Female reproductive competition within families in rural Gambia

Abstract

Many studies show that the extended human family can be helpful in raising offspring, with maternal grandmothers, in particular, improving offspring survival. However, less attention has been given to competition between female kin and co-residents. It has been argued that reproductive conflict between generations explains the evolution of menopause in cooperatively breeding species where females disperse, and that older females are related to the offspring of younger females through their sons, whereas younger, incoming females are unrelated to older females. This means the pattern of help will be asymmetric, so older females lose in reproductive conflict and become ‘sterile helpers'. Here, we seek evidence for female reproductive competition using longitudinal demographic data from a rural Gambian population, and examine when women are helping or harming each other's reproductive success. We find that older women benefit and younger women suffer costs of reproductive competition with women in their compound. But the opposite is found for mothers and daughters; if mother and daughter's reproductive spans overlap, the older woman reduces her reproduction if the younger woman (daughter) reproduces, whereas daughters' fertility is unaffected by their mothers' reproduction. Married daughters are not generally co-resident with their mothers, so we find not only competition effects with co-resident females, but also with daughters who have dispersed. Dispersal varies across human societies, but our results suggest reproductive conflict could be influencing reproductive scheduling whatever the dispersal pattern. A cultural norm of late male marriage reduces paternal grandmother/daughter-in-law reproductive overlap almost to zero in this population. We argue that cultural norms surrounding residence and marriage are themselves cultural adaptations to reduce reproductive conflict between generations in human families.

1. Introduction

Studies of cooperative and communal breeding in animals focus strongly on reproductive conflict between individuals [1], whereas human studies tend to focus more on the cooperative benefits of communal life, and its relevance for the evolution of menopause [2]. Human life history is characterized by a long childhood, followed by a rapid reproductive phase, and then a long post-reproductive life, at least for females. Human menopause might be selected for by kin selection favouring older mothers investing in their grandchildren rather than continuing to reproduce themselves [2]. There is now considerable evidence that grandmothers enhance the reproductive success of their offspring (reviewed in Sear & Mace [3]), including evidence from our own study in rural Gambia [4–6]. Evidence that maternal grandmothers benefit the survival of their daughter's offspring has been found across a wide range of societies, although kinship norms may influence which residence patterns are most favourable [7]. However, there is disagreement about whether the grandmother effect, on its own, could select for menopause. Verbal models have emphasized other factors, such as the importance of mothers survival during childhood [8], or the importance of an extended lifespan when fertile span is phylogenetically constrained [9]. Some mathematical models informed by data either fail to predict any fitness benefit associated with terminating reproduction so long before death [10] or predict small fitness benefits [11].

Implicit in all the various models of the ‘grandmother hypothesis’ is the notion that mothers and daughters are in reproductive competition, as it is assumed that only by becoming non-reproductive can a grandmother really help her daughter's reproduction. It is striking in humans how little human female generations overlap; as a daughter reaches reproductive age, her mother reaches menopause, and as she reaches menopause, her mother dies. However, while reproductive conflict predicts that reproductive generations should reduce overlap [12], it not does address why it is the older woman who is foregoing reproduction rather than the younger one, as is generally the case in most cooperatively breeding birds or mammals [1,13]. Johnstone & Cant [14] argue that relatedness to group members, and hence the power of kin selection, depends on mating and dispersal patterns. They argue that female–female competition is particularly intense in species in which females disperse, as it means that older females are not closely related to younger breeding females in their group. Female dispersal is unusual among mammals, but is thought to be the most common condition in chimpanzees (Pan troglodytes) and in ancestral humans, although there is variation in contemporary hunter–gatherer populations [15]. Johnstone & Cant [14] have modelled how relatedness to the group changes with age depending on dispersal patterns. Under patrilocal residence (i.e. when females disperse at breeding age and males do not), adult female relatedness to the group will be low on arrival in a new group, but will gradually increase with age as her own offspring (particularly sons, who do not disperse) are born and grow up to reproduce themselves. When older women find themselves in competition with their sons' spouse for reproductive resources, there is an essential asymmetry in which the older woman is related to her son's offspring (and will thus suffer a fitness cost in harming her son's wife's reproduction) but the son's wife is not related to the older woman's offspring, so natural selection does not favour helping her husband's mother reproduce. Johnstone & Cant's [14] models show that the younger woman is more likely to win the competition, and the older woman is destined to become the sterile helper. They argue this explains late life low fertility in whales and primates in which either females disperse or males and females mate outside the group; and humans usually fall into the former category.

Here, we re-examine historical demographic data from rural Gambia to seek evidence for some of the assumptions and predictions of the model that within group female–female competition may be occurring in a patrilocal human group; in particular, we look for the differing effects of age on competition as this is assumed to be important in some models for the evolution of menopause in humans. We examine how a female's relatedness to those with whom she is co-resident changes over her lifespan. We then seek evidence that females may be in competition with each other while co-residing in the same compound. Finally, we examine, in particular, whether reproductive competition exists between mothers and grandmothers.

Our earlier studies suggest that the benefits of grandmothering may occur only when the grandmother is non-reproductive [5], although this trend falls short of statistical significance. However, the reverse effect—that becoming a grandmother has any effect on the older woman's fertility—has not previously been investigated.

2. Data and methods

(a). Data source and studied population

Our data come from a longitudinal dataset collected by the MRC in rural Gambia, initiated by the late Sir Ian McGregor in 1949. He studied four villages, collecting demographic and some anthropometric and health data on all inhabitants for nearly 25 years [16]. This is a population of Mandinka farmers, who farm mainly maize, groundnuts and some rice, with a few livestock. They were originally chosen for study by the MRC owing to the high disease burdens they experienced. Until 1975, this population lived in a largely ‘natural fertility’ and ‘natural mortality’ environment, albeit with some medical care available on occasions when McGregor and others were residing in the village. Deaths owing to malaria and other infections were commonplace, with only 45 per cent of children surviving to their fifth birthday. After 1975, a permanent medical centre opened in one of the villages and mortality declined markedly [17]. Modern contraception was also introduced, and has gradually been adopted [18,19]. Our analysis here focuses primarily on the period 1950–1975, when mortality was still high, and it covers the two villages in which records were most complete. Compounds were areas where people lived together in a cluster of households. Compounds often comprise a patrilocal group (fathers, brothers and sons, and their spouses and children), although more distant relatives or unrelated families may reside in the same compound (ranging between 1 and 49 adults in this sample). The society is polygynous and co-wives usually inhabit the same compound. Divorce and remarriage do occur, in which case women move compounds (but often leave weaned children in their natal compound with their patrilineal kin). Because these compounds are matters of historical record, we do not know exactly how individuals shared work on fields or how they shared food within these compounds or who owned what land.

The nature of the data allows us to track how closely related individuals are to those living around them, at both the scale of the compound, and of the village. Household residential units were usually a hut occupied by one married woman, her husband and children, so we have not examined relatedness within the household. McGregor recorded the compound in which all children were born (marriages are not recorded, but genealogy is, so we take the birth of a child in a compound as an indication that its mother is residing in that compound, which is generally that of the father of her child). We assume that women move from their natal compound when they first give birth. Women often remained in their husbands households for life, but occasionally started to reproduce with another man in another compound even if their first husband was alive, at which point we assumed divorce, remarriage and relocation to the new husband's compound had occurred. Women do not remain single for long in this society, and on widowhood remarry quickly, often marrying their late husband's brother through the levirate system.

(b). Assessment of age-varying relatedness to others

Relatedness of each village member to each other village member was calculated, using Descent (Hagen 2002–2005; http://itb.biologie.hu-berlin.de/~hagen/Descent/), for each year of data that each person was observed in one of the four villages. Data are censored when women are last seen or die. Relatedness is genealogical, measured back to however many generations are available. Relatedness connections were occasionally available to common ancestors four generations back but, in most cases, links up to two generations were available. For each woman, we derived the number of reproductive (aged 15–49) unrelated women living in the same compound for each year, from the age of 15 to 49 (or age of death or last seen if it occurred before the age of 49).

(c). Event-history analyses

We investigated the consequences of reproductive competition (i) with unrelated females of reproductive age who are co-resident in the same compound, (ii) with mothers of reproductive age, and (iii) with daughters of reproductive age; mothers and daughters are generally not co-resident. In all cases, we tested whether reproductive competition (as approximated by either the number of unrelated women of reproductive age in the compound or the presence/absence of potential reproductive overlap between mother and daughter) is a predictor of (i) the risk of a birth and (ii) the risk of child death. We focused on the risk of mortality to age 5, as most deaths occur prior to that age in preindustrial societies [20].

We used event-history analyses [21,22] to predict the instantaneous risk of an event occurring (i.e. the probability of occurrence at a given time given that it has not already occurred) at different time intervals or episodes that describe the duration of exposure to the risk. We performed both (i) single-event-history analyses to predict the risk of child death during the first 5 years of life and (ii) multiple-event-history analyses to predict the risk of birth. To describe the sample distribution of event occurrence, data were converted to a person-period dataset (or life table) where, for each individual, data were expanded to describe event occurrence (0 or 1) on discrete time intervals or episodes within individual, from the beginning of time where no one has experienced the event to the time of censoring (i.e. until the event occurs or that person leaves the dataset for some other reason, such as reaching the maximum age of interest). In the single-event-history analysis predicting risk of child death, an episode is the number of years since births, up to the age of 5 (which is the time of censoring if a death has not already occurred). In the multiple-event-history analysis of interbirth interval, the episode is the number of years since the last birth. Women are followed from the age of 15 to 49, although intervals longer than 10 years were not considered. We specified a discrete time function and included a factor term describing time intervals (fixed effect and multiple intercepts) which allow us to estimate the mean risk of event occurrence at each time interval, thereby accounting for the longitudinal design of the data. Finally, we used mixed-level logistic regression to account for the multi-level structure of the data (two levels: mother and child).

In all analyses, we were able to include time-varying variables for age and potential reproductive conflict (e.g. number of unrelated women in the same compound; presence/absence of a reproductive overlap with either a daughter or a mother). The inclusion of time-varying predictors allows us to investigate within-individual variation across time (e.g. age) in the risk of an event (e.g. reproduction and death). All analyses also controlled for between-individual variation in the time of birth (e.g. cohort effect), which enables us to control for the fact that we know less about the ancestors of those who were old people in 1950s than we do of younger generations (the reference category is always the youngest generation). The analysis thus allows us to disentangle the influence of age (time varying) from that of cohort (constant over time).

(d). Multi-model inference

To make statistical inferences about the population given the data, we used an a priori model formulation approach [23]. For each question, we considered a set of a priori candidate models (see §2e), for which a measure of each model's fit scaling to its complexity is derived using the Akaike information criterion (AIC; [23–25]). The model for which AIC is minimized is selected as the best approximation for the empirical data at hand. The evidence for each alternative model is carried out by rescaling AIC values relative to the model with the minimum AIC, which subsequently allows models to be ranked according to their ability to account for the data. In addition, a measure of weight of the evidence that a given model is the best approximation in the set of models considered is calculated (Akaike weight, ω) [23]. Subsequently, rather than base inferences on a single best approximating model, inferences are calculated by the entire set using model-averaged-based estimators taking into account model uncertainty (estimates are balanced using Akaike weights and averaged across models [23]). The use of model-based average estimators allows better precision and reduced bias compared with the estimator of that parameter only for the best approximating model [23]. Following Anderson et al. [25], we present models that collectively account for only 95 per cent of the available model weight. All analyses were carried out using R v. 2.12.2 software [26] and the R packages lme4 [27] and AICCMODAVG [28].

(e). Candidate models sets

(i). Competition between unrelated women

To investigate reproductive competition with unrelated women in the compound, and whether such effect changes with age, we considered a set of six models in both fertility and child mortality analyses. We first considered (i) a null model,  including only the variable describing the duration of exposure to the risk Episode; (ii) a control model C: Episode + Confounding variables; (iii) a model considering competition: C + NURWC, the number of unrelated women in the compound at the time of the episode; (iv) a model considering a nonlinear effect for competition: C + NURWCsquare; (v) a model considering that the effect of competition varying depending on the episode considered: C + Episode × NURWC. Finally, we included a model considering that competition with unrelated women changes with age; and (vi) C + Age × NURWC.

(ii). Competition between mothers and daughters

To investigate reproductive competition within mother–daughter dyads, for both the older women (mothers) and younger women (daughters), we considered a set of five models in both fertility and child mortality analyses: (i) a null model, including only the variable describing the duration of exposure to the risk Episode; (ii) a control model C: Episode + Confounding variables; (iii) a model including control variables and reproductive overlap: C + Overlap; (iv) a model including control variables and the interaction between age and reproductive overlap: C + Overlap × Mother's Age at birth, to investigate in any age-specific reproductive competition; and (v) a model, including control variables and the interaction between time and reproductive overlap: C + Overlap × Episode, to investigate any time-specific effect.

(iii). Confounding variables

In fertility analyses, control variables were age, age squared, survival status of the previous child (binary), order of events (i.e. birth order) and cohort (electronic supplementary material, table S1). The inclusion of the information on whether or not the previous birth has survived allow us to control for replacement effects, i.e. that fertility is caused by mortality. In child mortality analyses, control variables were child sex, mother's age at birth, birth order and child cohort (electronic supplementary material, table S2).

3. Results

(a). Changes in relatedness to the residential group with age

Figure 1a shows the general patterns of how a breeding female's average relatedness to other adult females and other adult males in her compound and in her village changes with age. We define adults as those aged 15–49 (15 is the youngest age of reproduction observed). Data cover 318 men over a mean period of 11 years (±s.d. = 6.2) and 541 women over a mean period of 13.2 years (±s.d. = 7.5).

Figure 1.

(a) Relatedness of adult females (solid line) to other adult males (dashed line) and females co-resident in the compound and (b) in the village. Female relatedness to other adults is very low, but increases as sons and daughters mature; relatedness to co-resident adult males is higher, and increases much more with age than relatedness to adult females. (c) Relatedness of adult males to other adult males and females co-resident in the compound or (d) the village. Males are much more closely related to each other than to females co-resident in their compounds but relatedness does not change greatly with age. Relatedness is calculated using relationships over as many generations as possible, usually three generations, maximum four.

First, we show that average genealogical relatedness to other adult members of the village is very low, and there is little difference in the average relatedness to other adult males or adult females in the village (figure 1b,d). This suggests that the village is not a useful unit of analysis for our purposes: as villages are large (several hundred people) and marriage usually but not always occurs within the village [18], so patrilocal residence generates few sex differences in relatedness at village level. Relatedness within compounds provides a better description of relatedness to the individuals a person will see and interact with every day.

Figure 1a,c shows that adult males spend their lives more closely surrounded by relatives than do adult females. An adult female's relatedness to co-residents in her marital compound starts low but does increase with age, especially her relatedness to males because they do not disperse, fulfilling the predictions from assuming low male high female dispersal in Johnstone and Cant's model. This may be relevant not only to potential reproductive conflict, but also to women's status and security. At no age are women as closely related to males or females in their marital compounds as males are, which may make it harder for them to exert political influence or experience much autonomy, especially when young.

(b). Competition between unrelated women

We examine whether having co-resident unrelated reproductive-age females (15–49) in the compound influences female reproductive success, as measured by both risk of a birth (n women = 277, 1630 birth intervals) and risk of child death (n women = 520, n children = 1300). We found that co-residency with unrelated women has a variable effect on risk of reproduction depending on age. Female co-residency is associated with a reproductive cost at young ages but with reproductive benefits when older (electronic supplementary material, table S1; Age × NURWC: β = 0.003; 95% CI [0.001; 0.01]; figure 2). Model selection reveals that the age-specific effect of female co-residency is relatively robust, i.e. the model considering such an effect (w = 0.89; table 1) is found to be the best in the set considered. Using estimates from the best model and adjusting for control variables, we found that a woman aged 20 is at a higher risk of reproduction if she co-resides with only one unrelated woman (i.e. 0.56) than with 28 unrelated women (maximum observed in the population) living in the same compound (i.e. 0.43). By the age of 40, however, a woman will be at a higher risk of reproduction if she lives with 28 unrelated women (i.e. 0.38) than if she lives with only one unrelated woman (i.e. 0.18; figure 2).

Figure 2.

Female co-residency and risk of reproduction. High female co-residency (dashed line) indicates predicted risk of reproduction if 28 other women live in the same compound (maximum value observed). Low female co-residency (solid line) indicates predicted risk of reproduction if only one other woman lives in the same compound. Effect of female co-residency on risk of reproduction varies with age (age × Nb of unrelated women in the compound: β = 0.003; 95% CI [0.001; 0.01]). Co-resident females are costly for young women and beneficial for older women. Predicted values taken from best model (Akaike weight = 0.89) controlling for cohort (born after 1940), order of birth (second birth) and status of the previous birth (alive).

Table 1.

Summary of the best 95% a priori models for the data, including the total number of estimable parameters (K), the log-likelihood (LogLik), AIC differences relative to the minimum value in the set (dAIC), and the Akaike weight (ω_i). See §2 for details on each analysis and associate set of candidate models. In (1), NURWC refers to the number of unrelated women in the compound (time-varying). In (2), overlap refers to whether or not a woman's mother is still reproducing; in (3), overlap refers to whether or not a daughter has started reproducing. Mabyrs refers to mother's age at birth. All parameter estimates for the best fitting models for risk of reproduction (table S1) and risk of child death (table S2) are shown in the electronic supplementary material.

models	K	LogLik	dAIC	ωi
(1) competition with unrelated women
(a) risk of reproduction
control + age × NURWC	20	−2374.98	0	0.89
control	20	−2380.07	6.17	0.04
control + NURWC + NURWCsquare	21	−2378.48	6.48	0.03
(b) risk of child death
null	7	−648.30	0	0.43
control	13	−642.85	1.10	0.25
control + NURWC + NURWCsquare	15	−641.22	1.82	0.17
control + NURWC	14	−642.81	3.01	0.09
control + NURWC × mother's age	15	−642.35	4.10	0.06
(2) competition with mothers
(a) risk of reproduction
control	18	−3054.73	0	0.55
control + age + overlap	19	−3054.45	1.44	0.27
control + age × overlap	20	−3053.82	2.19	0.18
(b) risk of child death
control	13	−1106.44	0.00	0.39
control + overlap	14	−1105.59	0.29	0.33
control + overlap × Mabyrs	15	−1104.82	0.75	0.27
(3) competition with daughters
(a) risk of reproduction
control + overlap × age	20	−2928.76	0	0.99
(b) risk of child death
control + overlap × child's age	19	−2306.58	0	0.60
control + overlap	14	−2312.80	2.45	0.18
control	13	−2314.00	2.84	0.14

Regarding risk of child death, model selection reveals that the null model is the best model, and  relatively high uncertainty in model selection (Akaike weight of the best model is 0.43; table 1). Averaged estimates indicate that unrelated females in the compound do not influence risk of child death (β = −0.07; 95% CI [−0.44; 0.11]). In addition, most of child deaths occur during the first year of life, and are largely independent from mother's characteristics (electronic supplementary material, table S2).

That younger women lose out in competition with older, unrelated women, goes in the opposite direction to that predicted by Johnstone & Cant [14] for patrilocal populations. However, their models highlight reproductive conflict between paternal grandmothers and their son's mates (to whom they are unrelated, but to whose children they are related), and in this population the number of occasions that mothers and paternal grandmothers were both of breeding age was minimal (figure 3), as males marry late in this society [29]. Most co-resident unrelated breeding-age women were husband's brothers' wives, co-wives or unrelated individuals, and did not generally include the husband's mother. Therefore, the finding that older women appear to win in reproductive competition with younger unrelated women, but not with younger related women (i.e. daughters) means our results do not necessarily contradict their model's predictions.

Figure 3.

Distribution of the age at the birth of a child for mothers (dashed dotted line), maternal grandmothers (long dashed line) and paternal grandmothers (short dashed line). Women are more likely to face a reproductive overlap with mothers than with mothers-in-law. Mean age at which women give birth = 24.82 s.d. 6.88 [13; 46]; mean age at which women's daughters have a child (i.e. are maternal grandmothers) = 51.06 s.d. 8.91 [32; 77]; mean age at which women's sons have a child (i.e. are paternal grandmothers) = 60.84 s.d. 9.36 [39; 91].

(c). Competition between mothers and grandmothers

Finally, we look at competition between mothers and grandmothers from both mothers' and grandmothers' point of view. We know from our earlier studies that maternal grandmothers help their daughters' children survive in this [4] and other populations (reviewed by Sear & Mace [3]); paternal grandmothers are not helpful in this population, in terms of child survival although do promote the fertility of their daughters-in-law [6], perhaps because paternal grandmothers are less concerned about their daughters-in-law incurring costs of reproduction [30]. While there is little inter-generational overlap in reproduction, some women do become maternal grandmothers at an age when they are still capable of reproduction and here we test for the first time whether becoming a grandmother reduces reproductive rate in women under 50.

In modelling reproductive competition with either a mother or a daughter, we chose 30 years as a cut-off (for upper and lower limit, respectively) because the maximum age of reproductive overlap with a mother observed was 29 years and the minimum age of reproductive overlap with a daughter is 32 years.

(i). Competition with mothers (women aged 15–29)

First, we examine whether a daughter suffers competition if her mother is still reproducing after the daughter starts to reproduce. Regarding daughters risk of a reproductive event (n mothers = 664, 2209 birth intervals), we find little evidence of competition with mothers; model selection reveals relatively high uncertainty on which is the best approximating model (Akaike weight of the best model is 0.55). That reproductive competition between mothers and daughters has little impact on daughters' risk of reproduction is further suggested by the averaged estimate (β = −0.36; 95% CI [−2.06; 1.34]; electronic supplementary material, table S1). We also did not find strong evidence that reproductive overlap has an impact on the risk of child death (n mothers = 520, n children = 1300, β = 0.75; 95% CI [−1.07; 2.56]; electronic supplementary material, table S2) and model selection is again rather uncertain (table 1, weight of the best model = 0.39). Thus, there is little evidence that daughters' fertility or child mortality is influenced by their mothers' reproductive behaviour.

(ii). Competition with daughters (women aged 30–49)

We found that becoming a maternal grandmother (i.e. your daughter becoming a mother) does indeed slow down reproductive rate (both increasing the length of the birth interval and/or decreasing the probability of another birth) and tends to increase risk of infant death, although evidence for this latter effect is weak.

Regarding risk of reproduction (n mothers = 525, 1809 birth intervals), model selection reveals that the best approximating model in the set includes an interaction between age and becoming a grandmother (Akaike weight = 0.99; table 1). Women becoming grandmothers are more likely to reproduce than others (β = 3.18; 95% CI [0.78; 5.59]; electronic supplementary material, table S1). This suggests that women who become young grandmothers are extremely fertile women. However, we found evidence that becoming a grandmother at about 35 or older decreases their probability of conception, especially at later ages (interaction between age and grandmothering, β = −0.09; 95% CI [−0.15; −0.03]; figure 4). Regarding risk of child death (n mothers = 891, n children = 2760), model selection reveals that the best approximating model is the one including control variables and an interaction between child's age and reproductive overlap. However, there is uncertainty in model selection (weight = 0.60). While averaged estimates indicate that risk of child death during the first year of life tends to be stronger for women who are grandmothers, there is low confidence for this effect (β = 0.49; 95% CI [−0.73, 1.70]; electronic supplementary material, table S2). Thus, there is a negative effect on the fertility of women on becoming grandmothers between ca 35 and 49, and a small effect on infant mortality if they do continue to reproduce.

Figure 4.

Reproductive overlap with daughters and risk of reproduction. Becoming a grandmother correlates with a faster reduction in reproductive rate with age: at 30 years of age, women becoming grandmothers have a risk of reproduction 1.4 times higher than non-grandmothers. At 45 years of age, grandmothers are 57% less likely to reproduce as compared with non-grandmothers. Predicted values taken from best model (Akaike weight = 0.99) and controlled for cohort (born after 1944), order of birth (second birth) and status of previous birth (alive). Dashed line, reproductive overlap; solid line, reproductive separation.

(iii). Maternal versus paternal grandmothers

We predict that competition would be more intense when becoming a paternal grandmother owing to co-residence with an unrelated female, whereas competition with daughters would be less, owing to higher shared genetic interests and separate residence. The most likely cost of reproduction while daughters are reproducing elsewhere would be the cost of not being able to devote so much energy to helping her daughter reproduce. It was not possible to determine whether the effect of becoming a paternal grandmother was greater than for becoming a maternal grandmother in this population as nearly all cases of grandmotherhood prior to age 50 were cases of becoming a maternal grandmother, and there were not enough cases of women under 50 becoming paternal grandmothers to evaluate the effect (n = 8 only). The mean age of fatherhood (mean ± s.e.m. = 38.96 ± 0.18) was significantly later than mean age of motherhood (mean ± s.e.m. = 25.29 ± 0.14; paired Wilcoxon signed-rank test, V = 288; n = 2465; p < 0.001). This suggests that natural and cultural selection may have already occurred that effectively prevents overlap between older women and their son's wives reproductive activity (figure 3).

4. Discussion

The changes in relatedness to the co-resident group with age that we show here accord with those predicted by Johnstone and Cant for groups with low male, high female dispersal. We also show that the effects of competition depend on age—young women suffer costs from having more unrelated reproductive-age women in their compound, whereas older reproductive-age women appear to benefit, at least in terms of fertility. Women in different households within these compounds do not usually eat communally, and therefore the compound, while indicating physical proximity centred around patrilineal kin, does not necessarily imply intense competition for resources. Association of households into compounds may provide benefits in terms of childcare, cooperative labour in farming or domestic tasks, socializing or even security, all of which might mitigate any costs of reproductive conflict. We do not have data on land ownership, but if larger compounds were associated with larger total field holdings, this would also mitigate any effects of competition. Nonetheless, it does appear that older reproductive-age women are better able to compete for resources than are younger women; this could be due to their higher status enabling them to coopt the help of younger women, or any other reason why they might be more successful at securing any shared resource.

It is in the case of mother–daughter dyads that we see the strongest reproductive competition, but again the competition is asymmetric. It is only the older women, over 35, who pay a cost of intergenerational competition in terms of their fertility. The fact that the asymmetry with age is in the opposite direction of that found between unrelated women suggests that mothers may be willing to pay a cost to help daughters reproduce, whereas they are more likely to harm rather than help unrelated young women who reproduce in their compounds. But it is notable that costs are incurred by grandmothers even though their daughters reside elsewhere. In this population, daughters do not disperse far, often marrying within the village; so grandmothers may visit and help with chores, and daughters may take children back to visit their mother for food and care, and first births may be in the grandmother's house. We have observed this kind of helping even in populations, such as the Ethiopian Oromo, where females disperse much further [31]. There may also be cultural norms that reduce the expectation and/or obligation to reproduce after becoming a grandmother. While this is a proximate explanation, it could have a functional basis by enabling mothers to help their daughters, and preserving their health to enable them to perform these roles throughout their old age (we have already shown that having a surviving post-menopausal grandmother is beneficial [4]).

The effect sizes of reproductive conflict seen here do not necessarily tell us whether they are enough for menopause to evolve, partly because menopause has already evolved and, therefore, we can observe only overlapping reproduction in those women with unusually high fertility. Given that high fertility is associated with a range of positive phenotypic characteristics and is thought to be heritable in humans [32,33], it is expected that these mothers and daughters would have above average reproductive success. We do find support for the hypothesis that age and relatedness influence how women fare when in reproductive competition. Our finding that older women dominate when unrelated women compete, especially in larger groups, is in line with patterns found in studies of cooperative breeding in other species [34]. The finding that mothers give up potential fertility for their daughter's reproduction, but daughters do not give up reproductive benefits for their mother, is not described in other species and has not previously been shown in humans. These findings are consistent with the wider hypothesis that co-residence and dispersal patterns are important in determining patterns of help and harm among women in reproductive competition.

(a). Kinship and co-residence norms as cultural adaptations to reduce reproductive conflict

In this population, we are unable to evaluate how costly competition would be if mothers and daughters co-resided, or if mothers and son's wives had reproductive overlap. However, we argue that these scenarios rarely arise precisely because co-residence patterns and marriage norms are themselves likely to be cultural adaptations to minimize reproductive conflict in human families.

Reproductive conflict in communally breeding species generally leads to attempts by dominant breeders to control less dominant ones. Human parents are often in a strong position to control offspring reproduction, especially if parental wealth is heritable and needed to facilitate marriage. In this community, men need access to farmland and brideprice to marry, for which they need the cooperation of their patriline, and, at least traditionally, children's first marriages were arranged by parents, especially fathers. After marriage, it is less likely that parents or other kin can exert much control on the rate of birth [19], but see our earlier finding that post-menopausal paternal grandmothers increase fertility in their sons’ wives, [6]. Therefore, reproductive conflict within two generations of the same family is partly avoided by cultural means dictating when marriage occurs, combined with the existence of menopause. Parental control over sons' marriage can reduce intergenerational reproductive conflict between mothers and sons' wives almost to zero.

Unlike sons, daughters are often married prior to the end of their mothers' reproductive lives. We have shown previously that early age at first birth is associated with a prevalence of male kin (especially brothers) in the natal household [35], so a need to gain brideprice to help male kin marry may be influencing the decision as to when daughters marry. Furthermore, there is reluctance to delay marrying off daughters as they become less marriageable with age. One would expect intergenerational competition with mothers and competition with siblings for household resources to be reduced by daughters moving out and reproducing elsewhere. But we have shown that virilocal residence norms do not remove all reproductive costs of becoming a grandmother. This result suggests that intergenerational conflict may be important in shaping patterns of reproductive scheduling whether or not it is females that disperse. Cultural norms of marriage and residence may themselves be co-evolving with reproductive scheduling, as adaptations to reduce reproductive conflict within families.