1.1. Child mortality and social support

Although child mortality in the developed world is rare (6 deaths per 1000 live births), the developing world exhibits a 29-fold higher rate (175 per 1000 for sub-Saharan Africa; Black, Morris, & Bryce, 2003). This mortality level reflects the coincidence of an evolved set of life history traits, specifically the rapid production of altricial young requiring high levels of parental investment (Kaplan & Lancaster, 2003), and the poor socioeconomic conditions typical of many regions within the developing world, including food insecurity, high pathogen exposure, low education, and negative effects of global markets (Armelagos et al., 2005, Cesar et al., 2003). While within-population variation in child welfare and survival are clearly a function of household income, parental education, season of birth, maternal age, and child and maternal nutrition, public health scholars have recently extended their investigations to the importance of social support networks, often made up of kin, in promoting positive health outcomes through buffering households against risk of food shortages (Cohen, Underwood, & Gottlieb, 2000).

As regards such familial networks, biologists expect kin to assist their relatives in successfully raising offspring, even if at personal cost, because of inclusive fitness benefits. Hamilton's kin selection theory (Hamilton, 1964) provides an explanation for such altruism: altruistic behavior is favored wherever rb−c>0, where r is the genetic relatedness between actor and beneficiary, b is the benefit of receiving the altruistic behavior, and c is the cost of performing the behavior. Evolutionary anthropologists have made specific tests of Hamilton's rule regarding nepotism and investment within human families (e.g., Alexander, 1979, Chagnon, 1982, Chagnon & Bugos, 1979) and, in some cases, found quite close fits between predicted patterns of altruism and empirical data (Bowles & Posel, 2005).

It is often forgotten that Hamilton also recognized the potential for competition among kin in viscous populations where dispersal is limited (Hamilton, 1967). In studies of nonhumans, competition between relatives over resources has been convincingly shown to reduce selection for cooperation among relatives (Griffin et al., 2004, West et al., 2001) and to bias sex ratios away from the competing sex (Clarke, 1978, Gowaty, 1993); this is because although limited dispersal raises levels of relatedness among interacting individuals, it can also lead to more local competition among relatives (Frank, 1998). In a separate theoretical literature on reproductive skew, some of the transactional models in which dominants are assumed to control the reproduction of subordinates, specifically the concessions model (Johnstone, 2000, Keller & Reeve, 1994), predict that relatedness within a group can exacerbate reproductive differentials among kin, leading to the prediction that some relatives can suffer from living in high aggregations of kin.

In humans, there is clear evidence that within-family inequities exist (Boone, 1986, Hrdy & Judge, 1993) and that families can be suffused with tension (Alexander, 1979, Emlen, 1995). Furthermore, psychologists have explored health and educational differentials among siblings (Hertwig, Davis, & Sulloway, 2002), in some cases within the framework of resource dilution (Downey, 2001). Nevertheless, there has been no systematic examination of the implications of local resource competition for constraining the role of relatives as cooperators or competitors, with the exception of Hadley's (2004) study of Tanzanian Pimbwe in which he examined statistical interactions between socioeconomic status and kin effects. Ultimately, a perspective focusing on the intersections of kin effects and resources could be brought to bear on explaining cross-cultural variation in kinship systems, but in this analysis, I focus on the more tractable question of within-population variability.

1.2. Predictions for a Kipsigis community

The Kipsigis of Kenya are a Nilotic agropastoral population, inhabiting the fringes of the White Highlands, living on now demarcated plots of land on which they keep livestock and grow crops for subsistence and cash. They have, at least, since the late 18th century, been patrilineal, polygynous, and patrilocal (Mwanza, 1977). Marriages are sealed with a bridewealth payment from the groom to the bride's lineage, whereupon a woman moves from her natal to her husband's homestead and, as a new lineage member, becomes entirely dependent on her husband's resources (Peristiany, 1939). A woman's sons inherit land exclusively from their father.

The study area on the borders of Kericho and Narok Districts was settled in the 1930s and 1940s by Kipsigis men and their families escaping the overcrowded “Native Reserves” established by the British colonial government (Borgerhoff Mulder, 1990). The nontraditional practice of land enclosure (through fencing, Manners, 1967) that emerged throughout the century was exacerbated by a 1980s World Bank initiative for land registration in the area. Land ownership, together with rapid population growth and interethnic territorial tensions, have, to some extent, inhibited the traditional pattern of married sons establishing farms adjacent to that of their father, and, increasingly, all sons settle on their father's land or on a plot of land purchased by their father. Unsurprisingly, many demographic traits are affected by the size of a man's land holding, including polygynous marriage (Borgerhoff Mulder, 1987a, Borgerhoff Mulder, 1990), fertility, and offspring survival (Borgerhoff Mulder, 1987b). Perhaps, unsurprisingly, there can be intense competition among brothers over inheritances and reproductive opportunities, including bridewealth (Borgerhoff Mulder, 1998). As cooperation among kin may be restricted when local competition is intense, I predict that the effects of paternal kin on child survival will be stronger in wealthier than in poorer households. This hypothesis predicts wealth interaction effects, not necessarily a reverse in the sign of the statistical relationship across wealth categories.

2.1. Sample

Analyses are based on 785 births to 129 women and 107 men between 1945 and 1990, collected from retrospective interviews with members of all households in four neighborhoods (kokwetinik; singular, kokwet) intensively studied in 1982–1983 and 1990–1991. Of these live births, 15% failed to reach their fifth birthday, a percentage that varied over time: 19% of those born prior to 1980 (n=392) and 10% of those born in 1980 or later (n=393). As shown elsewhere, the principal cause of mortality for children below 5 years old is disease—malaria, diarrhea, measles, and influenza (Borgerhoff Mulder, 1987b)—all of which are treated with traditional and modern medicine. Consistent with analyses showing no gender-biased survival (Borgerhoff Mulder, 1989), child sex dropped out of all the models.

Kipsigis households, as defined here, consist of a woman, her children, and her husband. They are typically located within larger extended residential units of patrilineally related kin, units that include the households of a woman's cowives, of sons and their wives, and of the husband's father and his wives and that retain ultimate rights of ownership to land and livestock. Each woman owns her own hut; after a year or so of marriage, she obtains rights of use to land and livestock and builds her own granary. Reproductive histories were taken for all women in the four neighborhoods who had produced one or more children. Births to individuals deceased at the time of the interview were recorded [10 men (9.3% of the sample) and 12 women (9.3% of the sample)] by means of questions with a living spouse who fell in the sample.

2.2. Variables

Survival status or age at death was determined for each birth. Of special interest were the survival status and residence of the father's mother (FM) and father (FF) and the number of the father's brothers (FB), as well as those for maternal kin—mother's mother (MM), mother's father (MF), and mother's brothers (MB). The number and/or status of kin in these categories was determined from the full demographic sample (Borgerhoff Mulder, 1987a, Borgerhoff Mulder, 1987b). This was straightforward for paternal kin, given the strong patrilineal residence patterns among the Kipsigis, with men and their brothers very commonly living on the same or adjacent plots of land. Fortuitously, systematic data on wives' relatives had been collected as part of a bridewealth study (Borgerhoff Mulder, 1995). Additional kin effects considered in the model are the survival status of the focal child's mother and father.

Each parent and each grandparent of a child were coded as dead if this individual had died before or within the focal child's first 5 years of life. Because the date of the specified relative's death could not always be narrowed down to month and year, this approximation seemed more appropriate than restricting analysis to better educated individuals with more precise recall of exact dates. The number of a child's FB or MB was determined on the basis of questions regarding the number of currently living (1982–1983) brothers of the father and mother and their residential histories; it is therefore fixed for a sib set and can be considered only an approximation for any given child; these variables are accordingly treated as categorical variables (FBCAT, MBCAT). For FM, FF, MM, and MF, it was possible to determine whether this relative lived in the same set of adjacent kokwetinik as the child (generally within a 15-km range, an easy day's walk, termed “local”) or at a further distance (generally more than an easy day's walk, termed “distant”); even where relatives lived beyond the adjacent kokwetinik that constituted the sample of this study, the distance rarely exceeded 25 km. In this sample, the mode, median, and range of the ego's FM's residence and MM's residence are 0, 0, and 0–11 km and 0, 3, and 0–128 km, respectively. Unless other information was available, it was assumed that residence patterns observed in 1982–1983 and confirmed in 1991 had been constant. Very infrequent land sales and minimal labor migration justify such an assumption.

To study the independent effects of kin factors on focal child survival and their interactions with wealth, these measures were linked to the full demographic and socioeconomic records. Wealth is measured as land available to the child's mother, because of the significance of this factor in previous analyses of this population (Borgerhoff Mulder, 1987b). Used as control variables were data on year of birth (for secular changes in survival), gender, twin status, mother's age at birth of child, previous birth interval, birth order, polygynous status of mother, mother's education, wealth (as defined above), and the number of the focal child's brothers (both full and half). These variables allow for both a finer determination of the effects of kin and a more nuanced interpretation of the correlational results.

2.3. Analysis

Cox's regression, a form of event history analysis (Allison, 1984), was used to determine probability of survival within the first 60 months of life, using STATA (v 8.1). This method is appropriate in that it allows for the inclusion of censored cases, in this case, children who have survived to various ages but not yet reached their fifth birthday. Four analytical steps were taken. First, to identify unobserved variability between siblings born to the same mother, all independent variables were investigated for shared frailty using the STATA “share” command and entering the mother's ID as a covariate. Theta values (a single variance parameter that measures heterogeneity in survival times across the children of different mothers) were consistently very small (0.00–0.086) and never significant, indicating that there is no intermother variance in frailty; in other words, all full sib sets have the same unobserved frailty (Gutierrez, 2002). This results in part because many of the covariates of interest, such as wealth, status of grandparents, number of siblings, and so forth, vary much less within families than between families. The same analysis with similar outcomes was conducted for shared fathers and shared lineages. Each of these terms was investigated in the full model and, again, was observed to be very small, reflecting the fact that multiple independent variables capture much of the variation and leave little residue to be explained as unobserved heterogeneity. Since shared frailty parameters never contributed significantly to any of the models, they were dropped from the presentation of the final analyses.

The second step entailed identifying the effects of a range of control variables, defined as variables expected to affect survival times but not of primary interest in the current analysis. This was done by determining the increased probability of dying in the first 5 years of life as a function of a change in the specified independent variable. Models were fit using the Cox's proportional hazards model in STATA (see Table 1). Model A drops individuals for whom there are missing data, whereas Model B uses the full sample, dropping variables for which there is incomplete information. The results are largely congruent. Interpretation of the proportional hazard ratio is as follows. The estimated hazard ratio of 0.974 for birth year (in the reduced sample, Model A) indicates that an individual born in any given year has a 2.6% reduced likelihood of dying than does an individual born in the preceding year (i.e., the ratio of their respective hazards is 0.974). This ratio does not differ significantly from 1, on the basis of a Wald χ2 statistic (3.117), although it shows a statistically significant trend. Now, consider birth intervals, coded as a categorical variable. Compared to children born after the shortest interval, the 23- to 25-month group has a nonsignificant 37% greater chance of dying, the 26- to 36-month group a 36% reduced chance of dying, and the longest category an even more (61%) reduced chance of dying. While only the ratio of the first to the last category is significantly different from 1, there is an overall significant effect as indicated by the Wald χ2 statistic (11.741). Hazard ratios show that children with a later birth year are less likely to die (Table 1, Panel 1) and that twins have higher mortality than single births (Panel 3). Mortality declines with the length of the preceding birth interval (Panel 5), among children of middle birth order (Panel 6, Model A only), among children born to women with no cowives (Panel 7, B only), with each year of maternal education (Panel 8), with an intermediate number of full and half brothers (Panel 9), and as the size of the land holding increases (Panel 10). These are well-established and commonly observed effects found previously both in this and many other developing nation populations (Rutstein, 1984, Rutstein, 2000) and receive no further discussion here except with respect to their interactions with kin variables.

Table 1. Nonkin variables and child survival—multivariate analyses

	Variable	Model Aa			Model Bb
		n	Proportional hazard ratio and Z-score significance level	Wald χ2 and significance level	n	Proportional hazard ratio and Z-score significance level	Wald χ2 and significance level
	1. Birth yearc	582	0.974	3.117†	785	0.957	15.975⁎⁎⁎
	2. Sex			0.002			0.492
	Maled	300			404
	Female	282	1.012		381	0.882
	3. Twins			7.419⁎⁎			31.987⁎⁎⁎
	Singled	568			756
	Twin	14	3.793⁎⁎		29	6.304⁎⁎⁎
	4. Mother's age (in years) at birthe			0.571			1.341
	12–20d	105			253
	21–27	249	0.795		286	0.696
	28–47	228	0.975		246	0.715
	5. Previous IBI (in months)f			11.741⁎⁎
	10–22d	147
	23–25	147	1.373
	26–36	164	0.640
	37–150	124	0.389⁎
	6. Birth orderg			6.050†			3.382
	1–2d	99			255
	3	94	0.696		111	0.989
	4–6	228	0.363⁎⁎		248	0.649†
	7–14	161	0.601		171	1.152
	7. Polygyny (no. of cowives)e			3.211			10.120⁎⁎
	0d	254			391
	1	264	1.555†		317	1.841⁎⁎
	>1	64	1.804		77	2.376⁎⁎
	8. Education (in years)ch	582	0.930	3.463†
	9. Number of full and half brothers			7.213⁎			11.028⁎⁎
	0–3	112			140
	4–8	338	0.557⁎		411	0.600†
	9–15	132	1.130		234	1.330
	10. Acresc	568	0.978	6.183⁎⁎	785	0.974	11.606⁎⁎⁎
	Full model χ2 likelihood statistics	60.371, df=17, p<.001			78.715, df=13, p<.001

“Previous IBI” and “education” are dropped from Model B, leaving an n of 582 (there is an overlap in cases with missing values).

a Reduced sample: individuals with missing data were dropped. b Full sample: variables with missing data were dropped. c Covariate. d Reference category. e Calculated at the time of birth of the child. f Data on previous IBI are missing for 146 observations. g Birth order was categorized this way to maximize capture of inverted U shape survival by parity. h Data on education are missing for 87 observations. ⁎ p<.05. ⁎⁎ p<.01. ⁎⁎⁎ p<.001. † p<.10.

The third step entailed single-variate analyses of kin effects for the full sample of births (Table 2, Model C). The raw survival curves are presented in Fig. 1, Fig. 2, Fig. 3, Fig. 4. The fourth step examined a series of multivariate models with control and kin variables considered together (Table 3, Model D), split by wealth (Model E), and with wealth interactions (Model F). The third and fourth steps are presented in Section 3.

Table 2. Kin variables and child survival—single-variate analyses

		n	Model C
			Proportional hazard ratio and Z-score significance level	Wald χ2 and significance level
	1. Parental survival
	Father
	Deada	19		0.09 (ns)
	Alive	766	1.235
	Mother
	Deada	23		19.85⁎⁎⁎
	Alive	762	0.228⁎⁎⁎
	2. Paternal kin
	FM
	Deceaseda	117		29.73⁎⁎⁎
	Distant	214	0.294⁎⁎⁎
	Local	454	0.354⁎⁎⁎
	FF
	Deceaseda	193		6.98⁎
	Distant	219	0.581⁎
	Local	373	0.600⁎
	FBCAT
	0–1a	215		3.95
	2	177	0.892
	3	202	0.604†
	4–6	191	0.745
	3. Maternal kin
	MM
	Deceaseda	84		3.93
	Distant	381	0.582†
	Local	320	0.740
	MF
	Deceaseda	147		4.77†
	Distant	342	0.672
	Local	296	1.054
	MBCAT
	0–1a	112		6.23†
	2	320	0.732
	3	292	0.550⁎
	4–7	61	0.448†

a Reference category. ⁎ p<.05. ⁎⁎⁎ p<.001. † p<.10.

Fig. 1. The effects of (top) FM, (center) FF, and (bottom) the number of FB on child survival (<60 months). For statistical tests, see Tables 2 (single-variate analyses) and 3 (multivariate model).

Fig. 2. The effects of the number of FB on child survival (<60 months) splitting the population into a poorer and a richer half. Compared to the baseline (FBCAT 0 to 1), the proportional hazard ratio for 2, 3, and 4–6 FB are 0.996, 1.161, and 1.973 (Wald χ2=6.173) for the poorer households and 0.861, 0.232, and 0.124 (Wald χ2=18.603, p<.001) for the richer households, yielding a significant interaction (Wald χ2=20.823, p<.001). For the multivariate model statistics, see Table 3.

Fig. 3. The effects of (top) MM, (center) MF, and (bottom) the number of MB on child survival (<60 months). For statistical tests, see Tables 2 (single-variate analyses) and 3 (multivariate model).

Fig. 4. The effects of the number of MB on child survival (<60 months) splitting the population into a poorer and a richer half. Compared to the baseline (MBCAT 0 to 1), the proportional hazard ratio for 2, 3, and 4–7 MB are 0.730, 0.409, and 0.197 (Wald χ2=9.539, p<.05) for the poorer households and 1.264, 1.425, and 1.837 (Wald χ2=0.777, ns) for the richer households, yielding a significant interaction effect (Wald χ2=6.681, .05<p<.10). For the multivariate model statistics, see Table 3.

Table 3. Kin variables, control variables, and child survival (Model D), with data split by wealth (Model E), and with wealth interaction terms (Model F)

		n	Model D		Model E				Model Fa
					Poor		Rich
			Proportional hazard ratio and Z-score significance level	Wald χ2 and significance level	Proportional hazard ratio and Z-score significance level	Wald χ2 and significance level	Proportional hazard ratio and Z-score significance level	Wald χ2 and significance level	Wald χ2 and significance level
	1. Parental survival
	Father
	Deadb	19		1.49		1.39		0.01	2.40
	Alive	766	2.350		2.660		4.269
	Mother
	Deadb	23		20.72⁎⁎⁎		2.73†		18.07⁎⁎⁎	3.51†
	Alive	762	0.164⁎⁎⁎		0.357†		0.053⁎⁎⁎
	2. Paternal kin
	FM
	Deceasedb	117		19.50⁎⁎⁎		10.58⁎⁎		2.74	6.40⁎
	Distant	214	0.294⁎⁎⁎		0.173⁎⁎⁎		0.537
	Local	454	0.337⁎⁎⁎		0.481†		0.227
	FF
	Deceasedb	193		3.26		1.12		2.04	0.358
	Distant	219	0.574†		0.589		0.455
	Local	373	0.713		0.629		1.304
	FBCAT
	0–1b	215		7.35†		3.29		10.11⁎⁎	15.25⁎⁎
	2	177	0.721		1.620		0.186⁎
	3	202	0.502⁎		1.683		0.079⁎⁎
	4–6	191	0.494⁎		0.747		0.162⁎⁎
	3. Maternal kin
	MM
	Deceasedb	84		3.59		0.30		3.65	1.39
	Distant	381	0.727		0.781		0.503
	Local	320	0.478†		0.720		0.189†
	MF
	Deceasedb	147		3.35		3.27		0.25	2.05
	Distant	342	0.873		0.664		0.698
	Local	296	1.544		1.485		0.963
	MBCAT
	0–1b	112		10.82⁎		7.14†		4.75	6.77†
	2	320	1.790		1.314		2.189†
	3	292	0.838		0.561		1.321
	4–7	61	0.588		0.214		0.602
	4. Control variables
	No. of cowives
	0b	391		10.18⁎⁎		4.58†		2.25	1.21
	1	317	2.012⁎⁎		1.843†		2.007
	>1	77	2.359⁎⁎		2.332		1.481
	Birth year	785	0.940⁎⁎⁎	26.35	0.948	7.56⁎⁎	0.922⁎⁎	9.87⁎⁎	3.56†
	Singleb	756		27.00⁎⁎⁎		9.97⁎⁎		18.70⁎⁎⁎	2.46
	Twin	29	5.802⁎⁎⁎		6.760⁎⁎		9.314⁎⁎⁎
	Acres	785	0.968⁎⁎⁎	11.36⁎⁎⁎	0.925	2.31	0.954	3.16†	NA
	No. of full or half brothers
	0–3b	140		4.34†		5.65†		0.86	4.46†
	4–8	411	0.998		1.393		1.164
	9–15	234	1.813		3.318		2.078
	Full model log likelihood χ2	785	138.001, df=23, p<.001		78.34, df=23, p<.001		107.27, df=23, p<.001

a Model F is not strictly a single model. It results from a rerunning of Model D 12 times, each time with a wealth interaction term added for the specified variable. b Reference category. ⁎ p<.05. ⁎⁎ p<.01. ⁎⁎⁎ p<.001. † p<.10.

3.1. Parental effects

As shown in the first panel of Table 2, a child is 77% less likely to die if his or her mother is alive during the first 5 years of life than if the mother dies during this period; living fathers have no significant effect on survival. This pattern is retained in the full model (Table 3, Model D) and holds in both the richer and the poorer (at marginal significance) halves of the population (Table 3, Model E). There is a marginally significant interaction between mother and wealth, indicating that mothers' survival is more strongly associated with child survival in rich than in poorer households, an unpredicted trend we return to in Section 4.

3.2. Paternal kin effects

The effects of paternal kin on survival (Fig. 1) are shown in the second panel of Table 2. Single-variate analyses (Model C) show that children with a living FM and children with a living FF are significantly less likely to die than those without these paternal kin.

The principal paternal grandparental effect lies between those who are deceased and those who are alive and live either locally or distant. This indicates that for this category of kin, the distance at which they live is not a major factor in providing support, conforming to ethnographic observations: paternal kin amass at times of conflict over land, interlineage conflicts, or family crises in current times, as they did in the early years of the 20th century (Peristiany, 1939).

Examining paternal kin effects in a full model in which all demographic and economic variables are entered (Table 3, Model D) shows that both a surviving FM and the number of FB are retained as significant factors reducing child mortality, although the latter is only marginally statistically significant. In other words, paternal kin raise survival chances, independent of the positive effect associated with the number of MB (Panel 3, and see below) and of control variables (see Table 3, Panel 4), including wealth.

Splitting the sample by wealth shows clear wealth interactions. As expected, the positive association between FB (FBCAT) and child survival is stronger in the richer than in the poorer section of the population (Fig. 2). Children with a large number of FB show reduced mortality in the wealthier but not in the poorer half of the population. These results hold in the fully controlled model (Table 3, Model E), and there is a significant interaction effect between number of FB and wealth (Model F). The reasons why the FB might enhance survival in wealthy but not in poor households are examined in Section 4. A further wealth interaction indicates that the association between the FM and child survival is stronger in the poorer than in the richer half of the population, another unexpected result addressed in Section 4.

3.3. Maternal kin effects

The effects of maternal kin on survival (Fig. 3) are shown in the third panel of Table 2. Single-variate analyses (Model C) reveal that the status of the MM has no overall effect on child survival and that there is only a marginally significant trend for the MF and MB.

These marginally statistically significant effects reflect low mortality (high proportional hazard ratio) in cases where the MM and MF are alive but living beyond the neighborhood compared to those who are either deceased or alive and present. This suggests an unusual but ethnographically quite plausible maternal kin effect—a woman benefits if her parents live at some distance from her marital home. This “refuge” effect is explored in Section 4.

Examining the effects of maternal kin in the full analysis (Table 3, Model D) reveals that neither the effect of MM nor the MF retains significance. There is, however, a strong independent and positive effect of the number of MB on child survival. In other words, children whose mothers have three or more brothers experience a significantly reduced mortality risk, independent of whether the mother herself is alive (Panel 1), of the effects of the FM and FB (discussed above, Panel 2), and of other control variables (Panel 4).

Splitting the sample by wealth shows that the apparently beneficial effects of the MB are, in fact, most marked in the poorer half of the population (Fig. 4). Statistically, this is preserved in the full model (Table 3, Model E) where the number of MB is marginally significantly associated with child survival only in poorer households, yielding a marginally significant interaction between MB and wealth (Table 3, Model F). Adding to the model a further interaction term that identified whether the MB lived locally or not (as gauged by the location of the mother's parents) shows that the number of MB is associated with higher survival in poor households when these brothers live at a distance (Wald χ2=9.17, df=3, p<.05).

3.4. Control variables and the role of competition among kin

Analyses showing that nepotism toward grandchildren, nephews, and nieces within patrilineages varies by wealth provide only indirect evidence that kin are motivated to help their relatives as a function of potential competition over resources within the patrilineage. The most convincing evidence for this interpretation comes, unsurprisingly, from ethnographic observations (see Sections 1 and 4). Further indication of competition contributing to offspring survival in resource-stressed situations is found in the patterning of the effects of the control variables in these analyses. First, consistent with previous analyses, the number of a child's mother's cowives is negatively associated with child survival (Borgerhoff Mulder, 1990, Borgerhoff Mulder, 1997), an effect that is observed in poorer but not in richer households (although the interaction term is not significant). Second, again consistent with previous analyses (Borgerhoff Mulder, 1998, Table B1, and here just focusing on male siblings—the most likely competitors), survival chances are negatively associated with the number of a child's full or half brothers. As with the analysis of mother's cowives, this effect is stronger in poorer than in richer households (Model F).

Examining control variables in this way provides statistical support for the claim that competition within poorer patrilineages is modulated by wealth availability. On the other hand, it is important to stress that these imputed patterns of competition are not entirely accounted for by wealth since both the main effects and the wealth interaction terms are independent of the wealth covariate (acres). Furthermore, an anticipated interaction between birth date, wealth, and child survival (reflecting increased competition over time with escalating land shortages) was not found, probably because, in recent years, child mortality has declined substantially and wealth nowadays shows higher correlations with fertility than with offspring survival (Borgerhoff Mulder, 1987b).

The motivations for maternal kin to assist with the raising of grandchildren, nephews, and nieces are more difficult to evaluate since we know less about their circumstances; this is because many of them reside outside of the study area. The present analyses do, however, indicate that these lineages help out their daughters when their daughters find themselves married to men with limited resources.

4.1. Evidence for local resource competition

Why do paternal kin, and specifically FB, enhance survival in wealthy households but not in poor households? A man and his brothers and half brothers are direct competitors over their father's land and livestock. Kipsigis custom dictates an equal division of resources among sons, a practice that worked well in a traditional system in which livestock was the principal wealth source. During the colonial period, the ownership of cultivable land became important. Unlike cattle, land is an inelastic resource, and plot sizes are shrinking across generations as population grows. Violent cases of strife over land among brothers, among brothers' wives, and among the cowives married to polygynously married men were observed during fieldwork, including the following: a man burning down the house of his brother, with wife and children inside; a woman attempting to poison the children of her husband's elder brother; attempts to “move” fences (effectively brush piles) over night prior to the formal delineation of fields; witchcraft accusations; and local court cases. Equally notable were instances of supreme altruism and fraternal cooperation (such as shouldering the medical expenses of a brother's children or providing secondary school fees for a nephew or niece). The unexpected observation that living paternal grandmothers improve child survival in poor but not in rich households suggests that these grannies can perhaps buffer young children from mortality in patrilineages wrought with resource conflict.

The suggestion that brothers cooperate more in richer than in poorer patrilineages likely results from two specific features of Kipsigis society and ecology. First, patrilineally related kin tend to be equivalent in wealth status, reflecting the traditional egalitarian inheritance system and limited access to off-farm employment in this population. Second, the relationship between resource availability and fitness shows diminishing marginal benefits as wealth increases (Borgerhoff Mulder, 1987a). Under such circumstances, a poor man might find that the inclusive fitness benefit (b) of making a resource transfer to a sick nephew or niece is eclipsed by the direct fitness cost (c) of the lost resource; conversely, for a rich man, the marginal benefit of transferring resources to save the life of an ailing nephew or niece could outweigh the small, direct fitness cost. If this is the logic underlying these empirical patterns, nepotism would indeed be more likely to trump direct fitness costs in richer than in poorer lineages. This framework helps to explain why, in a study of the Tanzanian Pimbwe, Hadley (2004) found exactly the opposite pattern to the one reported here. The presence of mother's sisters in the village was positively associated with child weight-for-age scores only in families of low socioeconomic status, a more intuitively understood and possibly widespread pattern (e.g., Stack, 1997, for African Americans). Why is the Pimbwe pattern so different from the Kipsigis? With little to no inherited possessions, poorer families have few material assets to compete over and have much to gain from cooperation. In fact, because close relatives can vary enormously in wealth, it is the rich individuals who report being pestered by their poorer relatives into providing a costly resource stream (Hadley, 2004).

An alternative but unlikely explanation for the Kipsigis results is that it is not the b and c terms in Hamilton's equation that vary with wealth but rather r. Thus, relatedness among agnatic kin might be higher in richer than in poorer households, reflecting greater mate guarding potential on the part of husbands or lower temptations to infidelity on the part of women. However, extramarital affairs are heavily sanctioned in this population and lead to acute punitive outcomes for women, which (anecdotally at least) appear unrelated to wealth. Insofar as richer people tend to be more educated and cosmopolitan, one might expect (if anything) richer households to show lower r, although this is pure supposition.

These results can be placed in the broader framework of family conflict (Emlen, 1995). Hill and Hurtado (1996), after failing to show effects on survival associated with individual relatives, turn to the use of the ratio of related helpers to potential recipients in the Ache to indicate kin assistance in the context of competing demands on their time. In Dominica, Quinlan (2005) shows how the number of brothers in a community negatively affects reproductive success, a pattern elsewhere linked to competition over inheritance and marriage payments (East Africa, Mace, 1996) and land inheritance (agrarian Sweden, Low, 1994). Furthermore, in many parts of the world, strong negative effects of the number (Knodel, Havanon, & Sittitrai, 1990) and sex (Parish & Willis, 1994) of siblings are seen on health and educational outcomes (Downey, 2001, Garg & Morduch, 1998) as parents strategize within offspring sets. Specific links can be drawn between parental strategies and resource shortages in a historical German peasant population, where differential investment in sons and daughters is shown to vary in relation to habitat saturation (Voland & Dunbar, 1995). The demographic evidence presented here on the Kipsigis, a population with a strong ethos of equal inheritance to sons and an almost exclusive reliance on inherited resources of land and livestock, demonstrates how the balance between apparent kin assistance and kin competition pivots on the availability of resources. On these grounds, we might also expect that as rural populations become integrated into a cash economy (with wealth ultimately held individually in bank accounts), the importance of kin networks for resource transfers will decline, as is occurring in India (Shenk, 2005); kin networks may nevertheless continue to play an important political role (Munshi & Rosenzweig, 2005).

The implication of these results is that for humans, as with other species (Griffin et al., 2004, West et al., 2001), patterns of dispersal and resource distributions are key to the development of more precise predictions regarding altruism. This might explain why, in many domains of human behavior, altruism is only loosely explained by relatedness; for example, only one third of the variability in cash remittances in South Africa is attributable to a model of kin selection that includes reproductive value (Bowles & Posel, 2005). Furthermore, local competition among kin over resources may explain why, in some situations, kin of equal relatedness but different categories, such as those related through maternal or paternal ties, may evince very different patterns of nepotism (Alvard, 2003, Quinlan, 2005).

4.2. The role of other kin in contributing to child welfare

We now turn to the effects (and lack thereof) of parents, of the maternal grandmother, and of other maternally related kin on child survival among the Kipsigis and implications for understanding the role of kin as caretakers in human societies.

First, consider parents. Loss of a mother but not of a father is a strong determinant of offspring survival in the Kipsigis, as would indeed have been expected. Mother loss is an important influence on mortality in the first years of life in many populations (recently reviewed in Sear & Mace, 2006). While mother loss does not precipitate infanticide in the Kipsigis (unlike in some populations, e.g., the Paraguayan Ache, Hill & Hurtado, 1996), breast-feeding children are inevitably weaned rapidly, triggering potentially serious health complications that threaten survival, and weaned children move to the house of the deceased mother's cowife or sister-in-law where they are in direct competition with other children. Father loss, in contrast, has no evident survival costs. High mortality risks among children with a deceased father are most commonly attributed to the mother's new marriage, specifically the stepparent (Daly &Wilson, 1985, Voland, 1988); indeed, the stress that accompanies divorce may be more detrimental than the death of a father per se (Sear, Steele, McGreggor, & Mace, 2002). In the Kipsigis, remarriage does not occur. Widowed mothers are subject to “widow inheritance,” a practice whereby the woman is effectively passed on to one of her brothers-in-law; the fact that widowed women do not remarry but cohabit with their late husband's brother likely contributes to the finding that widowed Kipsigis women do not experience higher child mortality than nonwidowed women. Finally, the anomalous and unanticipated finding that maternal loss is more prejudicial to child survival among the wealthy than among the poor most likely reflects the fact that, in this small sample, children in the wealthy sector were somewhat older at their mother's death.

Second, we examine the absence of a maternal grandmother effect. Generally, maternal grannies are thought to be natural caretakers of their daughters' children as opposed to their sons' children because of the greater certainty of relatedness, and across many studies, their presence is found to be associated with positive survival outcomes (Sear & Mace, 2006). In the Kipsigis, however, it is the paternal grandmother whose presence has a beneficial effect on the survival of small children, no doubt reflecting the strongly patrilineal organization of Kipsigis (also see Gibson & Mace, 2005, Leonetti et al., 2005, Lycett et al., 2000). This suggests a facultative component not only to paternal grandparental behavior (as often surmised) but also, more surprisingly, to maternal grandparental roles.

We turn now to other maternal kin. In the Kipsigis, the status of maternal kin, specifically maternal brothers, is associated with child survival only in the poorer half of the population. This suggests that maternal relatives provide help primarily in situations where limited nepotism from a child's father's lineage can be expected. This is the case despite the fact that there is relatively little cultural emphasis on matrilineal as compared with patrilineal relatives in this population. These results speak to the obvious but often ignored fact that inheritance rules do not dictate behavior (Dickemann, 1982), as well as to the more specific point that patrilineal inheritance rules do not preclude an important role for maternal kin under certain circumstances, a phenomenon much appreciated by earlier ethnographers of African kinship (Radcliffe-Brown & Forde, 1950) and, indeed, by Peristiany who singled out a mother's true brother as a person “of astounding importance” (Peristiany, 1939, p. 100) for the Kipsigis. The data presented here do, indeed, suggest that maternal kin can play a key role in situations where they observe a patrilineage unable or unwilling to provide adequate care to their grandchildren. A similar effect is found among the patrilineal Ethiopian Oromo, where maternal grannies work hard in their daughters' households; accordingly, living maternal grannies tend to have taller granddaughters, although they have no significant effects on the survival of their daughters' children (Gibson & Mace, 2005).

Interestingly, the role of maternal kin as indirect providers of child welfare is only in evidence when they live at a distance (over a day's walk) from the mother's marital household. This is opposite to the Ethiopian pattern where child survival is enhanced by the presence of unspecified maternal kin in the village (Gibson & Mace, 2005). An effect superficially similar to the Kipsigis is found for non-coresiding maternal grandmothers among the Northern Indian Khasi (Leonetti et al. 2005) but is attributed to self-selection, specifically the decisions of grannies to live with otherwise disadvantaged daughters. The explanation in the Kipsigis is rather different (since women change residence at marriage, not in grandmotherhood) and suggests a “refuge” hypothesis. In a cultural group where divorce is practically nonexistent (Orchardson, 1961, Peristiany, 1939), women are typically unhappy during at least some period of their marital life. In such circumstances, and particularly when violence ensues, women run away (or at least take their children away) to the home of their parents, where their brothers also usually live. Several such instances were observed during fieldwork. When the natal homestead is near to the marital homestead, the husband and his kin arrive in person to demand the wife and her children back. When the natal homestead is at a distance, such delegations demanding the wife's return, to the extent that they occur, are generally less successful. This is primarily because a patrilineage typically lacks political allies beyond the area where it is localized. I therefore view this effect as evidence of the wife finding safe haven, for herself and her children, with her natal family. To what extent absent kin play this role in other populations is not known.