In almost all of the communities studied, wealthier sharing units had more mutual support connections and tended to be linked to relatively wealthier units. Across communities, greater wealth inequality was also associated with stronger separation between wealth groups in social networks. The data show relationships, not whether wealth changed people's connections or the other way around.
The findings come from an arXiv preprint, version 1, dated 26 Aug 2026. It covers approximately 3,500 sharing units, the household-like resource-pooling groups used in the analysis, in 46 small, mostly rural communities across 28 countries. The sites were at different stages of market integration.
How the networks were mapped
Researchers collected demographic and economic information, asset inventories, and support nominations from virtually all sharing units. Those nominations were used to build a directed, weighted network, keeping track of who was connected to whom and how the ties were represented. The study then compared wealth per person with network connections inside communities and compared community-level network measures with wealth inequality between communities.
The clearest result appeared within individual communities. In almost all of them, wealth was positively associated with the number of mutual support ties and with the relative wealth of the sharing units connected to each unit. That pattern is a form of economic homophily, the tendency for wealthier units to be linked with wealthier units.
The size of the association was estimated in varying-slope multilevel models, allowing the relationship to differ by site. Using pooled estimates across 46 sites, the slope was 0.220 for support access and 0.234 for support provisioning. For Average Alter Wealth, a measure of the typical wealth of linked units, pooled slopes were 0.191 for supporters and 0.179 for supportees.
The pattern did not depend equally on every network measure. Average Alter Wealth and support-provisioning results remained robust across the specification changes reported by the study, while support-access results were less consistently robust.
The picture changes across communities
The cross-community results were more mixed. A community's wealth inequality was not consistently associated with inequality in support access or support provisioning. The analysis described a positive, borderline association for support access, but it was not robust across alternative wealth specifications.
Relative Average Alter Wealth, which captures the relative wealth of a unit's network contacts, was lower in communities with greater wealth inequality. The correlation was -0.47, with a 95% confidence interval from -0.67 to -0.20. That is a cross-community association, not evidence that one measure drives the other.
The study also found that normed wealth modularity was positively correlated with the wealth per capita Gini, the study's measure of wealth inequality. Modularity measures whether ties stay within wealth groups rather than cross between them. The result is consistent with greater wealth-class network segregation in more unequal communities, but no numerical estimate was reported in the supplied analysis.
At the community level, more time in wage or salaried labor and more widely shared private-property rights were associated with greater wealth inequality. A marginal positive association was also reported for the Human Influence Index. These comparisons involved many possible community-level variables and a limited number of sites, so other relationships are difficult to identify with confidence.
A snapshot, not a verdict
The researchers used percentile-rank correlations, two-sided tests of positive correlations, random-effects meta-analysis, varying-slope multilevel models, and simple bivariate correlations across sites. Because the study was cross-sectional, it captured a snapshot rather than showing which came first. The design cannot identify the causal mechanisms behind the observed patterns.
The sample was also a convenience sample. It depended on researchers with established field sites and on near-complete coverage of sharing units, which limits how broadly the findings can be applied. The study data cannot currently be shared because of privacy and consent assurances, but site-level aggregate measures and analysis code with simulated data are available.
The fieldwork was supported by National Science Foundation awards 1743019, 2218860, and 2218861, with additional support from the Santa Fe Institute and the Omidyar Network's Emergent Political Economies project. The authors declared no competing interests.
Paper data and sources
Original title: Social Network Structure, Wealth, and Wealth Inequality Across Cultures
Authors: Eleanor A. Power, Monique Borgerhoff Mulder, Samuel Bowles et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text