PaperlayerView exact revision on arXiv

Source-linked AI summary

A cross-cultural dataset on the structure of social networks: The ENDOW project

Eleanor A. Power, Daniel Redhead, Tom Rutter, Sahana Subramanyam, Jeremy Koster, Monique Borgerhoff Mulder, Samuel Bowles, Matthew O. Jackson, Nurul Alam, Sarah Alami, Alexandra Alvergne, Curtis Atkisson, Michele Barnes, Bret Beheim, Christine M. Beitl, Madeline Brown, Mark Caudell, Wendy Chávez-Páez, Komal Chauhan, Joshua Cinner, Siobhán Cully, Augusto Dalla Ragione, Angelina L. DeMarco, Ivan Deschenaux, Federico Fernandez, Juan Pablo Ferreiro, Drew Gerkey, Matthew Gervais, Christopher Golden, Gianluca Grimalda, Werner Hertzog, Karen Kramer, Geoff Kushnick, Banrida Langstieh, Rodrigo Lazo, Sheina Lew-Levy, Shane Macfarlan, Emmanuel Maliti, Karl J. Mertens, Madalena Monteban, Rafael Morais Chiaravalloti, Daniel Murphy, Kathryn Oths, Alejandro Pérez Velilla, Emily Post, Sean Prall, Cody Ross, Brooke Scelza, Michael Schnegg, Edmond Seabright, Mary K. Shenk, Kathrine E. Starkweather, Chun-Yi Sum, Cole Tobin, Bram Tucker, Bapu Vaitla, Vivek Venkataraman, Justin Weltz, John P. Ziker

arXiv:2610.28793v1cs.SIecon.GNq-fin.EC

TL;DR

Evidence on social-network structure has largely come from limited settings, leaving the cross-cultural reach of common assumptions uncertain. ENDOW presents support-network data from diverse communities; its measures show substantial variation, suggesting that familiar rules of thumb may merit reconsideration.

  • Problem

    Many assumptions about social-network structure rest on a limited range of settings, while the rationale for multiple prompts is that they may elicit distinct alters and relationships.

    Source: Network Layer Overlap · p. 9, Discussion · p. 13

  • Method

    ENDOW collected comparable demographic, economic, and social-network data using ten support prompts answered for sharing units, then constructed networks from respondents’ nominations.

    Source: Methods · p. 15, Social Networks · p. 16

  • Results

    Substantial variation typifies nearly every measure examined, and ENDOW network measures differ in meaningful ways from friendship-network benchmarks.

    Source: Discussion · pp. 13–14

  • Takeaways & Limitations

    The findings suggest tacit network-analysis rules of thumb may merit reconsideration and add reference points for studying human social-structural diversity.

    Source: Discussion · pp. 13–14

    Differences in respondent sampling and partial concordance may mean observed network differences reflect sampling as well as local social patterns.

    Source: Discussion · p. 14

Abstract

from arXiv · show

We present multilayer social network data from 50 communities, located in 29 countries around the world, gathered as part of the "ENDOW" project. The communities show exceptional diversity, from fishers in Bangladesh and Vanuatu, to horticulturalists in Fiji and Bolivia, to pastoralists in Namibia and Argentina, to farmers and wage labourers in India and Colombia. Data collection embraced this variation, while still aiming for comparability. Respondents answered ten network prompts on behalf of their "sharing unit" (household), covering different types of material and immaterial support. From these, we construct networks representing flows of support between sharing units in each community. We compare nomination patterns and network summary statistics across the ENDOW communities and to other benchmark datasets. The summaries exhibit considerable heterogeneity, suggesting a need to revisit existing ``rules of thumb'' about social network structure. These measures contribute novel reference points for research on the diversity of human sociality and social structure.

Results

Across ENDOW communities, nomination patterns and network structure vary substantially by community, prompt, and support domain. Overlap between layers is uneven, and kin are commonly named as support partners despite wide differences in access to co-residing kin.

Nomination patterns

  • Respondents named an average of two to three alters per prompt, while aggregate within-community support prompts named about 10 individuals from about six sharing units across communities.Mean nominations varied substantially across communities, and global summaries pooled all respondents, giving greater weight to larger or more-sampled communities.Source: Nomination patterns · p. 4
  • About one quarter of named alters were non-residents overall, and external alters averaged around 20% even for aggregate within-community support prompts.About 10% of nominated alters were members of the respondent’s own sharing unit; that share was notably high for behavioural support and low for external support and household-goods borrowing or lending.Source: Nomination patterns · pp. 5–6

Network Structure

  • ENDOW support networks have lower median degree, reciprocity, and transitivity than friendship-network benchmarks, while generally short paths and higher-than-random clustering coexist with frequent negative assortativity.Borrowing and lending money layers more often show negative degree assortativity; many communities show negative values in every layer.Source: Network Structure · pp. 7, 9
  • Community identity explains 61% (95% CI: 49 −70%) of remaining variance in average degree after accounting for network layers, with substantial between-community variation also in other measures.After controls for community size, sampling, and external alters, between-site variation remains prominent.Source: Network Structure · p. 9

Network Layer Overlap

  • Sharing-unit nominations aligned more across prompts than individual nominations, but overlap varied by relationship type and was low across sharing units for double-sampled borrowing and lending questions.Cross-unit concordance averaged 13% for monetary loans and 21% for household-item exchanges; principal components captured basic connectedness first, while later patterns varied by community.Source: Network Layer Overlap · pp. 9–10
  • There are no consistent cross-cultural domains of support: which support layers align varies across communities, including among ENDOW communities in the same region.The two external-tie prompts named quite distinct individual alters, while the double-sampled domains typically loaded closely together in principal component analyses.Source: Network Layer Overlap · pp. 9–11

Kinship

  • In aggregate networks, about a quarter of other sharing units named were close kin, and over 60% of sharing units with close kin were named as providing some support.Access to co-residing kin varied substantially: some communities had a majority of sharing units with no kin among neighbours, while others averaged more than five close-kin sharing units.Source: Kinship · p. 11

Discussion

Across ENDOW communities, nearly every network measure varies substantially, challenging reliance on simple structural rules of thumb. Comparisons also require care because elicitation, sampling, and network construction can affect measured patterns.

Discussion

  • Substantial variation across nearly every measure suggests that many tacit network “rules of thumb” may merit reconsideration.The authors attribute some variation to measurement choices and much of it to meaningful differences in social relations.Source: Discussion · pp. 13–15
  • Respondents named over fifteen alters on average in some communities but closer to five in others, even for more standardized prompts.Source: Discussion · p. 13
  • Support alters and kinship ties vary across communities and prompts, although close kin present in a community are frequently named as support alters.Source: Discussion · p. 13
  • Different support types often share alters, but which types overlap varies across ENDOW communities.The authors suggest local norms and economic production and exchange as possible influences, not established causes.Source: Discussion · p. 13
  • Measurement choices complicate comparisons: roster-only elicitation would miss about a quarter of nominations, prompt order may affect overlap, and respondent sampling may shape aggregate networks.The paper notes that who represents a sharing unit can influence the resulting network.Source: Discussion · pp. 13–14
  • ENDOW networks differ from friendship-network benchmarks, with lower median reciprocity and transitivity and negative aggregate-network degree assortativity in about half of communities.Disassortativity also characterizes many communities in networks closest to friendship.Source: Discussion · p. 14

Methods

ENDOW collected social, demographic, and economic data across diverse communities using structured interviews designed to support comparison while fitting local contexts. The authors build sharing-unit networks from multiple support prompts and analyze their overlap and variation.

Methods

  • ENDOW’s fundamental unit of analysis is the sharing unit, where resources are principally pooled and shared; it often, but not always, aligns with a household.Communities had a median of about 60 sharing units, a mean of about 90, and units averaged five members.Source: Methods · p. 15

Data Collection

  • Interviews aimed for near-complete coverage of sharing units, averaging 96%, while intended respondents varied across communities.Targets ranged from all adult residents to one household head; researchers tailored question wording to communities while facilitating comparison.Source: Data Collection · p. 15

Social Networks

  • Ten network prompts covered multiple forms of support, including double-sampled borrowing and lending and gender-specific prompts for behavioral and informational or affective support.Source: Social Networks · p. 16
  • Prompts covered wage loans, locally salient goods, gender-specific behavioral and social support, and external institutional or economic support.Source: Social Networks · pp. 16–17

Kinship

  • The kinship network links sharing units that share close kin, defined here as a parent-child or full-sibling relationship.Researchers used genealogical information to identify relatedness among residents, though data collection varied by community.Source: Kinship · p. 17

Network Construction

  • Networks collapse individual responses to sharing-unit nodes and focus on within-community ties; aggregate edges sum weights from selected support layers.Each layer’s edge weight divides the number of alter-nominations from a sharing unit by its number of respondents for that prompt.Source: Network Construction · pp. 17–18

Network Overlap

  • The authors compare layers using alter-level Jaccard overlap, dyad-level correlations, and community-specific principal component analyses.Source: Network Overlap · p. 18

Variance Decomposition

  • Random-intercept models partition network-statistic variation into within-site across-layer and across-site components, with and without site-level controls.The authors summarize between-site variance using intraclass correlation coefficients and bootstrap confidence intervals.Source: Variance Decomposition · p. 18

Ethics

The study received local and overarching ethics review, and respondents gave informed consent. Because relational data pose privacy and deanonymisation risks, only network-level summary statistics are currently shared; the findings-generating code is available.

Ethics

  • Each community obtained ethics approval through its contributing researcher, alongside an overarching University of Cincinnati IRB review; respondents gave informed consent.Source: Ethics · p. 25

Data availability

  • The full network data are not currently shareable because of varied privacy assurances and deanonymisation risks; network-level summary statistics are provided instead.Source: Data availability · p. 25

Code availability

  • The code used to generate the findings is available in the project’s public GitHub repository.Source: Code availability · p. 25

A.1 ENDOW Communities

ENDOW communities vary in their modes of subsistence, and Table A1 summarizes community and sharing-unit characteristics alongside sampling information.

A.1 ENDOW Communities

  • The reported primary and secondary modes of subsistence provide one view of variation across ENDOW communities.The modes were reported by the contributing ENDOW researcher.Source: §A.1 · p. 28, Figure A1 · p. 31
  • Table A1 reports each community’s sharing-unit characteristics and sampling, including member counts and the proportion of sharing units with network data.Its Members columns give the average, standard deviation, and maximum number of members; Sampling columns report average people surveyed per sharing unit and data coverage.Source: Table A1 · p. 29
  • Figure A1 colors community site codes by United Nations geoscheme region, with only represented regions shown on the map.Figure 1 provides more exact locations.Source: Figure A1 · p. 31

A.2 Survey Respondents

Sharing-unit coverage was high on average, although the number of people sampled within each unit varied substantially.

A.2 Survey Respondents

  • Sharing-unit coverage averaged over 96%, while the number of people sampled within each sharing unit varied substantially.Table A2 summarizes sharing-unit coverage and respondents for each community.Source: §A.2 · p. 32, Table A2 · p. 33
  • Table A2 reports sharing-unit coverage and respondent information separately for each community.Source: Table A2 · p. 33

A.3 Nominations

Nomination patterns varied across communities and support prompts, with respondents naming roughly two to three alters per question on average. Gender and age patterns were generally heterogeneous, while external nominations and the potential effects of nomination caps varied by prompt.

A.3 Nominations

  • Across pooled respondents, informational/affective prompts produced the most nominations, government/NGO ties the most zero responses, and behavioural support the most within-unit nominations.Borrowing/lending household items had few within-unit nominations; prompts eliciting ties beyond the community had higher external-nomination proportions.Source: §A.3 · p. 35, Table A3 · p. 35
  • Comparisons with prior datasets found nomination counts varied across populations and support types, including relatively higher counts for socialisation questions in one personal-networks study.ENDOW’s external-tie proportions were broadly comparable to the Kenya and Uganda dataset’s average of 27.9%, while Thailand rice-harvest ties extending beyond villages averaged 46.7%.Source: §A.3 · pp. 35, 38
  • Across nearly every prompt, some respondents named very many individuals; a hypothetical cap of five would constrain a notable number, especially for borrowing/lending goods and informational/affective support.A cap of ten would only rarely constrain respondents.Source: §A.3 · p. 36, Figure A2 · p. 36
  • Respondents cumulatively named around 9 or 10 individuals across aggregate-network prompts, ranging from a median of 3 in DJ to 18 in BY.The maxima were 56 individuals named by a man in UP and 36 sharing units named by a woman in AV.Source: §A.3 · pp. 36–37, Figure A3 · p. 39
  • Gender differences in total nominations were generally slight and inconsistent across communities, with men naming more in some settings and women in others.Some variation reflects communities where gender-specific prompts differed by respondent gender.Source: §A.3 · p. 37
  • Age patterns were heterogeneous: some communities suggested nomination counts rising into middle age and declining later, while others showed little age-course change.Across communities, evidence for strong age differences was generally small; age and age-squared regressions had only a small handful of significant correlations.Source: §A.3 · pp. 37–38, Figure A5 · p. 41

A.4 Network prompts for each network layer

The appendix documents community-specific network prompts across layers, covering material support, labor, socializing, and social interaction and institutional contacts.

Prompt wording

  • In some communities, intended respondents shaped gender-specific wording; for example, AH prompts referred to the male respondent’s relations or his wife’s relations depending on the support type.The appendix notes that the intended respondent is identified in Table A1.Source: §A.4 · p. 52

Network prompts for borrowing and lending money

  • Money prompts asked whom households could borrow from in an urgent need and whom they would lend to, often using locally specified sums and medical emergencies.Examples include loans measured as a week’s wages, local currency amounts, or costs tied to healthcare.Source: §A.4 · pp. 53, 55

Network prompts for borrowing and lending goods

  • Goods prompts covered receiving and giving food, basic necessities, household items, and tools, with some questions specifying recent sharing or borrowing.Examples range from food and groceries to farm equipment and household items.Source: §A.4 · pp. 57, 59

Network prompts for labor/behavioral support

  • Labor prompts asked about women’s help and cooperation in tasks such as childcare, cooking, and farming, and men’s help with fishing, livestock, fieldwork, and repairs.Some prompts focused on working together; others asked who could provide help when needed.Source: §A.4 · pp. 61, 63

Network prompts for female-oriented informational/affective support

  • Female-oriented social prompts asked whom women talked with, visited, or spent leisure time with, including at gatherings and during routine visits.Prompts variously referred to casual conversation, socializing, and visiting in free time.Source: §A.4 · p. 65

Social interaction

  • Social-interaction prompts ask about household members’ conversations, visits, friendships, and time spent socializing, often focusing on men.Examples refer to casual conversation, regular visits, social gatherings, and time spent together at venues such as coffee shops, markets, or religious services.Source: §A.4 · p. 67
  • Some social prompts specify particular settings or periods, such as the past month, Sundays, evenings, or leisure time.Several prompts ask about talking, visiting, or spending time together in these contexts.Source: §A.4 · p. 67

Institutional ties

  • Institutional-tie prompts ask which government, NGO, or other institutional contacts household members know and could approach for support.Variants mention help with favors, financial needs, advice, paperwork, conflicts, or urgent situations; some identify contacts by role or organization.Source: §A.4 · p. 69

A.5.1 Alter overlap

Alter overlap is summarized with Jaccard similarities across prompts, both pooled across respondents and separately by community. Pooled borrowing/lending and informational/affective similarities are lower than a reported friendship-financial comparison, while some community comparisons are unavailable because prompts differed or were omitted.

A.5.1 Alter overlap

  • 0.11–0.13 is the Jaccard similarity between borrowing/lending money and informational/affective ties, below the reported value of about 0.27 between friendship and financial networks in Honduras.The ENDOW values pool respondents across communities, giving more weight to communities with more respondents and nominations.Source: §A.5.1 · p. 72
  • Figures A6–A8 show Jaccard similarity of alters named across prompts separately for each community, distinguishing individuals from sharing units.Lower triangles show all individuals named; upper triangles show sharing units named.Source: §A.5.1 · p. 72, Figure A6 · p. 73, Figure A7 · p. 74
  • Some community comparisons are blank where different respondents received different questions or some questions were not posed as expected.Gender-specific questions were sometimes asked only of respondents of that gender; external-alter and other prompts were also omitted in some communities.Source: §A.5.1 · p. 72

A.5.2 Dyad overlap

Dyad overlap is assessed by correlating binary or weighted edge values for the same dyad across network layers. Figures show pooled correlations and community-specific versions, with some cells blank where prompts were omitted or yielded no nominations to other residents.

A.5.2 Dyad overlap

  • Figure A9 reports Pearson correlations between the same dyad’s binary or weighted edge values across network layers, pooling dyads across all communities.The analysis places dyads in rows and network layers in columns before correlating layer entries across dyads.Source: §A.5.2 · p. 76, Figure A9 · p. 76
  • Some cells are blank when prompts were absent, mismatched, or produced no nominations to other community residents.The stated cases include omitted external-alter questions and communities with no nominations on particular external-tie prompts.Source: §A.5.2 · p. 76
  • The captions for Figures A10, A11, and A12 each describe correlations of the dyads, considering weighted edges in the lower triangle and binary edges in the upper triangle.Source: Figure A10 · p. 77, Figure A11 · p. 78, Figure A12 · p. 79

A.5.3 Principal Component Analysis

The appendix presents weighted-edge principal-component loadings for each community. The figures show the first and second components, while the table provides loadings by layer and variance explained by component.

A.5.3 Principal Component Analysis

  • Figures A13–A15 show loadings on the first and second principal components for each community, constructed using weighted rather than binary edges.Table A.5.3 reports the variance explained by each principal component and the loading of each layer onto each principal component.Source: §A.5.3 · p. 80
  • In Table A24, each entry is the correlation (loading) between a network layer and a principal component.Loadings near ±1 indicate strong association with that component; values near 0 indicate little association.Source: Table A24 · pp. 84–89
  • Read Table A24 across rows to see how a layer’s variance is distributed across components, and down columns to interpret each component.Components are ordered by descending variance explained.Source: Table A24 · pp. 84–89

A.6 Concordance

The analysis assesses respondent agreement in two ways: among members of the same sharing unit and across the two double-sampled questions.

A.6 Concordance

  • Concordance is assessed both among respondents from the same sharing unit and between respondents answering the two double-sampled questions.The first comparison applies when a sharing unit has multiple respondents; the second compares respondents across the double-sampled questions.Source: §A.6 · p. 93

A.6.1 Concordance within Sharing Units

Within-unit concordance compares nominations made by multiple respondents from the same sharing unit, across individual prompts and prompts contributing to the aggregate network.

A.6.1 Concordance within Sharing Units

  • Figures A16 and A17 show within-unit nomination concordance by prompt and for prompts contributing to the aggregate network, using individual and sharing-unit nominations, respectively.The analysis includes communities with at least 10 cases of sharing units with multiple respondents.Source: §A.6.1 · p. 93, Figure A16 · p. 94, Figure A17 · p. 95
  • Empty figure entries mark communities with fewer than 10 multiply sampled sharing units or where questions were not asked or not asked of multiple respondents.This includes cases where gender-aligned questions were asked of only one gender.Source: Figure A16 · p. 94, Figure A17 · p. 95
  • Figure A17 uses a lower concordance standard: naming different members of the same sharing unit counts as concordant.Figure A16 instead concerns concordance in nominations of individuals.Source: Figure A16 · p. 94, Figure A17 · p. 95

A.6.2 Concordance for Double-Sampled Networks

Double-sampled request and give layers allow concordance to be estimated across sharing units. Concordance is greater for household item exchange than for monetary loans.

A.6.2 Concordance for Double-Sampled Networks

  • Household item exchange layers show greater concordance than monetary loan layers.Figure A18 summarizes the overall proportions reported in Table A25.Source: §A.6.2 · p. 96, Figure A18 · p. 99
  • A tie in one sharing unit’s borrow-money layer is expected to match a tie in the other unit’s lend-money layer in the opposite direction.The analysis also compares request and give layers for household item exchange.Source: §A.6.2 · p. 96
  • Table A25 reports concordant-tie proportions overall and by layer, with overall concordance calculated over the union of request and transposed give networks.Request denotes reported receipt of support; Give denotes reported provision of support.Source: §A.6.2 · p. 96, Table A25 · p. 97

A.7 Summary Statistics for Social Support Networks

This section examines ENDOW support networks across individual layers, aggregate networks, and close kin through network statistics, tables, and visualizations, including how respondent sampling and close-kin availability relate to support ties. It considers variation across communities and excludes external-focused prompts from within-community analyses.

A.7.1 Mean Degree

  • Aggregate mean degree ranges from 1.7 in Colombian community IH to 15.3 in Indian community UP, with a cross-community median of 6.6; layer medians range from roughly 1.5 to 2.4.Weighted degree follows similar patterns but is higher; respondent-count associations seen in unweighted degree are absent from weighted measures.Source: §A.7.1 · p. 102, Figure A19 · p. 102

A.7.2 Degree Distribution; A.7.3 In-Degree Centralization

  • Degree distributions are right-skewed with heavy tails but rarely strictly scale-free; in-degree centralization is generally low, with NI and AP outliers driven by hubs in borrow-money networks.The men’s informational/affective support network has the highest median degree and in-degree centralization.Source: §A.7.2 · p. 104, Figure A20 · p. 104, Figure A21 · p. 105

A.7.4 Degree Assortativity; A.7.5 Density

  • Degree assortativity is near zero or slightly positive overall, with negative monetary-loan medians; aggregate density ranges from 0.31 in 18-unit KT to 0.01 in 730-unit PQ.Density varies with community size, though similarly sized communities can also differ substantially; the aggregate network is much denser than individual layers.Source: §A.7.4 · p. 107, Figure A23 · p. 107, §A.7.5 · p. 108

A.7.6 Average Path Length

  • Aggregate networks have median average path length 2.61 and median diameter 6, but isolates and components complicate layer-level reachability; observed aggregate path lengths exceed random-graph expectations.Behavioral-support layers can have short paths yet substantial isolates, with median isolate proportions of 13–20% across communities.Source: §A.7.6 · pp. 110–111, Figure A26 · p. 110, Figure A27 · p. 111

A.7.7 Reciprocity

  • Layer reciprocity medians range from 0.2 to 0.3, but cross-layer aggregate reciprocity exceeds 50% in most communities and has a median of 0.46.Reciprocity is generally higher for exchanges of household items and lowest for monetary loans.Source: §A.7.7 · pp. 112–113, Figure A29 · p. 113

A.7.8 Transitivity

  • Median clustering is lowest for borrow-money networks (global 0.18; local 0.28) and highest for borrow-goods networks (global 0.26; local 0.36).Aggregate median global clustering is 0.34, compared with 0.26 in the Karnataka household-network benchmark.Source: §A.7.8 · p. 114, Figure A30 · p. 115

A.7.9 Consequences of Sharing Unit Sampling

  • Weighted degree removes the respondent-count association seen in unweighted degree, while respondent-count patterns in local clustering occur within sharing units but are not consistent across communities.Weighted and unweighted degree, path length, and local clustering track closely, so the analysis largely retains simpler binary-edge measures while noting sampling-coverage caveats.Source: §A.7.9 · p. 116, Figure A32 · p. 118, Figure A33 · p. 119

A.7.10 Variance Decomposition

  • The variance-decomposition models use network-layer fixed effects and community random effects to explain variation in network summary statistics.The passage introduces the modeling approach but provides no model results in this section part.Source: §A.7.10 · p. 123
  • Between-community variation is substantial across the modeled network statistics and is attenuated, but persists, after controls.The reported variance decomposition uses layer fixed effects and community random effects; adjusted models also control for community size, mean respondents per prompt, and nominations directed outside the community.Source: §A.7.10 · p. 123, Table A27 · p. 124, Figure A37 · p. 125

A.7.11 Network-wise summary statistics

  • Community-level summary statistics are tabulated for each network layer, the aggregate network, and the close kin network; tables are omitted for two external-tie networks with few nominations.The tables report measures including degree, centralization, assortativity, path length, isolates, components, clustering, and reciprocity.Source: §A.7.11 · p. 128, Table A29 · p. 129, Table A37 · p. 145

A.8.1 Social Support Network Layers

A.8.2 Aggregate Network

A.9 Kinship

  • Kin availability varies substantially: MY sharing units average eight close-kin neighbours, while 70% of FF households have none and the rest have one.Indirect close-kin links suggest several community patterns, from nearly one extended kin group to mixed clusters or limited local kin; this measure may underestimate extended kin ties.Source: §A.9 · p. 185, Figure A98 · p. 187
  • Close kin are important support alters in every network layer, and women-specific networks show higher reliance on kin than men-specific networks.The reported pattern is consistent with observations in other settings.Source: §A.9 · p. 186, Table A39 · p. 188
  • Close kin are named as support alters more often than their availability alone predicts, and close-kin dyads are more likely to have support ties than non-kin dyads.These comparisons are shown for the aggregate network across communities.Source: §A.9 · p. 186, Figure A99 · p. 188, Table A40 · p. 189
  • More co-resident kin is broadly associated with greater reliance on kin, with the association tempering at higher counts, but not with more overall support alters.The pattern suggests kin reliance may saturate, while other routes to supportive relationships remain available.Source: §A.9 · p. 190, Figure A100 · p. 191, Figure A101 · p. 192
  • Exogamous communities tend to have a lower share of close kin among support alters than highly endogamous communities, though the differences are slight and suggestive.The comparison groups communities by ethnographers’ reports of local endogamy patterns.Source: §A.9 · p. 190, Figure A102 · p. 193

Related papers

Loading 2610.28793v1…