Over the past decade, teen birth rates in the U.S. have consistently fallen, yet large geographic disparities remain (Sedgh et al., 2015). In 2015, county-level teen birth rates varied between 3 and 119 births per 1000 females ages 15–19 years old (Division of Reproductive Health, 2020) with clearly evident clusters of high rates in the southern U.S. and clusters of low rates in the Northeast (Khan et al., 2017; Maslowsky et al., 2019; Romero et al., 2016; Ventura et al., 2001). Recent research has suggested that social determinants of health (SDOH) are a key driver of this spatial variation (SmithBattle, 2012; Viner et al., 2012; Maness et al., 2016). As defined by the World Health Organization, SDOH are “the conditions in which people are born, grow, live, work, and age” (Marmot et al., 2008) and include a multitude of factors, such as poverty, healthcare access, and housing affordability, that can influence health outcomes. These conditions are themselves influenced by, and often the consequence of, more upstream structural, cultural, and place-based factors—what we collectively refer to in this paper as intrinsic geographical contextual effects—that together produce environments that shape individual and community-level behavior (Viner et al., 2012; Koh et al., 2011; Braveman, 2023; Mays, 2021). Disentangling the extent to which each of these sources, the downstream SDOH conditions versus the upstream intrinsic geographical context, influences teen birth rates is important for helping policymakers and health practitioners develop and implement tailored policies and interventions targeted toward addressing the persistent spatial disparities in health outcomes that we observe. We argue here that if the upstream effects of place-based values, belief systems, and cultural norms are not properly incorporated into models linking health disparities to SDOH, such linkages may well be misspecified.

The central argument that geographical context can have a major impact on people’s beliefs, preferences, and actions, has been promulgated by many authors in many different application areas (inter alia, Agnew, 1996, 2014; Duncan et al., 1998; Enos 2017; Golledge, 1997; Gould, 1991; Harvey and Wardenga, 2006; Pred, 1984; Tuan, 1979; Winter and Freksa, 2012; Link and Phelan, 1995). Several theories have been proposed to account for such a relationship. For instance, a link between place and behavior can arise if a person’s actions or beliefs are influenced by the people that person talks to on a regular basis (“social imitation”), or by the local media, or by long-term conditions that are peculiar to certain locales and which shape a person’s outlook on certain issues (Beck et al., 2002; Huckfeldt and Sprague, 1995; Huckfeldt et al., 1995). Equally, traditions, customs, lifestyles, and psychological profiles common to an area can affect social norms, which in turn affect individual behavior. Several studies, for example, have commented on personality differences across regions and how these can explain behavioral differences (inter alia, Rentfrow et al., 2015; Rentfrow et al., 2013). These “values, beliefs and norms” (VBN) transcend individual demographic characteristics and can manifest themselves in a variety of ways, such as how people feel about government control, how much they believe in scientific evidence, and to what extent they view teen pregnancy in a negative light (Stern et al., 1999). Within the extensive literature on health and place, neighborhood social environments are often cited as potential pathways through which geographical context might influence health-related outcomes (Diez Roux and Mair, 2010). Features of the neighborhood social environment, such as the strength of social connections, social cohesion, and social capital present in an area, are thought to contribute to the enforcement of certain norms and the transmission of particular behaviors (“social contagion”) and have been linked to variation in health outcomes, including all-cause mortality, self-rated health, epilepsy, and asthma (Szaflarski, 2014; Gold et al., 2002; Sullivan and Thakur, 2020). More recently, these aspects of the neighborhood social environment along with a broader set of structural factors, including local policies and governance practices, have become known as structural determinants of health. Structural determinants of health are thought to influence specific health-related outcomes by creating, configuring, and maintaining social hierarchies that lead to social stratification and ultimately give rise to SDOH (Mays, 2021; Zuckerman, 2021; Diez Roux and Mair, 2010; Viner et al., 2012; Koh et al., 2011; Braveman, 2023).

Since the early 1990s, a growing body of empirical research has sought to measure the geographical contextual effects on health (inter alia Macintyre et al., 2002; Cummins et al., 2007; Diez Roux and Mair, 2010; Decker et al., 2018). Much of the early work in these literatures focused on separating area effects into geographical contextual effects versus compositional effects and often found residual geographical contextual effects on health outcomes after controlling for compositional effects (Decker et al., 2018; Cummins et al., 2007). Within the teen pregnancy literature, several studies have identified significant associations between teen birth rates and various measures of socioeconomic disadvantage, including receipt of public assistance, low levels of education, high unemployment rates, and neighborhood deprivation (Moore, 1995; Penman-Aguilar et al., 2013; Harding, 2003; Sucoff and Upchurch, 1998; Yee et al., 2019; Wei et al., 2005; Fuller et al., 2018). The vast majority of this research has relied on global regression models to assess the relationship between SDOH and teen childbearing (Yee et al., 2019; Orimaye et al., 2021; Bickel et al., 1997; Kirby et al., 2001; Gold et al., 2001), however such models assume that these relationships are constant across space. Despite established theories suggesting that processes by which SDOH influence outcomes might vary spatially and operate at different geographic scales (Cummins et al., 2007), only one study (Shoff and Yang, 2012) to our knowledge has ever examined local variation in the relationships between teen birth rates and SDOH. Using geographically weighted regression (GWR) models, Shoff and Yang (2012) identified spatially nonstationary associations between teen birth rates and several ecological factors, including the demographic composition and rate of religiosity, in both metropolitan and nonmetropolitan counties. Although this work provided crucial insights into how the influence of SDOH on teen birth rates might vary across space, these models relied on a strong and likely incorrect assumption that all measured relationships operated over the exact same spatial scale. As a result, these associations might be biased.

Despite the extensive research into the roles of geographical context, and more recently, SDOH and their influence on health-related behaviors and outcomes, little is still known regarding how much each of these sources contributes to spatial disparities in outcomes and whether these relationships vary not only across space but at different spatial scales, particularly in the context of teen pregnancy. Consequently, what is needed is a means of separating these two potential drivers of geographic variations in teen birth rates (downstream SDOH conditions versus intrinsic geographical context) and to assess the relative contributions of each. Of course, if this can be done for teen birth rates, it could be done for any other health variable that exhibits geographic variation in any country, so the results of this analysis have universal appeal. The general question that needs to be answered, and which is the focus of this analysis, is: “To what extent are observed health variations due to differences in the SDOH and to what extent are they due to the intrinsic geographical context?”

Between 2010 and 2015, the U.S. Office of Adolescent Health and the Centers for Disease Control invested considerable resources to help reduce disparities by raising awareness around this connection between SDOH and teen pregnancy (Romero et al., 2016). Research conducted by Fuller et al. (2018) as part of this effort proposed that eliminating disparities in teen birth rates requires addressing SDOH that contribute to teen pregnancy, implying that the geographic variation in teen birth rates could be ameliorated, and possibly reduced altogether, if community-level variations in SDOH were eliminated. However, as discussed above, a great deal of literature across the social sciences suggests that geographical context can affect behavior and that this is upstream and separate from variations in SDOH. Here, we suggest that both spatial variations in SDOH and the impacts of geographical context lead to the levels of spatial variation in teen birth rates we observe across the U.S.

What is needed therefore is a means of identifying and quantifying the relative impacts of spatial variations in SDOH and intrinsic geographical context on teen birth rates. To do this, we use a local regression technique, multiscale geographically weighted regression (MGWR), to estimate the effects of SDOH on teen birth rates. MGWR has been used to estimate spatially varying associations in obesity rates (Oshan and Smith, 2020; B Neelon et al., 2017; Chi et al., 2013; Dwicaksono et al., 2018), HIV (Zhou et al., 2015; Nakaya et al., 2005; W Wabiri et al., 2016), and other public health topics (Schooling et al., 2011; Ribeiro and Pereira, 2018; Tu et al., 2012). An important output from MGWR is the estimation of a local intercept, which can be used to identify and measure the intrinsic geographical contextual effects independent from other effects related to SDOH (Fotheringham et al., 2021; Fotheringham and Li, 2023).

By modeling the relationship between SDOH and county-level teen birth rates using MGWR, this paper can thus examine the important question:

“To what extent can the observed spatial variation in teen birth rates across the U.S. be ascribed to variations in SDOH and to what extent can they be ascribed to intrinsic geographical contextual effects?”

This raises two further intriguing questions which are answered in this paper:

“Would disparities in teen birth rates remain if all counties had identical SDOH? That is, if there were no variations in population over space, would we still observe significant, geographically patterned, variations in teen birth rates?”; and

“How would teen birth rates be spatially distributed if intrinsic geographic contextual effects played no role in affecting teen behavior? That is, if teen birth rates depended solely on SDOH, what would be the spatial variation in these rates across space – would it be identical to that observed?”