Demographic Evidence That Human Ovulation Is Undetectable (At Least in Pair Bonds) 1

465

Reports

Demographic Evidence ThatHuman Ovulation Is Undetectable(At Least in Pair Bonds)1

alexandra brewis and mary meyerDepartment of Anthropology/Department of Statistics,University of Georgia, Athens, GA 30602, U.S.A.([email protected]). 16 xi 04

Human sexual behavior has classically been portrayedas distinct among the primates. Hallmark features in-clude continuous female receptivity and an undetecta-ble2 (“concealed,” “sequestered,” “cryptic,” or “unsig-naled”) ovulation, with sexual behavior neither peakingnor even intensifying with its occurrence (Small 1996;see, e.g., Alexander and Noonan 1979, Campbell 1985,Fedigan 1986, Fisher 1982, Geary and Flinn 2001, Lovejoy1981, Pilbeam 1980, Strassman 1981). The origins andimplications of a sexuality so disentangled from repro-ductive biology demand special explanation: historicallymany models have posed the emergence and mainte-nance of undetected ovulation and constant receptivityas central to the course of Homininae and human evo-lution, including the process of hominization itself, bi-pedalism, the relative success of species, and the emer-gence of such social-behavioral strategies as moremonogamous pair-bonding, commitment to fathering,family-focused social organization, and within-group co-operation (Daly and Wilson 1983; Etkin 1963; Geary andFlinn 2001; Hrdy 1981; Lovejoy 1981; Symons 1979;Tooby and DeVore 1987; Turke 1984, 1988; Washburnand Lancaster 1968).

Critiques of the basic assumption of the singularity ofhuman sexuality in the primate order have been devel-oping for some time (e.g., Hrdy and Whitten 1987; Small1998:table 2), with growing consensus that humans mayrepresent one end of a primate continuum rather than atruly distinct form. In relation to the claim that onlyhuman females are continuously receptive, more recentfield observational studies have shown that females insome other primate species can be sexually active when

� 2005 by The Wenner-Gren Foundation for Anthropological Re-search. All rights reserved 0011-3204/2005/4603-0005$10.00

1. We thank Agustin Fuentes and Carolyn Ehardt for generouslysharing their knowledge of the primate evidence and four anony-mous reviewers for very helpful comments and suggestions.2. “Nondetected,” “undetected,” and “undetectable” have beensuggested as the most neutral and useful terms. “Hidden” or “con-cealed” semantically implies female manipulation (Agustin Fu-entes, personal communication, September 2003).

anovulatory, including during pregnancy and lactation.Under some specific social conditions in some species,such as Japanese macaques, this can reach quite highrates (see Soltis 2002:table 1 for a recent summary; alsoHeistermann et al. 2001). By contrast, increasingly fieldevidence is showing that sexual activity proliferates andusually peaks in the ovulatory phase of the fertile cyclein probably all nonhuman primate species (Strassman1999:268), including chimpanzees and bonobos (see of-ferings in Boesch, Hohmann, and Marchant 2002). Thisindicates that, from a practical, behavioral perspective,nonhuman primate sexual behavior is ovulation-linkedand detectable.

The human case for ovulation-linked sexual behaviorremains more contentious, in part because, as the oneprimate that is not publicly sexually active, we are lim-ited to demographic data to characterize it. The findingsto date are equivocal (Steklis and Whiteman 1989, Wood1994). Studies have identified proliferation or peaking ofcoitus around ovulation but also at all other phases ofthe cycle except during menses (Adams, Gould, and Burt1978; Bancroft et al. 1983; Burleson, Trevathan, andGregory 2002; McCance, Luff, and Widdowson 1937;Matteo and Rissman 1984; Spitz, Gold, and Adams 1975;Udry and Morris (1968, 1970, 1977, 1978); cf. James 1971;and see Hill 1988, Pawłowski 1999, Steklis and White-man 1989, and Wood 1994 for summaries and reviews).Other studies have found the distribution of coitus acrossthe cycle to be uniform or random in relation to ovu-lation or judged their findings ambiguous (James 1971,Morris, Udry, and Underwood 1977, Persky et al. 1978).Others interpret any identified preovulatory or posto-vulatory rise as possibly a response to menstrual absti-nence or its anticipation respectively rather than nec-essarily related to ovulation per se (Ford and Beach 1951,Gold and Adams 1981, Wood 1994).

Much of the disparity in findings is attributable tostatistical problems, specifically small sample sizes re-ducing the power and choice of tests and lack of propercontrol for confounding variables such as reproductivestatus, partner absence, rhythm, or chemical contracep-tive method use, women’s pregnancy goals and their em-pirical understanding of their own cycles, and male con-trol of sexual decision-making (Burleson, Trevathan, andGregory 2002:409; Hill 1988; Steklis and Whiteman1989; Wood 1994:312). Another problem with the exist-ing studies is that most are based on U.S. samples, pro-viding a limited basis for panhuman generalization (Hill1988, Steklis and Whiteman 1989, Wood 1994). Here wereport the findings of a study designed to overcome theseproblems. Using demographic data for 20,304 potentiallyovulating and sexually active women in stable relation-ships from 13 countries in three world regions, we testwhether coitus is more likely to occur around ovulation.

466 F current anthropology

The null hypothesis is that human coitus is randomlydistributed across the phases of the cycle excluding men-struation. If sexual behavior bears no relation to the tim-ing of ovulation and the finding is consistent across eco-logically and culturally diverse populations, then thisadds weight to the proposition that human ovulation—atleast in human pair-bonded coitus—is behaviorally un-detected, a pattern distinct from that in other primates.

Clearer demographic findings on this point are im-portant because ambiguous ones have lent themselvesto divergent evolutionary interpretations. In a recent re-consideration of the demographic literature, Pawłowskiconcludes that “intensification of sexual activity in theproliferative and ovulatory phase of the menstrual cycleconfirms the view that ovulation in humans is not com-pletely concealed” (1999:262) and therefore “assumingthe existence of direct selection pressures on sexual be-havior in the Plio/Pleistocene evolution of the Homi-ninae seems unjustified” (p. 257). Wolf agrees that “hu-man sexual behavior is not completely independent ofthe hormonal fluctuations of the menstrual cycle” andtherefore “the overall pattern of human sexual behavioris not unique among the anthropoid primates” and “re-productive biology and sexual behavior are not a productof recent sexual selection or a new reproductive strategy”(p. 269). Steklis and Whiteman (1989:430) argue that “itis premature to define what is unique about human re-productive behavior and physiology and for evolutionarymodels to assume that ‘loss of estrus’ and ‘concealedovulation’ are characteristics selected for in the courseof human evolution.” Strassman notes that “data to re-solve this debate will be difficult or impossible to obtain”(1999:268). Difficult, perhaps, not impossible.

methods

Data sets. The data provided by the most recent (post-1998, Measure DHS�) Demographic and Health Sur-veys—the product of a worldwide research project ini-tiated 20 years ago by the U.S. Agency for InternationalDevelopment—present an unprecedented opportunity totest hypotheses of coitus-ovulation relationships. De-mographers generally consider the DHS the most validand reliable source of cross-sectional data on the repro-ductive status and experience of women in developingcountries. The surveys use nationally representativehousehold sampling, sample sizes are extremely large,and the survey methods are known to be both highlyvalid and reliable. DHS� details, including the samplingstrategy, questionnaires, and data sets, are available on-line (http://www.measuredhs.com).

To improve interpretability of results we limited anal-yses to women who were theoretically at risk of ovu-lation occurring in the month in question, coitus oc-curring within the preceding 24 hours of the interview,and reporting of it when it occurred. We thus includedonly cases of women (a) who were married, (b) whosespouse was currently in residence, (c) who were not preg-nant, (d) who had menstruated within the preceding six

weeks, (e) who reported having had sex within the pre-vious 12 months and were not currently practicing post-partum or permanent abstinence, (f) who were not usingeither the rhythm method or a chemical contraceptivemethod (pill, injectable, Norplant), and (g) who were be-tween the ages of 18 and 40 years. We removed chemical-contraceptive users because if sexual behavior does re-spond to hormonal variations around ovulation, thentheoretically the effect should be hidden in women usinga method designed to inhibit ovulation (James 1971). Wealso removed all cases in which women reported thatthey had begun menstruating on the day of the interviewor the preceding one because we cannot determinewhether the event should be properly classified to a pre-menstrual or a menstrual cycle phase.

The selection of the specific DHS� country data setsto be included in this study was based on three criteria:availability in early 2003, availability and identical cod-ing of the variables of interest, and a sample exceeding700 once it was limited as described above. This resultedin inclusion of data sets from Armenia (1,761), Benin(960), Ethiopia (1,638), Guinea, (860), Guatemala (918),Haiti (1037), Khazakhstan (1,343), Mali (2,247), Malawi(1,865), Nigeria (1,183), Nepal (2,220), Peru (3,502), andRwanda (770).

Study variables. The dependent variable was whethercoitus had occurred within the preceding 24 hours (1) ornot (0). The time frame was limited to one day becauseinaccurate recall is known to be a particular problem insexual behavior studies (e.g., James 1971). The indepen-dent variable was the woman’s current cycle phase, cat-egorized on the basis of women’s reports of time sincemenstrual bleeding last began. With the first day of men-strual bleeding defined as day 1, cycle days 2–5 wereclassified as menstrual, days 6–11 as preovulatory (fol-licular), days 12–17 as ovulatory, and days 18 � aspostovulatory (luteal). Basic reproductive research showsthat ovulation is extremely likely to occur between cycledays 12 and 17 if women’s cycles are between 26 and 31days in length: median cycle length for human popula-tions across diverse ecologies is in the range of 29–30days, and the mode is around 27–28 days (Wood 1994).Median reported cycle day for this sample was 14, whichindicates that average cycle length would be around 28days and so within this expected range.

Covariates included in the model to account for var-iance in the relationship between occurrence of coitusand cycle phase are woman’s age, partner’s age, country,breast-feeding status, union type (monogamous or po-lygamous), number of children, woman’s immediate re-productive goals, her empirical knowledge of the exis-tence and timing of cyclic variation in pregnancy risk,whether she is currently using contraception (as notedabove, chemical and rhythm method users are alreadyexcluded), and whether she identifies mid-cycle as themost fertile time. This last variable is included to providesome control for women’s facultative adjustments intheir sexual behavior based on their understanding of themenstrual cycle. Covariates were entered categoricallyinto the model.

Volume 46, Number 3, June 2005 F 467

table 1Odds Ratio Estimates and Their Confidence Intervals for Variable of Interest and Se-lected Covariates as Significant Predicators of Occurrence of Coitus within the Past Day

Variable of Interestand Covariatesa Odds Ratio Estimate

95% Wald ConfidenceInterval P

Cycle phasePreovulatory vs. postovulatory 1.019 0.934–1.112 NSOvulatory vs. postovulatory 1.026 0.939–1.122 NSOvulatory vs. preovulatory 0.993 0.899–1.097 NSMenses vs. postovulatory 0.666 0.601–0.738 ∗∗∗

Menses vs. preovulatory 0.654 0.585–0.731 ∗∗∗

Menses vs ovulatory 0.649 0.601–0.738 ∗∗∗

Breast-feeding statusCurrently breast-feeding vs.

not breast-feeding0.821 0.752–0.896 ∗∗∗

AgeUnder 21 vs. over 35 years 1.502 1.261–1.790 ∗∗∗ b

21–25 vs. over 35 years 1.214 1.053–1.401 –26–30 vs. over 35 years 1.231 1.092–1.388 –31–35 vs. over 35 years 1.158 1.038–1.292 –

Partner’s ageUnder 31 vs. over 50 years 1.279 1.050–1.557 ∗∗∗ b

31–40 vs. over 50 years 1.077 0.903–1.285 –41–50 vs. over 50 years 1.033 0.864–1.234 –

Number of childrenNo living children vs. 4 or

more1.780 1.525–2.077 ∗∗∗ b

1 child vs. 4 or more 1.274 1.121–1.448 –2 or 3 children vs. 4 or more 1.115 1.012–1.228 –

Currently wishes to becomepregnant vs. not

1.147 1.045–1.259 ∗∗

Knows mid-cycle is fertile timevs. not

0.989 0.912–1.074 NS

Using barrier contraception vs.not

0.908 0.825–1.000 ∗

Polygamous vs. monogamousunion

0.868 0.776–0.971 ∗

note: NS, not significant (p 1 0.05), ∗∗∗ p ! 0.001, ∗∗ p ! 0.01, ∗ p ! 0.05.aCovariates of women’s country were highly significant predictors of odds of the occurrence of coitus (p !

0.001); we have not included those results here because they are voluminous.bThe significant differences by age, partner’s age and number of children indicate that there is at least onesignificantly different categoric comparison but does not indicate which one(s).

It is possible that male control of sexual timing damp-ens or eliminates cyclic shifts in female proceptivity,receptivity, attractivity, responsiveness, or sexual satis-faction in relation to ovulation (Burleson, Trevathan, andGregory 2003, Clayton et al. 1999, Gold and Adams 1981,Hedricks et al. 1994, Slob et al. 1996). To address thiswe included two “negotiation” covariates that reflectedwhether a woman reported that she felt any control ofsexual timing. One related to whether it was acceptableto refuse sex when tired or not in the mood and the otherwhether she believed a husband was justified in hittinghis wife if she refused to have sex. These data were avail-able for only 7 and 6 of the 13 countries respectively,and the model including these variables was run for justthis subsample.

Analyses. We used logistic regression predicting theoccurrence of coitus within the previous day by cyclestage. This solved the problem of specifying a distribu-tion of the dependent variable, because, while the dis-tribution is unlikely to be linear, it could also be incor-

rect to impose a quadratic or sinusoidal shape. Inreporting findings the maximum likelihood estimates ofthe model coefficients do not have particularly intutiverelationships and therefore can be difficult to interpret(Hosmer and Lemeshow 2000). Consequently, we usedthe odds ratios comparing the odds of the occurrence ofcoitus between any two cycle phases. A 95% confidenceinterval was applied: if the confidence interval contained1.0, then there was no significant difference between cat-egories. All analyses were done using SAS version 8.2(SAS Institute, 1999–2001).

results

Odds ratio estimates and confidence intervals for thecomparisons of coital occurrence by cycle phase and allcovariates are presented in table 1, and odds ratios forthree comparisons across the four cycle phases overalland by country are shown graphically in figures 1, 2, and3. Given the very large sample size and control for po-


Fig. 1. Odds ratio for the comparison of occurrence of coitus in ovulatory vs. preovulatory cycle phases for 13countries and overall. The samples are ordered from left to right by increasing sample size. RW, Rwanda; GN,Guinea; GU, Guatemala; BJ, Benin; H, Haiti; NG, Nigeria; KK, Kazakhstan; ET, Ethiopia; AM, Armenia; MW,Malawi; NP, Nepal; ML, Mali; PE, Peru; Overall, all country samples combined.

Fig. 2. Odds ratio for the comparison of occurrence of coitus in ovulatory vs. postovulatory cycle phases for 13countries and overall.

tentially confounding variables, confidence intervals areextremely small, and the power of all tests thereforeproved very high. Contrary to our predictions, in the totalsample odds ratios comparing the occurrence of coitusin the preovulatory, ovulatory, and postovulatory phaseswere not statistically different. These same results wereobserved for every country sample with just one excep-tion (1 in 20 results would be expected to be spuriouswhen hypotheses are being tested at the 95% confidenceinterval). There is, by contrast, a significant difference

in the odds of the occurrence of coitus during mensescompared with other cycle phases (see table 1 and figure3): for example, in the total sample the odds ratio of theoccurrence of coitus in the menstrual vs. the preovula-tory phase is 0.654 with a 98% confidence interval(0.585–0.731).

Table 1 shows that most covariates were highly sig-nificant predictors of the occurrence of coitus. Currentlybreast-feeding, being older, having an older partner, andhaving more children were all highly significantly pre-


Fig. 3. Odds ratio for the comparison of occurrence of coitus in menses vs. postovulatory cycle phases for 13countries and overall.

dictors of lower odds of reporting coitus. Wanting to be-come pregnant and monogamous marriage were mod-erate predictors of higher odds of coitus. Currentcontraceptive use was of borderline significance (but re-member that this includes only those using barrier andtraditional methods). The only covariate that did notprove significant was that concerning whether womencould identify the mid-cycle as the most fertile time.3

When we ran the model separately on the basis of thenegotiation variables, we found that the overall patternsfor the two groups were very similar to those of the totalsample. As can be seen in table 2, we observe no signif-icant variation in the odds of the occurrence of coitus ina particular nonmenstrual cycle phase in terms of per-ception of control of sexual timing.

discussion and conclusion

The question of whether human sexual behavior occursrandomly with respect to ovulation or is patterned by itis significant. Primate evolutionary models require aproper comparative species characterization of humans.Many of our standard narratives pose the emergence ofundetected ovulation as central to our evolutionary his-tory, but it has been very difficult to determine empir-ically whether this assumption is valid.

Examining a very large and diverse data set of womenin stable unions, we find no evidence of any variation inthe occurrence of coitus according to cycle phase except

3. We could not specify the exact ovulatory status of each womanat the time of her interview, but large sample size allowed us torepeat the analyses using a subsample made up of only those womenwho could be classified to cycle phases with the most certainty.When the same model was developed using the constrained cate-gories of cycle phase (women at days 2–3 vs. 8–11 vs. 14–16 vs.20–27), the results were the same.

that it is much reduced during menses. The high signif-icance of so many covariates shows that the model issufficiently powerful to detect differences in the occur-rence of coitus by cycle phase if indeed it exists. In evo-lutionary terms, this suggests that human ovulation isbehaviorally undetectable, at least in pair-bonded rela-tionships.

This raises a further issue: undetectable by whom? Evo-lutionary models variously propose that human ovulationis undetectable by both men and women (e.g., Geary andFlinn 2001) or perhaps detectable to some degree bywomen while “hidden” from men (e.g., Small 1996). Ouranalyses support the former. We draw this suggestion fromthe finding that when we differentiate women who havethe means and motivation to make facultative adjust-ments to the timing of intercourse across the cycle (i.e.,have technically correct knowledge about ovulation andstated pregnancy goals) and then those who perceive sex-ual timing as negotiable, the findings are the same in bothcases for other women who do not.4 This does not precludethe possibility of mid-cycle elevations in human femalesexual mood, but analysis of this sample indicates thatthese may have little practical influence on the basic pat-terning of marital coitus.5 Perhaps behavioral cyclicity

4. Sixty-seven percent of Cornell undergraduates accurately identifythe mid-cycle as the time ovulation occurs (Small 1996). Smallinterprets this as evidence that extant humans can detect ovulation,although it could also be argued to show that surprisingly manycannot, even when provided with the very best bio-scientific ed-ucation. Although rhythm-method users were excluded from thisstudy, we also ran the model including only those women and foundthat their odds of coitus across different cycle phases were notsubstantially different from those of other women.5. Some even suggest that cyclic variations in desire and arousalindependent of ovulation are evidence of the ability of human fe-males to manipulate males through nonreproductive sex (Thiessen1997). Our findings do not contradict this possibility.


table 2Odds Ratio Estimates and Their Confidence Intervals for Variable of Internet as Significant Pre-dictors of Occurrence of Coitus within the Past Day by Perceived Control of Sexual Timing

Perceived Control of SexualTiming and Cycle Phase Odds Ratio Estimate 95% Wald Confidence Interval P

Acceptable to refuse sex whentired or not in the moodPostovulatory vs. preovulatory 1.067 0.870–1.309 NSOvulatory vs. postovulatory 1.086 0.939–1.253 NSOvulatory vs. preovulatory 1.206 0.962–1.513 NS

Not acceptable to refuse sexwhen tired or not in the moodPostovulatory vs. preovulatory 1.067 0.870–1.309 NSOvulatory vs. postovulatory 1.130 0.922–1.386 NSOvulatory vs. preovulatory 1.206 0.962–1.513 NS

Wife-beating for refusing tohave sex not justifiedPostovulatory vs. preovulatory 0.992 0.840–1.772 NSOvulatory vs. postovulatory 1.159 0.982–1.368 NSOvulatory vs. preovulatory 1.150 0.954–1.386 NS

Wife-beating for refusing tohave sex justifiedPostovualtory vs. preovulatory 1.039 0.822–1.314 NSOvulatory vs. postovulatory 0.980 0.770–1.246 NSOvulatory vs. preovulatory 1.018 0.781–1.328 NS

note: NS, not significant. Seven countries for which the “mood” variable is available (Armenia, Benin, Haiti,Mali, Malawi, Peru, Rwanda) are combined in the model, and the total sample is 11,931. Six countries for whichthe “wife-beating” variable is available (Armenia, Benin, Haiti, Mali, Malawi, Rwanda) are combined in the model,and the total sample is 8,996. (All co-variates used in the model presented in table 1 were included in the analysesbut are not given here because the results were similar.)

would be more evident in nonmarital matings (short-term,extrapair, or nonheterosexual), in which the evolutionarycontext of sexual behavior should be quite different. Theseare not captured in this data set.

In summary, while ovulation-linked proliferations ofhuman sexual behavior may occur under some socio-eco-logical conditions, this study provides improved demo-graphic confirmation that heterosexual human pair-bonded sexual behavior may be appropriately charac-terized by an absence of proliferation in sexual activity inconcert with ovulation—that is, by undetectable ovula-tion. A limitation of this study is that we rely on women’sreports of menstrual-cycle day as a proxy for classifyingwomen by cycle phase. Ideally, longitudinal, biologicallybased indicators would be used to identify precisely thetiming of ovulation for each woman. It is not currentlypossible to combine such biological data collection witha sample size sufficient to allow this level of statisticalconclusiveness, but this is what will be required to de-termine whether our sexuality is utterly and uniquely dis-entangled from our reproductive biology.

References Citeda d a m s , d . b . , a . r . g o u l d , a n d a . d . b u r t . 1978.

Rise in female-initiated sexual activity at ovulation and itssuppression by oral contraceptives. New England Journal ofMedicine 299:1145–50.

a l e x a n d e r , r i c h a r d d . , a n d k a t h l e e n m .n o o n a n . 1979. “Concealment of ovulation, parental care,

and human social evolution,” in Evolutionary biology and hu-man social behavior. Edited by Napoleon Chagnon and Wil-liam Irons, pp. 436–53. North Scituate: Duxbury Press.

b a n c ro f t , j . , d . s a n d e r s , d . d a v i d s o n , a n d p .w a r n e r . 1983. Mood, sexuality, hormones, and the men-strual cycle, 3: Sexuality and the role of androgens. Psychoso-matic Medicine 45:509–16.

b o e s c h , c h r i s t o p h e , g o t t f r i e d h o h m a n n , a n dl i n d a m a r c h a n t . Editors. 2002. Behavioural diversity inchimpanzees and bonobos. Cambridge and New York: Cam-bridge University Press.

b u r l e s o n , m a ry, w e n d a t r e v a t h a n , a n d w.l a r ry g r e g o ry. 2002. Sexual behavior in lesbian and het-erosexual women: Relations with menstrual cycle phase andpartner availability. Psychoneuroendocrinology 27:489–503.

c a m p b e l l , b e r n a r d . 1985. 3d edition. Human evolution.Chicago: Aldine.

c l a y t o n , a n i t a , g a i l c l a v e t , e l i z a b e t h m cg a rv e y, j u l i a w a r n o c k , a n d k . w e i s s . 1999. As-sessment of sexual functioning during the menstrual cycle.Journal of Sex and Marital Therapy 25:281–91.

d a l y, m a r t i n , a n d m a r g o w i l s o n . 1983. 2d edition.Sex, evolution, and behavior. Boston: Willard Grant Press.

e t k i n , w i l l i a m , 1963. Social behavioral factors in the emer-gence of man. Human Biology 35:299–310.

f e d i g a n , l i n d a . 1986. The changing role of women in mod-els of human evolution. Annual Review of Anthropology 15:25–66.

f i s h e r , h e l e n . 1982. The sex contract: The evolution of hu-man behavior. New York: William Morrow.

f o r d , c l e l l a n , a n d f r a n k b e a c h . 1951. Patterns ofsexual behavior. New York: Harper and Row.

g e a ry, d a v i d , a n d m a r k fl i n n . 2001. Evolution of hu-man parental behavior and the human family. Parenting: Sci-ence and Practice 1:5–61.

g o l d , a . r . , a n d d . b . a d a m s . 1981. Motivational fac-


tors affecting fluctuations of female sexual activity at menstru-ation. Psychology of Women Quarterly 5:670–80.

h e d r i c k s , c y n t h i a , l i n d a p i c c i n i n o , j . r i c h a r du d ry, a n d t o m t . k . c h i m b i r a . 1987. Peak coital ratecoincides with onset of luteinizing hormone surge. Fertilityand Sterility 48:234–38.

h e i s t e r m a n n , m i c h a e l , t h o m a s z i e g l e r , c a r e l p .v a n s c h a i k , k r i s t i n l a u n h a r d t , p a u l w i n k l e r ,a n d j . k e i t h h o d g e s . 2001. Loss of oestrus, concealedovulation, and paternity confusion in free-ranging Hanumanlangurs. Proceedings of the Royal Society of London B 268(1484):2445–51.

h i l l , e l i z a b e t h m . 1988. The menstrual cycle and compo-nents of human female sexual behavior. Journal of Social andBiological Structures 11:443–55.

h o s m e r , d a v i d , a n d s t a n l e y l e m e s h o w. 2000. 2dedition. Applied logistic regression. New York: Wiley.

h r d y, s a r a h b . 1981. The woman that never evolved. Cam-bridge: Harvard University Press.

h r d y, s a r a h b . , a n d p . l . w h i t t e n . 1987. “Patterningof sexual activity,” in Primate societies. Edited by BarbaraSmuts, D. L. Cheney, R. M. Seyfarth, R. W. Wrangham, and T. T.Struhsaker, pp. 370–84. Chicago: University of Chicago Press.

j a m e s , w i l l i a m h . 1971. The distribution of coitus withinthe human intermenstrum. Journal of Biosocial Science 3:159–71.

l o v e j o y, o w e n . 1981. The origin of man. Science 211:341–50.

m c c a n c e , r . a . , m . c . l u f f , a n d e . e . w i d d o w -s o n . 1937. Physical and emotional periodicity in women.Journal of Hygiene 37:571–611.

m a t t e o , s . , a n d e . f . r i s s m a n . 1984. Increased sexualactivity during the midcycle portion of the human menstrualcycle. Hormones and Behavior 18:249–55.

m o r r i s , n a o m i , j . r i c h a r d u d ry, a n d l o u i s u n -d e r w o o d . 1977. A study of the relationship between coitusand the luteinizing hormone surge. Fertility and Sterility 28:440–42.

p a w ł o w s k i , b o g u s ł a w. 1999. Loss of oestrus and con-cealed ovulation in human evolution: The case against the sex-ual selection hypothesis. current anthropology 40:257–75.

p e r s k y, h . , n . c h a r n e y, h . i . l i e f , c . p . o ’ b r i e n ,w. r . m i l l e r , a n d d . s t r a u s s . 1978. The relationshipof plasma estradiol level to sexual behavior in young women.Psychosomatic Medicine 40:523–35.

p i l b e a m , d a v i d . 1980. “Major trends in human evolution,”in Current argument on early man. Edited by L. K. Konigsson,pp. 261–85. Oxford: Pergamon.

s l o b , a . k . , c . m . b a x , w. c . h o p , d . l . ro w l a n d ,a n d j . j . v a n d e r w e r f f t e n b o s c h . 1996. Sexual ar-ousability and the menstrual cycle. Psychoneuroendocrinology21:545–58.

s m a l l , m e r e d i t h . 1988. Female primate sexual behaviorand conception: Are there really sperm to spare? current an-thropology 29:81–100.

———. 1996. “Revealed” ovulation in human? Journal of HumanEvolution 30:483–88.

s o l t i s , j o s e p h . 2002. Do primate females gain non-procrea-tive benefits by mating with multiple males? Theoretical andempirical considerations. Evolutionary Anthropolgoy 11:187–97.

s p i t z , c . , a . g o l d , a n d d . a d a m s . 1975. Cognitive andhormonal factors affecting coital frequency. Archives of SexualBehavior 4:249–63.

s t e k l i s , h o r s t , a n d c a t h e r i n e w h i t e m a n . 1989.Loss of estrus in human evolution: Too many answers, too fewquestions. Ethology and Sociobiology 10:417–34.

s t r a s s m a n , b e v e r l y. 1981. Sexual selection, paternal care,and concealed ovulation in humans. Ethology and Sociobiology2:31–40.

———. 1999. Comment on: Loss of oestrus and concealed ovula-tion in human evolution, by Bogusław Pawłowski. currentanthropology 40:268.

s y m o n s , d o n a l d . 1979. The evolution of human sexuality.Oxford: Oxford University Press.

t h i e s s e n , d . 1997. False estrus in human females: Sexual ma-nipulation without conception. Perceptual and Motor Skills84:385–86.

t o o b y, j o h n , a n d i rv e n d e v o r e . 1987. “The recon-struction of hominid behavioral evolution through strategicmodeling,” in Primate models of hominid behavior. Edited byW. Kinzey, pp. 183–237. Albany: SUNY Press.

t u r k e , p a u l . 1984. Effects of ovulatory concealment and syn-chrony on protohominid mating systems and parental roles.Ethnology and Sociobiology 5:33–44.

———. 1988. “Concealed ovulation, menstrual synchrony, andpaternal investment,” in Biosocial perspectives on the family.Edited by E. E. Filsinger, pp. 119–36. Newbury Park: Sage.

u d ry, j . r i c h a r d , a n d n a o m i m o r r i s . 1968. Distribu-tion of coitus in the menstrual cycle. Nature 220:593–96.

———. 1970. Effect of contraceptive pills on the distribution ofsexual activity in the menstrual cycle. Nature 227:502–3.

———. 1977. Distribution of events in human menstrual cycle.Journal of Reproduction and Fertility 51:419–25.

———. 1978. Pheromonal influences on human sexual behavior:An experimental search. Journal of Biosocial Science 10:147–57.

w a s h b u r n , s h e r w o o d , a n d c h e t l a n c a s t e r . 1968.“The evolution of hunting,” in Man the hunter. Edited byRichard Lee, pp. 293–303. Chicago: Aldine.

w o o d , j a m e s w. 1994. Dynamics of human reproduction:Biology, biometry, demography. Hawthorne, N.Y.: Aldine deGruyter.

Local Resource Enhancement andSex-biased Breastfeeding in aCaribbean Community1

robert j . quinlan, marsha b. quinlan, andmark v. f l innDepartment of Anthropology, Ball State University,Muncie, IN 47306-0435, U.S.A. ([email protected]).(M. Quinlan and R. Quinlan)/Department ofAnthropology, University of Missouri, Columbia, MO65211, U.S.A. (Flinn). 17 xi 04

[Supplementary material appears in the electronic edition of thisissue on the journal’s web page (http://www.journals.uchicago.edu/CA/home.html).]

Parents often treat sons and daughters differently. Boysare favored in some societies (Messer 1997), but in others


1. We thank the people of Bwa Mawego for their good-natured tol-erance of our presence in their community. D. Shillingford, R. Na-siiro, P. Rickets, Sister D. James, and the Dominica Ministry ofHealth provided valuable assistance. Research was supported bygrants from the National Science Foundation (BNS 8920569, SBR9205373, and SBR 0136023), the University of Missouri ResearchBoard (to MVF), and the Earthwatch Center for Field Research (toMVF, MBQ, and RJQ).


table 1Studies Showing Sex-biased Breastfeeding

Sex Bias Country/Culture PI Hypothesis Reference

Female Switzerland – Bouvier and Rougemont (1998)Female Hungary, Gypsy TW∗ /LRE∗ Bereczkei and Dunbar (1997, 2002)Female Kenya, Mukogodo TW∗ Cronk (2000)Female U.S., Hutterites TW/LRE∗ Margulis, Altmann, and Ober (1993)Female Dominica TW Quinlan, Quinlan and Flinn (2003)Female Australia – Scott et al. (1999)Female United Kingdom – Whitehead, Paul, and Ahmed (1986)Female Norway – Pande, Unwin, and Haheim (1997)Male Northern Thailand – Jackson et al. (1992)Male India – Nath and Goswami (1997), Rao and Kanade (1992)Male Bangladesh – Brown et al. (1982)Male Egypt – Ahmed (1990)Depends on father’s status Poland TW∗ Koziel and Ulijaszek (2001)

note: PI, parental investment; TW, Trivers-Willard; LRE, Local resource enhancement. Results consistent with the hypothesis areindicated by ∗ .

girls receive preferential treatment including better med-ical care and more frequent breastfeeding (Cronk 2000).Cross-culturally, breastfeeding is a major component ofearly parental care that directly affects child health. Herewe examine female-biased breastfeeding in Bwa Ma-wego, a horticultural community in the Commonwealthof Dominica, and test hypotheses from parental invest-ment theory about the reasons for such bias (reviewedin Clutton-Brock and Godfray 1991). Multiple linear re-gression suggests that sex differences in children’s timeallocation to productive activity account for female-bi-ased breastfeeding in this community. Ethnographic datafrom Bwa Mawego suggest that female-biased maternalcare results from male socioeconomic marginality andits effects on sex-specific risks of parental investment.

Breastfeeding is a primary mechanism among mam-mals for provisioning dependent offspring and is a keycomponent of parental investment—care benefiting oneoffspring at a cost to parents’ ability to invest in othercomponents of fitness (Clutton-Brock 1991:9).2 Compo-nents of fitness include well-being of existing offspring,parents’ future reproduction, and “indirect fitness”through aid to kin (see Hamilton 1964, Beatty 1992).Nursing can be an important building block of themother-child bond, associated with positive emotionsand intense attachment that are linked to maternal hor-mones including prolactin and oxytocin (Ellison 2001:83–126). Because of its reliability as a mechanism forbonding, it may have evolved regulatory functions af-fecting other aspects of human parental investment(Hrdy 1999). Its infant health advantages are well known(Oddy 2001) and it may also protect against diseases ap-pearing in adulthood (Cunningham 1995). Other benefitsto children’s long-term cognitive and psychomotor de-velopment are apparent (Horwood, Darlow, and Mo-gridge 2001, Feldman and Eidelman 2003, Mortensen et

2. ”Fitness” is defined as an individual’s genetic representation infuture generations.

al. 2002, Pollock 1994, Vestergaard et al. 1999) and mayhave long-term fitness consequences.

For mothers, lactation entails energetic and opportu-nity costs and suppresses fertility (Valeggia and Ellison2001, Tracer 1996, Vitzthum 1994). The energetics ofhuman lactation requires 670 kcal/day during exclusivebreastfeeding (Dewey 1997), which may be 50% of awoman’s total energy budget in undernourished popu-lations (Valeggia and Ellison 2001:86). Increased energydemand may tax household production and delay futurereproduction through reduced fecundity among under-nourished mothers (Jasienska 2001). Opportunity costsmay also be substantial (Quinlan, Quinlan, and Flinn2003). Nursing can interfere with work that benefitsother children or kin and direct maternal care for olderoffspring. Finally, length and frequency of breastfeedingsessions are associated with postpartum fecundity, pos-sibly through effects on gonadotrophin-releasing hor-mone (Vitzthum 1997). Such “structural” costs of breast-feeding (associated with nursing-session length,frequency, and age at weaning) should be expressed infuture reproduction or “residual reproductive value.”

In general, the costs of lactation for mother’s futurereproduction easily fit within a life-history framework(see Hill and Kaplan 1999, Roff 2002). Delayed repro-duction is “discounted” in growing populations becauselater-born offspring represent a smaller proportion of thetotal population and, hence, reduced fitness. Delay alsoincreases the likelihood that a woman will die before thebirth of her next child. Costs to existing offspring aremore complicated to model in terms of residual repro-ductive value, but formal life-history models can be ex-tended to include special aspects of parental investment(Hill 1993). In sum, we expect the timing of weaning toreflect a trade-off between benefits to offspring and coststo mother, which among humans may be influenced byfamily cooperation.

Sex-biased breastfeeding is evident in many humanpopulations (table 1). Evolutionary theory predicts that


sex-biased parental investment will be adaptive in cer-tain circumstances. In general, if parental expendituresin one sex yield higher fitness than expenditures in theother sex, then parents should bias investment towardthe former. Socioecological conditions may affect sex-specific fitness returns on parental investment. Matingsystem, competition, and cooperation among kin are im-portant factors (see Chagnon, Flinn, and Melancon 1979,Clutton-Brock 1991, and Sieff 1990 for reviews).

One evolutionary hypothesis for sex-biased parentalinvestment posits that mothers in poor condition (usu-ally measured in terms of socioeconomic status) maybias investment toward daughters while mothers in goodcondition bias investment toward sons (Trivers and Wil-lard 1973). This prediction is logical if males have morereproductive variance than females and if offspring re-production is sensitive to parental resources. Parents ingood condition provide more resources, leading to po-tentially higher fitness through sons, and parents in poorcondition provide fewer resources, leading to higher fit-ness through daughters. Among humans, males oftenhave higher reproductive variance than do females, andcorrelates of reproductive success (wealth, social com-petence, etc.) may be sensitive to parental influence(Geary and Flinn 2001).

Recent studies of sex-biased breastfeeding show con-flicting results for a test of the Trivers-Willard hypoth-esis. (Studies of nonhuman primates [Brown 2001] andother mammals [Hewison and Gaillard 1999] also pro-duce conflicting results for such tests.) Among the rel-atively low-status Mukogodo of Kenya, girls were morelikely than boys to be observed nursing (Cronk 2000).Compared with ethnic Hungarians, low-status Gypsiesweaned girls later than boys (Bereczkei and Dunbar1997). In Poland, Koziel and Ulijaszek (2001) found aweak paternal-status # child’s-sex interaction effect onbreastfeeding duration but no maternal-status # sex in-teraction. Research in the contemporary United Statesfound no status # child’s-sex interaction effect on ageat weaning (Keller, Nesse, and Hofferth 2001), and astudy of Hutterites concludes that female-biased breast-feeding cannot be attributed to a Trivers-Willard effect(Margulis, Altmann, and Ober 1993).

Absence of the effect in some populations may be dueto local factors that alter the sex-specific costs of children(Brown 2001, Sieff 1990). If girls help parents (or relatives)more than boys do, then parents may bias investmenttoward daughters because expenditures on them are lesscostly to other components of fitness. Conversely, if sonsare more helpful, then we expect male-biased care. In-deed, among foragers, sex differences in subsistence con-tributions are associated with the juvenile sex ratio, in-dicating sex-biased parental investment (Hewlett 1991:23–28). Male-biased parental investment among theInuit, for example, may be related to hunting and malecontributions to family well-being (Smith and Smith1994). If reproduction is constrained by maternal work,then investment biased toward the more helpful sex canincrease parental fitness. Among Hungarian Gypsies, forexample, girls take on child-care duties, apparently in-

creasing their mothers’ fertility, and are weaned laterthan boys (Bereczkei and Dunbar 2002). Similarly, in ru-ral Dominica girls spend more time in household workthan boys (Quinlan 1995), and this may offset the costsof investing in girls. Sex-specific repayment of parentalcare has been dubbed “local resource enhancement” (Go-waty and Lenartz 1985). Here we test the hypothesis thatfemale-biased breastfeeding in rural Dominica is attrib-utable to sex differences in children’s time allocation toproductive activities.

methods

Data were collected in the village of Bwa Mawego from1993 to 1999 as part of an ongoing study of family en-vironment and child health in rural Dominica begun in1987 (see Flinn and England 2003; Quinlan and Flinn2003, 2005; Quinlan 2003). Age at weaning was deter-mined through retrospective health interviews con-ducted in 1993, 1994, 1997, and 1999 and cross-checkedby review of individual clinic health cards when avail-able (Quinlan 2004). Although the weaning process usu-ally begins with the introduction of solid food, here wedefine weaning as the complete cessation of breastfeed-ing. Cessation of breastfeeding often marks the point atwhich mothers are no longer willing or able to continuenursing because of other demands on their time and at-tention. Completed weaning is therefore a good indicatorof shifts in parental investment behavior. Another ad-vantage of this measure is that, in contrast to that forother infant feeding transitions, recall of complete ces-sation of breastfeeding is quite reliable (Quandt 1987).Weaning ages recorded on clinic health cards were highlycorrelated with mother’s recall (r p .91). (Quandt [1987]reported a similarly high correlation (r p .89) betweensuch measures in another population.) Children’s timeallocation to productive activity was measured using in-stantaneous behavioral scans. The method was a mod-ified version of “spot-check” behavioral sampling (Flinn1988, Hames 1992, Johnson and Sackett 1998, Quinlan1995). The sample included 1,074 observations of 21 boysand 22 girls between age 5 and 16 years living in twohamlets of Bwa Mawego.

Breastfeeding duration is the child’s age in months atweaning (complete cessation of breastfeeding). Sex iscoded as 1 p male, 0 p female. Time allocation to pro-ductive activity is the percentage of observations of achild coded as child care, chores, and subsistence activ-ities. Age is the mean of a child’s age in years in thesummers of 1993 and 1994 and is included as a controlvariable because children’s time allocation to productiveactivity is a linear function of age (Quinlan 1995).

Multiple linear regression with breastfeeding durationas the dependent variable was performed in two steps.In the first model control variables for age and fatherpresence were entered with sex to estimate sex-biasedbreastfeeding duration. Other studies in the communityhave shown that father presence is associated with ageat weaning (Quinlan, Quinlan, and Flinn 2003) and chil-


dren’s time allocation to productive activity (Quinlanand Flinn 2003). Children’s time allocation to productiveactivity and an age # sex interaction term entered thesecond model to test the hypothesis that sex differencesin children’s productive activity mediate sex-biasedbreastfeeding duration: If children repay parental invest-ment through household productive activity, then thereshould be little association between sex and breastfeed-ing duration after time allocation to productive activityenters the equation. The age # sex term was necessaryto control for sex differences in the association betweenage and time allocation to productive activity. All otherinteractions were tested, and nonsignificant terms wereexcluded from the final analysis. Alpha was set at .10 tocompensate for reduced power due to the small sample.Increased alpha is justified by the directional hypothesesfor female-biased breastfeeding and the predicted posi-tive association between breastfeeding duration and timeallocation to productive activity.

ethnographic setting

The Commonwealth of Dominica is a small, rural islandnation located between Guadeloupe and Martinique(15�N, 61�W). The island is mountainous and relativelyundeveloped. Dominica’s population of 65,000 is ofmixed African, European, and Island Carib descent. MostDominicans are bilingual in English and French-Patois.Bwa Mawego is one of the least-developed villages onthe island. It has approximately 700 full- and part-timeresidents. Economic activities include subsistence agri-culture, fishing, and, for a few villagers, wage labor orpenny capitalism. Cash crops from smallholdings are animportant source of income for many families. Averageannual household income in Bwa Mawego is approxi-mately $5,000 E.C. (US$1,850). (For more detailed de-scriptions see Quinlan, and Flinn 2005, Quinlan 2004.)

Mean surviving offspring for a completed-fertility sam-ple (130 males and 124 females born between 1900 and1955) is 4.4 for men (95% CI p 3.8–5.1) and 5.0 forwomen (95% CI p 4.4–5.6), suggesting a low-fertilitybut growing population. Multigenerational data show nodifference in the number of descendents producedthrough sons or daughters. Adult sex-ratios are balancedfor these individuals (Quinlan and Flinn 2005).

Breastfeeding for the first three months is nearly uni-versal in rural Dominica. Physicians and nurses in healthcenters recommend exclusive breastfeeding for at least6 months, conforming to World Health Organization(WHO) recommendations. Many women comply, in partbecause commercial baby formulas are expensive andhard to obtain except in the capital and a few large vil-lages. There is little social stigma attached to breast-feeding, and nursing in public is traditional. Babies some-times drink a mixture of water and arrowroot starch,occasionally fortified with condensed cow’s milk, butthis formula is recognized as inferior to breast milk andis used mainly as a supplement. The mean age for in-troducing food (porridge or bananas) is 6.5 months witha median of 6 months (rated “good” by WHO [2003] stan-

dards). Mean age at completed weaning is 12.5 monthswith a median of 12 months (Quinlan, Quinlan, andFlinn 2003) (rated “poor”). This is slightly higher thanin other Caribbean populations (Perez-Escamilla 1994),albeit lower than in many societies (Dettwyler 1995, Sel-len 2001). Mothers in Bwa Mawego expressed variousreasons for weaning their children, but most gave nospecific reason for the precise timing. Some simply re-ported that a doctor or nurse had instructed them tobreastfeed for at least 6 months and they had tried tocomply. When mothers did give specific reasons forweaning, the most common were breastfeeding’s inter-ference with work and an insufficient milk supply (Quin-lan, Quinlan, and Flinn 2003). Women living with a long-term mate tended to nurse about 4.5 months longer thansingle mothers, and women living with multiple adultfemale kin weaned their children relatively early. Age atweaning was negatively associated with householdwealth. These associations are apparently due to differ-ences between families in women’s work that interfereswith nursing (Quinlan, Quinlan, and Flinn 2003).

Finally, three tests of the Trivers-Willard hypothesisfailed to support its predictions (Quinlan, Quinlan, andFlinn 2003). Our predictions of male-biased breastfeedingamong (1) women living with their mother or adult sis-ters, (2) women living with their children’s father, and(3) women from wealthier households were not sup-ported in multiple regression analyses. These analysesdid show that girls were weaned on average about 3.5months later than boys.

Women and their daughters in Bwa Mawego do mostof the family’s domestic chores, and women typicallyspend more time in productive activity than men (53%versus 30% of daylight hours [Quinlan 1995]). Mothersare primarily responsible for child care, but grandmoth-ers, sisters, and older daughters often help. Girls aged8–10 years frequently take some responsibility for caringfor younger siblings. Girls’ time spent in productive ac-tivity is a linear function of age (r p .80, between 1 and16 years), and by about 15 years most girls have a fullyadult time allocation pattern. Men are responsible formore periodic tasks such as repairing buildings, collect-ing large amounts of firewood, and hauling bay leaf. Ingeneral men spend 20% of their time in productive ac-tivity until they enter a long-term conjugal union, whentime allocation to production increases to about 40%(Quinlan 1995). Differences between married men’s andwomen’s time allocation to production may reflectgreater energetic demands for men’s work. Boys, how-ever, spend less time in productive activity and moretime playing than girls (Quinlan 1995), and their pro-ductive activity does not significantly increase with age(r p .12, between 1 and 16 years).

Household composition is associated with children’stime allocation. Number of adults in the household andtime spent with mother are negatively associated withchildren’s time spent roaming the village. Additionally,children living with their father spend significantly moretime in productive activity than father-absent children(Quinlan and Flinn 2003).


Fig. 1. Boxplots showing sex difference in (A) breastfeeding duration (BFD), (B) time allocation to productiveactivity (TAPA), and (C) age for 22 girls and 21 boys. The horizontal line in the center of each box is the me-dian. Upper and lower segments of the box indicate the 25th and 75th percentiles respectively. Small squaresindicate outliers.

results

Descriptive statistics show sex differences and potentialconfounding variables. Girls were weaned later than boys(fig. 1, A; one-tail p p .042 by Mann-Whitney test),which is consistent with previous research in a largersample from this community (Quinlan, Quinlan, andFlinn 2003). Girls also spent more time in productiveactivity than boys (fig. 1, B; one-tailed p p .003). Therewas no sex difference in age, however (fig. 1, C; two-tailed p p .971), nor was age significantly associated withbreastfeeding duration, although a nonsignificant trendwas apparent (r p �.258, two-tailed p p .113; fig. 2, A).Age was correlated with time allocation to productiveactivity for girls (r p .750, one-tailed p ! .001) but notfor boys (r p .002, one-tailed p p .497; see fig. 2, B).Finally, productive activity was positively correlatedwith breastfeeding duration (r p .362, one-tailed p p.001; fig. 2, C).

Regression diagnostics suggested problematic outliers.Descriptive statistics revealed that breastfeeding dura-tion and time allocation to productive activity were sig-nificantly skewed (breastfeeding skewness p 2.04, 90%CI p 1.45–2.634; time allocation skewness p 0.70, 90%CIX0.11–1.29), which is likely with small samples. Plotsof residuals and diagnostic statistics identified four po-tentially influential outliers, two brothers and two sis-ters from one family (shown in fig. 1, A). The sisters (aged7 and 11 years) were weaned at 38 and 42 months andwere engaged in productive activity in 29% and 25% ofobservations. Their two brothers (aged 13 and 14 years)were weaned at 24 months and were engaged in produc-tive activity in 0% and 8% of observations; hence, thisfamily with unusually late weaning conformed to thepredictions of the repayment hypothesis. When the fouroutliers were removed from the sample, diagnostic anal-yses indicated approximately normal errors and nonsig-nificant skew. Four sibling outliers point to potentialhierarchical family effects. Thirty-one children in thereduced sample also had at least one sibling in the sam-

ple. An additional multilevel analysis, however, showedlittle clustering of breastfeeding duration within fami-lies. Further inspection of the raw data showed no ob-vious pattern of residuals.

Two multiple regression models tested the hypothesisthat local resource enhancement among females was as-sociated with sex-biased parental investment. Sex wasassociated with breastfeeding duration in the first re-gression model, which included age and father presenceas control variables. Mean breastfeeding duration wasabout 3.5 months longer for girls, which is comparableto differences reported in Quinlan, Quinlan, and Flinn(2003:9). The second model tested the repayment hy-pothesis by entering children’s time allocation to pro-ductive activity as a predictor. If including time alloca-tion to productive activity appreciably decreases theassociation between sex and breastfeeding duration, thenfemale-biased breastfeeding duration is associated withgirls’ greater contribution to household production,which is consistent with the hypothesis. Children’s timeallocation to productive activity was positively associ-ated with breastfeeding duration (table 2; fig. 3, A) andmediated the association between breastfeeding durationand sex. Child’s sex was not significantly associated withbreastfeeding duration after productive activity enteredthe model (table 2; fig. 3, B), indicating that sex-biasedparental investment in the form of breastfeeding wasassociated with girls’ contribution to household labor.Age # sex was included as a control variable becausegirls’ time allocation to productive activity was stronglyassociated with age and boys’ was not (fig. 2, B). If age# sex entered model 1 without time allocation to pro-ductive activity, then sex was still significantly associ-ated with breastfeeding duration and age # sex was not.Conversely, if time allocation to productive activity en-tered without age # sex, then time allocation was sig-nificant and sex was not. This pattern shows that theage # sex interaction was significant because of its as-sociation with time allocation to productive activity (fig.


Fig. 2. Scatter plots showing bivariate correlations between variables in the multiple regression analysis.White squares, girls; black squares, boys. Lines show linear association. A, a non-significant correlation be-tween age and breastfeeding duration (BFD) (r p �.26). B, age # sex interaction for time allocation to produc-tive activity (TAPA): solid line, linear association for girls; dashed line, association for boys. Age is significantlyassociated with girls’ (r p .75) but not boys’ (r p .002) TAPA. C, a significant association between BFD andTAPA (r p .36).

2, B). Other interactions (including sex # time allocationto productive activity) were not significant and were ex-cluded from the final model.3 In sum, children’s resi-dualized time allocation to productive activity ac-counted for 29% of the variance in residualized breast-feeding duration, and sex accounted for nonsignificantvariance.

In a separate analysis including the four outliers,breastfeeding duration and time allocation to productiveactivity were recoded as 20% Winsorized variables,which may increase accuracy and power (Wilcox 2001).The Winsorized mean for breastfeeding duration of 10.6months was lower than the non-Winsorized mean of 12.4but consistent with age at weaning in other Caribbeanpopulations (Perez-Escamilla 1994). Diagnostic analysesindicated normally distributed errors using the completeWinsorized sample (N p 43). Winsorized regressionyielded more conservative estimates, but the pattern wassimilar to that of the reduced sample (table 2). Initiallysex was significantly associated with breastfeeding du-ration, but the association was not significant after pro-ductive activity entered the model.

Sample size analysis demonstrated that a significantassociation between sex and breastfeeding duration of bp �.16 adjusted for covariates (table 2) would require asample of 450� children (alpha p .05 and power p .95),which is impossible given the community size and theconstraints of formal direct observation. Given theseconstraints, it is appropriate to consider that the asso-ciation between sex and breastfeeding was largely me-diated by productive activity.

3. Here interaction terms are the product of two first-order variablescentered around their means (e.g., age # sex p [age � meanage]∗[sex � mean sex]). Interaction terms did not substantially in-crease variance inflation factors.

discussion

The results suggest that mothers in Bwa Mawego maymake weaning “decisions” anticipating that childrenwill repay parental investment through household work.The completed weaning variable in this well-nourishedlow-fertility population indicates maternal costs ofbreastfeeding primarily as lost opportunities to engagein other activities.

Other research in Bwa Mawego showed potential in-direct fitness benefits from daughter-biased parental in-vestment. Reproductive success was higher among menfrom kin groups with a female-biased sex ratio, thoughthe effect may be due to competition among male kinrather than female resource enhancement (Quinlan andFlinn 2005). Number of sisters in the community, how-ever, influenced neither male nor female fitness, sug-gesting that kinswomen’s mutual aid is diffuse among abroad kin group. Previous research did indicate that hav-ing sisters coresident in a household had effects on com-ponents of female fitness (Quinlan 2001). In sum, indi-rect fitness gains in Bwa Mawego may be rooted inreciprocity among mothers, daughters, and sisters.

Weaning decisions may not be entirely conscious, andthe psychological mechanisms for sex-biased parental in-vestment are unknown. Some informants claim to giveequal treatment to their boys and girls; however, wean-ing data for children of those informants show female-biased breastfeeding duration by as much as five months.(Senior [1991:33] reports similar unstated female bias inseveral Caribbean populations.) Other village motherssay that, compared with boys, “girl-babies are easy.” Cul-turally patterned relations among and between sexesmay affect willingness to continue nursing a particularchild.


table 2Multiple Linear Regression Showing Effects of Sex and Time Allocation to Productive Activity on Age atWeaning

B SE b t P

Breastfeeding duration (N p 39)Model 1 (adjusted R2 p 16, P p .028)

Intercept 16.13 2.62 – 6.22 0.000Age �0.47 0.23 �0.316 �2.09 0.044Father present 1.49 1.39 0.162 1.07 0.291Sex �3.53 1.38 �0.384 �2.56 0.015

Model 2 (adjusted R2 p .38, P p .001)Intercept 16.02 2.30 – 6.31 0.000Age �0.90 0.23 �0.603 �4.00 0.000Father present 0.17 1.27 0.019 0.14 0.892Sex �1.47 1.36 �0.160 �1.08 0.289Time allocation to productive activity 0.25 0.07 0.702 3.68 0.001Age # sex 1.18 0.45 0.400 2.61 0.014

Winsorized breastfeeding duration (N p 43)Model 1(adjusted R2 p .17, P p .017)

Intercept 12.92 1.73 – 7.45 0.000Age �0.24 0.15 �0.230 �1.61 0.115Father present 2.24 0.94 0.341 2.39 0.022Sex �2.05 0.92 �0.317 �2.24 0.031

Model 2 (adjusted R2 p .28, P p 004)Intercept 11.55 1.80 – 6.41 0.000Age �0.37 0.15 �0.345 �2.49 0.018Father present 1.33 0.96 0.203 1.39 0.174Sex �1.10 1.01 �0.169 �1.09 0.283Winsorized time allocation to productive activity 0.16 0.07 0.438 2.45 0.019Age # sex 0.69 0.30 0.331 2.27 0.029

note: B, unstandardized regression coefficient in months of breastfeeding duration; SE, standard error of B; b, standardized re-gression coefficient; t,B/SE; P, significance; R2, proportion of variance in breastfeeding duration explained by the model. Log-transformed breastfeeding duration was regressed on variables in model 2 in an additional analysis (not shown) that included theoutliers and yielded similar results to the Winsorized regression.

Family life in Bwa Mawego is strongly matrifocal. Aselsewhere in the Caribbean, reciprocity between mothersand daughters often forms the core of stable family re-lations. Households in Bwa Mawego are usually linkedthrough women. Villagers often recognize a man as theowner of a house (e.g., “down past Roger’s house”), butthey identify households with women (e.g., “ she stayswith Margaret and them”). Female kin organize and carryout economic enterprises and maintain reciprocal childcare arrangements that may entail costs to reproduction(Quinlan 2001).

In contrast, relations between Caribbean men andwomen are sometimes antagonistic (Senior 1991:166–84). Caribbean mothers also appear to be stricter andmore demanding with daughters than with sons, whichmay reflect indifference as well as indulgence towardsons (cf. Senior 1991:33–36; Moses 1985). These rela-tionship qualities may be due to unbalanced reciprocitybetween the sexes (e.g., Barrow 1999:72–79). Extreme andunpredictable variability in men’s contribution to fam-ilies suggests local motivation for daughter-biased in-vestment. Some men in Bwa Mawego contribute signif-icantly to household wealth, but access to resourcesthrough wage labor and commercial agriculture is lim-ited and unreliable. In impoverished Caribbean areas,skilled men often emigrate, and those who stay behind

may have little earning potential or pursue risky illegaltrade (Williams 2002). Further, men’s income is oftenchanneled into luxury items and alcohol. Rum drinking,a common problem among village men, interferes withproductivity and family relations. Lack of opportunityand widespread alcohol abuse can make a man unreliableand a periodic drain on family material and emotionalresources. Stable conjugal families exist, but thosehouseholds often develop into a matrifocal variant asdaughters are unwilling (or unable) to enter a neolocalor patrilocal conjugal relationship (Quinlan and Flinn2003). Sons of respected village families are not guar-anteed to escape the risks of Caribbean social ecology.Male economic and social marginality (perhaps associ-ated with antagonistic relations between the sexes) is apart of life that village women cannot ignore, and thismay lead them to bias their attention and energy to morereliable relationships among female kin. We propose,then, that local environmental risks that can be reducedthrough parental care (see Borgerhoff Mulder 1992, Quin-lan 2003) may be gender-specific and therefore affect sex-biased parental investment. In general, environmentalunpredictability associated with sex differences in re-productive value tends to reduce sex-biased parental in-vestment (West and Sheldon 2002), but gender culturesamong humans (see Helman 2001:109–14) allow for dif-


Fig. 3. A, partial regression plot showing the association between breastfeeding duration (BFD) and time allo-cation to productive activity (TAPA) after both variables have been adjusted (“residualized”) for the other vari-ables in table 2, model 2 (r2 p .291). B, sex differences in residualized BFD. Comparison of B with figure 1, A,shows that sex-biased breastfeeding is statistically mediated by children’s time allocation to productiveactivity.

ferent levels of predictability for men and women. Thispossibility is not a variant of the Trivers-Willard hy-pothesis (no gradient of parental condition is necessary)but rather illustrates the complexity of parental invest-ment in humans.

An alternative interpretation is that sex-biased breast-feeding and children’s time allocation covary in ways thatmask a Trivers-Willard effect. For example, women at oneend of a resource gradient may be more likely to havedaughters, breastfeed them longer, and demand more workfrom them. Inspection of the raw data, however, showsthat female-biased breastfeeding is evident in many fam-ilies with balanced sex ratios. Rather, female-biased pa-rental investment appears to be related to expected sexdifferences in contributions to family well-being.

Studies in other populations indicate complex rela-tions among socioeconomic status, household produc-tivity, and sex-specific costs of children that may resultin misleading conclusions. For example, one study com-paring Gypsies with ethnic Hungarians found that fe-male-biased parental investment among Gypsies wasconsistent with the Trivers-Willard model, predicting fe-male-biased investment in low-status families (Bere-czkei and Dunbar 1997). Subsequent research, however,found that Gypsy girls spent more time engaged in childcare than ethnic Hungarians and that low-birth-orderdaughters decreased mothers’ interbirth interval and in-creased fertility, consistent with local resource enhance-ment (Bereczkei and Dunbar 2002). An apparent Trivers-Willard effect may, then, be due to different patterns ofchildren’s time allocation in the two ethnic groups.

Multiple factors may work to create patterns of pa-rental investment that are difficult to tease apart. Forexample, among Gabbra pastoralists of Kenya, male re-productive success is more sensitive to household wealththan female reproductive success, males with olderbrothers tend to have reduced fertility, and females repay

a portion of parental investment through bride-price(Mace 1996). Three models of sex allocation (Trivers-Willard, resource enhancement, and resource competi-tion [see Sieff 1990]) apparently account for different fac-ets of parental investment in one population. Thecross-cultural trend for girls to contribute more to do-mestic work than boys (Whiting and Edwards 1988) sug-gests, again, that parental investment strategies may bedifficult to observe in many populations, and domesticwork is just one factor that may influence sex-specificcosts of children. Despite the possibility of multiple in-fluences on sex-biased parental investment, previous re-search in Bwa Mawego found female-biased breastfeed-ing inconsistent with a Trivers-Willard effect (Quinlan,Quinlan, and Flinn 2003) or its variant the “advantageddaughter” hypothesis (see Hiraiwa-Hasegawa 1993), butthe findings reported here are consistent with local re-source enhancement.

This study has limitations, and the small sample war-rants caution. The results, however, address previouslyunexplained female-biased breastfeeding in Bwa Mawegoand justify further research and replication in similar com-munities. Factors influencing gender relations and paren-tal care in Dominica (poverty, male marginality, and co-operative female kin) may be common to many ruralcommunities in developing nations—precisely wherebreastfeeding may be most important for child well-being.

In sum, evolutionary ecology offers new avenues forunderstanding gender relations and child well-being inthe Caribbean and elsewhere. This study shows that girlsspend more time in productive activity than boys andthat this difference statistically mediates sex-biasedbreastfeeding in Bwa Mawego, a horticultural village inthe Commonwealth of Dominica. Empirical advancesconcerning parental investment and components of fit-ness await research into locally determined householdlabor demands, residence patterns, relations among kin


groups, gender differences in environmental risks, andsex-specific costs of children. Such research depends onlongitudinal multimethod, multilevel analyses.

References Cited

a h m e d , f . a . 1990. Gender difference in child mortality.Egyptian Population and Family Planning Review 24(2):60–79.

b a r ro w, c h r i s t i n e . 1999. Family in the Caribbean. Prince-ton: Markus Wiener.

b e a t t y, j o h n . 1992. “Fitness: Theoretical contexts,” in Keywords in evolutionary biology. Edited by E. F. Keller and E. A.Lloyd, pp. 115–19. Cambridge: Harvard University Press.

b e r e c z k e i , t . , a n d r . i . m . d u n b a r . 1997. Female bi-ased reproductive stategies in a Hungarian Gypsy population.Proceedings of the Royal Society of London, Series B: Biologi-cal Sciences 264:17–22.

———. 2002. Helping-at-the-nest and sex-biased parental invest-ment in a Hungarian Gypsy population. current anthropol-ogy 43:804–9.

b o r g e r h o f f m u l d e r , m o n i q u e . 1992. “Reproductive de-cisions,” in Evolutionary ecology and human behavior. Editedby Eric A. Smith and Bruce Winterhalder, pp. 339–74. NewYork: Aldine de Gruyter.

b o u v i e r , p . , a n d a . ro u g e m o n t . 1998. Breast-feedingin Geneva: Prevalence, duration, and determinants. Sozial- undPraventivmedizin 43:116–23.

b ro w n , g . r . 2001. Sex-biased investment in nonhuman pri-mates: Can Trivers and Willard’s theory be tested? Animal Be-havior 61:683–94.

b ro w n , k . h . , r . e . b l a c k , s . b e c k e r , s . n a h a r ,a n d s . s a w y e r . 1982. Consumption of foods and nutrientsby weanlings in rural Bangladesh. American Journal of ClinicalNutrition 36:878–89.

c h a g n o n , n a p o l e o n , a . , m a r k v. f l i n n , a n d t .m e l a n c o n . 1979. “Sex ratio variations among the Yano-mamo Indians,” in Evolutionary biology and human social be-havior: An anthropological perspective. Edited by N. A. Chag-non and W. Irons, pp. 290–320. Boston: Duxbury Press.

c l u t t o n - b ro c k , t . h . 1991. The evolution of parentalcare. Princeton: Princeton University Press.

c l u t t o n - b ro c k , t . h . , a n d c . g o d f r a y. 1991. “Pa-rental investment,” in Behavioural ecology: An evolutionaryapproach. Edited by J. R. Krebs and N. B. Davies, pp. 234–62.Boston: Blackwell.

c ro n k , l e e . 2000. “Female-biased parental investment andgrowth performance among the Mukogodo,” in Adaptationand human behavior: An anthropological perspective. Editedby L. Cronk, N. Chagnon, and W. Irons, pp. 203–21. New York:Aldine de Gruyter.

c u n n i n g h a m , a . s . 1995. “Breastfeeding: Adaptive behaviorfor child health and longevity,” in Breastfeeding: Bioculturalperspectives. Edited by P. Stuart-Macadam and K. A. Dettwy-ler, pp. 243–63. New York: Aldine.

d e t t w y l e r , k a t h e r i n e a . 1995. “A time to wean: Thehominid blueprint for the natural age of weaning in modernhuman populations,” in Breastfeeding: Biocultural perspec-tives. Edited by P. Stuart-Macadam and K. A. Dettwyler, pp.39–73. New York: Aldine.

d e w e y, k a t h ry n g . 1997. Energy and protein requirementsduring lactation. Annual Review of Nutrition 17:19–36.

e l l i s o n , p e t e r t . 2001. On fertile ground: A natural historyof human reproduction. Cambridge: Harvard University Press.

f e l d m a n , ru t h , a n d a r t h u r i . e i d e l m a n . 2003. Di-rect and indirect effects of breast milk on the neurobehavioraland cognitive development of premature infants. Developmen-tal Psychobiology 43(2):109–19.

f l i n n , m a r k v. 1988. Step and genetic parent/offspring rela-tionships in a Caribbean village. Ethology and Sociobiology 9:

335–69.f l i n n , m a r k v. , a n d b a r ry g . e n g l a n d . 2003.“Child-

hood stress: Endocrine and immune responses to psychosocialevents,” in Social and cultural lives of immune systems. Ed-ited by James M. Wilce, pp. 107–47. London: Routledge.

g e a ry, d a v i d , a n d m a r k v. fl i n n . 2001. Evolution ofhuman parental behavior and the human family. Parenting:Science and Practice 1:5–61.

g o w a t y, p a t r i c i a a . , a n d m . r . l e n a r t z . 1985. Sexratios of nestling and fledgling red-cockaded woodpeckers (Pi-coides borealis) favor males. American Naturalist 126:347–53.

h a m e s , r a y m o n d . 1992. “Time allocation,” in Evolutionaryecology and human behavior. Edited by E. A. Smith and B.Winterhalder, pp. 203–36. New York: Aldine de Gruyter.

h a m i l t o n , w i l l i a m d . 1964. The genetical evolution ofsocial behavior. Journal of Theoretical Biology 7:1–52.

h e l m a n , c e c i l g . 2001. 4th edition. Culture, health, andillness. New York: Oxford University Press.

h e w i s o n , a . j . , a n d j . m . g a i l l a r d . 1999. Successfulsons or advantaged daughters? The Trivers-Willard model andsex-biased maternal investment in ungulates. Trends in Ecol-ogy and Evolution 14:229–34.

h e w l e t t , b a r ry s . 1991. Demography and childcare in prein-dustrial societies. Journal of Anthropological Research 47(1):1–37.

h i l l , k i m . 1993. Life history theory and evolutionary anthro-pology. Evolutionary Anthropology 2(3):78–88.

h i l l , k i m , a n d h i l l a r d k a p l a n . 1999. Life historytraits in humans: Theory and empirical studies. Annual Re-view of Anthropology 28:397–430.

h i r a i w a - h a s e g a w a , m . 1993. Skewed birth sex ratios inprimates: Should high-ranking mothers have daughters orsons? Trends in Ecology and Evolution 8:395–400.

h o r w o o d , l . j . , b . a . d a r l o w, a n d n . m o g r i d g e .2001. Breast milk feeding and cognitive ability at 7–8 years.Archives of Disease in Childhood: Fetal and Neonatal Edition84:F23–27.

h r d y, s a r a h b . 1999. Mother Nature: A history of mothers,infants, and natural selection. New York: Pantheon Books.

j a c k s o n , d . a . , s . m . i m o n g , l . w o n g s a w a s d i i , a .s i l p r a s e r t , s . p r e u n g l a m p o o , p . l e e l a p a t , r . f .d r e w e t t , k . a m a t a y a k u l , a n d j . d . b a u m . 1992.Weaning practices and breast-feeding duration in northernThailand. British Journal of Nutrition 67:149–64.

j a s i e n s k a , g r a z y n a . 2001. “Why energy expenditurecauses reproductive suppression in women: An evolutionaryand bioenergetic perspective,” in Reproductive ecology and hu-man evolution. Edited by P. T. Ellison, pp. 59–84. Hawthorne,N.Y.: Aldine de Gruyter.

j o h n s o n , a l a n , a n d ro s s s a c k e t t . 1998. “Direct sys-tematic observation of behavior,” in Handbook of methods incultural anthropology. Edited by H. R. Bernard, pp. 301–31.Walnut Creek: AltaMira Press.

k e l l e r , m . c . , r . m . n e s s e , a n d s . h o f f e r t h . 2001.The Trivers-Willard hypothesis of parental investment: No ef-fect in the contemporary United States. Evolution and HumanBehavior 22:343–60.

k o z i e l , s . , a n d s . j . u l i j a s z e k . 2001. Waiting for Triv-ers and Willard: Do the rich really favor sons? American Jour-nal of Physical Anthropology 115:71–79.

m a c e , ru t h . 1996. Biased parental investment and reproduc-tive success in Gabbra pastoralists. Behavioral Ecology and So-ciobiology 38:75–81.

m a r g u l i s , s . w. , j . a l t m a n n , a n d c . o b e r . 1993.Sex-biased lactational duration in a human population and its re-productive costs. Behavioral Ecology and Sociobiology 32:41–45.

m e s s e r , e l l e n . 1997. Intra-household allocation of food andhealth care: Current findings and understandings, introduction.Social Science and Medicine 44:1675–84.

m o r t e n s e n , e . l . , k . f . m i c h a e l s e n , s . a . s a n d -e r s , a n d j . m . r e i n i s c h e . 2002. The association be-tween duration of breastfeeding and adult intelligence. Journalof the American Medical Association 287:2365–71.

m o s e s , y o l a n d a t . 1985. “Female status, the family, and


male dominance in a West Indian community,” in The blackwoman cross-culturally. Edited by Filomina Chioma Steady,pp. 499–514. Cambridge: Schenkman.

n a t h , d . c . , a n d g . g o s w a m i . 1997. Determinants ofbreast-feeding patterns in an urban society of India. Human Bi-ology 69:557–73.

o d d y, w. h . 2001. Breastfeeding protects against illness andinfection in infants and children: A review of the evidence.Breastfeeding Review 9:11–18.

p a n d e , h . , c . u n w i n , a n d l . l . h a h e i m . 1997. Factorsassociated with the duration of breastfeeding. Acta Paediatrica86:173–77.

p e r e z - e s c a m i l l a . r . 1994. Breastfeeding in Africa and theLatin American and Caribbean region: The potential role of ur-banization. Journal of Tropical Pediatrics 40:137–43.

p o l l o c k , j . i . 1994. Long-term associations with infant feed-ing in a clinically advantaged population of babies. Develop-mental Medicine and Child Neurology 36:429–40.

q u a n d t , s a r a a . 1987. Maternal recall accuracy for dates ofinfant feeding transitions. Human Organization 46:152–60.

q u i n l a n , m a r s h a b . 2004. From the bush: The front line ofhealth care in a Caribbean village. Belmont,Calif.: Wadsworth.

q u i n l a n , ro b e r t j . 1995. Father-absence, maternal care,and children’s behavior in a rural Caribbean village. M.A. the-sis, University of Missouri, Columbia, Mo.

———. 2001. Effect of household structure on female reproduc-tive strategies in a Caribbean community. Human Nature 12:69–89.

———. 2003. Father absence, parental care, and female reproduc-tive development. Evolution and Human Behavior 24:376–90.

q u i n l a n , ro b e r t j . , a n d m a r k v. f l i n n . 2003. Inter-generational transmission of conjugal stability in a Caribbeancommunity. Journal of Comparative Family Studies 43:569–83.

———. 2005. Kinship, sex, and fitness in a Caribbean commu-nity. Human Nature. In press.

q u i n l a n , ro b e r t j . , m a r s h a b . q u i n l a n , a n dm a r k v. f l i n n . 2003. Parental investment and age at wean-ing in a Caribbean village. Evolution and Human Behavior 24:1–16.

r a o , s . , a n d a . n . k a n a d e . 1992. Prolonged breastfeedingand malnutrition among rural Indian children below 3 years ofage. European Journal of Clinical Nutrition 46:187–95.

ro f f , d e r e k . 2002. Life history evolution. Sunderland, Mass.Sinauer Associates.

s c o t t , j . a . , a . a i t k i n , c . w. b i n n s , a n d r . a .a ro n i . 1999. Factors associated with duration of breastfeedingamongst women in Perth, Australia. Acta Paediatrica 88:356–58.

s e l l e n , d a n i e l w. 2001. Comparison of infant feeding pat-terns reported for nonindustrial populations with current rec-ommendations. Journal of Nutrition 131:2707–15.

s e n i o r , o l i v e . 1991. Working miracles: Women’s lives in theEnglish-speaking Caribbean. Bloomington: Indiana UniversityPress.

s i e f f , d a n i e l a f . 1990. Explaining biased sex ratios in hu-man populations: A critique of recent studies. current an-thropology 31:25–35.

s m i t h , e r i c a l d e n , a n d s . a b i g a i l s m i t h . 1994. In-uit sex-ratio variation. current anthropology 35:595–624.

t r a c e r , d a v i d p . 1996. Lactation, nutrition, and post-par-tum amenorrhea in a lowland Papua New Guinea. Human Bi-ology 68:277–92.

t r i v e r s , ro b e r t l . , a n d d a n e . w i l l a r d . 1973. Nat-ural selection of paternal ability to vary the sex-ratio of off-spring. Science 179:90–92.

v a l e g g i a , c l a u d i a r . , a n d p e t e r t . e l l i s o n . 2001.“Lactation, energetics, and postpartum fecundity,” in Repro-ductive ecology and human evolution. Edited by P. T. Ellison,pp. 85–105. Hawthorne, N.Y.: Aldine de Gruyter.

v e s t e r g a a r d , m . , c . o b e l , t . b . h e n r i k s e n , h . t .s o r e n s e n , e . s k a i a a , a n d j . o s t e r g a a r d . 1999.Duration of breastfeeding and developmental milestones duringthe latter half of infancy. Acta Paediatrica 88:1327–32.

v i t z t h u m , v i r g i n i a j . 1994. Causes and consequences of

heterogeneity in infant feeding practices among indigenous An-dean women. Annals of the New York Academy of Sciences709:221–24.

———. 1997. “Flexibility and paradox: The nature of adaptationin human reproduction,” The evolving female: A life historyperspective. Edited by M. E. Morbeck, A. Galloway, and A. L.Zihlman, pp. 242–59. Princeton: Princeton University Press.

w e s t , s . a . , a n d b . c . s h e l d o n . 2002. Constraints inthe evolution of sex ratio adjustment. Science 295:1685–88.

w h i t e h e a d , r . g . , a . a . p a u l , a n d e . a . a h m e d .1986. Weaning practices in the United Kingdom and variationsin anthropometric development. Acta Paediatrica Scandinav-ica (suppl.) 323:14–23.

w h i t i n g , b e a t r i c e b l y t h , a n d c a ro l y n p . e d -w a r d s . 1988. Children of different worlds: The formation ofsocial behavior. Cambridge: Harvard University Press.

w h o ( w o r l d h e a l t h o r g a n i z a t i o n ) . 2003. Infant andyoung child feeding: A tool for assessing national practices,policies, and programmes. Geneva.

w i l c o x , r a n d r . 2001. Fundamentals of modern statisticalmethods: Substantially improving power and accuracy. NewYork: Springer.

w i l l i a m s , s i a n . 2002. “The effects of structural adjustmentprograms on the lives of children in Jamaica,” in Globalizationand children. Edited by N. H. Kaufman and I. Rizzini, pp.151–60. New York: Kluwer Academic/Plenum Publisher.

Human Migrations in ContinentalEast Asia and Taiwan: Genetic,Linguistic, and ArchaeologicalEvidence

peter bellwood and alicia sanchez-mazasSchool of Archaeology and Anthropology, AustralianNational University, Canberra, ACT 0200, Australia([email protected])/Department of Anthro-pology and Ecology, University of Geneva, 12, rue deGustave-Revilliod, CH-1227 Carouge, Geneva, Switzer-land ([email protected]). 23 xii 04

A multidisciplinary conference held in Geneva in June20041 brought together geneticists, linguists, and ar-chaeologists to debate a wide range of hypotheses con-cerning the origins and early dispersal histories of a num-ber of major human populations in East and SoutheastAsia. This conference followed a predecessor entitled“The Peopling of East Asia,” held in Perigueux in 2001


1. Held at the University of Geneva, June 10–13, 2004, thanks toa joint grant of the Swiss National Science Foundation and theNational Science Council of Taiwan (# IB3220-101318) to AliciaSanchez-Mazas and Marie Lin. The conference was also supportedby the Santa Fe Institute (U.S.A.), by Mackey Memorial Hospital,Council of Aboriginal People, Ministry of Foreign Affairs and Min-istry of Education (Taiwan), and by the University of Geneva andthe Academic Society of Geneva (Switzerland).


and now published as Sagart, Blench, and Sanchez-Mazas(2004).

The concept of an ancient “population” differs from onediscipline to another. Comparative linguists are concernedwith populations of related languages, and human beingsfigure for them precisely to the extent that they can beattached to languages as their users and transmittersthrough space and time. Archaeological populations com-prise connected groupings of material culture, but ar-chaeologists are often instrumental in the recovery ofskeletal populations, the most direct survivals of any an-cient biological subgroup of mankind. The populationsstudied by geneticists are mostly still living and may begeographically or linguistically defined. To what extentany groups defined by geography, language, archaeology,and skeletal and genetic biology in the past may have beencorrelated with each other is, of course, an issue best de-bated independently for each individual case.

Unfortunately, much recent literature reveals that thescope for terminological confusion and disagreementamongst those who try to transcend the boundaries be-tween the data sets of these intersecting yet independentdisciplines is very large, as was pointed out in their pres-entations at the conference by Roger Blench, FrancoisJacquesson, and Murray Gell-Mann. One obvious way toreduce such confusion is to bring together the practition-ers of the different disciplines so that they can explainjust what they do and how they come to their conclusions.In the past decade, this kind of multidisciplinary gatheringhas become rather fashionable, a major example (albeitwith a thematic rather than, as here, a geographical focus)being the collection of archaeological, linguistic, and ge-netic papers on the farming/language-dispersal hypothesisedited by Bellwood and Renfrew (2003).

In the Geneva conference, a number of East Asian pop-ulations were discussed, either as categories based on thelanguages they speak (the basis of a majority of presen-tations) or as geographical groupings.

austronesian- and tai-speaking peoples

As a linguistic category, the Austronesian languages havea history of dispersal from Taiwan through the Philip-pines into Island Southeast Asia and on to Oceania andMadagascar. Malcolm Ross explained why the majorityof linguists accept Taiwan as the Proto-Austronesianhomeland and in what directions the ancestral languagesspread and emphasized their transmission through in-heritance rather than language shift, implying that theAustronesian language dispersal was associated with anactual movement of Austronesian-speaking people. Henoted the phylogenetic strength of the Proto-Malayo-Pol-ynesian and Proto-Oceanic nodes in Austronesian pre-history, each representing a relative standstill in disper-sal associated with an accumulation of discrete inno-vations, followed in turn by rapid geographical radiationthrough Island Southeast Asia and western Oceania re-spectively. Paul Li presented internal Formosan evidencesupporting the view that the linguistic ancestors of the

Formosan peoples came to Taiwan in one event and thendiscussed possible chronologies for the separations of themajor Formosan subgroups within Taiwan itself. (For in-stance, eastern Taiwan may have been settled from thesouthwestern coast between 4000 and 3000 BP.)

A dissenting linguistic opinion by Ilia Peiros doubtedthe importance of Taiwan in Austronesian prehistory,questioning the phonological innovations which otherstake as part of the evidence for the Malayo-Polynesiansubgroup. Peiros suggested that (1) there is little lexicalevidence that the Austronesian languages of Taiwanform several groups rather than just one and (2) there are20 or more distinct Malayo-Polynesian groups in thewestern area of the family rather than a single “WesternMalayo-Polynesian” group.

Laurent Sagart presented what was perhaps the mostunexpected historical suggestion of the meeting—thatthe Tai-Kadai (or Daic) languages descend in cousinlyfashion with Malayo-Polynesian from an eastern For-mosan (i.e., Austronesian) ancestor. This is contrary toreceived wisdom that sees the Tai languages as derivedfrom southern China and unrelated to Austronesian. Sa-gart suggested that the Proto-Malayo-Polynesian (Extra-Formosan) numerals seven (∗ pitu), eight (∗ walu), and nine(∗ Siwa) were derived from older additive forms of whichtraces survive in some western Formosan languages, thisobservation giving strong support to the recognition ofTaiwan, specifically northwestern Taiwan, as the loca-tion of Proto-Austronesian. Because cognate numeralsand other forms occur in Buyang, a Tai-Kadai languagein northern Vietnam, Sagart suggested that the Tai-Kadaifamily as a whole originated in northeastern Taiwanfrom an Austronesian homeland region very closely re-lated in time and space to that which gave rise to theMalayo-Polynesian languages. Roger Blench likewiseagreed on possible Malayo-Polynesian linkages (via thenorthern Philippines?) for early Tai. Li Jin, Bo Wen, andHui Li presented Y-chromosome data in support of aclose relationship between Formosan- and Tai-speakers,seemingly supporting the linguistic observations of Sa-gart. A Formosan-Tai linkage is perhaps in need now ofsome archaeological commentary, and this is an issuethat is likely to generate considerable multidisciplinaryinterest in the future.

Alicia Sanchez-Mazas compared HLA markersamongst Austronesians and other East Asian popula-tions, concluding that Formosans are rather more dif-ferentiated than Extra-Formosan Austronesians—thisperhaps supporting the notion of a greater Formosan timedepth. Pointing to GM haplotype variation and linkagedisequilibrium amongst HLA loci in Formosan popula-tions, Sanchez-Mazas also reported that genetic diversitydecreases while allelic association increases from thewest coast to the south and east regions of Taiwan. Un-less western populations have been subjected to a greaterlevel of gene flow from continental Asia, Formosan pop-ulations had a western origin, as predicted by Sagart’slinguistic theory based on numerals, and lost diversityas they moved south and east.

Jean Trejaut discussed mtDNA haplotypes in Formo-


san populations, reinforcing the observation that the an-cestral form of the “Polynesian motif” mtDNA haplo-type was transmitted through Taiwan during theHolocene, perhaps within chronological range of theAustronesian dispersal within the past 6,000 years (seealso Cox n.d.). Erika Hagelberg pointed out that mtDNAand HLA polymorphisms support an “express train”model of Austronesian dispersal (i.e., relatively rapidmovement from China and Taiwan), whereas Y-chro-mosomes indicate a much higher Melanesian compo-nent in the Polynesian gene pool and thus a higher degreeof male population admixture.

These observations, reflecting long-term and enor-mously widespread population relationships, are not tobe confused with evidence on a more local scale thatmight indicate gender-based trends in postmarital resi-dence and patterns of endogamy versus exogamy. Socialissues of this type were discussed by Steven Lansing andTatiana Karafet, who are analysing Y-chromosome mi-crosatellites amongst Balinese to try to identify pasttrends towards increasing endogamy. Rake phylogeniesare taken as clues to high levels of past endogamy, thetopic of interest here being the origins of the Balinesesubak cooperating groups concerned with rice field ir-rigation management. Subaks promote endogamy be-yond the level expected for less specialized corporategroups, and therefore the appearance of a rake-like Y-chromosome haplotype phylogeny can have social im-plications, particularly if the appearance can be datedusing data on genetic mutation rates.

Two archaeological papers threw further light on earlyAustronesian prehistory, both utilizing the assumptionthat the appearance of Neolithic cultures in Taiwan andthe Philippines can be correlated in terms of recon-structed economic and material cultural items with theintroduction of Austronesian languages. Tsang Cheng-hwa equated the appearance of the Neolithic Dabenkengculture in Taiwan at about 3500 BC with the initial ar-rival of (Pre-)Austronesian-speaking populations. He de-scribed new data on this culture, with prolific evidencefor rice and foxtail millet cultivation (both reconstruc-tible to Proto-Austronesian) newly excavated from the3000-BC pile-dwelling site of Nanguanli in the TainanScience-Based Industrial Park in southwestern Taiwan.Tsang also discussed features such as painted pottery andshouldered adzes which, when considered in compara-tive typological detail, could indicate a source for theDabenkeng Neolithic in the Pearl Delta region ofGuangdong.

Peter Bellwood then carried the archaeological recordinto more recent times with a discussion of new researchin the Batanes Islands, between Taiwan and Luzon. ANeolithic culture using red slipped pottery and other ar-tifacts of eastern Taiwan origin was established here byabout 1600 BC, after which a trade in ornaments madeof Taiwan jade, from a source near Hualien, continuedthroughout the Philippines until well within the IronAge. A workshop for making ear ornaments of Taiwanjade, dating from about 2,000 to 1,300 years ago, has beenfound at the site of Anaro on Itbayat Island. The oldest

Neolithic sites on Itbayat and Batan Islands are taken byBellwood to represent the arrival of Malayo-Polynesian-speaking populations from eastern Taiwan as one of thefirst stages in the eventual Malayo-Polynesian dispersalthrough the Philippines and Indonesia to as far as Sa-moa—a dispersal over at least 8,000 km achieved in per-haps as little as 500 years.

Subsequent Taiwan-Batanes-Philippines biological re-lationships were discussed in a paper on Batanes mtDNAby June-Hun Loo, albeit without invoking specific di-rectionality, as is perhaps understandable given the evi-dence for long-continuing trade and interaction betweenthese regions in later prehistory. Marie Lin presentedmultilocus genetic data to indicate relationships be-tween Formosans and Island Southeast Asians and inparticular to reinforce genetic relationships between Tai-wan eastern coastal Formosans and the Ivatan of Batanes.

sino-tibetan- and altaic-speaking peoples

Roger Blench discussed the rather complex issues ofSino-Tibetan subgrouping, observing that the expansionof the Sinitic languages could make identification of ahomeland very difficult, a problem reinforced by uncer-tainties over the positions of the Sinitic and Tibeto-Bur-man languages within the Sino-Tibetan phylogeny. Heregarded the Miao-Yao (Hmong-Mien) languages as pre-Sinitic within central China, overlain by Sinitic lan-guages and now deprived by replacement of any clearmarkers of deep linguistic diversity (Miao-Yao migrationinto Southeast Asia is historically very recent). Likewise,he identified the Austroasiatic languages as pre-Siniticwithin southern China, perhaps equatable archaeologi-cally with the pre-Neolithic archaeological occurrencesof very coarse cord-marked pottery, with Hoabinhianlithics, in certain early Holocene southern Chinese cavessuch as Xianrendong (Jiangsi) and Zengpiyan (Guangxi).He also noted the existence of difficult-to-classify lan-guages in China, suggesting very deep histories for pre-Sinitic populations.

The issue of Sino-Tibetan origins was also discussedby Henry Wright in a paper on the early Holocene ar-chaeology of China, touching again on Xianrendong,Zengpiyan, and other sites. Wright’s view, based on aprojection back into time of historically attested Siniticcultures such as the Shang, is that the Beixin-Dawenkou-Longshan lineage of Shandong enshrined the ancestry ofthe Sinitic peoples. Peiligang and Yangshao are regardedas a different cultural lineage (Starostin, below, regardsthe latter as perhaps Altaic) replaced in the process ofSinitic (Longshan) expansion. Wright also drew attentionto several important pre-Longshan sites in China in east-ern and southern China, including Xihe, a site of theHouli (Pre-Beixin culture in Shandong dating to ca. 6500BC, with pit houses and both rice and foxtail millet. Asdid Blench, Wright discussed the early Holocene potteryassemblages associated with a hunting and gatheringeconomy in the caves of Xianrendong and Zengpiyan,concluding that these sites, with Xihe, were probablynon-Sinitic in linguistic affiliation.


Li Jin, Bo Wen, and Hui Li placed the homeland of thespeakers of Sino-Tibetan languages in southeasternGansu in terms of Y-chromosome data. They also dem-onstrated massive demic expansion of northern Chineseinto southern China in post-Qin times, particularlyamongst males (see also Bo et al. 2004). Many southernChinese (Han) populations appear to have up to 60% Y-chromosome haplotypes of northern Chinese origin.There is clear historical evidence that supports suchmassive quantities of migration (Ge, Cao, and Wu 1993,1997; LaPolla 2001), with records of 5 million peoplemoving south during the early second millennium AD,particularly during the Southern Song dynasty.

Sergei Starostin presented data reflecting borrowinginto Old Chinese, as spoken during the Western Zhouand perhaps also Shang periods (late second to early firstmillennium BC), from within the ancestral nexus of theTungus-Manchu, Korean, and Japanese languages, clas-sified within the Altaic macrofamily. Forty-one suchloans are identified by Starostin, including a term thatseems to refer to either broomcorn or foxtail millet, bothcrops grown in the early Neolithic of the Manchurianregion as well as in the Yellow River Basin. This obser-vation will surely be of interest to archaeologists, giventhat Starostin’s linguistic calculation of the date of Proto-Altaic (5000 BC) predates that for Proto-Sino-Tibetan(4250 BC, as presented at the meeting). Starostin is astrong supporter of the existence of Altaic as a geneticgrouping including Tungus-Manchu, Korean-Japanese,Mongolian, and Turkic.

Roy Andrew Miller described how the involved historyof scholarship involving Korean-Japanese linguistic com-parison both in Japan and in the West has led to theprobably misleading grouping “Korean-Japanese.” Heproposed that, though much remains to be done here,confronting data from these languages establishes thetwo as separately descended from Starostin’s “Altaic” atthe same time that it reinforces many of the other pos-tulations in the Etymological Dictionary of the AltaicLanguages (Starostin, Dybo, and Mudrak 2003).

Satoshi Horai presented mtDNA and Y-chromosomedata on the Ainu of Hokkaido, stressing their separationfrom Koreans, Japanese, and Ryukyuans in general com-pared with their fairly close mtDNA relationships withSakhalin and Kamchatka populations, especially theNivkhi and Koryaks. This suggests an entry of ancestralAinu into Japan via Sakhalin from the north rather thanvia Kyushu from the south.

siberian populations

Tatiana Karafet and Mike Hammer noted high degreesof genetic drift between Siberian populations, reflectingsmall population sizes and long-term processes promot-ing isolation. In such situations, it is not surprising thatclear relationships between Y-chromosome genotypesand languages can be documented. The spread of Y-chro-mosome haplogroup P through Siberia to become themajor haplogroup in the New World was also noted. Lud-mila Osipova discussed the threats to Siberian popula-

tions, particularly in health matters, resulting fromglobal trends in social change.

comparative genetic histories

Several genetics papers focused on the histories of a num-ber of non-recombining mtDNA and Y-chromosome hap-lotypes, analysed across large numbers of unrelated pop-ulations. Peter Underhill presented an Asiawide surveyof Y-chromosome lineages, discussing especially mattersof time depth and distribution. Estella Poloni and BarbaraArredi summarized the evidence for a north-to-south dis-tinction between East Asian populations in terms of GM,mtDNA, and Y-chromosome data. They concluded thatthe most likely explanation of the observed patterningwas two separate early modern human populationspreads into East Asia—one ancestral to populationswithin the Sino-Tibetan and other Southeast Asian lan-guage families, the other ancestral to Altaic-speakers andmany Tibetans—with subsequent gene flow betweenthese two groups of populations mainly through themovement of women. Finally, Andre Langaney raised theimportant point that coalescence times given by genetrees should not be taken automatically to be times ofpopulation origins or migrations, a view shared by manypresent at the meeting.

concluding thoughts

Conferences such as this and its predecessor in Perigueuxpresent data on a scale that can often seem overwhelm-ing, particularly to social scientists not versed in thecomplexities of population genetics and vice versa. Thepast has been complex in the extreme, and to recover allthis complexity event by event would be impossible. Weneed to propose overarching hypotheses that can accountfor the comparative data from linguistics, genetics, andarchaeology with as little stress as possible. Many ofthese hypotheses will inevitably include major compo-nents of migration as well as interaction. The migrationsdiscussed in the Geneva conference are significant ex-amples of past expansions that have left indelible tracesin the modern world. We need to analyse them if we areto understand our history.

References Citedb e l l w o o d , p e t e r , a n d c o l i n r e n f r e w. Editors. 2003.

Examining the farming/language dispersal hypothesis. Cam-bridge: McDonald Institute for Archaeological Research.

b o w e n , h u i l i , d a ru l u , x i u f e n g s o n g , e t a l .2004. Genetic evidence supports demic diffusion of Han cul-ture. Nature 431:302–5.

c o x , m u r r a y. n.d. Indonesian mitochondrial DNA and its op-position to a Pleistocene-era origin of Proto-Polynesians in Is-land Southeast Asia. Human Biology. In press.

g e q i a n x i o n g , c a o s h u j i , a n d w u s o n g d i . 1993.Jianming Zhongguo Yimin Shi. Fuzhou: Fujian Renmin.

———. 1997. Zhongguo Yimin Shi. 6 vols. Fuzhou: FujianRenmin.

l a p o l l a , r . 2001. “The role of migration and language con-

https://www.researchgate.net/publication/200033709_Genetic_evidence_supports_demic_diffusion_of_Han_culture?el=1_x_8&enrichId=rgreq-7ff15151-6ba9-466b-91ee-b78ad22d0c99&enrichSource=Y292ZXJQYWdlOzI0OTE3OTU0MztBUzoxNDU0MjUzODIyNTI1NDRAMTQxMTY4MzUyMTE5Mw==




tact in the development of the Sino-Tibetan language family,”in Areal diffusion and genetic inheritance. Edited by R. Dixonand A. Aikhenvald, pp. 225–54. Oxford: Oxford UniversityPress.

s a g a r t , l a u r e n t , ro g e r b l e n c h , a n d a l i c i a s a n -

c h e z - m a z a s . Editors. 2004. The peopling of East Asia. Lon-don: RoutledgeCurzon.

s t a ro s t i n , s e r g e i , a n n a d y b o , a n d o l e g m u -d r a k . 2003. Etymological dictionary of the Altaic languages.3 vols. Leiden: Brill.

Demographic Evidence That Human Ovulation Is Undetectable (At Least in Pair Bonds) 1

Documents

Transcript of Demographic Evidence That Human Ovulation Is Undetectable (At Least in Pair Bonds) 1