Social Connectedness: Measurement, Determinants, and Effects
Author
Abstract
social networks can shape many aspects of social and economic activity: migration and trade, job-seeking, innovation, consumer preferences and sentiment, public health, social mobility, and more. In turn, social networks themselves are associated with geographic proximity, historical ties, political boundaries, and other factors. Traditionally, the unavailability of large-scale and representative data on social connectedness between individuals or geographic regions has posed a challenge for empirical research on social networks. More recently, a body of such research has begun to emerge using data on social connectedness from online social networking services such as Facebook, LinkedIn, and Twitter. To date, most of these research projects have been built on anonymized administrative microdata from Facebook, typically by working with coauthor teams that include Facebook employees. However, there is an inherent limit to the number of researchers that will be able to work with social network data through such collaborations.
Measuring Social Connectedness
The Social Connectedness Index is constructed using aggregated and anonymized information from the universe of friendship links between all Facebook users as of April 2016. Duggan, Ellison, Lampe, Lenhart, and Madden (2015) report that as of September 2014, more than 58 percent of the US adult population and 71 percent of the US online population used Facebook. The same source reports that, among online US adults, Facebook usage rates are relatively constant across income groups, education groups, and racial groups. Usage rates among online US adults are declining in age, from 87 percent of 18-to-29 year-olds to 56 percent of above-65 year-olds.
In the United States, Facebook mainly serves as a platform for real-world friends and acquaintances to interact online, and people usually only add connections on Facebook to individuals whom they know in the real world (Jones et al. 2013; Gilbert and Karahalios 2009; Hampton, Goulet, Rainie, and Purcell 2011). Establishing a friendship link on Facebook requires the consent of both individuals, and the total number of friends for a person is limited to 5,000. As a result, Facebook data have a unique ability to provide a large-scale representation of US friendship networks.
To measure the social connectedness between geographies, we map Facebook users to their respective county and country locations, and thus obtain the total number of friendship links between these geographies. Locations are assigned to users based on the users’ information and activity on Facebook, including the stated city on their Facebook profile, and device and connection information. We only consider friendship links among Facebook users who have interacted with Facebook over the 30 days prior to the April 2016 snapshot.1 We treat each friendship link identically.
We then construct the Social Connectedness Index between all pairs of 3,136 US counties, and between every US county and every foreign country, as the normalized total number of friendship links for each geographic pair. In particular, the Social Connectedness Index is constructed to have a maximum value of 1,000,000, and relative differences in the index correspond to relative differences in the total number of friendship links. The highest Social Connectedness Index value of 1,000,000 is assigned to Los Angeles County–Los Angeles County connections (Los Angeles County is where people have the most friends with other people in their county).
2
The Determinants of Social Connectedness
The Social Connectedness Index can be used to analyze the correlates of the intensity of social connectedness between US counties. We first analyze the role of geographic distance in shaping social connectedness in the United States. The effects of geographic proximity on friendship formation and social interactions have been studied in a number of papers, including Zipf (1949), Verbrugge (1983), and Marmaros and Sacerdote (2006).
County-Level Friendship Maps
A: Relative Probability of Friendship Link to San Francisco County, CA
B: Relative Probability of Friendship Link to Kern County, CA
Note: The heat maps show the relative probability that a Facebook user in each county j has a friendship link to San Francisco County, CA (Panel A) and Kern County, CA (Panel B). Darker colors correspond to counties in which there is a higher probability of a friendship link between a person in home county i (San Francisco or Kern) and county j. The “relative probability of friendship” is constructed by taking the Social Connectedness Index between counties i and j and dividing it by the product of the number of Facebook users in the two counties.
Distance and Friendship Links: Across-County Summary Statistics for the United States
Note: Table shows across-county summary statistics for the share of friends of a county’s population living within a certain distance of that county as well as the share of the US population living within those distances. P5, P10, P90, and P95 are the 5th, 10th, 90th, and 95th percentiles, respectively. Counties are weighted by their populations.
Determinants of Social Connectedness across County Pairs
Note: Table shows results from a regression of the log of the Social Connectedness Index on a number of explanatory variables. The log of the geographic distance between the counties is the explanatory variable in column 1. In column 2, we include an additional control indicating whether both counties are within the same state. In columns 3 and 4, we restrict the sample to county-pairs that are more and less than 200 miles apart, respectively. The unit of observation is a county-pair. Standard errors are given in parentheses. The online Appendix (http://e-jep.org) provides more details on the data sources and exact specifications.
*, **, and *** indicate significance levels of p < 0.1, p < 0.05, and p < 0.01, respectively.
Connected Communities within the United States—20 Units
Note: Figure shows US counties grouped together when we use hierarchical agglomerative linkage clustering to create 20 distinct groups of counties.
3
Concentration of Social Networks and County Characteristics
The geographic concentrations of the friendship networks of different counties reveal a great deal of heterogeneity: for example, the earlier Table 1 shows that the 5th–95th percentile range across population-weighted counties in the share of friends living within 100 miles is 46.0 percent to 76.9 percent. Existing theoretical work suggests that the diversity of social networks is an important determinant of economic development; conversely, tightly clustered social ties can limit access to a broad range of social and economic opportunities (for example, Granovetter 1973). However, empirical studies of the relationship between the structure of social networks and economic outcomes of communities are rare. One exception is Eagle, Macy, and Claxton (2010), who use UK cellphone data to document that the diversity of individuals’ social networks is correlated with regional economic well-being. In this section, we provide evidence that the geographic dispersion of friendship links across US counties is highly correlated with social and economic outcomes at the county level, such as average income, educational attainment, and social mobility.
If we define the concentration of a friendship network as the share of friends who live within 100 miles, then friendship networks in the South, the Midwest, and Appalachia are the most geographically concentrated. Counties in the Rocky Mountains have the smallest share of friends living within 100 miles, in large part because these areas are often less-densely populated. Among the western United States, Utah and inland California have the most geographically concentrated friendship networks. The online Appendix shows heat maps of this and other measures of the geographic concentration of friendship networks.
What are the effects of differentially structured social networks on county-level outcomes? As a first step toward answering this question, we correlate our measure of the concentration of friendship links with county-level characteristics. Figure 3 presents county-level binned scatterplots using the share of friends living within 100 miles and a number of socioeconomic outcomes. The overall message is that counties where people have more concentrated social networks tend to have worse socioeconomic outcomes along a number of dimensions: on average, they have lower income, lower education, higher teenage birth rate, lower life expectancy, less social capital, and less social mobility.
These correlations cannot be interpreted as causal (although the online Appendix discusses a number of causal mechanisms proposed by the literature that are consistent with our findings). Our goal here, as in the rest of the paper, is to document patterns that can guide future research investigating the causal effects of social network structure on socioeconomic outcomes, and to describe the Social Connectedness Index data that can help with such analyses. More generally, the strong correlation between social connectedness and socioeconomic outcomes suggests that controlling for the geographic concentration of social networks is important to minimize omitted variables bias across a number of research agendas that study economic and social outcomes at the county level.
4
Social Connectedness and Cross-County Activity
Social connectedness between two regions may be related to other economic and social interactions between these regions. Indeed, we next document correlations between the number of friendship links and trade flows, patent citations, and migration patterns. As before, we illustrate some salient patterns in the data rather than providing full-fledged causal analyses. For each of the patterns documented below, the online Appendix (http://e-jep.org) provides more details on the variables, data construction, specifications, and additional exploration.
Network Concentration and County-Level Characteristics
Notes: Panels show binned scatterplots with counties as the unit of observation. To generate each binned scatterplot, we group the x-axis variable into 50 equal-sized bins. We then compute the mean of the x-axis and y-axis variables within each bin and create a scatterplot of these 50 data points. The horizontal axes measure the share of friends of the county population that live within 100 miles. On the vertical axes are a number of county-level measures of socioeconomic outcomes: the mean county income in Panel A; the share of the population with no high school degree in Panel B; the teenage birth rate as provided by Chetty, Hendren, Kline, and Saez (2014) in Panel C; the life expectancy of males in the first quarter of the national income distribution from Chetty et al. (2016) in Panel D; the measure of social capital in 2009 as defined by Rupasingha, Goetz, and Freshwater (2006) in Panel E; and the absolute measure of social mobility from Chetty et al. (2014) in Panel F. The red line shows the fit of a quadratic regression. The online Appendix (http://e-jep.org) provides more details.
Social Connectedness and Across-Region Economic Interactions
Note: Table shows the relationship between bilateral economic activity across geographic units and thegeographic distance and social connectedness between these units. “SCI” stands for Social Connectedness Index. In Panel A, the unit of observation is a state-pair, and the dependent variable is the log of the value of 2012 trade flows between the states. All specifications include state fixed effects, dummies for own state, and dummies for neighboring states; column 4 also controls for differences across states on important socioeconomic indicators. In Panel B, the unit of observation is a patent-pair. The dependent variable is an indicator of whether patent i cites patent j. All specifications control for the county and technology category fixed effects, and column 4 also controls for patent fixed effects and other differences across the counties of the patents on important socioeconomic indicators. In Panel C, the unit of observation is a county pair, and the dependent variable is the log of across-county migration between 2013 and 2014. All specifications control for county fixed effects, and column 4 also controls for other differences across counties on important socioeconomic indicators. Standard errors are given in parentheses. The online Appendix (http://e-jep.org) provides more details on the data sources and exact specifications.
*, **, and *** indicate significance levels of p < 0.1, p < 0.05, and p < 0.01, respectively.
Conclusion
First, many contagious illnesses and diseases, such as the flu or tuberculosis, spread through human contact. Combined with localized data on the prevalence of the flu, data on social connectedness might allow researchers and public health officials to better predict where to expect future outbreaks of the flu (Cauchemez et al. 2011; Christakis and Fowler 2010).
Second, the Social Connectedness Index data could also be used to track whether measures of sentiment—for example, those tracked by the Michigan Survey of Consumers or through geo-coded Twitter feeds—spread along social networks.
Third, sociolinguistic research has argued that social networks are an important force determining how languages evolve over time (for example, Milroy 1987). The Social Connectedness Index data would allow researchers to study the extent to which linguistic development in the United States is associated with patterns of social connectedness.
Fourth, the relationships between transportation networks and social connectedness may prove interesting. For example, significant social connectedness between two regions might be a strong indicator that providing transportation infrastructure between these regions, such as direct airline routes, is profitable. Using the Social Connectedness Index as a measure of the potential demand for various routes could address some of the identification issues in the literature analyzing airline scheduling in operations research and industrial organization. Moreover, increased transportation links might also have a causal effect on social connectedness. One approach using the SCI data is to compare the social connectedness of two counties that happen to lie on the straight line between two major cities, and which are therefore connected by a highway, to the connectedness of two similar counties that do not lie on the straight line between major cities (see Bailey et al. 2018).
Finally, the SCI might prove useful in testing theoretical models of network formation (Jackson 2014). Specifically, in models of geographic strategic network formation models, the costs of network formation are directly related to distance (for example, Johnson and Gilles 2000). Using data from the National Longitudinal Survey of Adolescent Health on close friends of individuals, Patacchini, Picard, and Zenou (2015) show that students living in central locations have higher levels of social interactions. Our estimates of the elasticities of friendship links with respect to distance often map directly into the parameters of these models and can be used to parameterize them.
资源仅供学术交流使用，严禁商用！如有侵权，请联系小编微信：yi_na_na。
文献来源：doi=10.1257/jep.32.3.259
推文期数：2018122
责任编辑：易娜 张茜茜 谢蒙利 张晶飞
王小军 马龙
推文审核：张天舒 梁龙武 骆丹云
总审核：学术无界顾问团