinical examinations, individuals’ lack of access or unwillingness to seek testing, and the high prevalence of asymptomatic cases (CDC, 2019; Mayer et al., 2012).

Community Sociodemographics

We account for several community area sociodemographic variables using the 2000 Decennial Census, and the 2008–2012 five-year estimate ACS. We include measures of community disadvantage, residential stability, and racial and ethnic diversity. We also assess the importance of the number of workers who work locally and do not commute for work using LODES data. We create a standardized index of community disadvantage and residential stability using factor weighted principal component analyses. The measures included in our index of disadvantage are the percentage of residents living below the poverty line, unemployed, receiving public assistance, and the percentage of female-headed families with children. The measures included in our index of residential stability are the percentage of residents over the age of five who have lived in the same house for the past five years and the percentage of owner-occupied housing units. We use a Herfindahl concentration index to calculate a community’s level of racial and ethnic diversity. This index is equal to one minus the sum of squares of the population proportions of each racial or ethnic group living in the community area; these racial/ethnic groups include: non-Hispanic Whites, non-Hispanic Blacks, Hispanics, Asians, Native Americans, and Others. We standardize this index, where higher numbers indicate greater diversity. Lastly, we measure the number of local workers by standardizing the proportion of jobs located in the community which are occupied by individuals who also live in the community. We also standardize the measure of local workers in a community so that all variables in the model are standardized. Table 1 shows our descriptive statistics. The variables are measured from multiple different sources, so we present statistics by data sources. Additionally, as many of our variables are indexed measures, we show values for the variables that make up these measures

Table 1.

Descriptive Statistics.

	Mean (S.D.)	Min, Max
STI Prevalence (Incidence) (2002–2014)
Chlamydia (Females)	2705.62 (2358.25)	89.59, 8340.05
Chlamydia (Males)	1093.90 (1049.56)	0, 3813.21
Gonorrhea (Females)	804.25 (918.90)	0, 3275.16
Gonorrhea (Males)	771.33 (851.25)	0, 2999.44
Disadvantage (Percent) (2000 Decennial Census)
Living Below Poverty Line	20.12 (13.09)	2.40, 56.31
Unemployed	11.72 (7.31)	2.80, 33.53
Receiving Public Assistance	8.12 (6.83)	0.23, 29.03
Female-Headed Families with Children	14.79 (10.60)	1.52, 50.28
Disadvantage (Percent) (2008–2012 ACS)
Living Below Poverty Line	23.30 (12.12)	2.95, 58.32
Unemployed	15.38 (7.53)	4.74, 35.87
Receiving Public Assistance	4.54 (3.50)	0.57, 19.81
Female-Headed Families with Children	21.07 (12.74)	2.75, 55.08
Stability (Percent) (2000 Decennial Census)
Residents Living in Same House in Past Five Years	57.30 (11.19)	30.14, 77.03
Owner-Occupied Housing Units	48.30 (22.10)	8.94, 91.14
Stability (Percent) (2008–2012 ACS)
Residents Living in Same House in Past Five Years	88.77 (4.73)	78.42, 97.66
Owner-Occupied Housing Units	49.11 (18.93)	12.88, 90.65
Diversity (Percent) (2000 Decennial Census)
Non-Hispanic Whites	31.20 (29.87)	0.32, 93.33
Non-Hispanic Blacks	40.89 (41.10)	0.17, 98.09
Hispanic	21.77 (25.15)	0.59, 88.90
Asians	4.34 (8.62)	0.03, 60.71
Native Americans	0.14 (0.07)	0.03, 0.40
Others	1.66 (1.09)	0.34, 5.42
Diversity (Percent) (2008–2012 ACS)
Non-Hispanic Whites	28.42 (27.88)	0, 92.22
Non-Hispanic Blacks	39.33 (40.24)	0, 99.43
Hispanic	25.42 (27.94)	0, 90.20
Asians	5.47 (9.80)	0, 66.95
Native Americans	0.10 (0.13)	0, 64.19
Others	1.31 (0.93)	0, 4.89
Local Workers (Percent) (2002–2014)	1.94 (3.45)	0.07, 26.38

Note: 2002-2014 values are averaged over time.

Analysis

In our analyses, we model both spatial and network autoregressive models. Spatial methods account for spatial dependence, the tendency for variables measured in spatially proximate areas to be correlated. Spatial lag models assess processes of spatial spillover by examining if the dependent variable in neighboring places has a spillover effect into the focal spatial unit. Spatial error models assess whether there is spatial autocorrelation in the error term which indicates there are spatially correlated omitted variables which influence the outcome of interest. Because communities who share commuters or public transit users are also linked through geographic space, it is likely that spatial spillover processes occur across commuting and public transit boundaries, not only across spatially contiguous boundaries⁶.

We run our autoregressive models using three row-standardized spatial weights matrices, on all of which we find significant spatial dependence of our dependent variable, STI prevalence, using the global Moran’s I test. Our first spatial weights matrix is based on geographic contiguity, with neighbors defined as communities which are immediately proximate the focal community of interest (Queen 1 criterion). Our second spatial weights matrix is based on communities which share lines of public transportation. Two communities are considered connected if they share either a bus or train line. Our third spatial weights matrix is defined by commuting ties. Two communities are considered connected by a commuting tie if at least 0.5% of the home community commutes to the work community in 2002, the first year in our study⁷. As we conceptualize the commuting and public transit spatial weights as inter-neighborhood networks representing potential social ties between communities, we will refer to all three spatial weights matrices as networks in the remainder of this paper: spatial network, commuting network, and public transit network.

As data in the current study are longitudinal, we estimate fixed effects spatial models which include both a spatial lag and spatial error term⁸. We estimate our models using the spxtregress command in Stata (StataCorp, 2019). We estimate fixed effects models to account for unobserved neighborhood variation, as well as to examine within unit change of the rate of STIs in neighborhoods across time. The fixed effects spatial autoregression model is represented by the following equation:

$$y_{nt}=\lambda W{y}_{nt}+\beta X_{nt}+c_n+u_{nt}$$

$$u_{nt}=\rho M{u}_{nt}+v_{nt}$$

where the subscript n represents the spatial unit for time t. In our three separate models, our spatial weighting matrices are represented by W and M. X represents a vector of time varying covariates, c represents individual effects, u represents the spatially lagged error term, and v represents a vector of innovations. In addition to our three fixed effects spatial autoregressive models, we estimate a fixed effects model without the inclusion of any spatial terms. Our fixed effects models include dummy variables for time.

Results

Maps and Network Graphs

Although many sexual partnerships occur due to spatial proximity, Laumann et al. (2004) assessed the spatial distribution of dating ties between their study sites and areas across Chicago to understand how factors such as organizations, social networks, and urban spaces lead to partnerships that can be geographically distant. They refer to these dating ties as “sex-market ties” because individuals must navigate social and structural barriers to find sexual partnerships. In Figure 1, we reproduce the map of Laumann and collaborator’s study of sex-market ties, shown on the left, and compare it to a map using our data of commuting ties and STI prevalence, shown on the right. Laumann et al.’s map charts the ties between four communities in their study and the areas where residents’ sex partners reside by different percentage cutoffs. For comparison, we adopt similar cutoffs and highlight the same sample communities but show how they are connected through their residents’ commuting flows. Laumann et al. show that sexual partnerships connect Shoreland (i.e., Lakeview) and Southtown (i.e., Roseland) to many other communities in the city both near and far. Though Erlinda (i.e, Hermosa) and Westside (i.e., Lower West Side) are connected to fewer communities, they are also connected to geographically distant ones. Their study demonstrates that sexual partnerships are not limited to geographically proximate communities. Our comparison map using our commuting data shows that there is an important overlap in which communities are connected through commuting and sexual relation ties. The commuting ties also demonstrate that communities are connected both near and far. Often but not always, commuting connects communities with similar STI rates. Overall, this visualization indicates that commuting networks may be an avenue for individuals to meet their sexual partners, as people may meet their sexual partners at their work destinations and influence the spread of STIs⁹.

Figure 1.

Map of Chicago Commuting Ties in Comparison to Laumann et al. (2004) Dating Ties Map

Notes: Leftmost Map. Adapted with permission from Laumann et al. (2004). Rightmost map: Nodes are positioned based on the geographic coordinates of the centroids of community areas, represented as polygons. The community areas are colored based on the quartile category of their STI levels. Showing only commuting ties with origin or destination in any of Laumann’s et al.’s (2004) four study communities, and only ties with values between 3% and 5% of commuters (thin lines); and over 5% (thick lines).

Figure 2 presents our three networks: spatial, commuting, and public transit. Across networks, the nodes represent neighborhoods and are colored based on their STI tercile with blue denoting bottom tercile, yellow for middle tercile, and orange for upper tercile. Nodes are sized by outdegree, meaning the number of ties a community sends to another based on the tie definition (i.e., geographic contiguity, commuting, or public transit). The leftmost network presents our spatial network which shows the ties between communities based on geographic contiguity. Although we see clear spatial patterning of STIs, it is evident that geographic contiguity does not completely drive these trends. We see highly infected areas (orange nodes) occur in the bottom and top areas of Chicago with communities of varied STI rates in between.

Figure 2.

Map of Chicago’s Spatial, Commuting, and Public Transit Networks by Network Degree and STI Rate

Notes: Community areas are represented as nodes positioned according to geographic coordinates of their centroids. The color of nodes is based on STI tercile categories in 2002 (blue for bottom tercile; yellow for middle tercile; orange for upper tercile). Across all three maps, the size of nodes represents outdegree (number of ties to different connected areas based on the corresponding tie definition). The leftmost’s graph ties are based on spatial contiguity. The middle graph’s ties are based on the 0.5% commuting cutoff and shown are only ties between STI clusters (removed are ties within each of the STI tercile categories). The rightmost map shows ties based on the public transit network and shown are only ties between STI clusters.

The middle network graph represents the 2002 commuting network but, to minimize visualization crowdedness, shows only ties between clusters. Although ties between communities within the same STI cluster are significant, in this map we see that communities are highly connected, through commutes, to other areas with varied STI rates. This figure highlights how commuters may potentially expose their residential communities to STI risks from their work environments. They also indicate the potential for low STI work communities to drive down STIs in high STI residential areas.

The network on the right shows the public transit network, focusing again on the between cluster links. We observe a large concentration of nodes with a high degree value that also have a high infection rate. This finding indicates that infected communities are highly connected by individuals through public transit connections. In comparison, communities with lower STI rates are connected to fewer communities by public transportation, making them less reachable.

Figure 3 highlights the difference between social space and geographic space when using commuting and public transit networks compared to the geographic contiguity ties. The leftmost network graph shows geographic proximity ties. The middle graph shows commuting ties but excludes ties between geographically proximate communities. The rightmost graph shows transportation ties, excluding those between geographically proximate communities. All graphs use the same spring embedding procedure, Kamada-Kawai in Pajek (Kamada & Kawai, 1989). This is a force-directed layout using a random starting point and optimizing within clusters which determine the optimal location of communities in geometric space relative to each other based on their ties to one another. In contrast to Figure 2, the geographic location is ignored. Note that in the geographic space, high STI communities are rarely tied to low STI communities, whereas in the social space graphs such ties are much more common. Communities of lower STI levels, like O’Hare and the Loop often function as employment hubs that connect many higher STI communities. The transportation network exhibits the most connections between communities of different socioeconomic levels, demographic composition, and varying STI levels. Additional tests support the visual indications that commuting exhibits more of a core-periphery structure than the transportation network. Notably, the denser groups in both networks exhibit a lower STI prevalence on average than the weaker-connected group. This indicates the need for a deeper understanding of the link between connectivity and STI.

Figure 3.

Comparing Connectivity in Geographic and Social Space and Links to STI

Notes: Community areas are represented as nodes positioned in geometric space using the Kamada-Kawai spring embedding procedure, with within-clusters optimization and using a random starting point to determine the location of communities in geometric space relative to each other based on their connections (i.e., the actual geographic location is ignored). The color of nodes is based on STI tercile categories in 2002 (blue for bottom tercile; yellow for middle tercile; orange for upper tercile). The leftmost’s graph ties are based on spatial contiguity. The middle graph’s ties are based on the 0.5% commuting cutoff (excluding commuting ties between geographically proximate communities). The rightmost map shows ties based on the public transit network (excluding transportation ties between geographically proximate communities).

Spatial and Network Autoregressive Models of STI Diffusion

To determine how STIs can spillover and exhibit autocorrelation across geographic and social space, we estimate spatial and network autoregressive models. Our models include spatial lag and spatial error terms. The spatial lag term is a spatial lag of the dependent variable and represents the correlation between the focal communities STI prevalence and connected communities’ average STI prevalence, as defined by the spatial weights matrix used. We refer to this as “Network STI Risk,” and represents a spatial and social spillover process. Including a spatial lag of our dependent variable modifies the effects of our independent variables. A change in an independent variable of a community will modify the STI prevalence of that community, which will in turn modify the STI prevalence in all the communities to which that community is connected to through spatial and social spillovers. This leads to all the independent variables having direct and indirect effects on STI prevalence. The spatial error term represents spatial dependence in our error terms which indicate spatially dependent omitted variables predicting similar STI rates in the focal community and the neighborhoods it is connected to as defined by the spatial weights matrix. Table 1 presents the dynamic fixed effects models that examine data for all the time periods in our study. The table starts with models predicting STI prevalence without any spatial terms and then moves to models that use the three different spatial weights matrices. We estimate each spatial and network model with and without disadvantage.

Model 1 presents a fixed effects model without accounting for spatial or social dependence. Models 2 and 3 examine how spatial dependence in geographically contiguous areas influences STI prevalence. Models 4 and 5 examine how communities connected through work commuting ties influence STI prevalence. Models 6 and 7 examine how communities connected through public transit lines influence STI prevalence. As expected, a community’s previous STI prevalence is the strongest community-level predictor of continued STI prevalence. Prior high STI rates influence the persistence of high STI rates in consecutive years. Residential diversity is also a consistent predictor of STI prevalence except in Models 3–5, with an increasing level of diversity associated with a lower STI rate. These findings support other research identifying assortative mating as a driver of STI prevalence. As diversity increases, the chances that individuals might choose interracial sexual partners also increases, which might reduce the flow of STIs within a neighborhood. Interestingly, we find that higher levels of disadvantage decrease STI prevalence in Model 1 which includes no terms for spatial autocorrelation and in Model 2 which accounts for spatial dependence in contiguous communities. However, this result should be interpreted cautiously as the inclusion of time-lagged STI rates may account for this slight negative effect. Due to these models being fixed effects models, they only assess variation across time within a neighborhood and levels of disadvantage do not vary much over time. Disadvantage is quite persistent across Chicago and few communities would be expected to have substantial variation over time.

We find that the spatial lag and error term are significant across Models 2–7. A higher prevalence of STIs in geographically contiguous communities is associated with an increase in a focal community’s STI rate. These results suggest STIs can diffuse across spatially contiguous neighborhoods, as neighborhood “boundaries” do not physically exist to stop the flow of people and subsequently STIs across space. Interestingly there is a negative spatial error term in our spatial model. This result implies there may be omitted variables which serve to decrease STIs in a focal community when its geographically contiguous neighbors have a high STI prevalence.

The public transit and commuting models show also significant effects of connected communities influences on a focal community’s STI prevalence. We also find a positive and significant spatial error term in the transit and commuting models suggesting that there is unexplained spatial variance in the models which serves to increase STI prevalence among communities who are connected via transportation and commuting ties. The model fit statistics indicate that the public transit model is the best model to explain the diffusion of STIs across space. The fit estimates also imply that the commuting model better explains the diffusion of STIs across space than the spatial proximity model.

Supplementary Analyses

Our results demonstrate that not only do spatially contiguous neighbors affect changes in STI prevalence over time, but also so do the communities that are connected through public transit and worker’s commutes, which may be stronger influences. We further assess the importance of neighborhood connections in supplementary analyses that combine the three main spatial weight matrices into their four varying combinations that build on the concept of spatial proximity and extend it to the broader idea of social proximity (Kelling et al., 2020). The spatial lag term of STI prevalence is significant across all groupings of combined weight matrices. Models that include the public transit and the commuting network in various combinations also fit the data better than models that only account for spatial interdependencies¹⁰. These findings highlight the benefit of an expanded view of inter-neighborhood connections, above and beyond geographic contiguity ties to better understand the diffusion of risk for infectious diseases like STIs.

As noted in our discussion on the potential role of selection in tying communities together by commuting ties, we explore bootstrapped temporal exponential random graph models (TERGMs) in Appendix C. We estimate TERGMs to better understand how communities’ STI prevalence predict the presence of commuting ties. We find a significant homophily effect where commuters tend to work in environments that have similar rates of STIs as their residential neighborhood. These results support our main results which find a significant error variance parameter with our commuting network, indicating significant social autocorrelation between communities connected through commuting.

As the prevalence of STIs varies dramatically across various demographic groups, we present additional analyses in Appendix E using more detailed information on the racial composition of the community, and the neighborhood’s age structure, marriage rates, and average household size. We also consider how the teen birth rate, the age-adjusted total fertility rate, total population logged, and population density could influence neighborhood STIs. Our results remain robust to these additional controls. We retain the more parsimonious models in the main tables because of consistently better model fit scores. Additionally, the inclusion of prior STI rates naturally absorbs the effects of the additional STI determinants from these longer models.

In Appendix G we explore using both weak and strong commuter tie thresholds to examine how the relationship between STI prevalence and commuting varies by tie strength. We incorporate two commuting networks with a weaker tie threshold cutoff, 0.1% and 0.25%, as well as two commuting networks with a stronger tie threshold cutoff, 1% and 2.5%. We find with smaller cutoffs there is a stronger effect and with stronger cutoffs there is no effect. This is likely because the smaller commuting tie thresholds lead communities to be connected to most other communities in the network, while the stronger commuting tie thresholds lead communities to only be connected to a select few hub communities such as O’Hare and the Loop.

Discussion

The results of the current study suggest that spatial spillovers of STIs and STI preventive information and risk behaviors occur not only between geographically proximate communities but also between communities that are socially connected within the city, even at a distance. They reveal that the commuting and public transit networks explain STI transmission across space better than the geographic contiguity model. These results suggest that as a neighborhood’s residents travel beyond their immediate residence to work or conduct their daily routines using public transportation, they facilitate the spillover of STIs and related risk factors across communities. These results illustrate two key ways in which communities become connected by the flow of individuals moving across space, with important consequences in influencing community processes such as the spread of STIs and STI-related norms and behaviors.

Our findings advance prior research that has largely focused on individual-level determinants (Burstein et al., 1998; Chatterjee et al., 2006; Kelley et al., 2003; Laumann & Youm, 1999) and residential neighborhood effects (Ellen et al., 2004; Jennings et al., 2012). Importantly, these results suggest that structural network influence is likely an important part of the long-term reproduction of health disadvantages at the community level. The significance of the commuting and the public transit networks for STI risk spillovers indicates that, beyond spatial proximity, social proximity related to broader patterns of mobility and activities are at play in the spread of STI risk across urban space. Although we can map connections between neighborhoods via transit lines, we are unable to document the unique transit patterns of individuals. Some individuals might use multiple transit lines during a trip and connect to neighborhoods that may not be directly tied to their residential neighborhood. This data limitation might allow for us to not exactly estimate the effects of transit lines on STI rates. However, we can precisely map where commuters work and live, which we find similar effects for on STIs as transit lines. Laumann et al. (2004) showed that dating ties can stretch across a city as individuals sometimes find their sexual partners through their social activities. Our findings are consistent with this work and further advance the literature by documenting how community networks defined by commuting and public transportation can contribute to the flow of STIs across Chicago.

Implications

The current findings have theoretical implications for advancing the scholarship on neighborhood effects, residential segregation, and population mobility by demonstrating the value of connections beyond residential neighborhood boundaries and geographic proximity space, to better understand the effects of population-level mobility flows for local health and other demographic outcomes. These findings are consistent with the growing body of research on activity space exposures and residential mobility (which often focuses on individuals), and further advances existing knowledge by demonstrating that inter-neighborhood ties are significant in shaping the health and wellbeing of entire communities.

This study also contributes to current knowledge methodologically, by showing the value of combining longitudinal spatial and network autoregressive models to address questions important to demographers and social epidemiologists alike. The results help us better understand how population mobility and socially connected communities contribute to changes over time in STI patterns, with relevance for other infectious diseases. Future research might also look more closely at the diffusion of infections across neighborhoods. Although our study can assess the autocorrelation in STI rates among neighborhoods, more precise data and methods might be able to assess the diffusion of these infections more accurately across space. Additionally, our research documents connections between neighborhoods as dichotomous, which future research might build on by using weighted networks to understand important nuances by tie strength.

Past research has indicated the importance of spatial clustering for STI patterns in an urban environment (De et al., 2004; Potterat et al., 1985; Risley et al., 2007); however, focusing only on these areas has proven ineffective in STI targeted interventions (Jolly & Wylie, 2013; Rothenberg et al., 2005). Our results suggest that interventions would benefit from considering how people interact with their environment and the implications connected communities pose for infectious diseases. Instead of focusing only on highly infected communities, future interventions should consider contact tracing to better track and treat STIs across communities. Additionally, information about STI prevention might be circulated in communities that are highly connected by commuters and public transit, as exposure to this information can decrease STI incidence (Warner et al., 2008).

Although prior studies document the effects of residential neighborhoods on health (Arcaya, Tucker-Seeley, et al., 2016; Cubbin et al., 2019, 2005; Sampson, 2003), the current study highlights a great need for future research to explore the implications of connected communities on health outcomes. It would be valuable for future studies to explore how exposures to different racial and ethnic groups in people’s work environments influence inter-racial marriage, above and beyond such exposures in residential neighborhoods or perhaps even despite strong segregation patterns in residential neighborhoods. Future research might also pay closer attention to the ways race and ethnicity shape STI patterns across cities, as some groups are more predominantly affected by STIs than others (Adimora & Schoenbach, 2005, 2013; Harling et al., 2014; Thomas & Thomas, 1999).

A social epidemiology approach that highlights the importance of inter-neighborhood connections will be particularly valuable in also understanding the unequal spatial distribution of other infectious disease patterns, such as COVID-19 (Jia et al., 2020) or the seasonal influenza virus, across commuting, public transportation, and other population mobility pathways. Other major population outcomes that likely depend not only on physical exposures to risks or resources, but also on behavioral and normative influences through mobility pathways are violent victimization and crime rates (Kelling et al., 2020; Levy et al, 2020), asthma, obesity, or smoking prevalence (Christakis & Fowler, 2008, 2013; Zhang & Centola, 2019), and infant mortality and birthweight through exposure to factors such as pollution or food environments.

Our research investigates the structural mechanisms of socio-spatial spillovers, specifically public transportation networks and worker commute networks. Future research would benefit from further investigation of how these networks facilitate the underlying social interactions and mechanisms that influence health behaviors and outcomes. Connected communities likely play an important role in shaping STI rates through either exposure to infected sexual partners or through social contagion of health behaviors (Christakis & Fowler, 2013). Beyond physical transmission, STIs are likely dependent on social learning and contagion of preventive health behaviors, which occur more slowly, through social reinforcements from multiple sources (Zhang & Centola, 2019). Health behaviors and the normativity of risky health behaviors are influenced by reinforcing messages from multiple network ties about the acceptability and safety of contraceptives (Behrman et al., 2002; Guilkey et al., 2020; Kohler, 1997; Valente et al., 1997). This is particularly relevant for behaviors that involve other people and are subject to normative pressure (Christakis & Fowler, 2008; de Vaan & Stuart, 2019). Additionally, better knowledge of how these connections shape or constrain sexual partnerships might highlight inequality in opportunities to partner with uninfected individuals. Our research identifies the structural networks that contribute to STI spillovers across communities, but future research might investigate how specific mechanisms such as dating, and the spread of sexual norms or behaviors contribute to these patterns.

More broadly, our research highlights the importance of going beyond the standard approach to neighborhood effects to better understand population health patterns and social behaviors. Our results find that socially connected communities are key drivers of STI infection patterns in Chicago. Future research would benefit from assessing the role of connected communities in shaping other population health and demographic patterns and to better understand the mechanisms underlying the social connections that shape them.

Table 2.

Spatial Lag and Error Models Predicting STI Rates

	Without Network	Spatial Network		Commuting Network		Public Transit Network
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
Disadvantage	−0.060* (0.029)	−0.064** (0.022)		−0.038 (0.031)		−0.054 (0.030)
Stability	−0.000 (0.019)	−0.004 (0.013)	−0.004 (0.013)	−0.020 (0.023)	−0.019 (0.023)	−0.021 (0.023)	−0.023 (0.023)
Diversity	−0.047* (0.018)	−0.037* (0.016)	−0.013 (0.014)	−0.039 (0.021)	−0.029 (0.020)	−0.053** (0.019)	−0.037* (0.017)
Local Workers	0.008 (0.021)	0.002 (0.018)	0.005 (0.018)	−0.007 (0.018)	−0.008 (0.018)	−0.001 (0.021)	−0.002 (0.021)
Prior STI Rate	0.390*** (0.030)	0.296*** (0.031)	0.289*** (0.032)	0.380*** (0.032)	0.373*** (0.032)	0.385*** (0.033)	0.375*** (0.032)
Network STI Risk		0.603*** (0.064)	0.589*** (0.070)	0.960*** (0.011)	0.960*** (0.011)	0.883*** (0.006)	0.883*** (0.006)
Error Variance Parameter		−0.629*** (0.108)	−0.588*** (0.116)	0.961*** (0.010)	0.962*** (0.010)	0.883*** (0.006)	0.883*** (0.006)
Time Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
AIC	−894.57	−921.79	−915.45	−1268.55	−1269.08	−3391.84	−3390.63
BIC	−806.21	−823.62	−822.18	−1170.38	−1175.81	−3293.66	−3297.36
Neighborhoods=77; Observations=1,001

Note: Standard errors in parentheses. All models include year dummies. The Wald test of spatial autocorrelation is significant across all models at p<0.001.

^*p<0.05,

^**p<0.01,

^***p<0.001