While it is commonly thought that patrilocality is associated with worse outcomes for women 27 and their children due to lower social support, few studies have examined whether the structure 28 of female social networks covaries with post-marital residence. Here we analyze scan sample 29 data collected among Tsimane forager-farmers. We compare the social groups and activity 30 partners of 181 women residing in the same community as their parents, their husband’s parents, 31 both or neither. Relative to women living closer to their in-laws, women living closer to their 32 parents are less likely to be alone or solely in the company of their nuclear family (OR: 0.6, 33 95%CI: 0.3-0.9), and more likely to be observed with others when engaging in food processing 34 and manufacturing of market or household goods, but not other activities. Women are slightly 35 more likely to receive childcare support from outside the nuclear family when they live closer to 36 their parents (OR=1.8, 95%CI 0.8 - 3.9). Their social group size and their children’s probability 37 of receiving allocate decrease significantly with distance from their parents, but not their in-laws. 38 Our findings highlight the importance of women’s proximity to kin, but also indicate that 39 patrilocality per se is not costly to Tsimane women.

Data collection took place between March 2002 and November 2007 in 9 separate Tsimane communities. In each community, households were sorted into clusters of multiple physically close houses from within which researchers could easily monitor the activity of all inhabitants. Clusters were then selected for data collection at random without replacement until all clusters were sampled. Data collection involved monitoring each member of the cluster households for 2-to-3-hour blocks between 7AM and 7PM, with point scans every half hour. During point scans, the location, activity, and objects of interaction of each individual was recorded. Individuals were coded as being in the same social group if they were either a) engaged in active conversation or b) within 3 meters of each other, and in the same activity group if they were engaged in the same activity. When household members from the sampled cluster were absent, their whereabouts, activity, and (where possible) companions were ascertained by asking their family members. 

For this analysis, we selected scans of mothers of children under the age of 14, excluding visitors to the communities. This resulted in a total sample of 11940 observations of 181 Tsimane women, ranging in age between 15 and 59 with an average age of 32 (see table 1). For each woman’s scan we examined the list of individuals aged 14 or over who were in a) her social group or b) her activity group during the scan, excluding her husband and children. From this list we then calculated the total number of individuals engaged in: (1) Any activity; (2) Hunting, fishing, or gathering food; (3) Manufacturing cloth, bags, or jatata thatch; (4) Garden labor or wage labor, and (5) Processing or preparing food. Next, we selected observations of the children under the age of seven, which corresponds to the age range when Tsimane children require most supervision. This amounted to a sample of 21,938 observations of 351 children (52% male). Children were coded as receiving extra-familial childcare either if they were recorded receiving direct care (e.g., holding, playing, feeding, teaching, etc.) or if they were engaged in some social interaction with an individual 11 years or older, at which age we determined any social interaction with a child under seven could be reasonably construed as childcare based on ethnographic insights and existing literature in traditional societies [48,49]. Siblings were excluded as providers of childcare in this analysis since their presence is not tied to the post-marital residence choices of their parents. Because Tsimane adults often supervise children passively rather than actively caring for them, we also tested whether residence patterns affected children’s probability of being unsupervised, which we coded as being in a social group with no adults.

The post-marital residence choices of the women in our sample were coded in two ways. First, we categorized the women as being either patrilocal, matrilocal, bilocal, or neolocal based on the known residences of their parents and parents-in-law, following Gruitjers and Ermisch [50] . Couples for whom no information existed for either set of parents were assigned according to the presence of siblings in their home community. Accordingly, women coded as neolocal lived in communities where none of their or their husbands’ nuclear family lived. Bilocal families had at least 1 parent of each of the husband and the wife living in the same community. As a robustness check we also analyzed a sub-sample of families for whom GPS data existed for at least one parent of both the husband and the wife. Starting in 2007, the THLHP and its collaborators have collected GPS data for every household, which we used to reconstruct a subsample of the households where data was collected. When the precise GPS location was unavailable, but the community was known, which generally occurred when the parent or in-law lived in a non-sample community, we took their location to be the central point of their community, which given the distances between communities is a fairly accurate estimate on the log scale. This sub-sample included 83 women and 180 children, which corresponds to ~50% of the total sample. Using these data, we were able to model women’s social group size and children’s probability of receiving allocare as a function of the (ln-transformed) distance from the woman’s parents (the child’s maternal grandparents) and her in-laws (the paternal grandparents). 

All analyses were conducted in R version 4.1.2. We fit generalized linear multilevel models (GLMMs) using the glmmTMB package, which allows for mixed-effect hurdle and zero-inflation models. To account for the possible overdispersion of the count data, specifically the observed size of women’s social and activity groups, we compared multilevel Poisson, negative binomial, and zero-inflated Poisson models, all adjusting for mothers’ age, age squared, and the time of day of the observation block (morning or afternoon), with random intercept terms to control for repeated observations of individuals as well as the communities. Mother’s age was selected because of its possible causal influence over both residence and social group size. Including age squared significantly improved model fit according to likelihood ratio tests (Chi-square=9.06, p= 0.011). Time of day also had a significant effect on group size in many models, and due to sampling randomness may have varied across residence patterns, so was included in the model as a control.

Likelihood ratio tests confirmed that the zero-inflated models were much better fit to the data than Poisson and negative binomial models (Electronic Supplementary Materials [ESM], table S1). Accordingly, each model fit two sets of parameters, one for the zero-inflation component and one for the count component. For the analyses of children’s probability of receiving non-sibling childcare, we fit multilevel Bernoulli logit models controlling for child’s age, with random intercepts terms for the child’s ID, their mother’s ID, and the community.

R and R Studio.