Inclusive fitness and differential productivity across the life course determine intergenerational transfers in a small-scale human society

Transfers of resources between generations are an essential element in current models of human life-history evolution accounting for prolonged development, extended lifespan and menopause. Integrating these models with Hamilton's theory of inclusive fitness, we predict that the interaction of biological kinship with the age-schedule of resource production should be a key driver of intergenerational transfers. In the empirical case of Tsimane’ forager–horticulturalists in Bolivian Amazonia, we provide a detailed characterization of net transfers of food according to age, sex, kinship and the net need of donors and recipients. We show that parents, grandparents and siblings provide significant net downward transfers of food across generations. We demonstrate that the extent of provisioning responds facultatively to variation in the productivity and demographic composition of families, as predicted by the theory. We hypothesize that the motivation to provide these critical transfers is a fundamental force that binds together human nuclear and extended families. The ubiquity of three-generational families in human societies may thus be a direct reflection of fundamental evolutionary constraints on an organism's life-history and social organization.

between harvesters and those contributing labor to preparing and maintaining the fields from which products were harvested. The relative weight of credit to each laborer was assigned in proportion to the hours of labor they contributed, multiplied by the expected age-and sex-specific efficiency of labor.
To reduce potential biases resulting from uneven sampling across seasons, production rates were calculated separately for each month, then averaged with equal weighting across months; individuals and families with insufficient sampling during the critical harvest months (February-May) were excluded from the analysis, as described in Section S2. Given the present focus on subsistence production, cash-cropped horticultural goods were not included in the tallies. (Futher methodological details are given in Ref. [3].) Daily consumption requirements (in cals/day) were estimated on the basis of age, sex, and weight according to FAO formulae [7]. Net production (in cals/day) was calculated by subtracting consumption from gross production, while net need (equal to negative net production) was calculated by subtracting gross production from consumption.

S1.4 Transfers
Caloric transfers were calculated from the redistribution of food products recorded in the production-and-sharing dataset. Donor credit for transfers was attributed to primary producers according to the methods described above. The share received by gift recipients was assigned according to the weight of gifts recorded in the interviews; once gift quantities were subtracted, the product's remaining calories were divided among meal recipients in proportion to estimated consumption requirements.
For each type of food t (game, fish, or horticulture), mean gross calories transferred from individual i to individual j per day were calculated in two steps. First, the fraction of i's production of t received by j was calculated by dividing the raw total of calories of t transferred from i to j by the raw total of calories of t transferred from i to all recipients (including i and j). This fraction was then multiplied by the measured daily production rate of t by i, to yield mean gross calories of t transferred per day. (More formally: the gross amount transferred from i to j for each food type t is G ijt = p it g ijt / k g ikt , where p ijt is i's daily production rate for that type, g ijt is the raw total of calories of type t transferred from i to j, and k g ikt is the raw total of calories of type t transferred from i to all recipients.) This method ensures that transfers reflect an individual's total productivity, as well as specific patterns of sharing for each food type. Calories transferred were summed across food types to yield gross total calories transferred per day.
Net transfers from i to j were calculated as gross transfers from i to j minus gross transfers from j to i. Net transfers from nuclear family i to nuclear family j were calculated by summing the net tranfers from each member of family i to each member of family j.

S1.5 Kinship
Consanguineous (e.g. parent, sibling) and affinal (e.g. sibling's spouse, spouse's parent) kinship categories and genetic relatedness (r) were calculated between individuals on the basis of ≥3 generations of genealogy derived from census and demographic interviews [2]. Mean relatedness between nuclear families i and j was calculated as the mean relatedness of each member of family i to each member of family j.

S2.1 Transfers between individuals
1,047 of 1,254 individuals in the full sample had sufficiently detailed information on the production of seasonal horticultural goods to be included in the individual-level analysis. Outliers with net transfers over 7 SDs from the mean were excluded from the analysis (which excluded one observation from the 'parents → child' model, and one observation from the 'grandparents → grandchild' model), an action which had no effect on the direction or significance of results.
In the individual-level analysis (Tables E1-E7 and Fig. 2), two models were estimated for each type of relationship: first, a sex-and age-stratified model, with 10-year age categories for focal donors (parents, grandparents, spouses, and parents-in-law) and 4-year age categories for focal recipients (children, grandchildren, and children-in-law); and second, an all-ages model stratified by sex.
Random effects for community identity were included in the individual-level models of net transfers in order to capture heterogeneity in transfers across the study communities. The standard deviation and significance of these random effects are reported in Table E8.
The statistical significance of estimated values was bootstrapped by comparing observed values against the values produced from an ensemble of randomly resampled 'null datasets'. To construct the ensemble of null datasets, the sum of net transfers between focal individuals and different categories of kin was recomputed after randomly reshuffling the net transfer values across all individual-individual dyads in each category (e.g. parents and children). The reported p-values represent the fraction of null datasets yielding estimated values ≥ observed values.

S2.2 Transfers between nuclear families
The family-level analyses (Tables 1 and 2) were motivated by the fact that, due to regular pooling of food within nuclear families, the amount that family members give to and receive from others is expected to be determined not so much by their own hunger as by the hunger of their families. Thus the effects of need should be plainly observable at the level of the nuclear family. The individual-and familylevel analyses in this way are complementary, providing insight into both distinct and overlapping aspects of the sharing system across two scales.
194 of 239 nuclear families in the full sample had sufficiently detailed information on the production of seasonal horticultural goods to be included in the family-level analysis. Of the 194 nuclear families included in the family-level analysis, 175 (90%) contained only biological children of the reproductive-age adults, or no children at all, while 19 included one or more "adopted" dependents. Of the 32 "adoptees" in these families, 19 were grandchildren, 2 were younger siblings, and 11 were more distant kin or non-kin. For this analysis, individuals that changed family membership during data collection due to marriage, divorce, or migration were assigned membership in the family where they resided for the majority of the sample period. Two types of family-level models were estimated: the first estimated the relationship between transfers and interactions between mean genetic relatedness (r) and the net need of each family (Table 1); the second examined differences in patterns of transfers according to need across four categories of relationship: parent-offspring family pairs, sibling-sibling family pairs, other genetically related family pairs, and unrelated family pairs (Table 2).
Two different but related variables of net caloric need (consumption minus production) were used in the family-level analysis. First, a family's estimated net need was calculated on the basis of its demographic composition and population mean age-and sex-specific rates of consumption and production. This can be interpreted as an instrumental variable representing the sum of the expected net need of all members of a nuclear family. Second, measured net need was calculated by summing the individual-specific, measured consumption minus production rates of each family member.
Estimated net need was utilized in addition to measured net need because of the possibility of correlated error between measured net need and transfers, since both are calculated using individual-specific production rates. Estimated net need thus allows an evaluation of the effects of need in the absence of this potential source of bias. The two measures may also capture different aspects of the relationship between need and transfers. Estimated need should reflect to a greater extent transfers occuring on the basis of the long-term expected economic and demographic state of families; while measured need should capture to a greater extent idiosyncratic differences in productivity across families, or vagaries of fortune within the period of sampling. Estimated need is also less affected by individual-level sample/measurement error, since it is based on expectations for age/sex classes.
Net transfer and net need terms were standardized to have mean = 0 and standard deviation = 1. Random effects were included for community identity and the identity of each family, i and j. The standard deviation and significance of these terms are reported in Table E9. The statistical significance of estimates was computed by randomly reshuffling net transfer values across family-family dyads, with p-values representing the fraction of reshuffled null datasets yielding estimates equal to or more extreme than the observed values.          Table E9: Standard deviation and significance of random-effect terms for community, donor famliy identity i, and recipient family identity j in the mixed-effect models presented in Tables 1 and 2.