The Osteological Paradox Reconsidered

by Mark Nathan Cohen, James W. Wood, George R. Milner
The Osteological Paradox Reconsidered
Mark Nathan Cohen, James W. Wood, George R. Milner
Current Anthropology
Start Page: 
End Page: 
Select license: 
Select License

Volume 35, Number 5, December 1994 1 629


ter, and Sontz 1971; Gallagher 1977; Hayden 1977, 1979) that the rules may be so lax [at least with regard to the overall morphology of the lithic artifacts) that the ar- chaeologist may be unable to ascertain from the lithics that they were made according to such rules. I suspect [although I do not speak as a lithics specialist) that we recognize the symbolic nature of the archaeological rec- ord of early Upper Paleolithic Europe more from its dec- orative and representational art than from its lithics.

Nevertheless, I am in complete agreement with Byers's interpretation of the Middle-to-Upper-Paleolithic transition in Europe with the one exception that there are more kinds or levels of symbolic behavior than he mentions and that the origins of language are as impor- tant to understand as the origins of symbolic culture. Above all. I am encouraged to see a scholar from outside


Paleolithic archaeology taking a serious and anything but naive interest in what archaeology has to offer. If we archaeologists can return the compliment by taking a serious and ideally not too naive interest in what other disciplines have to tell us about the evolution of human culture and of the human mind, our discipline will bene- fit enormously.

References Cited
AXELROD, R. 1984 The evolution of cooperation. New York: Ba- sic Books. CHASE, P. G. 1987 Specialisation de la chasse et transition vers le Paleolithique superieur. L'Anthropologie 9 I: I 75-88.

-. 1988. "Scavenging and hunting in the Middle Paleolithic: The evidence from Europe," in Upper Pleistocene prehistory of Western Eurasia. Edited by H. L. Dibble and A. Montet-White, pp. 226-32. Philadelphia: University Museum, University of Pennsylvania.

1989. "How different was Middle Paleolithic subsistence? A zooarchaeological perspective on the Middle to Upper Paleo- lithic transition," in The human revolution. Edited by P. R. Mellars and C. Stringer, pp. 32 1-37. Edinburgh: Edinburgh Uni- versity Press.

-. 1991. Symbols and Paleolithic artifacts: Style, standardiza- tion, and the imposition of arbitrary form. Journal of Anthropo- logical Archaeology 1o:193-214.

CHASE, P. G., AND HAROLD L. DIBBLE. 1992. Scientific archae- ology and the origins of symbolism: A reply to Bednarik. Cam- bridge Archaeological Iournal 2:42-5 I.

GALLAGHER, 7. 1977. Contemporary stone tools in Ethiopia: Implications for archaeology. Journal of Field Archaeology 4: 407-14. GAMBLE, CLIVE. I $182. Interaction and alliance in Paleolithic so- ciety. Man 17:92-102.

COULD, R., D. KOSTER, AND A. SONTZ. 1971. The lithic as- semblage of the Western Desert Aborigines of Australia. Ameri- can Antiquity 36: 149-46.

HAMILTON, W. D. 1964. The genetical evolution of social behav- ior. Iournal of Theoretical Biology 7:1-16.

HAYDEN, B. 1977. "Stone tool functions in the Western Desert," in Stone tools as cultural markers: Change, evolution, and complexity. Edited by R. V. S. Wright, pp. 178-88. Canberra: Humanities Press.

-. 1979. Paleolithic reflections: Lithic technology and ethno- graphic excavation among the Australian Aborigines. Atlantic Highlands: Humanities Press.

STRINGER, C. B., AND C. GAMBLE. 1993. In search of the Ne- anderthals: Solving the puzzle of human origins. London: Thames and Hudson.

WAAL, F. DE. 1989. Chimpanzee politics: Power and sex among apes. Baltimore: Johns Hopkins University Press.

The Osteological Paradox Reconsidered

Department of Anthropology, State University of New

York College at Plattsburgh, Plattsburgh, N.Y. 12901,

U.S.A. 15 VII 94

In 1982 George Armelagos and I [Cohen and Armelagos 1984) collected studies of pathology from skeletons and mummies of prehistoric hunter-gatherers and farmers from every region of the world that had produced compa- rable data. We found a number of fairly consistent trends: [I)that the frequency of nonspecific chronic in- fection displayed by skeletons was commonly higher among farmers than among earlier hunter-gatherers; (2) that the frequency of specific infections such as yaws and tuberculosis or tuberculosis-like infection usually increased as groups became larger and more sedentary; 13) that the frequency of intestinal infections and para- sites increased with group size and sedentism when mummies or feces were studied [see Allison 1984; see also Reinhard, Hevly, and Anderson 1987); (4) that the frequency of porotic hyperostosis, the skeletal lesion of [childhood?) anemia, was almost always higher among farmers than among earlier hunter-gatherers; (5) that other signs of malnutrition [retarded growth and osteo- porosis in children, premature adult osteoporosis, re- duced tooth size, etc.) were more common among farm- ers than earlier hunter-gatherers; (6) that the average stature of measured adult individuals declined through- out the Old World from the Paleolithic through the Neo- lithic period; and (7) that signs of systemic stress visible in teeth, including macroscopic enamel hypoplasia and microscopic Wilson bands, were usually more frequent and pronounced in farmers than among earlier hunter- gatherers.

I have interpreted these data as supporting the conclu- sion that biological stress increased with farming [Co- hen 1989). I continue to interpret them this way, and I note from at least two papers in the 1990s that Arme- lagos and colleagues also continue to make this interpre- tation [Armelagos 1990, Armelagos, Goodman, and Ja- cobs 1991; see also Goodman 1993). This interpretation makes the implicit assumption that skeletons in a cem- etery, at least on the average, are reasonably representa- tive of the living populations that produced them and therefore that changes in skeletal assemblages reflect real changes in the health of once-living populations. This is an assumption made implicitly or explicitly, with varying degrees of caution, by most quantitatively oriented paleopathologists.

The skeletal data and our interoretation of them have been challenged, however. In elaborating what they call "the osteological paradox," Wood et al. [CA 33:343-70; see also Harpending 1990) suggest that a number of fac- tors including nonstationarity, hidden heterogeneity, differential frailty, and selective mortality can bias the


sample of skeletons in a cemetery, making it an unrepre- sentative sample of a once-living people and rendering conclusions about the impact of economic change on human health unreliable. The points that they raise seem theoretically to be valid, and I note that they are discussed positively by a number of my colleagues in skeletal analysis with reference to the analysis of indi- vidual populations (see CA* comments). However, their theoretical arguments lead them to offer a reinter- pretation of health trends associated with the origins of agriculture which, from a broad geographic and temporal perspective, 1 find untenable and refutable.

Wood et al. begin by conceding that Armelagos and I may be correct in our interpretation of the consequences of early farming, but they argue that other interpreta- tions of the skeletal material are equally possible (and equally difficult to prove or disprove). They argue, in particular, that the apparent increase in pathology as- sociated with early farming populations in fact could reflect an improvement in health. They suggest that farming populations may normally have been better- nourished and longer-lived than their hunting-andgathering forebears and thus better able to record stresses in their skeletons. By this interpretation, the frequency of infection (or other pathology) did not in- crease with farming, but its record was more fully pre- served in the skeletons of better-nourished, more resil- ient, longer-lived farming populations. The skeletons of foragers were relatively pathology-free not because the foragers were healthy but because they died of insults before their skeletons could record them. Using this logic, Wood et al. produce a reevaluation of the Dickson Mounds archaeological sequence spanning the adoption and intensification of agriculture in Illinois. They sug- gest that low average ages at death and high frequencies of dental enamel hypoplasia in the latest, fully agricul- tural population, previously read as a record of declining health and longevity (Goodman et al. 1984)~ might actu- ally be indicative of increased fertility and biological privilege.

I respectfully disagree. Various types of evidence sug- gest that ours is the more probable interpretation. For one thing, the conclusions from paleopathology, as I have interpreted it, conform so closely to observations of health from ethnography and predictions from epide- miological theory that they should be taken as they ap- pear. For example, epidemiological theory predicts an increase in infection and parasite rates with sedentism and larger group size, supporting the direct interpreta- tion of this pattern of pathology in the skeletons. More- over, the pattern of increasing infection and parasites with group size and sedentism occurs repeatedly in com- parisons of historical and modern populations (Cohen 1989). Similarly, modern hunter-gatherers display very low rates of juvenile infection, malnutrition, and ane- mia, just as their prehistoric counterparts display low rates of porotic hyperostosis (see also Kent and Dunn's [1993] discussion of hypoferremia in newly settled San in the Kalahari). Contemporary hunter-gatherers also display relatively low rates of weanling diarrhea, thought to be a major contributor to enamel hypoplasia, so the low rates of hypoplasia in prehistoric foragers should not be surprising. Tuberculosis and related dis- eases occur primarily in archaeological samples from re- cent, relatively urban environments, mimicking the pat- tern of the disease of the present day (Cohen 1989). Goodman (1993) has pointed out, also, that enamel hy- poplasia, one of the few skeletal pathologies that can readily be seen in living individuals, has repeatedly been found to occur among living people in the pattern that our hypothesis predicts-it is regularly more common in lower-class than in upper-class individuals, suggesting that it reflects relative stress rather than relative background nutrition or resilience. Can it really be mere coincidence that our direct explanation of paleopatho- logical data fits so well with these expectations?

Furthermore, it is arguable at best whether nutrition and health improved with farming and sedentism among prehistoric populations as Wood et al. assume or whether survivorship was significantly greater in early agricultural groups than in hunter-gatherers. Optimal foraging data (reviewed in Cohen 1989; see also Simms 1987, Russell 1988) suggest that prehistoric hunter- gatherers were in a position of descending to agriculture as once-superior economic strategies had to be aban- doned (see, in particular, Russell's description of the ef- ficiency of harvesting wild and then domestic einkorn wheat). Hunter-gatherers should typically have had bet- ter background nutrition (as well as fewer background infections) than farmers and should normally have been the more resilient even though the stresses of mobility itself might have worked against them.

The perspective of Wood et al, may be affected by their (and our) familiarity with the !Kung San of the Kalahari, whose caloric intake is marginal (at the low end of the modern hunter-gatherer spectrum and undoubtedly well below that of prehistoric hunter-gatherers [see Cohen 19891) and who gain weight and resilience when they settle down under 20th-century conditions with 20th- century benefits (Pennington 1992). Under these conditions they attain a success at rearing their children that should imply population growth rates greater than those displayed by either prehistoric hunter-gatherers or pre- historic farmers-so their use as an ethnographic anal- ogy for prehistory may be limited.

In general, populations settling down under zoth- century conditions often display either increased fertil- ity or increased survivorship or both. However, neither can have occurred to any significant degree with the be- ginning of the Neolithic, as I will show. Perhaps more to the point, neither fertility nor survivorship can have changed very much at the Neolithic transition unless the two changed in opposite directions. The best esti- mates typically suggest that the rate of growth for our species as a whole accelerated from an average of about 0.01% per year before the adoption of farming to only about 0.1% after it (see, e.g., Hassan 1981, Bentley, Goldberg, and Jasienska 1993). Even if we assume that this entire acceleration resulted from increased survivor- ship with no contribution from increased fertility, there simply was not enough of an improvement in survivor- ship on the average to account for the increase in visible pathology. For the !Kung San, for example, an increase from 44% to 45% of individuals surviving to the mean age of maternity would suffice to explain the accelerated Neolithic growth rate of 0.1%~ but it alone could not explain the apparent increase in pathology among early farmers-especially because both ethnographic compar- isons and some paleodemographic life tables suggest that fertility often increased with farming, further reduc- ing any possible average Neolithic increase in survivor- ship (Cohen 1989, Buikstra, Konigsberg, and Bullington 1986, Wood et al.)

Further, re historic farmers also commonlv have higher rates of dental caries than hunter-ga;herers. Should we conclude, as almost all scholars do, that the farmers' diet was more cariogenic than that of hunter- gatherers, or should we conclude that hunter-gatherers ~ormally'also suffered but died of acute cariogeiesis be- fore their teeth had had the chance to develop lesions while farmers lived long enough to develop caries? Con- versely, some prehistoric hunter-gatherers have greater skeletal robusticity and more arthritis than farmers. Is this simply because the hunter-gatherers lived long enough to develop robusticity and arthritis but farmers did not? We can reconcile these two contradictory pat- terns only by recognizing that each pathology is telling us something of its own character and the lifestvle of its victims by its pattern in the skeletons. I submk that caries and arthritis and also chronic afflictions like peri- ostitis and porotic hyperostosis should be interpreted as diseases whose skeletal pathology is fairly straightfor- ward. Diseases like measles, which are known to kill without scarring the skeleton, will unquestionably have to be dealt with in a different manner.

Perhaps we can resolve our differences by pointing out that heterogeneity, differential frailty, and selective mortality, although real, do not play quantitatively as important a selective role in the creation of cemetery samples as Wood et al. imply.

If the individuals in a population were completely equal in their risk of dying from all causes (or if all causes of death were strictly accidental), we would ex- pect a death cohort or a cemetery full of such cohorts to be a random and usually representative sample of the living group, as the our hypothesis assumes. As Wood et al. point out, however, populations are in reality het- erogeneous in various ways; individuals are not at equal risk of dying from each cause, and death is selective. But, as is true when we study natural selection and evo- lution, that other domain of differential frailty and selec- tive mortalitv, not all or even most deaths are necessar- ily selected, ;or are they all selected for the one trait or condition under consideration. For example, even in that famous model of natural selection, the British pep- pered moth whose color evolved to match the tree- trunks, presumably only a fraction of the moths are ac- tually eaten or spared because of their wing color. Others presumably are eaten or escape being eaten de- spite their color-get caught on the wing, hit the wind- screens of fast-moving trucks, fly too close to hot lights, get stepped on, starve to death, or suffocate in the smog that is darkening the trees. The visible effect of natural

Volume 35, Number 5, December 1994 / 631

selection for color in any one generation may be rather small.

In fact, the sample of deaths in a population will al- ways include both a selected and a random component. I suggest that under most circumstances the effects of selection will show up only as relatively minor statisti- cal currents against the background of competing fac- tors. I submit that, except under very extreme selective conditions, the actual death cohort for a population for any year will normally be a fairly good representation of the living population from which it came because of the random nature of the unselected deaths, with only a small bias for each of various slight selective advan- tages. In particular, most human deaths are probably only weakly related to the chronic illnesses that human skeletons display (or those pathologies make only a small percentage contribution to the probability of dy- ing), and skeletons may therefore be a relatively random sample with regard to visible skeletal pathology in the population. (In fact, accidents and zoonotic diseases [see Fiennes 19781, significant causes of death among hunter- gatherers and early farmers which tend to strike active adults, would arguably kill those who were otherwise the most fit and the least frail). In specific instances, of course, severe selection or random statistical departure from expected frequencies ("drift") may make any one skeletal sample a misleading sample of the parent popu- lation. I protect myself from this possibility by using only trends that occur repeatedly in different popula- tions.

One possible application of this reasoning is the change in stature from the Paleolithic to the Neolithic throughout the Old World. If we consider that only se- lected mortality affects the distribution of heights in the cemeteries, then, as Wood et al, claim, the declining stature of Neolithic populations might represent the overall good health and nutrition of the population from which they come. However, if we allow that a signifi- cant portion of the dead represent random deaths (with regard to stature), as they certainly must, the implica- tion is that the Neolithic skeletons represent a universe of smaller people and therefore presumably a population with declining nutrition-which is of course what the optimal foraging data suggest we should find.

Although there is ample room for further discussion, I suggest that for now, on the whole, the conclusions of Cohen (1989) and Cohen and Armelagos (1984) can stand.


Department of Anthropology, Pennsylvania State University, University Park, Pa. 16802,U.S.A. 8 VIII 94

Several responses to our paper on "The Osteological Par- adox" (CA 33:343-70) have now appeared in CURRENT ANTHROPOLOGY,

and this may be a good time to reply to them all.' The commentators are evenly divided be- tween those who heartily loathed the paper and those who basically liked it but wanted to make additional points or to suggest ways of tackling some of the prob- lems we discussed. Since our intention was to spark debate, we welcome all the comments, even those ac- cusing us of scientific snobbery, nihilism, and aiding and abetting sinister pro-state, pro-civilization forces- although how we can be both nihilist and pro-civ- ilization is something of a mystery to us.

Our paper highlighted several problems that can con- found inferences drawn from skeletal sam~les about the health of prehistoric populations. These iifficulties are now widely recognized in epidemiology and demography but have received insufficient attention in research on skeletons from archaeological sites. As Goodman (CA 34:281-88) and Saunders and Hoppa (1993) point out, and indeed as we pointed out ourselves, osteologists have long acknowledged that they deal with samples made up of life's failures at any particular age. But con- trary to Goodman we see only limited progress being made in the development of formal methods for building this insight into the interpretation of skeletal lesions. If problems of inference and interpretation have proven difficult for researchers working with living popula- tions, we cannot imagine that they can be any easier for those dealing with bones dug up from cemeteries.

Most of the problems we discussed in our paper stem from heterogeneous frailty and selective mortality, the confounding effects of which can be illustrated with a homely parable. Suppose that one of the four authors of our original paper has grossly elevated serum cholesterol levels, while the other three are normal, at least in that respedt. In the population at large, serui cholesterol has a measurable, monotonic association with the risk of death from coronarv heart disease. Thus, an individual's serum cholesterol ievel can be considered a component of frailty, and our tiny population of coauthors is het- erogeneous for frailty. Now suppose that our high-cholesterol colleague keels over dead one day on dis- covering that he has been labeled a pro-civilization, nihilistic, scientific snob in a major journal. An autopsy reveals that this ill-starred anthropologist suffered a myocardial infarction resulting from the plugging up of his coronary arteries by atherosclerotic plaques, no doubt aggravated by the arterial constriction that can accompany acute psychological stress. In addition, mul- tiple ischemic lesions are found on his heart, signs of past cardiac "events." We now have a mortality sample on our hands, albeit a small one. What can we infer about the health status of the living population that launched this lost soul on his journey across the Styx? Obviously not much, except that it contained at least one individual who was apparently at an elevated risk

I. Although this response is written by only two of the original coauthors, the other two have read and endorsed it. An additional article focusing on the issues raised in our paper has appeared in the Yearbook of Physical Anthropology [Saunders and Hoppa 1993).

of death by cardiovascular disease. Because of selective mortality, this one now-shattered vessel of clay, this kicker of buckets, tells us little about the three individu- als who did not die. And this sort of selective mortality must occur whenever (I)individuals differ in their bio- logical (or even "lifestyle") characteristics and (2)those characteristics bear some relationship to the likelihood of death. Despite Cohen's assertion that deaths are es- sentially random, we suggest (and we are by no means the first to do so) that these two conditions are universal in human populations.

This example allows us to lay to rest one point of confusion in Goodman's comment (see his n. 7). Goodman equates the words "biological" and "genetic" in a way that we do not. Surely one's serum cholesterol level is a biological characteristic. Yet our friend with the clogged arteries may have had elevated serum choles- terol because he had familial hypercholesterolemia (a genetic condition) or because he habitually ate fried egg sandwiches with pork drippings three times a day (the ever-popular Elvis Diet). In either case, elevated serum cholesterol is a frailty factor, and selective mortality will act upon it. If serum cholesterol is at least partially heri- table (as, in fact, it is), then selective mortality in the demographic sense will also constitute natural selection in the genetic sense. But demographic selectivity can occur in the complete absence of natural selection.

There are other lessons to be drawn from this parable. The fact is, the miserable corpse at our feet did not die in childhood from diphtheria, typhus, or smallpox, and he certainly did not die from marasmus precipitated by weanling diarrhea. Thus, the fact that cardiovascular disease is responsible for such a high fraction of the ob- served deaths (IOO% in our sample so far) is partly attrib- utable to the comparative unimportance of other dis- eases in the population-the paradox of proportional mortality. And the multiple ischemic scars on his now- stilled heart suggest that he lived under conditions that permitted him to survive with his disease for a consider- able period before the fatal heart attack that left him pining for the fjords. Indeed, had he died not from a heart attack but from chronic congestive heart failure-had he not perused the pages of CURRENT ANTHROPOLOGY that sore and dreadful morn-he would have left behind a heart in really terrible shape, and that hideous lump of myocardial tissue would have been a sign, in all its ugliness, that he had managed to live successfully for many years despite his illness. The worse the condition of the heart, the longer the inferred survival-a kind of myocardial paradox.

This example can be criticized, most obviously be- cause the sample is ludicrously small but more interest- ingly because the other three coauthors will eventually die, and we will then have an opportunity to bring their blighted carcasses under the pathologist's knife. That, one might think, would provide an unbiased view of the entire population. But each and every death will misrep- resent the remaining frailty distribution among the liv- ing at whatever age the death occurs-except, of course, for the very last death, for that particular coauthor will

Volume 35, Number 5, December 1994 1 633

be perfectly representative of himself. Acquired condi- tions, including those associated with aging, will partly regenerate the upper tail of the frailty distribution among the survivors, so that selective mortality will al- ways produce corpses that look terrible, no matter what the general health of the population. The fact that one in five Americans will eventually die of cancer does not mean that 20% of living Americans have cancer right now.

Byers [CA 35:282-84) makes an interesting point in this connection. He suggests that it may be possible to assess the magnitude of the selective mortality acting on a trait such as stature by examining shifts in the distribution of the trait among individuals who die at later [presumably postselection) ages. The specific tests he advocates, however, are lacking in statistical power, as he is careful to point out; they also make a strong presumption that the trait of interest is normally distrib- uted at the outset of the selection process, for otherwise measures of skew and kurtosis by themselves are unin- formative about selection. An assumption of normality may be approximately correct for stature but is unlikely to hold for many frailty factors (see Schork, Welder, and Schork 1990). Still, Byers's basic idea is sound. Here we sketch an alternative approach to the same problem, an approach that may be both more general and more pow- erful. If z is a frailty factor, be it stature or whatever, then.the conditional distribution of z among survivors at age t [i.e., among those whose life span T is greater than t) is

where f [z) is the density function of frailty at the outset of the selection process and S(tl z) is the survival func- tion conditional on z. Given appropriate specifications of f[z) and S[tl z), the equation could be fitted to data on skeletal samples by straightforward maximum-likelihood methods, thus providing a direct estimate of the effect of selective mortality on the trait. Of course, for a trait that changes with age, as does stature, we would have to rescale the f [zl T > t) function properly, perhaps by using the residuals from one of the Preece- Baines model growth curves rather than raw stature it- self (see Preece and Baines 1978). But that is a secondary point. We believe that Byers's suggestion has consider- able merit and deserves further exploration. Inciden- tally, the equation tells us explicitly what we have to worry about in this whole area of inquiry: how frailty varies, indicated by f [z), and how frailty affects mortal- ity, captured by S[tl z).

The deeper problem, alluded to in our paper, is what to do when we don't have a readily measurable frailty factor such as stature or when measurable variables do not capture all the variation in frailty. As we tried to convey in our paper, this is a tough nut but not, we think, an uncrackable one (see the extended discussion in Wood et al. 1992). But it can't be cracked by ignoring it, wishing it away, or pretending that it doesn't matter in the population of interest.

Pretending that it doesn't matter is precisely what Co- hen [above) tries to do. As he remarks, his preferred in- terpretation of the paleodemographic and paleopatho- logical evidence "makes the implicit assumption that skeletons in a cemetery, at least on the average, are rea- sonably representative of the living populations that produced them and therefore that changes in skeletal assemblages reflect real changes in the health of once- living populations. This is an assumption made implic- itly or explicitly, with varying degrees of caution, by most quantitatively oriented paleopathologists." That was precisely the point of our paper: paleopathologists do make this assumption, often without examining its implications for their inferences. But according to Co- hen, the assumption that the dead are representative of the living is justifiable because mortality is largely ran- dom with respect to individual characteristics: "I sub- mit that, except under very extreme selective condi- tions, the actual death cohort for a population for any year will normally be a fairly good representation of the living population from which it came because of the random nature of the unselected deaths." Frankly, we don't know whether to be terrified or relieved by this conclusion-terrified because even the healthiest among us can drop dead tomorrow for no good reason, relieved because we can all stop worrying about diet, exercise, and those annoying annual check-ups. We re- spectfully suggest that Cohen's assertion that mortality is mostly random (nonselective) with respect to individ- ual characteristics is one of the most remarkable state- ments in the history of population science. If it is true, epidemiologists may as well retire, for no relative risk will ever be distinguishable from I. And demographers can pretty well close down shop too: life tables are un- necessary because mortality cannot be affected signifi- cantly by age, and sex-specific life tables are even more of a waste of time because they foolishly assume that one's sex may influence the risk of death. The claim that mortality is largely random and nonselective is so profound in its implications that it needs to be sup- ported by evidence, not merely asserted. After all, the entire insurance industry is founded on the premise that the risk of death varies among individuals in a poten- tially predictable way.

But we agree with Cohen that mortality has stochas- tic [random) as well as deterministic [nonrandom) ele- ments. In fact, several of us have spent major portions of our professional lives formulating demographic models that include both stochastic and deterministic compo- nents [for a review, see Wood et al. 1992). With respect to mortality, such models are consistent in showing the following pattern: Within a given risk group, the vari- ance in life span is typically large, and in that narrow sense the random component of mortality can be consid- ered important [see Vaupel 1988 for an insightful discus- sion of this point). When risk groups are pooled, however, mortality at any given age will still be dominated by those individuals of highest risk who survived to that age [Manton, Stallard, and Vaupel I 986). In other words, demographic and epidemiological models make it clear


that substantial selective mortalitv can occur even in the face of nontrivial stochastic variation in life span2

Cohen's claim that mortality was largely nonselective in prehistoric societies appears to be based on his belief that most deaths in such societies were accidental. While reliable data on causes of death are notoriously difficult to obtain from nonliterate people, such data as exist suggest that the overwhelming majority of deaths in all preindustrial populations are infectious in nature, with undernutrition being a frequent contributary cause. In our work with the Gainj, a population of swid- den horticulturalists on the northern fringes of Papua New Guinea's Central Highlands, we found that about 70% of deaths were from infectious causes, with diar- rheal disease being more common in children and acute respiratory infections somewhat more common in adults; physical trauma accounted for only about 2% of deaths (Wood 1980:122). Broadly similar patterns have been reported for the !Kung [Howell 1979:69) and the Aka Pygmies (Hewlett, van de Koppel, and van de Kop- pel 1986:54-5 5). Infectious diseases are known to be highly selective, especially with respect to nutritional status and immune function. In a studv of the Turkana of northern Kenya, one of the most traditional groups of nomadic pastoralists in the world, Shell-Duncan (1993, 1994) has shown the overwhelming importance of gas- trointestinal and respiratory tract infections in determining patterns of morbidity and mortality (with malaria running a somewhat distant third); more important, she has shown that these diseases are power- fully selective on nutritional status and immunocompe- tence as assessed by delayed-type hypersensitivity tests. In none of these studies or any others that we are aware of have accidents been shown to be a leading cause of death. It is all very well for Cohen to dismiss evidence on the !Kung from the discussion, but the fact remains that there are very few studies of anthropological popu- lations that have yielded high-quality information on causes of death, and none of them indicates that acci- dents predominate. Perhaps people were clumsier in pre- history, but we doubt it.

Besides, there is no reason to believe that accidents are genuinely random: "accident-proneness" is a well- established epidemiological phenomenon. Nor can we assume that violent deaths as opposed to those from illness or any of the other vicissitudes of life are ran- domly distributed. The selectivity of violence can be seen even in one of the very few osteological samples in which a fairly large fraction (at least 16%) of individuals died from being hit with clubs or shot with arrows (Mil- ner, Anderson, and Smith 1991). In this skeletal collec- tion from a cemetery geographically and temporally close to Dickson Mounds, most of the people who had been slaughtered were adults, presumably because their daily activities put them at greater risk of being at- tacked. Furthermore, many of these skeletons showed

2. For outstanding reviews of the relevant models, see Manton and Stallard (1984: chap. 6; 19881, Mode (1985: chaps. 3 and 41, and Namboodiri (1991: chap. 9).

signs of debilitating injuries or active infections that would have reduced any chance they might have had to fight or flee successfully. In this particular prehistoric setting, then, violent death was not unrelated to an indi- vidual's age, prior life history, or current health status. In other words, even accidents and violence can be selec- tive on individual characteristics.

Jackes (CA 34:434-3 9) notes, quite rightly, that there are many serious problems that we did not highlight in our original paper, including but not limited to age- estimation and differential preservation. However, as we pointed out, those issues have received far more atten- tion in the literature than the ones we did raise.3 But that does not mean that the issues we deemphasized have all been settled. As long as we are enumerating additional problems to worry about, we would mention three more, without, however, meaning to imply that this is an exhaustive list. First, as many osteologists have noted, the disease processes that can leave ob- servable skeletal lesions represent only a tiny fraction of all the afflictions likely to be present in the living population; focusing on them will almost necessarily give a highly distorted view of overall health and disease in the population. The second problem is differential diagnosis-the difficulty of assigning a slzeletal lesion unambiguously to a particular cause; because of errors in differentiating diseases, skeletal lesions are often of low specificity as diagnostic markers. The third problem is that skeletal lesions usually occur only in rare, often extreme, cases of the diseases that produce them; as a result, such lesions are of low sensitivity as markers of disease. As it happens, there are formal statistical meth- ods for assessing both specificity and sensitivity (Kelsey, Thompson, and Evans I98 6: 28 6). If such assessments were done, we suspect that most skeletal lesions would prove to be poor diagnostic markers.

A superficial reading of our paper might suggest that we are being inconsistent on this point. Some of the time we seem to be saying that mortality samples over- estimate the prevalence of pathological conditions (be- cause of selective mortality), and some of the time we seem to be saying that mortality samples underestimate the prevalence of pathological conditions (because of low sensitivity). What we are actually saying is that both biases are likely to be operating simultaneously, and both make it difficult for us to link lesion frequen- cies in skeletons with disease prevalence in the living

3. In fact, biases arising from the differential preservation of bones in archaeological deposits were noted at least two centuries ago. On digging a mound in Virginia in the 18th century, Thomas Jeffer- son (1788:105) reflected that "the bones of infants being soft, they probably decay sooner, which might be the cause so few were found here." Observations on the differential preservation of soft tissue are even older. When Hamlet asked the grave maker, "How long will a man lie i'th'earth ere he rot!" he was told: "Faith, if 'a be not rotten before 'a die-as we have many pocky corpses now-a- days that will scarce hold the laying in-'a will last you some eight year or nine year. A tanner will last you nine year" (Act 5, Scene I).We only hope that these literary and antiquarian allusions are not too pro-civilization for some of our readers' tastes (cf. Cohen,

CA 33:359).

population in any straightforward way. Although the two biases operate in opposite directions, it would be foolish to assume that they must exactly cancel each other out for any disease.

In our view, a far more serious block to inference, which none of our critics addresses, is the fact that le- sion frequencies in the dead partly reflect proportional mortality, which in turn reflects the whole spectrum of conditions affecting the risk of death, not just those that caused the particular lesion or lesions in question. This problem is not solved by using Goodman's (p. 281) "mul- tiple indicators of health," because the full set of lesions potentially detectable in the skeleton still represents only a small portion of that spectrum. Moreover, selec- tive mortality may be operating on all the indicators simultaneously, and to differing degrees.

According to Goodman (p. 282), "Wood et al. miss the mark in their explication of the dynamics of selective mortality partly because they are committed to the no- tion that the goal of paleoepidemiology is understanding cause of death. Paleoepidemiologists are in fact seldom concerned with cause-of-death analysis, which is exceed- ingly difficult for a number of reasons and is not at all es- sential to saying something about health and adjustment in past populations." Surely paleopathologists and espe- cially paleodemographers are very interested indeed in causes of death, at least part of the time. But that is beside the point. If we want to say something about health and adjustment on the basis of skeletal samples, then we have to own up to the fact that we are looking at samples of dead people, and presumably their reason for dying had something to do with their health and adjustment. We would never draw inferences about the distribution of health characteristics in the general popu- lation by exclusively examining hospital patients, among whom serious disorders approach 100%. Serious disor- ders are even more common among the dead: such disor- ders are what most of them died of.

We cannot stress this point too much. If a skeletal lesion-or the condition responsible for it, or a trait pre- disposing to that condition, or another trait highly corre- lated with that condition-has any relationship whatso- ever to the risk of death, the skeletal collection must be a biased sample for the living population. The bias caused by selection may be large or small; it may be positive or negativej but mathematically it must exist. We would be more than happy to discover that the bias is consistently small and, hence, untroubling, as Saun- ders and Hoppa (1993) suggest for the special case of staturee4 But at present, we have neither theory nor em- pirical evidence showing that the bias can always be ignored. It will take a lot of hard work to develop the

4. Goodman (p.283) argues that the linkage between short stature and the risk of death is partly spurious, owing to the confounding effects of "the causal link of both with impoverished and stressful living conditions." Of course it is, and this fact reinforces our basic point: mortality is selective not only for conditions that directly cause death but also for any other traits (including socioeconomic characteristics) that are consistently correlated with such condi- tions, whatever the source of correlation.

Volume 35, Number 5, December 1994 1635

necessary theory and compile the necessary evidence; bald assertions and appeals to conventional wisdom won't take us very far in the right direction.

Readers of Goodman's comments might conclude that we somehow missed the fact that skeletal lesions attrib- utable to infection and malnutrition take some time to develop, that people often live with those conditions for extended periods, and that some fraction of them re- cover from their illness. Yet it is precisely this variation in experience that makes the interpretation of skeletal lesions in mortality samples such a vexing topic-and, if explored further, such a potentially enlightening one. While it goes without saying that Goodman (p. 282) is correct when he writes that there are many reasons to investigate the "cultural and social repercussions" of disease, it is difficult to see how morbidity and mortality can be disassociated from one another in studies based strictly on individuals of a particular age who failed to survive. After all, the ability to live with a condition causing a pathological bony response or to recover from it, perhaps with an impaired ability to withstand further illness, does not imply that the presence of these skele- tal lesions is not in some way associated with an ele- vated risk of death.

Goodman makes extended criticisms of several of our illustrative exercises. In doing so, he misses the spirit of those exercises, which was to illustrate potential prob- lems and not to hold a mirror up to nature. (Having said that, we would contend that the assumptions underly- ing our illustrations are no less realistic than those made by Goodman, Cohen, and others in their more "final" analyses.) Goodman spends a long time on one espe- cially simple-minded example that was merely intended to introduce some of the problems; but by doing so, he inadvertently uncovers one of the most fundamental dif- ficulties in this area of study. In our example, we posited three subgroups within a population and assumed that those subgroups were exposed to low, moderate, and high levels of "stress," respectively. Under not too unre- alistic assumptions, we showed that the high- and low- stress groups could display similar lesion frequencies and thus be difficult to distinguish from one another using skeletal samples alone. But Goodman says that we willfully ignored an important piece of evidence: had we examined the mortality patterns of the three groups, they would have fallen out from each other unambig- uously. This claim is unassailable-provided that we know beforehand how many subgroups there are, are able to assign skeletons to the subgroups without error, and can reliably reconstruct the mortality patterns of the subgroups. Otherwise we are presented with an un- known number of risk groups, each of unknown size, whose apparent pooled age-at-death distribution is con- founded by heterogeneous frailty and demographic non- stationarity. Alas, we fear that osteological samples more often approximate the latter set of circumstances than the former.

In his response to our alternative explanation for the Dickson Mounds findings, Goodman criticizes the as-


sumptions of our model, dismissing many of them as


unrealistic. While the reasonableness of our assumptions is difficult to assess from the osteological evidence available from Dickson Mounds (as Jackes so rightly em- phasizes), this criticism again misses the point of the exercise. Given the time, energy, and inclination, we suspect that we could come up with a much more realis- tic model (involving, for example, more than two sub- groups or continuously distributed frailty). But that would not alter our fundamental point: we can interpret the osteological record meaningfully only if we do in- deed have an explicit model for the formation of the mortality sample we are working with. As we said re- peatedly in our paper, the point is not that our model is right and other models are wrong but that models are in fact necessary and for the most part lacking.

It is clear from his very first sentence (p. 281) that Goodman misunderstands our position: we emphatically do not argue that "health inferences from paleode- mographic and paleopathological data are impossible" but merely suggest that such inferences are more diffi- cult than we all originally believed, and also more inter- esting. We are convinced that there are solutions to the problems we raised, but finding the solutions will re- quire some deep thinking about the processes linking frailty, stress, disease, tissue responses, and the forma- tion of mortality samples. It will also require formal modeling of those processes. As Goodman (p. 282) notes, "There is . . . a mathematical tethering of individual and group frailty; if group frailty changes, then either the size of subgroups or the frailty of one or more subgroups must change. Furthermore, one can begin to interpret the individual significance of aggregate frailty if one has a theory about the distribution of frailty and some idea of how groups might change in size and how exposures might change subgroups' health risks-in short, contex- tual information." Although apparently written in criti- cism of us, this is in fact precisely our position. Indeed, if one adds the further stipulation that we need models of the linkage between frailty and the risk of death, then Goodman's statement summarizes our view admirably.

We are pleased that Goodman underscores the impor- tance of "cultural context" in the interpretation of skel- etons from cemeteries; in doing so, he again reiterates our position as expressed in the original paper. It is strange, however, that the Dickson Mounds skeletons continue to be misperceived by Goodman and others (Goodman et al. 1984; Goodman and Armelagos 1985; Cohen 1989: 121, 126) as somehow "spanning the adop- tion and intensification of agriculture," to use Cohen's phrase. The full course of changes in diets and lifeways related to an increased reliance on cultivation actually spanned a much longer period than the few centuries when the Dickson Mounds burial ground was used, and maize was a late addition to diets already based in large part on several cultigens (see Smith 1989). To date, only a few Dickson Mounds skeletons have been analyzed for stable carbon isotopes-an indication of maize consumption-but skeletons from throughout the cemetery sequence yield figures that fall within the late prehis- toric, or agriculturalist, range for eastern North America (Buikstra and Milner 1991, Buikstra 1992). Thus, despite ~revious characterizations of the Dickson Mounds sam- ple, it does not encompass the full spectrum of changes from hunting-gathering through intensified agriculture. It is likewise difficult to argue from the available archae- ological evidence that the latest Dickson Mounds people were sick because they suffered the ill effects of an out- flow of food to other places (Goodman et al. 1984, Good- man and Armelagos 1985). At present, it is simply not clear what the changes over time in skeletal lesion fre- quencies observed at Dickson Mounds might mean and what might have caused them.

In general, both Cohen and Goodman seem to want to cast us as irredeemably opposed to the scientific posi- tions they have staked out. This reaction is as surprising as it is unwarranted. As we took pains to say, "It is important to emphasize that our reinterpretation of the health consequences of early agriculture is not necessar- ily more correct than previous interpretations. And, in- deed, we suspect that both interpretations may be cor- rect for different periods and locations. The point, however, is not that we are right and other authors are wrong but that the data [by themselves, i.e., unaided by models] support both interpretations equally well" (CA 3 3 :357).Cohen and Goodman also try to make us sound more pessimistic than we actually are. We believe that advances can be made, but we reiterate that a necessarv precondition for advancement is the development of st;- tistical models for the formation of mortality samples. Although we do not underestimate the difficulties of this work, we do not think they are insurmountable. And we a;e pleased to acknowledge that the work has already begun, thanks in part (we are gratified to say) to our paper. Byers's idea on measuring selectivity, for example, is a nice step in the right direction, as is the more extended analysis of Saunders and Hoppa (1993). While neither of these analyses can be considered the final word on the subject, both at least show what can be done if the issues are taken seriously. Jackes's (p. 435) trenchant comment is relevant here: "Despair should arise only if osteologists are not intellectually honest and able to withstand a reappraisal of their methods."

References Cited
ALLISON,M. 1984. "Paleopathology in Peruvian and Chilean

populations," in Paleopathology at the origins of agriculture.

Edited by M. Cohen and G. Armelagos, pp. 5 15-29. New York:

Academic Press. ARMELAGO S, G. J. 1990. "Health and disease in prehistoric pop-

ulations in transition," in Disease in populations in transition.

Edited by A. Swedlund and G. Armelagos, pp. 127-44 New


co B s. I 99I. The origins of agriculture: Population growth dur-

ing a period of declining health. Population and Environment


ENSKA.1993. The fertility of agricultural and non-agricultural

traditional societies. Population Studies 47:269-81. BUIKSTRA,J. E. 1992. "Diet and disease in late prehistory," in Disease and demography in the Americas. Edited by J. W. Ve- ran0 and D. H. Ubelaker, pp. 87-101. Washington. D.C.: Smith- sonian Institution Press.

BUIKSTRA,J. E., AND G. R. MILNER. 1991. Isotopic and archae- ological interpretations of diet in the central Mississippi Val- ley. Iournal of Archaeological Science 18:319-29.

BUIKSTRA, JANE, LYLE KONIGSBERG, AND JILL BULLINGTON. 1986. Fertility and the development of agriculture in the prehistoric Midwest. American Antiquity 5 1:528-46.

BYERS,S. N. 1994. On stress and stature in the "osteological par- adox." CURRENT ANTHROPOLOGY 35 :282-84. COHEN,MARK N. 1989. Health and the rise of civilization. New Haven: Yale University Press. -. 1992. Comment on: "The osteological paradox," by J. W. Wood et al. CURRENT ANTHROPOLOGY 33:35 8-59.

COHEN,MARK N., AND GEORGE J. ARMELAGOS. Editors. 1984. Paleopathology at the origins of agriculture. New York: Academic Press.

FIENNES,RICHARD. 1978 Zoonoses and the origins and ecol- ogy of human disease. New York: Academic Press. GOODMAN,ALAN. 1993. On the interpretation of health from skeletal remains. CURRENT ANTHROPOLOGY 34381-88. GOODMAN, H., A.
AND G. J. ARMELAGOS. 1985. Disease and death at Dr. Dickson's mounds. Natural History 94:12-18.

GOODMAN, A., J. LALLO, G. J. ARMELAGOS, AND J. C. RO sE. 1984 "Health changes at Dickson Mounds, Illinois (A.D.950-I~oo)," in Paleopathology at the origins of agricul- ture. Edited by M. Cohen and G. J. Armelagos, pp. 271-306. New York: Academic Press.

HARPENDING, 1990. Review of: Health and the rise of

HENRY. civilization, by Mark N. Cohen (New Haven: Yale University Press, 1989) American Ethnologist 17:799-800.

HASSAN,FEKRI. 1981. Demographic archaeology. New York: Academic Press.

HEWLETT, B. S., J. M. H. VAN DE KOPPEL, AND M. VAN DE KOPPEL,1986. "Causes of death among the Aka Pygmies of the Central African Republic," in African pygmies. Edited by

L. L. Cavalli-Sforza, pp. 45-63. New York: Academic Press. HOWELL,N. 1979. The demography of the Dobe !Kung. New York: Academic Press. JACKES,N. 1993. On a paradox and osteology. CURRENT ANTHRO


T. 1788. Notes on the state of Virginia. Philadel- phia: Pritchard and Hall.

KELSEY, J. L., W. D. THOMPSON, AND A. S. EVANS. 1986. Methods in observational epidemiology. Oxford: Oxford Uni- versity Press.

KENT, SUSAN, AND DAVID DUNN. 1993. Etiology of hypofer- remia in a recently sedentary Kalahari village. American Jour- nal of Tropical Medicine and Hygiene 48:554-67.


1984. Recent trends in mortality analysis. New York: Academic Press. -. 1988. Chronic disease modelling. New York: Oxford Uni- versity Press.

MANTON, K. G., E. STALLARD, AND J. W. VAUPEL. 1986. Alternative models for the heterogeneity of mortality risks among the aged. Iournal of the American Statistical Associa- tion 81:635-44.

MILNER,G. R., E. ANDERSON,AND V. G. SMITH. 1991. War- fare in late prehistoric west-central Illinois. American Antiq- uity 56:581-603.

MODE,C. J. 1985. Stochastic processes in demography and their computer implementation. Berlin: Springer-Verlag. NAMB OODIRI, K. 1991. Demographic analysis: A stochastic ap- proach. New York: Academic Press. PENNINGTON, 1992. Did food increase fertility? Eval-

RENEE. uation of !Kung and Herero history. Human Biology 64:497-


PREECE, M. A., AND M. J. BAINES, 1978. A new family of mathematical models describing the human growth curve. Annals of Human Biology S:I-24.

REINHARD, CARL, RICHARD HEVLY, AND GLENN ANDER-so N. 1987. Helminth remains from prehistoric Indian copro-

Volume 35, Number 5, December 1994 1 637

lites on the Colorado Plateau. Iournal of Parasitology 73:3:


1988. After Eden: The behavioral ecology of earlv food aroduction in the Near East and North Africa. British ~rchakolo~ical

Reports International Series 391.
SAUNDERS, 1993. Growth deficit in

S. R., AND R. D. HOPPA. survivors and non-survivors: Biological mortality bias in sub- adult skeletal samples. Yearbook of Physical Anthropology 36: 127-51.

the asymmetry of biological frequency distributions. Genetic
Epidemiology 7:427-46.


B. 1993. Cell-mediated immunocompetence among nomadic Turkana children. American Iournal of Hu- man Biology 5:225-35.

-. 1994. Determinants of infant and childhood morbidity among nomadic Turkana pastoralists of northwest Kenya. Ph.D. diss., Pennsylvania State University, University Park, Pa.

SIMMS,STEVEN R. 1987. Behavioral ecology and hunter- gatherer foraging. British Archaeological Reports International Series 38 I.

SMITH,B. D. 1989. Origins of agriculture in eastern North America. Science 246: 1566-71. VAUPEL,J. W. 1988. Inherited frailty and longevity. Demogra- phy 25:277-87.

WOOD, J. w. 1980. Mechanisms of demographic equilibrium in a small human population, the Gainj of Papua New Guinea. Ph.D., University of Michigan, Ann Arbor, Mich.

WOOD, J. W., D. J. HOLMAN, K. M. WEISS, A. V. BUCHANAN, AND B. LE FOR. 1992. Hazards models for human population biology. Yearbook of Physical Anthropology 35:43-47.

WOOD, JAMES, GEORGE MILNER, HENRY HARPENDING, AND KENNETH WEISS. 1992. The osteological paradox: Prob- lems in inferring prehistoric health from skeletal samples. CURRENT ANTHROPOLOGY 33 :343-70.

On the Emergence of Agriculture in the New World

Smithsonian Tropical Research Institute, Unit 0948, Apartado 2070, Balboa, Panama. 15 VI 94

Fritz (CA 35:305-9) argues for a much younger chronol- ogy for the emergence of agriculture in the New World than is generally accepted. She suggests that the 5th- millennium-B.P. radiocarbon dates obtained on the Te- huacan maize cobs (Long et al. 1989) should cause schol- ars to reevaluate the entire sequence of food production and associated early settlement in both Central and South America. I argue here that a major revision for New World agricultural beginnings (I)is premature as applied to Mesoamerica, (2)does not fit the existing evi- dence from lower Central America and northern South America, and (3) fails to appreciate the plants and eco- logical contexts around which New World tropical food production emerged.

Fritz sets a beginning date on agriculture in Meso- america no earlier than 6,ooo-5,ooo B.P., although ex- perts agree that the maize, squash, and other domesti- cates recovered from the Tehuacan Valley caves underwent incipient domestication and genetic manipu-

  • Recommend Us