Fig. 3: County-level correlations between upward income mobility and measures of social capital.

a, County-level univariate correlations of upward income mobility with social capital measures. Extended Data Table 2 lists the correlation coefficients plotted here. b, Estimates from a multivariable regression of upward income mobility on all variables in a together, standardizing the outcome and dependent variables to have a mean of zero and a standard deviation of one. Upward income mobility is obtained from the Opportunity Atlas⁷² and is measured as the predicted household income rank in adulthood for children in the 1978–1983 birth cohorts with parents at the 25th percentile of the national income distribution. Economic connectedness (EC) is twice the share of above-median-SES friends among below-median-SES people. Language connectedness is the share of friends who set their Facebook language to English among users who do not set their language to English, divided by the national share of users who set their language to English. Age connectedness is the share of friends who are aged 35–44 years among users who are aged 25– 34 years, divided by the national share of users aged 35-44 years. Clustering is the share of an individual’s friend pairs who are also friends with each other, averaged over all individuals in the county. Support ratio is the share of friendships between people in the county with at least one other mutual friend in the county. Spectral homophily is the second largest eigenvalue of the row-stochasticized network adjacency matrix, a measure of the extent to which the county-level friendship network is fragmented into separate groups. The Penn State index⁶³ is an index of participation in civic organizations and other measures of civic engagement. Civic organizations is the number of civic organizations with Facebook pages per 1,000 Facebook users in the county. Volunteering rate is the percentage of Facebook users in the county who are members of volunteering or activism groups. All correlations and regressions are weighted by the number of children in each county whose parents have below-national-median income. Intervals represent 95% confidence intervals calculated using standard errors clustered by commuting zone.