Is Polygyny a Risk Factor for Poor Growth PerformanceAmong Tanzanian Agropastoralists?Craig Hadley*

Department of Anthropology, University of California at Davis, Davis, California 95616

KEY WORDS East Africa; Sukuma; seasonal variation; nutritional status; marital status;mixed models

ABSTRACT Anthropologists and demographers havedevoted considerable attention to testing the fertility-po-lygyny hypothesis, which posits that polygynously mar-ried women have lower fertility than their monogamouslymarried counterparts. Much less attention has been paidto the potential impact of polygynous marriages on thehealth and well-being of children. The objective of thispaper was to assess whether polygynous marital status isa risk factor for poor nutritional status and growth per-formance among a cohort of young Tanzanian children.Using data collected from both wet and dry season peri-ods, we tested for an association from both cross-sectionaland longitudinal perspectives. Despite relatively high nu-tritional status compared to other agropastoralists and

horticultural populations, as well as the presence of vari-ous socioecological factors that were expected to mitigateany “costs” to polygynous marriage, we found that amongour target population, polygynous marital status is a riskfactor for poor growth performance. This association ismost pronounced in the wet season, and maintains evenafter allowing for the potential influences of child age andsex, and household characteristics. These results counterour original expectation, and suggest that polygyny iscostly to children in this population; this does not appearto be the result of difference in early child environment orhousehold wealth. Am J Phys Anthropol 126:471–480,2005. © 2004 Wiley-Liss, Inc.

Considerable research by demographers and an-thropologists has been devoted to testing the “polyg-yny-fertility hypothesis,” which posits that polygy-nously married women have lower fertility thantheir monogamously married counterparts. Anthro-pological interest in this hypothesis lies primarily inidentifying any fitness costs to polygyny in the hopesof understanding something about reproductive de-cision-making within societies, and more broadly,about the marked cross-cultural diversity in mar-riage systems (Borgerhoff Mulder, 1988; Chisholmand Burbank, 1991; Heath and Hadley, 1998; Jo-sephson, 2002; Sellen, 1999; Strassmann, 1997). Theresults from both demographical and anthropologi-cal studies are mixed, with many but not all studiesshowing a fitness cost to polygynously marriedwomen. (Josephson, 2002). Curiously, in spite ofthese groups’ considerable interest in the subject ofpolygyny, very few studies investigated the associa-tion between polygynous marital status and chil-dren’s nutritional status. This is especially surpris-ing, given the large numbers of children who may beat an elevated risk of poor nutritional status becauseof the high prevalence of polygyny in sub-SaharanAfrica countries (Timaeus and Reynar, 1998). Inaddition to the public health implications, knowl-edge of the links between polygyny and children’snutritional status is imperative to understandingthe processes that may underlie any association be-tween fertility and polygyny, and critical to the test-

ing of many hypotheses derived from evolutionarytheory.

The objective of this paper, therefore, is to assesswhether polygynous marital status is a risk factorfor poor nutritional status and growth performanceamong a cohort of young Tanzanian children. Thestudy subjects are Sukuma agropastoralists livingin the land-rich Rukwa region of Tanzania. The dataare presented for both the wet and dry seasons,periods roughly corresponding to the maxima andminima of household food security and morbidity.

In general, it is hypothesized that polygyny leadsto deleterious health outcomes because, holding re-sources constant, an additional wife leads to fewerresources for everyone in the family (Brabin, 1984).If true, this “resource dilution” should first manifestitself in terms of lower nutritional status and thenhigher mortality among children with currently po-lygynously married mothers (Brabin, 1984). How-

Grant sponsor: National Science Foundation.

*Correspondence to: Craig Hadley, Population Studies and Train-ing Center, Brown University, Providence, RI 02912.

Received 17 December 2002; accepted 22 May 2003.

DOI 10.1002/ajpa.20068Published online 26 July 2004 in Wiley InterScience (www.

interscience.wiley.com).

AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 126:471–480 (2005)

© 2004 WILEY-LISS, INC.

ever, like life-history tradeoffs in general (Lessells,1991; Tracer, 1991), any “cost” to children with po-lygynously married mothers may only be detectablein particularly marginal environments, or duringparticularly stressful periods of the year such as thepreharvest “hunger” periods. Thus, failure to iden-tify any costs in any one population may be due inpart to different overall levels of nutritional statusbetween populations, or within populations, to sam-pling during periods of high food security and lowprevalence of morbidity.

Despite the sound theoretical and public healthreasons for studying the relationship between polyg-yny and child health, very few studies have actuallyexplored this topic. This may be due to the fact thatstudies seeking to explore the relationship betweenpolygyny and maternal and child health with a pro-spective study design are often at a disadvantage, asprospective studies are expensive and time-consum-ing, and require large sample sizes. Anthropologistsinterested in the relationship between polygyny andchild health have thus relied on retrospective datacollection or family records, neither of which is en-tirely satisfactory. Besides issues of data quality, if arelationship is identified, often little or no evidenceexists to understand the processes that may under-lie it.

In contrast to the difficulties inherent in prospec-tive studies of child mortality, cross-sectional andlongitudinal designs using anthropometry offer rel-atively efficient and powerful means to investigatethe relationship between polygyny and child health.A large number of children can be measured in asingle field season, and the measures are informa-tive about the different living conditions of the chil-dren being measured. That is, anthropometric mea-surements, although nonspecific (Gibson, 1990), aresensitive to variation in parental care (Panter-Brick,1998). Children can also be measured in differentseasons to assess how their nutritional statuschanges with varying levels of food security andmorbidity (Tomkins, 1993; Ulijaszek and Strickland,1993a,b). Importantly, recent work suggests thatmortality, morbidity, fertility, and children’s nutri-tional status are all intimately interlinked phenom-ena, and thus derived anthropometric indices of chil-dren’s nutritional status can serve as a proxymeasure for risk of mortality (Pelletier, 1994;Schroeder and Brown, 1994; Tomkins and Watson,1989).

To date, two studies have empirically assessed theassociation between polygyny and the nutritionalstatus of children, both of which identified healthcosts to polygynously married women and their chil-dren. The prospective study by Strassman (1997)investigating child mortality also included some sec-ondary data on children’s nutritional status. Thedata from the Dogon suggest a weak relationshipbetween children’s nutritional status and theirmother’s marital status. Although the nutritionalstatus of children living in compounds with a high

ratio of married women to married men was slightlylower than that of children of monogamous mothers,the association was only marginally significant, andprompted Strassmann (1997, p. 693) to concludethat “the mechanisms causing low child survivor-ship in polygynous families were probably not nutri-tional.” Measuring children’s nutritional status wasnot the primary objective of this study, and she usedonly one measure of nutritional status, weight-for-height, which when low is a signal of short-termnutritional stress (WHO, 1998). Further, only a por-tion of the complete sample was measured, therebyintroducing a potential bias. The anthropometricdata were also collected during the dry season, andthus may not represent the lowest point of children’snutritional status

Using data collected in a cross-sectional survey ofDatoga mothers and their children in Tanzania,Sellen (1999) showed that even after controlling forthe influence of wealth, age, and maternal charac-teristics, Datoga children born to polygynously mar-ried first wives present with particularly poor nutri-tional status, and this is especially pronouncedamong children associated with poorer households.Compared to monogamous women, first and secondwives also presented with significantly lower mea-sures of energy balance (body mass index; BMI).This was a surprising result, given that Datogawomen appear to have considerable control over thehousehold herd, and Sellen (1999; see also Sellen etal., 2000) concluded that contrary to expectation,polygynous marriage is costly for some Datogawomen. The results of Sellen (1999) pertain to apolitically and geographically marginalized and in-creasingly impoverished segment of Tanzania (Sieff,1999), and therefore polygyny may exacerbate thealready low levels of nutritional status. For thesereasons, it is important to assess the external valid-ity of the Datoga results.

Here we ask whether polygyny is associated withdiminished nutritional status in a population whereaverage levels of nutritional status are relativelyhigh. That is, does polygyny affect nutritional statusregardless of overall levels of nutritional status, or isit only in particularly impoverished areas that po-lygyny is “costly” to women and children? To illus-trate the generally higher levels of nutritional sta-tus, in a recent study of Datoga pastoralists, Sellen(2000, p. 767) noted that 45% of all women measuredpresented with BMI below the 18.5 cutoff, the cur-rently accepted cutoff for being “at risk of chronicenergy deficiency” (Ferro-Luzzi et al., 1992). Inmarked contrast to the Datoga, previous workamong the Sukuma subjects of this paper showedthat less than 5% of the women presented with aBMI less than 18.5. Of course, such interethnicgroup comparisons should be used cautiously be-cause of possible genetic differences between groups.However, further evidence of the relatively high nu-tritional status of the Sukuma subjects of this papercomes from an analysis of a cross-sectional study,

472 C. HADLEY

which showed that Sukuma children enjoy muchgreater nutritional status than the Pimbwe, a groupof horticulturalists who also occupy the study area(Hadley, 2002). This may be because, compared totheir horticultural neighbors, the Sukuma appear toenjoy high levels of household food security thatbuffers them through wet seasons (Brockington, nodate; Hadley, 2002).

In addition to the relatively high nutritional sta-tus among the Sukuma, the presence of several so-cioecological factors is also expected to mitigate anynegative effects of polygyny on children’s health.Brabin (1984) pointed out that polygyny is expectedto be associated with poor nutritional status whenland and resources are scarce, and this will be exac-erbated when there are few alternative economicactivities available for women. For the Sukuma, thelatter is certainly true. Women have exceedingly fewalternative options, and stories abound aboutwomen stealing from their husbands to purchasehousehold goods like soap. Land, however, is notscarce in this frontier area of Tanzania’s Rukwaregion, and therefore is not likely to be a limitingfactor. There is also evidence that women have con-siderable control over their choice of husband. Po-lygyny is the preferred marital form and is highlyprevalent, but women appear to have considerablefreedom in their choice of husband (although theactual arrangements are done through their par-ents). One indication of the freedom women have inchoosing their husbands is the practice of chagu-laga. Chagulaga occurs after dance competitionsand other large gatherings, when a number of youngmales surround a young woman and chase her untilshe chooses a young man to discuss a possible ro-mantic meeting in the future. The process continues,and other young men are chosen until finally onemale is guaranteed a meeting with the youngwoman. Traditionally, this has been a formal mar-riage market for parents searching for husbands fortheir daughters.

Thus, Sukuma agropastoralists in this area enjoyhigh mean levels of nutritional status, and land forherding and farming is abundant. Moreover, ethno-graphic data and household surveys show an abun-dance of food and few problems with seasonal foodinsecurity, a result of their reliance on both agricul-ture and livestock. Lastly, women have considerablefreedom of choice in who they marry. For all thesereasons, we expected there to be no difference in thenutritional status of children of monogamously mar-ried and polygynously married mothers.

STUDY SITE AND SUKUMA ETHNOGRAPHY

The study site in the Rukwa region of southwest-ern Tanzania (Fig. 1). The region is considered oneof the more remote regions of the country in terms ofstate infrastructure. Throughout the Rukwa region,roads are dirt and villages have no electricity, andfew have running water. In the main village at thestudy site there are a few small stores, a health

clinic, and some government offices for the district.The area is marked by heavy rains (600–900 mm/year), beginning in November and culminating inMarch. Malaria is endemic in the area.

The Sukuma, the largest ethnic group in Tanza-nia, are a group of Bantu speakers who in recentyears have moved from Shinyanga throughout Tan-zania and, within the last 30 years, into the studyarea (Brockington, 2001; Galaty, 1988). The Su-kuma in the study area are sedentary and practice amixed economy, herding cattle and cultivating largeplots of maize and rice (Coppolillo, 2000). Herd sizesrange from 1–�200, and approximately half of thehouseholds who participated in this research had atleast some cattle. This means that many householdshad very few cattle, and about half had none; theSukuma in this area rely heavily on agriculture.Harvest of maize begins in June, but new maize isavailable as early as late March.

Households are widely scattered in the area sur-rounding settled villages. Sukuma households in thestudy area are extended patrilocal and patrilineal,and modal household size is 8 people, but can be aslarge as 45. Women move into to their husband’scompound upon marriage. Marriages come with abride wealth payment, usually in cattle, and divorceis extremely rare in this population. A typical set-tlement (Swahili: mji, compound/settlement) con-sists of a number of small buildings, constructed ofpoles and mud, which are organized in clusters nearacacia thorn-enclosed livestock corrals. Livestocksuch as goats and chickens often roam freelythroughout the compounds. Individuals occupying

Fig. 1. Map of study area.

POLYGYNY AND CHILDREN’S NUTRITIONAL STATUS 473

the same compounds work together and share fieldsand other household duties. Men are primarily re-sponsible for rice farming and the sale of any crops;women and older girls are responsible for childcare,household chores, and agriculture work; and smallboys (and to a lesser extent girls) are responsible fortending cattle and small stock. Among polygynoushouseholds, cowives reside in the same compound,often sharing the same house, and all women andchildren eat together from a common pot. Afterabout 1 year, children’s diets are dominated bymaize, spinach and other wild greens, milk, andsome meat.

DATA COLLECTION AND ANALYSIS

A cross-sectional anthropometric survey was con-ducted in the late wet-early dry season of 2001(April, survey 1) in the village of Kibaoni to collectbaseline data on children’s nutritional status as partof a study comparing the growth of horticulturalistand agropastoralist children. In the wet season of2002 (February, survey 2), these children were mea-sured again. Similar protocols were used in eachseason, except that in 2001, weighing was done at acentral location. The central location was used toreduce time in transport and setup. Village leaderswere asked to assist, and they called people to acentral location on a prespecified day for weighingand measuring. Because of weather (muddy trails,heavy rain) in the wet season (February 2002, sur-vey 2), many people were unwilling to walk to acentral location and, to avoid sampling bias, house-holds were visited individually.

Children’s height and weight were measured us-ing standard procedures (Frisancho, 1990; Gibson,1990; Shorr, 1986). Unfortunately, very young chil-dren frequently were not measured for length dur-ing the April 2001 study because of the mother’sreluctance to lay their children on the measuringboard. In contrast, nearly all of these children weremeasured for length in the 2002 study; in this sur-vey, mothers were shown a picture of a child beingmeasured for length, and this made mothers muchmore comfortable. Those children who were mea-sured for height were encouraged to stand tall, andthe chin support method was used. Children wereminimally clothed, and always shoeless. Heightmeasurements were made using a standard heightboard and were taken to the nearest 0.1 cm, andweight was measured to the nearest 0.1 kg, using aSeca 890 electronic scale (kindly provided byUNICEF). There is no indication that the height-for-age z-score (HAZ) of children who were measured forlength or height in the both seasons differed fromthose children who were measured for length orheight only in the wet season (t-test t � �1.30, P �0.19).

Ascertaining children’s age was made easier inthis semiliterate population because many childrenhad birth cards (approximately half of all childrenare born in a health clinic). In the absence of these

cards, discussions were held with mothers and/orother household members to ascertain the child’sage. This was done by referring to local events withknown dates, children with known birthdates, orreference to the agricultural cycle. For young chil-dren (approximately 5 years), obtaining a birth dateto the month was often quite easy, and reliable, butthe dates of older children are somewhat suspect, aproblem addressed below. All participants receiveda piece of candy and a photograph of their family.

A few weeks after the 2002 wet season anthropo-metric data were collected, each mother was revis-ited and asked a series of questions about her repro-ductive and marital status. No information wasobtained on bride price or other aspects of the mar-riage (e.g., official or unofficial). Thus, all anthropo-metric data and demographic data were collectedbefore the investigator knew the mother’s maritalstatus. Data on livestock holdings, corn and riceyields, food security, and ownership of various ma-terial items were also collected. These data were allhighly correlated, and therefore were reduced usingprincipal components analysis into a single factorcalled WEALTH in the following analyses.

A woman’s current marital status was defined onthe basis of how many cowives she had. A mother isconsidered polygynous if she currently has any ad-ditional cowives, and monogamous if she has noadditional cowives. Certainly there are many waysto describe the marital status of women in polygy-nous societies, but in this paper we define a mother’smartial status as either polygynous or monogamous.Further disaggregating to compare women by wiferank reduces statistical power by decreasing groupsample sizes. In addition to statistical concerns, eth-nographically there is no indication of hierarchicaltreatment of wives, further justification for not dis-aggregating the data. Across all outcomes, childrenof monogamous mothers had higher mean valuesthan did first and second wives in polygynous mar-riages.

The anthropometric data were converted toz-scores using Epi-Info, the CDC’s free program thatuses 2000 growth reference data. The following in-dices of nutritional status were created: height-for-age z-score (HAZ), weight-for-age z-score (WAZ),and weight-for-height z-score (WHZ). The z-scoresrepresent standardized deviations from the refer-ence median that is constructed by measuringhealthy age- and sex-matched children whosegrowth has been largely unaffected by undernutri-tion and frequent bouts of infection. The reference isnot a standard per se, but allows comparisons to bemade across a range of study sites and ethnicgroups. Low values of HAZ, or stunting, are inter-preted as evidence of long-term chronic undernutri-tion; low values of WAZ, or underweight, are consid-ered indicative of slightly shorter-term nutritionalstress. Low WHZ, or wasting, is also indicative ofshort-term nutritional stress (Gibson, 1990). Follow-ing standard public health protocols used when the

474 C. HADLEY

relationships between anthropometric measure-ments and functional impairment and mortality areunknown for a specific population, children present-ing with weight-for-age z-scores less than �2 WAZare considered underweight, children with �2 HAZare considered stunted, and children with a WHZless than �2 are considered wasted (WHO, 1998).

Two-tailed t-tests are used to explore differencesbetween mothers, and children’s anthropometric in-dicators by women’s marital status (and the non-parametric Wilcoxon test when sample sizes are be-low 20 per group). Chi-square tests are used toexplore differences in percent prevalence of stunt-ing, wasting, and underweight among children. Mul-tivariate regression models are fit to control for thepotential influence of children’s age, sex, and house-hold wealth. In addition to ordinary least squares(OLS) regression models, hierarchical level models(using the MIXED procedure in SAS; Littell, 1996)were also used to assess the influence of mother’smarital status on children’s nutritional status whilecontrolling for children’s age, sex, and other covari-ates. These models accommodate the clustering thatis inherent in the data because multiple childrenfrom one mother were weighed, as were multiplemothers from a single compound. This is necessary,because children of the same mother (and mothersfrom the same compound) are not truly independent,and thus potentially violate an important assump-tion of OLS regression; ignoring this issue increasesthe probability of rejecting the null hypothesis whenit is in fact true (Singer, 1998). Maternal and com-pound effects are neither presumed nor ignored, butrather treated as a hypothesis to be tested. All non-significant interactions were dropped from the mod-els. Results are considered significant at the 0.05level and marginally significant at or below the 0.10level.

RESULTS

Mothers were first compared to assess whetherpolygynously married women differed from monog-amously married women in ways that may affecttheir children’s nutritional status. To test this, t-tests were used to compare mothers with regard totheir age, education, height at survey 1, weight insurveys 1 and 2, household wealth (aggregate mea-sure of livestock, land, and material wealth), andparity and number of children who died. Monoga-mously married women were statistically indistin-guishable from their polygynously married counter-parts on each of these variables (Table 1). It isnotable that monogamously married women and po-lygynously married women also had access to simi-lar numbers of cattle and goats (Cattle Polygynous31 � 65 Range 200, monogamous 34 � 67 range 200;Goats: Polygynous 12 � 11 range 40, monogamous13 � 12 range 40). This is not surprising, given thatmany compounds housed both monogamous and po-lygynously married women, all of whom shared thecompounds’ resources.

Complete weight series were available for 105children (47 with monogamous mothers) from 29polygynously and 21 monogamously married womenwho lived in 32 different compounds. Completedheight series were only available for 54 children, 29with monogamous mothers. At baseline, the sampleof children with polygynously married mothers didnot differ in age and sex (Table 2). Nor did thenumber of children younger than 3 years differ be-tween the polygynous (poly) and the monogamous(mono) sample (51% poly vs. 53% mono, �2 � 0.02,P � 0.88); this is important because most growthfaltering occurs prior to age 3 (Table 2). Consider-ably more girls than boys were weighed and mea-sured (62% girls, binomial P � 0.008), but there wasno age difference between boys and girls (boys,38.21 � 40; girls, 38.21 � 32.0; t-test, P � 0.14).Because of the small sample sizes and because theage and sex distributions of the monogamous andpolygynous samples are statistically indistinguish-able, all analyses were conducted on the full sample.

Bivariate tests

In the dry season of 2001, 14% of 105 childrenwere underweight, 20% of 54 were stunted, and 6%of 54 were wasted. There were no differences in theprevalence of any indicator between sexes. Childrenof polygynously married mothers presented with aslightly greater prevalence of underweight (18%poly vs. 8% mono mothers), but this difference wasnot statistically significant. Children of monoga-mously married mothers and children of polygy-nously married mothers also did not differ in theirWAZ scores, although mean values for children inthe polygynous group were about 0.30 WAZ belowthe monogamous group. No difference was observedin the HAZ scores for all children, although thesample was quite small for this indicator.

In the wet season of 2002, weight data were avail-able for all 105 children. Height data were availablefor 100 of these children (43 with monogamousmothers). Overall, 17% of children were under-weight, 24% were stunted, and 3% were wasted.Again, there were no differences in prevalence be-tween sexes, but there were differences between mo-nogamous and polygynous children (Fig. 2). In thisseason, children of polygynously married motherspresented with a increased prevalence of under-

TABLE 1. Baseline characteristics of study mothers1

Variable Monogamous Polygynous

Education (years) 0.86 � 1.7 0.76 � 1.7Age (years) 25.19 � 4.1 26.48 � 6.1No. of children born 3.62 � 1.9 3.59 � 2.7No. of children died 0.67 � 0.9 0.48 � 0.8Weight in 2002 (kg) 57.9 � 7.1 55.3 � 6.8Weight in 2001 (kg) 57.77 � 9.0 55.47 � 6.6Height in 2001 (cm) 161.04 � 3.8 160.94 � 3.1Household wealth score 0.95 � 0.6 1.04 � 0.5

1 No differences are statistically significant (P �� 0.05). N � 21monogamous, 29 polygynous.

POLYGYNY AND CHILDREN’S NUTRITIONAL STATUS 475

weight (24% poly vs. 8% mono, �2 � 4.46, P � 0.03)and significantly lower WAZ scores (�1.37 � 1.26poly vs. �0.87 � 1.28 mono, t-test P � 0.05). Pairedt-tests comparing the dry season WAZ with the wetseason WAZ revealed highly significant declines inWAZ scores among children in the polygynous group(P � 0.001) but not among children in the monoga-mous group (P � 0.11). Children with polygynousmothers were also relatively shorter than childrenwith monogamously married mothers (monogamous�0.86 � 1.18 vs. polygynous �1.44 � 1.13, t-test t �2.47, P � 0.01; Fig. 2).

Growth velocity

The results above indicate that a mother’s maritalstatus may be a risk factor for young children, par-ticularly in the wet season survey. One issue thatmay cloud any clear association between mother’smarital status and children’s nutritional status isthat children’s ages may not have been accuratelyobtained. Focusing the analysis on younger childrenis one way around this, but a more powerful methodis to focus on growth velocity, or the change inweight and height between the two surveys. Suchintrasubject comparisons reduce the problem of in-accurate ages, and growth velocity potentially re-veals more about nutritional status and growth per-formance than do the static indices used above.Moreover, this gets around that fact that the staticmeasures may represent events when the motherhad a different martial status.

Children’s growth performance between the twosurveys was associated with their mother’s maritalstatus. Children with polygynously married mothersgained less weight and height between the two sur-veys than children with monogamously marriedmothers. Children gained on average 0.15 kg/monthbetween the two anthropometric surveys, but chil-dren with monogamously married mothers gainedsignificantly more weight than did those childrenfrom polygynous marriages (t-test t � 2.05, P � 0.04;Fig. 3). For the smaller sample of children withcompete height records, children with monogamousmothers also gained more height between the twosurveys (t-test t � 3.09, P � 0.003).

To assess whether the differences in nutritionalstatus between these two samples were the result ofslight differences in age and sex or householdwealth, several multivariate models were fitted tothe anthropometric and growth data. These modelsincluded mother’s marital status, child’s age at base-line (and an age-squared term), child’s sex, and thewealth score for each compound. Response variableswere WAZ in the dry season, WAZ in the wet season,

TABLE 2. Baseline age and sex characteristics of children in each sample

Age group N % of sample % of sample male Mean age � SD Minimum Maximum

Mono�36 months 25 53 36 17.2 � 10.6 4.5 35.8�36 months 22 46 31 78.2 � 36.4 36.9 192.0All 47 34 49.7 � 1.1

Poly�36 months 30 51 26 15.0 � 9.4 2.9 33.4�36 months 28 48 53 66.8 � 32.3 36.6 163.0All 58 39 41.8 � 35.4

Fig. 2. Children’s 2002 wet season nutritional status bymother’s marital status. Children of monogamous mothers aretaller and heavier for their age than are children of polygynousmothers. Error bars show one standard deviation.

Fig. 3. Longitudinal data for 105 children, showing modelage-adjusted weight change (kg) for 105 children, and change inheight for 53 of those children between April 2001–February2002. Children with monogamous mothers gained significantlymore weight and height. Error bars show standard error.

476 C. HADLEY

HAZ in the wet season, and change in height andweight. The age-squared term and the wealth termwere removed from the models when they were in-significant. Other factors such as maternal heightand child’s birth order were originally included inthe models but were not significantly related to chil-dren’s nutritional status in this sample.

In the multivariate models, none of the variables,including mother’s martial status, predicted chil-dren’s dry season weight-for-age z-scores (Table 3).The parameter estimate for a mother’s marital sta-tus, however, suggested that children of monoga-mous mothers were heavier for a given age. In con-trast, in the wet season, a mother’s martial statuswas a predictor of WAZ, as were the age and age-squared terms. Children with monogamously mar-ried mothers were also taller for their age than werechildren with polygynously married mothers, andthey gained more weight and height in the periodbetween the two anthropometric surveys. Thehousehold wealth term was a significant predictoronly of children’s gain in weight. This may be be-cause wealthy households were less affected by pre-harvest food insecurity.

Mixed models

As mentioned above, the hierarchical structure ofthe data introduces the potential for serious bias inOLS regression models and therefore incorrect tests

of the significance of the coefficients. To assess andcounter this problem, five other models were fittedthat initially included compound-specific effects.The models also included the following covariates:child’s age, age-squared, child’s sex, and mother’smarital status. The squared term was dropped whennot significant, although all other terms were re-tained. Children’s weight and height at baselinewere also added to the models for growth velocity,although in both cases they were nonsignificant andwere removed from the final models. Similarly, in nocase was there evidence of maternal effects, so thisterm was dropped from the models. The models werefit to the 2001 weight-for-age data, and the 2002 wetseason weight-for-age and height-for-age data, aswell as the dynamic measures of weight and heightgain (Table 4).

After accommodating the clustering issues by in-cluding the compound-specific random effect, themodels showed that a mother’s marital status wasstill an independent predictor of children’s nutri-tional status and growth performance, and this ef-fect was greatest among the more robust outcomemeasures of growth velocity (Table 4). Only the mod-els for weight-for-age failed to show a statisticallysignificant effect of a mother’s martial status, al-though the mean values were consistently loweramong the polygynous group. Children’s weight-for-age, however, was affected by the compound they

TABLE 3. Results of regression models1

Variables

WAZ dry WAZ wet HAZ wet Weight gain (kg/month) Height change (cm/month)

� � SE p � � SE p � � SE p � � SE p � � SE p

Moms MS(monogamous)

0.31 � 0.22 0.16 0.53 � 0.27 0.04 0.58 � 0.23 0.01 0.06 � 0.017 0.0001 0.10 � 0.03 0.0045

Months �0.004 � 0.003 0.12 0.02 � 0.01 0.008 �0.001 � 0.003 0.62 �0.0003 � 0.0002 0.1 �0.0057 � 0.001 0.005Child sex (F) 0.02 � 0.23 0.98 �0.16 � 0.25 0.53 0.09 � 0.24 0.68 �0.02 � 0.017 0.19 �0.004 � 0.03 0.9Months2 0.008 0.00002 � 9.569E-6 0.01Wealth 0.03 � 0.009 0.0009N 105 105 100 105 54

1 Impact of mother’s marital status on various measures of nutritional status while controlling for child age, sex, and household.

TABLE 4. Results of mixed models1

Variables

Model 1, WAZ(April 2001)

Model 2, WAZ(February 2002)

Model 3, HAZ(February 2002)

Model 4, weightvelocity (kg/month)

Model 5, heightvelocity (cm/month)

Parameter estimates (SE)

Intercept �0.83 (0.22)** �1.9 (0.37)** �1.47 (0.27)** 0.17 (0.02)** 0.70 (0.08)**Mother’s marital status

Monogamous 0.28 (0.25) 0.44 (0.28) 0.55 (0.26)** 0.04 (0.02)** 0.10 (0.04)**Child’s age (months) �0.005 (0.003)* 0.025 (0.009)** �0.002 (0.003) �0.0005 (0.0002)** �0.006 (0.002)**Child’s age2 — �0.00016 (0.00005)** — — 0.00003 (9.301E-6)**Child’s sex

Female 0.08 (0.22) �0.12 (0.25) 0.17 (0.24) �0.016 (0.02) �0.002 (0.04)Compound 0.29 (0.16)** 0.32 (0.19)** 0.27 (0.18)* 0.002 (0.0009)** 0.0013 (0.04)Weight at baseline — — — — —Height at baseline — — — — —n 105.0 105.0 100.0 105.0 54.0Model adjusted mean

valuesMonogamous �0.76 (0.20) �0.93 (0.21) �0.93 (0.21) 0.18 (0.02) 0.56 (0.03)Polygynous �1.04 (0.19) �1.37 (0.20) �1.48 (0.18) 0.14 (0.01) 0.46 (0.03)

* � P�0.10. **� P�0.05. —, nonsignificant and removed. 1 Impact of mother’s marital status on various measures of nutritional statuswhile controlling for child age, sex, and household.

POLYGYNY AND CHILDREN’S NUTRITIONAL STATUS 477

lived in, as evidenced by the statistically significantcompound term. Both the age and age-squared termwere significant, which suggests that older childrengained less weight between the two surveys. In the2002 survey, children with monogamous motherswere taller for their age than were children withpolygynous mothers. There was also an independenteffect of the compound in which a child lived, asevidenced by the statistically significant compoundterm.

The variables describing growth velocity showedeven stronger effects of a mother’s martial status.Mother’s martial status and the compound in whicha child lived were independently associated withchildren’s weight gain, with monogamy being asso-ciated with greater weight gain. There was also asignificant effect of age, but baseline weight and sexwere not associated with children’s weight gain.Children’s gain in height was also associated withtheir mother’s marital status, but not with com-pound where they lived. Together, the results fromthe mixed models highlight the appropriateness ofaccounting for the dependencies inherent in thedata. Most noticeably, P-values for the marital sta-tus term in the mixed models are substantially re-duced from what they were in the OLS models,suggesting that some unmeasured heterogeneity incompounds is accounting for much of the differencesobserved between children of monogamous and po-lygynous children.

Interestingly, including only those children aged 3years and younger still showed marginally signifi-cant effects of mother’s martial status (n � 39, P �0.10) and compound (P � 0.04) on children’s weightgain, suggesting that the difference between chil-dren with monogamous and polygynous childrenmay be set at a very early age. Fitting these modelsto the WAZ data also showed marginally significantassociations between mother’s marital status andchildren’s WAZ in the dry season (n � 50, P � 0.09)and the wet season (n � 39, P � 0.06). A similarassociation was observed in the wet season HAZdata; children less than 3 years old with polygynousmothers were shorter (n � 39, P � 0.04) for their agethan were similarly aged children of monogamouslymarried women. The effect of compound disap-peared in these samples, possibly because of thesmall sample sizes.

DISCUSSION

Why are children of polygynously marriedwomen worse off?

Contrary to initial expectations, the results showthat children of polygynously married mothers havepoorer nutritional status and growth performancerelative to children of monogamously marriedwomen, and these results maintain after controllingfor effects of age, compound, wealth, and child sex.Perhaps most surprisingly, given the abundance ofstudies linking wealth and fitness outcomes (Borg-

erhoff Mulder, 1987), there were no differences inwealth scores between monogamously marriedwomen and polygynously married women, and mo-nogamously married women were actually morelikely to report running out of food early during therainy season (results not shown). Wealth differen-tials, apparently, do not underlie the growth differ-ences reported.

There is also some indication that these differ-ences in growth performance appear to be set quiteearly and become more pronounced during the pre-harvest rainy season. These observations suggestthere may be key differences in the early child en-vironment for polygynous women and monoga-mously married women. In an attempt to under-stand more about the process that may lead to theresults presented above, we assessed this “early en-vironment” hypothesis by exploring six additionalvariables, all of which are known to be associatedwith poor growth performance. These were: 1)whether the child was born in a hospital or at home,2) whether the child received any nonhuman milkitem before breastfeeding, 3) where, if the child wasfed water, this water came from (well or river), 4)whether the mother self-expressed colustrum beforeinitiating breastfeeding, 5) when the child beganconsuming uji, a maize-based gruel which is typi-cally the first nonbreastmilk item to be introduced tochildren, and 6) whether the child was ill at any timeduring the week prior to the interviews.

Data to test the above predictions were availablefor 60 women and their youngest child. Overall, 54%of women reported giving birth to their last child athome, 78% reported feeding their child water orcow’s milk before they initiated breastfeeding, 78%reported self-expressing their colustrum becausetheir “milk had not started” or their “milk wasdirty,” 87% reported feeding their child water duringhis/her first week, and 82% of these women fetchedthis water from the river. Most (81%) women re-ported boiling the water before feeding their chil-dren. Taken together, infant feeding practicesamong these Sukuma women deviate substantiallyfrom current international recommendations (WHO,1998), but there were no differences between monog-amously married and polygynously married in anyof these variables.

To assess whether children born to polygynouslymarried mothers began consuming uji , the food thatis first introduced to children, earlier than childrenborn to monogamously married women, a mother’sresponse to the question “How old was your childwhen s/he began eating uji?” was compared to moth-er’s marital status. Because some children had notyet begun eating uji, a survival model was used tomaximize use of the available data (Allison and SASInstitute, 1995). Although these data are retrospec-tive, there is no reason to think that a mother’smarital status would affect maternal recall of thiskey event. Moreover, most children were still quiteyoung, and mothers were thus recalling events that

478 C. HADLEY

happened in the not-too-distant past, and consider-able time was devoted to pinning down the monthwhen supplementation began. Other studies of ma-ternal recall also suggest that mothers in developingnations are able to recall these events with consid-erable accuracy (Gray, 1995, 1996; Sellen, 1998).

For all mothers combined, the median age of in-troduction was 4 months (95% CI, 3, 5 months), andmonogamously married women reported a signifi-cantly earlier age of introduction than polygynousmothers (3 months vs. 4 months, log-rank �2 � 4.31,d.f. 1, P � 0.03). Given that delayed introduction ofcomplementary foods is often associated withimproved growth performance (Dewey, 1998), thisfinding is initially confusing. Because children ofmonogamously married women grow better, the ex-pectation is that they are consuming complementaryfoods later, not earlier. It is possible that mothersare using infant-based cues to assess when theyshould begin introducing solid foods (Marquis et al.,1997; Simondon et al., 2001).

Infections, which are known to slow growththrough a variety of mechanisms (Eveleth and Tan-ner, 1990; Stephensen, 1999), were assessed at thewet season anthropometric survey by 1-week mater-nal recall, and showed no differences when disaggre-gated by mother’s marital status. Of 60 children,30% were reported to be suffering from some illness(fever, coughing, and diarrhea) during the weekprior to weighing and measuring. There were nodifferences in the percent prevalence of illness bymother’s marital status (30% vs. 27% monogamousmothers), or in the number of days the child hadbeen ill (average � 7 days for each group). Althoughmaternal recall of illness may be a crude measure,there is little evidence of a difference in prevalenceby mother’s marital status. More problematic is therelatively small number of children who were re-ported ill and the small number of days of observa-tion. These are serious limitations which underminethe power of these results.

One of the more striking results is that when thecompound variable is allowed to contribute variationto the models, it is a significant predictor of chil-dren’s nutritional status in 4 of the 5 models. Onlythe model predicting children’s gain in height doesnot show an effect of compound, and this is also theindicator associated with the smallest sample size.It is noteworthy also that with the inclusion of thecompound term, a mother’s martial status is stillassociated with children’s nutritional status. Thismeans that even for mothers living in the samecompound and ostensibly sharing the same re-sources and home environment, children of polygy-nously married mothers suffer more, particularlyduring the preharvest wet season. It is thereforelikely that subtle asymmetries in the distribution offood and/or differences in maternal time constraints(which may influence breastfeeding schedules andinfant care) account for the observed differences inchildren’s health and well-being. Unfortunately,

data on the frequency and duration of feeding andmeasures of infant care are unavailable.

CONCLUSIONS

Despite the small sample sizes, we can tentativelyconclude that polygynous marriage status is a riskfactor for poor nutritional status among Sukumachildren, and that this is most pronounced duringthe wet season. Little evidence exists for differencesin the young child environment or in prevalence ofmorbidity between children with mothers of differ-ent marital status, although the latter test had ex-tremely low power. There is also little evidence thatwealth differentials offset the cost of polygyny; thewealth variables were often not a predictor of chil-dren’s nutritional status, and the marital statusterms are frequently still significant after account-ing for the mother’s compound. Contrary to initialexpectations, the results suggest that children ofpolygynously married women in this sample are atan elevated risk of poor nutritional status. This is asurprising result, given that land and food appear tobe abundant, and women are free to choose theirhusbands. These results imply that the potentialcosts of polygynous marriage are not offset by differ-ences in wealth, a finding running contrary to somepredictions of the polygyny threshold model (Searcyand Yasukawa, 1995).

ACKNOWLEDGMENTS

I thank the women and children who participatedin this study and my assistant Michael Sungula.Monique Borgerhoff Mulder, Dan Sellen, and KayDewey gave me comments on the manuscript, as didtwo very helpful anonymous reviewers.

LITERATURE CITED

Allison PD, SAS Institute. 1995. Survival analysis using the SASsystem: a practical guide. Cary, NC: SAS Institute.

Borgerhoff Mulder M. 1987. Resources and reproductive successin women, with an example from the Kipsigis. J Zool 213:489–505.

Borgerhoff Mulder M. 1988. Reproductive success in three Kipsi-gis cohorts. In: Clutton-Brock TH, editor. Reproductive success.Chicago: University of Chicago Press. p 419–435.

Brabin L. 1984. Polygyny: an indicator of nutritional stress inAfrican agricultural societies? Africa 54:31–44.

Brockington D. 2001. Communal property and degradation nar-ratives. Debating the Sukuma immigration into Rukwa region,Tanzania. Cah Afr 20:1–22.

Brockington D. No date. The politics and ethnography of ThirdWorld environmentalisms. Manuscript.

Chisholm JS, Burbank VK. 1991. Monogamy and polygyny insoutheast Arnhem Land—male coercion and female choice.Ethol Sociobiol 12:291–313.

Coppolillo PB. 2000. The landscape ecology of pastoral herding:spatial analysis of land use and livestock production in EastAfrica. Hum Ecol 28:527–560.

Dewey KG. 1998. Cross-cultural patterns of growth and nutri-tional status of breast-fed infants. Am J Clin Nutr 67:10–17.

Eveleth PB, Tanner JM. 1990. Worldwide variation in humangrowth. Cambridge: Cambridge University Press.

Ferro-Luzzi A, Sette SMF, James W. 1992. A simplified approachof assessing adult chronic energy deficiency. Eur J Clin Nutr46:173–186.

POLYGYNY AND CHILDREN’S NUTRITIONAL STATUS 479

Frisancho AR. 1990. Anthropometric standards for the assess-ment of growth and nutritional status. Ann Arbor: Universityof Michigan Press.

Galaty J. 1988. Pastoral and agropastoral migration in Tanzania:factors of economy, ecology and demography in cultural per-spective. In: Bennett J, Bower J, editors. Production and au-tonomy. Landham, MD: University Press of America. p 163–183.

Gibson RS. 1990. Principles of nutritional assessment. New York:Oxford University Press.

Gray SJ. 1995. Correlates of breastfeeding frequency among no-madic pastoralists of Turkana, Kenya—a retrospective study.Am J Phys Anthropol 98:239–255.

Gray SJ. 1996. Ecology of weaning among nomadic Turkanapastoralists of Kenya—maternal thinking, maternal behavior,and human adaptive strategies. Hum Biol 68:437–465.

Hadley C. 2003. Food security and nutritional status among twoTanzanian ethnic groups. Am J Phys Anthropol 36:107.

Heath KM, Hadley C. 1998. Dichotomous male reproductivestrategies in a polygynous human society. Mating versus pa-rental effort. Curr Anthropol 39:369–374.

Josephson SC. 2002. Does polygyny reduce fertility? Am J HumBiol 14:222–232.

Lessells CM. 1991. The evolution of life histories. In: Krebs JR,Davies NB, editors. Behavioural ecology, 3rd ed. Oxford: Black-well Scientific Publications. p 32–68.

Littell RC. 1996. SAS system for mixed models. Cary, NC: SASInstitute, Inc.

Marquis G, Habicht J, Lanata C, Black R, Rasmussen K. 1997.Association of breastfeeding and stunting in Peruvian toddlers:an example of reverse causality. Int J Epidemiol 26:349–356.

Panter-Brick C. 1998. Biological anthropology and child health:context, process and outcome. In: Panter-Brick C, editor. Bio-social perspectives on children. New York: Cambridge Univer-sity Press. p 66–95.

Pelletier DL. 1994. The relationship between child anthropome-try and mortality in developing countries—implications for pol-icy, programs and future research. J Nutr 124:2047–2081.

Schroeder DG, Brown KH. 1994. Nutritional status as a predictorof child survival: summarizing the association and quantifyingits global impact. Bull WHO 72:569–579.

Searcy WA, Yasukawa K. 1995. Polygyny and sexual selection inred-winged blackbirds. Princeton, NJ: Princeton UniversityPress.

Sellen D. 1998. Infant and young child feeding practices amongAfrican pastoralists: the Datoga of Tanzania. J Biosoc Sci 30:481–499.

Sellen D. 1999. Polygyny and child growth in a traditional pas-toral society: the case of the Datoga of Tanzania. Hum Nat10:120–147.

Sellen D. 2000. Seasonal ecology and nutritional status of womenand children in a Tanzanian pastoral community. Am J HumBiol 12:758–781.

Sellen D, Borgerhoff Mulder M, Sieff D. 2000. Fertility, offspringquality, and wealth in Datoga pastoralists: testing evolutionarymodels of intersexual selection. In: Cronk L, Chagnon N, IronsB, editors. Human behavior and adaptation: an anthropologicalperspective. New York: Aldine de Gruyter. p 91–114.

Shorr I. 1986. How to weigh and measure children: assessing thenutritional status of young children in household surveys. NewYork: United Nations Department of Technical Cooperation forDevelopment and Statistical Office.

Sieff D. 1999. The effects of wealth on livestock dynamics amongthe Datoga pastoralists of Tanzania. Agric Syst 59:1–25.

Simondon KB, Costes R, Delaunay V, Diallo A, Simondon F. 2001.Children’s height, health and appetite influence mothers’weaning decisions in rural Senegal. Int J Epidemiol 30:476–481.

Singer JD. 1998. Using SAS PROC MIXED to fit multilevel mod-els, hierarchical models, and individual growth models. J EducBehav Stat 23:323–355.

Stephensen CB. 1999. Burden of infection on growth failure. JNutr 129:534–538.

Strassmann BI. 1997. Polygyny as a risk factor for child mortalityamong the Dogon. Curr Anthropol 38:688–695.

Timaeus IM, Reynar A. 1998. Polygynists and their wives insub-Saharan Africa: an analysis of five demographic and healthsurveys. Popul Stud J Demogr 52:145–162.

Tomkins A. 1993. Environment, season and infection. In: Ulijas-zek SJ, Strickland SS, editors. Seasonality and human ecology:35th symposium volume of the Society for the Study of HumanBiology. Cambridge: Cambridge University Press. p 123–134.

Tomkins A, Watson F. 1989. Malnutrition and infection: a review.Geneva: ACC/ SCN.

Tracer DP. 1991. Fertility-related changes in maternal body com-position among the Au of Papua New Guinea. Am J PhysAnthropol 85:393–405.

Ulijaszek SJ, Strickland SS. 1993a. Nutritional anthropology:prospects and perspectives. London: Smith-Gordon.

Ulijaszek SJ, Strickland SS, editors. 1993b. Seasonality and hu-man ecology: 35th symposium volume of the Society for theStudy of Human Biology. Cambridge: Cambridge UniversityPress.

WHO. 1998. Complementary feeding of young children in devel-oping countries: a review of current scientific knowledge. Ge-neva: World Health Organization.

480 C. HADLEY