Social Network Structure, Wealth, and Wealth Inequality Across Cultures

• The study analyzes 46 mostly rural communities across 28 countries, spanning pastoralists, fishers, farmers, and hunter-gatherers at varied stages of market integration.The dataset combines household possessions with multiplex relationships involving food sharing, information exchange, money lending, and other daily livelihood ties.Source: §1 · pp. 2, 4
• The study addresses whether social-capital and economic-connectedness patterns generalize beyond predominantly WEIRD and online-social-media samples, where livelihoods, sociality, and wealth inequality differ.It focuses on supportive household exchanges that can buffer risks in rural subsistence economies and may be altered by market, state, and financial integration.Source: §1 · pp. 2–3
• In most communities, household wealth is positively correlated with soliciting support, provisioning support, and access to relatively wealthy others, significantly more often than expected by chance.These three network forms distinguish support received, support provided, and economic connectedness.Source: §1 · p. 3
• Associations vary partly with communities’ reliance on wage or salaried labor and the inclusiveness of private property rights, but the authors do not claim causality.The authors cite multiple potential causal pathways across highly distinct communities.Source: §1 · pp. 3–4

Fieldsites

• Data from 46 communities span substantial variation in ecology, infrastructure, livelihoods, market integration, governance, and exposure to shocks.Most communities are rural and remote, but some are densely populated or near highways and government services; livelihood strategies commonly combine subsistence, market enterprises, and wage labor.Source: §2 · p. 4

Units of analysis

• Sharing units pool resources within variable household-like arrangements, averaging four to five individuals across approximately 60 units per community.They may comprise nuclear families, polygynous compounds, or multigenerational extended families.Source: §2 · p. 5

Contribution

• The cross-cultural breadth and ethnographically grounded measures extend inequality and social-capital research beyond large economies, administrative tax data, and digitally mediated sources.The study captures locally meaningful connections across multiple interaction forms that are generally unavailable in other datasets.Source: §2 · p. 5

Data collection

• Researchers collected detailed demographic, economic, and social-connection data from virtually all sharing units in each community.Material wealth is derived from enumerated property and assets, while support nominations define a directed, weighted network of flows among sharing units.Source: §2 · p. 5

Measures

• The resulting network measures capture support access, support provisioning, and the average wealth of connected sharing units.Average Alter Wealth considers connections that either provide support to a sharing unit or receive support from it.Source: §2 · p. 5

3 Within-Site Results: Sharing Units’ Wealth and Social Capital

• The section evaluates whether sharing units’ social capital and material wealth are related within communities.Social capital is measured through support access, support provisioning, and the wealth of connected alters.Source: §3 · p. 5
• Across most communities, wealthier sharing units have greater support access and provisioning, and connect to wealthier alters; all four correlations are positive and significant across communities.The analysis uses percentile-ranked wealth and network measures, with cross-community effects estimated using random-effects meta-analyses.Source: §3 · pp. 5–7, Figure 3 · p. 6
• The wealth–support-access association is not always robust, whereas provisioning support and Average Alter Wealth remain consistently robust to alternative wealth measures, network constructions, and connectedness definitions.Alternative specifications include unweighted or adult-weighted wealth, controls for sharing-unit size, different network aggregation rules, and median or summed alter wealth.Source: §3 · p. 7

4.1 Wealth Inequality and Network Structures

• Support-access and support-provisioning inequality show no consistent association with communities’ material wealth inequality.A positive correlation for support-access inequality is borderline significant but does not remain consistent across alternative wealth accountings.Source: §4.1 · p. 7, Figure 4 · p. 8
• Wealthier communities are more economically connected across wealth groups: Relative Average Alter Wealth is strongly negatively correlated with wealth inequality, indicating poorer units are less connected to wealthier ones.The association holds across network aggregations, wealth-accounting choices, and connection measures; a value of 1 would indicate equally wealthy supporters for poor and wealthy units.Source: §4.1 · pp. 7–8, Figure 4 · p. 8

4.2 Wealth Inequality and Economic Connectedness

• Normed wealth modularity is positively correlated with wealth inequality, and analyses using four wealth classes likewise find the strongest association for connections between the poorest and wealthiest quartiles.The modularity approach identifies community-specific cohesive wealth classes rather than imposing a median split.Source: Figure 4 · p. 8, §4.2 · p. 9

4.3.1 Community Characteristics Beyond Social Structure

• The cross-site analysis considers kin connections, task-exclusion norms, wage labor, cash sales, shocks, biological productivity, human influence, GDP, property rights, mobility, and political corruption as potential correlates.These measures span community, local-area, and national characteristics.Source: §4.3.1 · pp. 9–10

4.3.2 Correlations of Community Characteristics with Inequality

• Greater wealth inequality is associated with more wage or salaried labor, broader private property rights, and marginally greater Human Influence Index values.Because the analysis includes many variables and limited correlations, the findings are descriptive and cannot confidently identify or rule out deeper relationships or causal contributions.Source: §4.3.2 · p. 10, Figure 5 · p. 10

Core findings

• Wealthier sharing units have more reciprocal support connections and tend to connect to wealthier units, while inequality is negatively associated with connections between poorer and wealthier units rather than connection counts.Material wealth and social capital are closely tied, but the substance of connections—who connects to whom by relative wealth—matters most for inequality.Source: §5 · p. 11

Potential mechanisms

• Connections to wealthier households could reduce inequality by buffering poor households against risk, enabling riskier high-return investment, and improving access to information, aspirations, and norms.Cross-sectional data do not establish causality; the discussion presents these as plausible channels supported by prior literature [8, 85, 86, 17, 52, 54].Source: §5 · p. 11

Contribution

• The findings broaden evidence linking economic connectedness to upward mobility across geographic and social contexts and beyond friendship networks.The results are consistent with evidence from Facebook users and Honduran villages, while incorporating relationships central to daily social and economic exchange.Source: §5 · pp. 11, 13

Wealth homophily

• Wealth homophily varies with social context, including caste segmentation, state policies, charity, and egalitarian norms, motivating new theories of how networks relate to poverty and inequality.The discussion notes that multiple mechanisms may operate together [90], and that communities without positive wealth homophily raise questions about whether social capital facilitates or impedes movement out of poverty.Source: §5 · pp. 11–12

Community attributes

• Property-rights protection and wage or salaried labor are positively associated with wealth inequality, potentially through wealth monopolization and the costs of shifting from informal production to paid employment.The wage-labor interpretation is consistent with a dual economy in which employer-paid wages exceed average informal-sector incomes.Source: §5 · p. 12

Open questions

• Future research should identify causal pathways and examine distinct support types, ties beyond communities, and intra-household differences in how social capital and wealth affect mobility.Longitudinal data are being collected to address causality, while external ties may provide resources, information, markets, and recovery from spatially distributed shocks.Source: §5 · pp. 12–13

6.1 Sample construction

• Surveys covered effectively all sharing units in each community, producing a convenience sample intended to span diverse human social and economic arrangements.Researchers relied on prior local experience and relationships to facilitate comparative data collection.Source: §6.1 · p. 21

6.2.1 Material Wealth

• Material wealth aggregates inventoried property and assets, is converted to US Dollars, and is primarily analyzed per capita and by percentile rank.Results were robust to alternative household-composition and wealth measures; percentile ranks support cross-site comparison and reduce outlier influence.Source: §6.2.1 · p. 21

6.2.2 Network Measures of Social Capital

• Social-capital networks combine support, assistance, socializing, information sharing, and kinship ties into normalized weighted directed networks among sharing units.Edge weights reflect the number of nominations across support types relative to surveyed respondents, with each survey question weighted equally.Source: §6.2.2 · p. 22
• Network outcomes include support access, support provisioning, and weighted Average Alter Wealth for supporters and supportees.Support access is the out-degree, support provisioning is the in-degree, and Average Alter Wealth measures the wealth of connected alters.Source: §6.2.2 · p. 22

6.2.3 Community-Level Measures

• Community-level measures use Ginis for wealth and support inequalities, while Relative Average Alter Wealth compares poor sharing units’ supporter wealth with that of richer units.A value of 1 indicates equally wealthy connections for poor and rich sharing units; alternative percentile-ratio inequality measures produced largely comparable results.Source: §6.2.3 · pp. 22–23

6.2.4 Wealth Modularity

• Normed wealth modularity identifies the wealth split yielding the strongest network division and subtracts the 95% null-distribution quantile from its maximum modularity.This approach avoids restricting analyses to a prespecified median split and assesses whether the observed division exceeds uniform randomness.Source: §6.2.4 · p. 23

6.2.5 Site-Level Variables

• Community characterization combines observed network and demographic aggregates, researcher-reported traits, geospatial measures, and national institutional variables.Examples include kin connectivity, exposure to shocks, task-related norms, biological productivity, and property-rights inclusiveness.Source: §6.2.5 · p. 24

6.3 Correlations

• Within- and cross-site analyses primarily use percentile-rank correlations, supplemented by random-effects meta-analysis, multilevel models, and multivariate regressions.The authors caution that significance levels should be interpreted carefully because no single clustering dimension is clearly appropriate across communities.Source: §6.3 · p. 24

A.1 Data Collection

• ENDOW was created to enable cross-cultural inequality comparisons by combining data from willing researchers working in their established field sites.Because the sample was opportunistic, inferences primarily concern a wide range of communities transitioning into market-based economies.Source: §A.1 · p. 28

A Supplementary Information

• The supplementary materials document community subsistence patterns, sampling, sharing-unit characteristics, and wealth distributions across ENDOW sites.Figure A1 reports primary and secondary subsistence modes, while Table A1 and Figure A2 summarize community and sharing-unit characteristics, sampling, and wealth.Source: §A.1 · p. 28, Figure A1 · p. 29, Table A1 · pp. 30–31

A.1 Data Collection

• The sharing unit is the fundamental resource-pooling unit, often but not always equivalent to a household, and averages four or five members across communities.Table A1 summarizes community, sampling, sharing-unit, and wealth characteristics; Figure A2 provides descriptive plots.Source: §A.1 · p. 28, Table A1 · p. 30, Figure A2 · p. 32
• ENDOW researchers covered, on average, 96% of sharing units and measured material, embodied, and relational wealth using culturally tailored prompts designed for cross-cultural comparison.Material wealth inventories were converted into comparable monetary estimates, while embodied wealth included food security, work ability, education, language fluency, and other locally valuable skills.Source: §A.1 · p. 32, Table A2 · p. 33
• The dataset was collected from 2017 to 2025, with timing and procedures adapted to community availability, researcher constraints, local knowledge, and logistical conditions.Researchers used different tools and sources, including pen-and-paper or tablet network surveys and existing or newly collected demographic and kinship data.Source: §A.1 · p. 32
• These adaptations produced heterogeneity in the resulting data, despite shared data requirements and submission templates across contributing researchers.The variation arose from differences in collection tools, available demographic and kinship information, seasonal constraints, and researcher decisions shaped by community discussions.Source: §A.1 · p. 32

A Supplementary Information

• Community data underwent completeness and error checks, followed by researcher clarification of ambiguities or missingness in contributed files.Each community also obtained local, national, and institutional ethical approval, with overarching review by the University of Cincinnati Institutional Review Board.Source: §A.1 · p. 32

A.2 Material Wealth Data

• Material wealth was measured by exhaustively enumerating locally relevant, economically productive assets and valuing them in their community context.Researchers identified items through resident discussions, home visits, and local-market information; item sets and granularity varied across communities.Source: §A.2 · p. 33
• Each sharing unit’s total wealth equals the summed value of observed item quantities, including baseline items owned by virtually all units.Baseline items reduced respondent burden, while remaining items were counted through direct observation and member reports.Source: §A.2 · p. 33
• The primary wealth measure is wi = Σqik·pk, with per-capita wealth defined as wpc_i = wi/ni.Here qik is unit i’s observed quantity of item k, pk is its unit value, and ni is sharing-unit size.Source: §A.2 · p. 33
• Missing asset data were imputed using multivariate chained equations, generating 50 datasets whose median supplied the paper’s baseline values.Robustness checks assess sensitivity to both imputation and valuation choices.Source: §A.2 · pp. 33–34
• Cross-site inequality patterns persist across total sharing-unit, per-capita, and per-adult wealth measures, whose site-level Gini coefficients are well-correlated.Figures A6–A10 present the corresponding Lorenz curves and cross-site Gini correlations.Source: §A.2 · p. 34, Figure A6 · p. 38, Figure A7 · p. 39

A.3 Social Network Data

• Relational wealth was measured with ten name-generator questions, combining standardized and community-tailored phrasing asked in a fixed order.The questions covered borrowing and lending money, sharing goods, behavioral and labor assistance, communication, affective support, and external support.Source: §A.3 · p. 43
• Gender-specific prompts captured potentially distinct assistance, labor, communication, and affective-support ties while eliciting relationships across entire sharing units.The design reflected gendered divisions of labor and aimed to move beyond respondents’ personal connections.Source: §A.3 · p. 43
• Kinship data identified close consanguineal and affinal relationships, producing a comprehensive multilayer community network alongside economic and support ties [13] [5].Coverage included parent-child and full-sibling relationships among adults and children, while broader genealogical information extended to first cousins.Source: §A.3 · p. 43

A.3.1 Network Construction

• Because sharing units differed in respondent counts and multiple respondents could nominate the same partner, the analysis checks robustness across alternative person-to-unit aggregation procedures.Sampling strategies varied with respondent availability and fieldwork constraints, producing unequal nomination opportunities across sharing units.Source: §A.3.1 · p. 43
• The main composite network is weighted and directed, with edge weights normalized by the number of surveyed individuals in the sending sharing unit and summed across support domains.For example, two respondents naming three and two people in another unit produce a layer weight of 2.5.Source: §A.3.1 · pp. 43–44
• The proportional composite network uses respondent-level nomination proportions, whereas the raw composite network retains unnormalized nomination counts without dividing by surveyed respondents.Both alternatives are used to assess whether findings are similar under different edge-weight definitions.Source: §A.3.1 · p. 44
• The analysis excludes double-sampled name generators involving lending wages or household items because their weighting complexities differ from the request-based prompts used elsewhere.These responses are reserved for future work.Source: §A.3.1 · p. 45

A.4 Measures Constructed from Network and Wealth Data

• The analysis ranks wealth within sites and combines node wealth attributes with directed, weighted network ties to construct measures of alter wealth and economic connectedness.Measures include AAW, AAWP, AAWR, RAAW, UEC, EC, and relative connectedness; tie direction distinguishes supporters from supportees.Source: §A.4 · pp. 46–47

A.5 Community-Level Information

• Community-level production, environmental, and work-arrangement variables are primarily treated as confounders of relationships between material inequality and network structure.The study supplements network and wealth data with national, local geospatial, and researcher-provided community attributes.Source: §A.5 · p. 48

A.5.1 National Attributes

• National covariates include GDP per capita, national inequality and literacy, governance indices, socioeconomic exclusion, property rights, domestic mobility, and state authority characteristics,,,,.These variables are compiled from the World Bank World Development Indicators, V-Dem, and Polity5.Source: §A.5.1 · p. 48

A.5.2 Local Geospatial Attributes

• Local geospatial covariates are constructed around community centroids using raster data and buffers, covering population density, nightlights, human influence, travel time, temperature, and precipitation.The analysis uses the Google Earth Engine API or existing open datasets, with 250m and 1000m buffers for selected measures.Source: §A.5.2 · p. 48

A.5.3 Researcher-Provided Attributes

• Researchers provide ethnographically grounded attributes on subsistence, residence, inheritance, political organization, exchange, labor, livelihoods, shocks, and norms restricting work or property access.A project questionnaire records output allocation, non-leisure time use, livelihood concentration, shock frequency, and strong non-gender, non-age, non-wealth work or property norms.Source: §A.5.3 · pp. 48–49

B.1 Different Network Constructions

• Results using proportional-composite and raw-composite networks are similar to main composite-network results for support access, support provisioning, and Average Alter Wealth analyses.Source: §B.1 · p. 50, Figure A12 · p. 50, Figure A13 · p. 51

B.2 Different Accountings for Sharing Unit Composition

• Alternative within-community wealth accounting produces broadly comparable results without weighting or with per-adult weighting, but size-residualized analyses preserve provisioning associations while weakening or eliminating access associations.The no-weighting specification is somewhat weaker specifically for the relationship between out-degree and wealth; residualization uses log(sharing unit size) or sharing-unit-size fixed effects.Source: §B.2 · p. 53, Figure A17 · p. 55, Figure A18 · p. 56

B.3 Different Measures of Economic Connectedness

• Unit Economic Connectedness and absolute alter-wealth measures remain positively associated with sharing-unit wealth across most or all communities, respectively.Unit Economic Connectedness is positive using either supporters or supportees and either absolute wealth or wealth per capita; median and summed alter wealth show consistent positive associations.Source: §B.3 · p. 58, Figure A21 · p. 58, Figure A22 · p. 59

B.4 Multilevel Models of Degree, AAW, and Wealth

• Multilevel varying-slope models yield positive average associations between wealth and support access, support provisioning, or Average Alter Wealth across all four specifications, with coefficients nearly identical to pooled estimates.Site-specific slopes vary around the average; except for Panel B, negative slopes lie within two standard deviations of the pooled slope.Source: §B.4 · p. 60, Table A3 · p. 61, Table A4 · p. 61

C.1 Different Accountings for Sharing Unit Composition

• Cross-site inequality analyses compare wealth Ginis with support provisioning, support access, Relative Average Alter Wealth, Economic Connectedness, and Relative Connectedness under per-capita, absolute, per-adult, and size-adjusted accounting.Figures A24–A28 present the corresponding bivariate associations and correlation coefficients with 95% confidence intervals.Source: §C.1 · p. 62, Figure A24 · p. 63, Figure A25 · p. 64

C.2 Using Incoming Nominations

• Using incoming nominations produces wealth-inequality associations for supportees that are quite similar to those based on outgoing nominations for Relative Average Alter Wealth, Economic Connectedness, and Relative Connectedness.The comparisons use alternative accounting of sharing-unit wealth, including absolute, per-adult, and size-adjusted measures.Source: §C.2 · p. 68, Figure A29 · p. 69, Figure A30 · p. 70

C.3 Using Absolute Wealth

• Cross-site analyses additionally construct ratios comparing below-median with above-median sharing units’ average summed or median alter wealth to relate absolute network wealth access to community wealth inequality.These measures are presented separately for alters who provide support and alters to whom support is provided.Source: §C.3 · p. 72, Figure A32 · p. 72, Figure A33 · p. 73

C.4 Using Different Measures of Inequality

• The 90th-to-10th and 80th-to-20th wealth-per-capita ratios are positively correlated with the Gini and significantly negatively correlated with Relative Average Alter Wealth.These alternative inequality measures assess whether results depend on extreme wealth outliers.Source: §C.4 · p. 74, Figure A34 · p. 75

C.5 Ruling Out Wealth-Band Tie Formation

• Across simulated bandwidths, observed RAAW–wealth-per-capita-Gini correlations generally lie far from simulated correlation distributions, except at 20%, where most mass remains distant; excess RAAW remains negatively correlated with the Gini.Source: §C.5 · p. 77
• Simulated networks with absolute-wealth-band or site-varying within-band connecting rules are compared with the observed cross-site correlation to assess wealth-band tie formation.The simulations vary the connecting rule, while the observed correlation is marked for comparison.Source: Figure A36 · p. 78, Figure A37 · p. 79, Figure A38 · p. 80

C.6 Relative Average Alter Wealth versus Assortativity

• Relative Average Alter Wealth has a clear, strong negative association with assortativity measured using either wealth per capita or its percentile rank.The comparison uses both wealth-per-capita assortativity and percentile-rank assortativity as alternative measures of homophily.Source: §C.6 · p. 81, Figure A39 · p. 81

C.7 Quartile Splits

• Greater wealth homophily across wealth quartiles is associated with greater wealth inequality, whereas poorer–wealthier connections are associated with less inequality.The quartile analysis finds positive correlations for wealthier quartiles and lower-diagonal connections, but negative correlations for poorer quartiles connected to the wealthiest quartile.Source: §C.7 · p. 82, Figure A40 · p. 83

C.8 Multilevel Model for Economic Connectedness

• The multilevel model estimates edge probabilities from wealth-quartile positions and ego household size, with economic connectedness linked to inequality through a regression coefficient controlling for site confounders.Separate models are fit for all 16 focal quartile pairs using posterior inference.Source: §C.8 · p. 84
• Communities where the wealthiest quartile was better connected to the poorest quartile had lower wealth inequality in the multilevel posterior analysis.The model estimates economic connectedness from wealth-quartile network formation while accounting for site-level confounders.Source: §C.8 · p. 84, Figure A41 · p. 85

C.9 Wealth Modularity

• The wealth-modularity analysis defines the strongest wealth-class structure by maximizing modularity over every possible binary division of each site's wealth distribution.Significance is assessed by permuting sharing-unit wealth while preserving the observed network structure.Source: §C.9 · p. 86
• Normalized maximum wealth modularity was strongly associated with site wealth inequality after comparing observed modularity with wealth-permutation null distributions.The modularity measure subtracts each site's 95% null-distribution quantile from its maximum wealth modularity.Source: §C.9 · p. 86, Figure A42 · p. 87, Figure A43 · p. 88

D Consequences of Wealth Mis-measurement

• Wealth estimates are imperfect proxies because asset inventories may omit holdings and nonmarket assets require value-equivalence judgments without market prices.Direct cross-validation against true wealth is infeasible because these estimates are the best available measures for each sharing unit.Source: D Consequences of Wealth Mis-measurement · p. 91
• Across communities, wealth Gini was unrelated to mean wealth, and Relative Average Alter Wealth was unrelated to median wealth, reducing concerns about systematic measurement bias.These relationships were evaluated using both absolute and per-capita wealth measures where reported.Source: D Consequences of Wealth Mis-measurement · p. 91, Figure A46 · p. 91, Figure A47 · p. 92

D.1 Noise Perturbation of Item Values

• Noise perturbations left wealth Ginis, within-site material–social wealth relationships, and the wealth-Gini–Relative Average Alter Wealth correlation largely robust.Across 50 trials, every item value was independently multiplied by 0.8 or 1.2 with equal probability, and downstream statistics were recomputed.Source: §D.1 · p. 93, Figure A48 · p. 93, Figure A49 · p. 94

D.2 Multiple Imputation of Missing Item Data

• Wealth Gini coefficients were fairly invariant across 50 stochastic imputations of incomplete item inventories.Figure A51 reports medians and approximate 95% ranges across the 50 imputed datasets.Source: §D.2 · p. 96, Figure A51 · p. 96
• Fifty imputed datasets per site were constructed with the mice package in R [96] when communities had incomplete item inventories.The procedure applied only to communities with missing inventory values for some sharing units.Source: §D.2 · p. 96
• Sites requiring item-inventory imputation did not systematically differ from sites where wealth imputation was unnecessary.Figure A51 highlights imputed cases in red and fully observed cases in gray.Source: §D.2 · p. 96, Figure A51 · p. 96

E Kinship

• Kin- and non-kin-based networks show broadly comparable cross-community correlations between social capital measures and material wealth.These patterns align with results from the overall composite network.Source: E Kinship · p. 97, Figure A52 · p. 98
• Kin-based and non-kin-based Relative Average Alter Wealth are uncorrelated across communities, with no consistent difference in which is higher.This suggests the relationship between RAAW and wealth inequality is not primarily driven by kinship connections.Source: E Kinship · p. 97, Figure A53 · p. 99, Figure A54 · p. 99
• Kinship provides baseline support across communities but is not the main explanation of the study’s results.The conclusion follows from the comparable kin and non-kin patterns.Source: E Kinship · p. 97

F The Relationship of Other Community Characteristics with Wealth Inequality and Economic Connectedness

• The analysis examines how additional community characteristics relate to both wealth inequality and economic connectedness.These characteristics are drawn from the variables presented in Table A5.Source: F The Relationship of Other Community… · p. 100, Table A5 · p. 100
• Exact values for the site-level variables are reported in Table A16 alongside other site-level variables in Section G.This extends the documentation of the variables introduced in Table A5.Source: F The Relationship of Other Community… · p. 100
• Table A5 provides basic descriptions, information sources, and measurement scales for the additional community-level variables.The variables characterize each community for subsequent analyses of inequality and connectedness.Source: Table A5 · p. 100, F The Relationship of Other Community… · p. 100

F.1 Correlations of Community Characteristics with Wealth Inequality

• LASSO results identify Relative Average Alter Wealth, private property rights, and wage or salaried labor as the leading predictors of the Gini of wealth per capita.Biologically-Effective Degree Days also remains among the strongest predictors after the Puffer transform.Source: §F.1 · pp. 101, 103, Figure A55 · p. 101
• The wealth-inequality analysis combines significant bivariate correlates, LASSO-highlighted variables, and the full variable set in stepwise regressions.Table A6 reports pairwise correlations corresponding to Figure 5, while Table A7 reports the stepwise models.Source: §F.1 · pp. 101, 103, Table A6 · p. 101, Table A7 · p. 104
• Relative Average Alter Wealth is significantly and strongly negatively associated with wealth inequality across all regression specifications, while wage or salaried labor and shared private property rights show positive associations.Better connections between poor and wealthy sharing units correspond to lower inequality; other covariates are not consistently informative.Source: §F.1 · p. 103, Table A7 · p. 104

F.2 Correlations of Community Characteristics with Economic Connectedness

• Economic connectedness is assessed with Relative Average Alter Wealth because it significantly predicts inequality and explains the greatest variation among the study’s variables.This analysis is motivated by comparisons with country-level social-capital research, including Otero et al..Source: §F.2 · p. 105
• No robust predictors of Relative Average Alter Wealth were found among the community characteristics considered.Time spent in wage or salaried labor has a marginal negative bivariate association but is not a consistent multivariate predictor.Source: §F.2 · p. 105, Table A8 · p. 105, Table A9 · p. 106

G Site-Level Variables

• The site-level appendix reports composition-adjusted measures and network-derived social-capital measures for supporters and supportees across the analyses.Tables A10–A14 cover wealth inequality, support inequality, Relative Average Alter Wealth, Economic Connectedness, Relative Connectedness, and Normed Wealth Modularity; Table A16 covers other site-level variables.Source: G Site-Level Variables · p. 107, Table A10 · p. 108, Table A11 · p. 109
• The site-level variable inventory includes absolute, per-capita, per-adult, and size-adjusted measures used throughout the analyses.The listed tables also include normed wealth modularity and other site-level variables considered in bivariate and LASSO regressions.Source: G Site-Level Variables · p. 107, Table A15 · p. 113, Table A16 · pp. 114–115