«John Komlos Professor Emeritus, University of Munich and visiting professor, Duke University. John.Komlos and Leonard Carlson Department of ...»
The Anthropometric History of Native Americans, c. 1820-1890
Professor Emeritus, University of Munich and visiting professor, Duke University.
Department of Economics, Emory University, Atlanta, Georgia 30322, U.S.A
Running Head: The Anthropometric History of Native Americans
The trend of the height of Indian scouts in the U.S. Army born between ca.
1825 and 1875 is analyzed. Their average height of ca. 170 cm (67 in.) confirms that natives were tall compared to Europeans but were nearly the shortest among the rural populations in the New World. The trend in their height describes a slightly inverted ”U” shape with an increase between those born 1820-34 and 1835-39 of ca.
1.8 cm (0.7 in.) (p=0.000) and a subsequent slight decline after the Civil War. This implies that they were able to maintain and perhaps even improve their nutritional status through the Civil War, though harder times followed for those born thereafter.
We also recalculate the heights of Native Americans in the Boas sample and find that the Plains Indians were shorter than most rural Americans. The trend in the height of Indians in the Boas sample is similar to that of the Scouts.
Acknowledgment: We thank Richard Jantz for providing a copy of the “Boas sample.”
1. Introduction Our knowledge of the material conditions of Native Americans in the 19th century is quite limited because of the scarcity of evidence. To be sure, we do know that there was a massive disruption of their way of life and a large decline in population following European contact. The North American Indian population (not just in the US) declined from 1,894,350 in a.d. 1500 to 530,000 in 1900 due to epidemics and other factors [Ubalker 1988]. Since 1900 native population has rebounded and exceeds its level in 1500.
Available evidence on their physical stature, however, does enable us to gain at least a glimpse of one important aspect of their biological welfare in the course of much of the 19th century. Human height is a widely used synthetic indicator of nutritional status, malnutrition, and biological living standards in many different settings, including but not limited to underdeveloped economies, in historical contexts, and in circumstances in which economic indicators are either unreliable or scarce as among slaves or Native Americans (Steckel 1995). Physical stature is positively correlated with net nutrition - the balance between the quantity and quality of nutrient intake and the demands on those resources by the human organism for growth, metabolic maintenance, work, and for resistance to diseases. Of course, individual heights depend as much on genetic potential as on nutrition, but at the population level environmental factors play a very substantial role in determining adult height (Bogin 1999). Hence, height of a population is eminently suitable to ascertaining the nutritional and epidemiological circumstances in which that population lived prior to reaching adulthood.
We analyze a newly collected data set on the heigth of Native American scouts in the U.S. army. Our paper is organised as follows: in section 2 we discuss prior estimates of the height of Native Americans; in section 3 we explain the regression technique we use in order to estimate mean of samples in which the height distributions are biased, i.e., are not normally distributed; in section 4 we focus on the history of Indian scouts in the U.S. army; in section 5 we present the newly discovered data; in section 6 we report the results of the analysis of these data; in section 7 we discuss our findings; in section 8 we connect with the history of Native Americans in light of our findings; and in section 9 we conclude.
2. Prior estimates of the Height of the Native Population of North
The main source on the height of North American natives hitherto analyzed was collected by the prominent anthropologist Franz Boas at the end of the 19 th century (Boas, 1895; Jantz 1995). Boas published the height distributions by tribe without noticing, however, that the samples were obviously biased insofar as the distributions were not symmetric as expected: there were almost always too few men in the sample left of the mean (or mode) (1895: 372). This is particularly evident among the Sioux and Crow, two tribes with the largest sample sizes which biased the averages in an upwardly direction (Figures 1 and 2). While a random sample of heights is always and everywhere normally distributed, the height samples of both of these tribes clearly suffer from a shortfall below c. 170 cm.
Although Jantz did state quite explicitly that “Boas’s samples of Native Americans cannot be regarded as random samples…” (1995), Steckel and Prince analyzed the Boas data set as though it were a random sample, concluding that the Plains Indians were the tallest populations in the world with a mean height of 172.6 cm (68.0 in) (Table 1 row 10) (Steckel, 2010: 267; Steckel & Prince, 2001: 289;
Prince & Steckel, 2003: 367). As a consequence of the sampling biases, this estimate is certainly too high (Table 1 row 17). Of the c. 1,700 observations that stem from tribes with a mean height above 170 cm, nearly two-thirds were from tribes whose height distribution did not pass the test of normality.1 If one excludes these tribes from the Boas averages, the mean height becomes 169.6 cm (66.8 in) or about
3.0 cm (1.2 in) below the Steckel-Prince estimates (Table 1, row 23). Of the eight tribes they included in their analysis 72% were from the Sioux and Crow with biased samples, but the height distribution of many of the others are also similarly distorted 2 (Figure 3). In other words, the mean height of Boas sample has to be calculated using techniques that account for the distorted nature of the sample: truncated regression, which has not been used up to now.
3. Truncated Regression3 Statistical analysis of height data from non-random samples is facilitated considerably by the biological law that height is approximately normally distributed within a population, and its standard deviation is practically constant, i.e., has a narrow range between ca. 6 and 7 cm among males and between ca. 5.3 and 6.5 cm among females even though mean heights can vary by as much as 20 cm within a population over time (Cole, 2003; Komlos & Baur, 2004). Consequently, variations in a population’s nutritional status affect mean heights, and not the form or dispersion of the distribution.
Height samples are frequently not representative of the population from which they are drawn, i.e., they are not random samples. Thus, the Boas sample as well as the scout sample about to be examined are hardly unique in this regard. The height distributions drawn from many historical military records (prior to the introduction of universal conscription) typically have a shortfall in the left tail,- fewer than expected observations - insofar as most armies imposed a minimum height requirement (Komlos, 2004). Thus, data are frequently available only for those individuals whose height exceeded the minimum height requirement (). In such cases, sample means and variances are biased estimators of the underlying population parameters, as are the coefficients of independent variables estimated by ordinary least squares regression (Komlos, 2003; Komlos & A’Hearn, 2004).
Suppose that we observe the latent normal random variable Y* with mean and variance 2: y* , ~ N (0, ) only if y* . Thus, sample Y is: y if y ; y missing otherwise. We observe y only if , or . Thus, conditional on being in the sample, E() 0, and is not normally distributed. Parametric methods for estimating build on the normal distribution of heights, enabling us to use the normal density as the likelihood function for (untruncated) heights. In the case of truncation, the area under the curve no longer integrates to unity without the lower tail. To correct for this, we can divide by the probability of being in the sample, i.e.
Pr(y ). This is the standard way to model conditional probability as it normalizes the area under the curve to unity.
The probability density function (pdf) of a truncated normal random variable is:
where denotes the standard normal pdf (Ruud 2000, Ch. 28; Greene 1993, Ch.
22). The log likelihood function of Eq. (1) can be formed and the parameter values that maximize it can be calculated using numerical methods. This maximum likelihood (ML) estimator has the usual ML properties of consistency, and asymptotic efficiency.
However, experience with actual samples demonstrated that the ML estimates can vary implausibly over time or cross-sectionally. This inference is based on the fact that there are biological limits to the variability in the physical stature of a population in the short run. The variability turns out to be particularly pronounced if sample sizes are small, if is close to the mode, or perhaps even to the right of it, or if it has been incorrectly identified. For this reason, it has been demonstrated that truncated regression is often more accurate if the standard deviation of the sample height distribution is simultaneously constrained to be the modern value (among men) of ca. 6.86 cm (2.7 in). The constrained truncated regression estimator with sigma thus constrained is frequently more reliable and has greater precision (A’Hearn 2004). As a consequence, we run the truncated regressions4 two ways: a) allowing the program to determine the standard deviation of the height distribution freely, and
b) constraining the standard deviations to be 6.86 cm. We refer to the former estimates as unconstrained and to the latter one as constrained.
When Congress authorized a force of 1,000 Indian scouts in 1866, the U.S.
Army began for the first time to formally include Indians in the military (Dunlay, 1982:
44). The use of scouts was the continuation of a long history of Indians serving as auxiliary troops or as allies fighting alongside American or other nations’ soldiers against enemy tribes. The years 1866 to1890 marked the end of warfare between Indians and the United States. Most of the roughly 270,000 Indians in the United States at that date were at peace5 (Utley, 1973:5).
Why were Indians willing to serve in the army or ally themselves with the American forces? Typically Indians saw themselves first as members of families, then clans, then a tribe, but they often saw members of other tribes as different from themselves and with good reason. Enemy tribes often raided for horses or slaves from neighbors; there were battles over territory, and revenge raids to retaliate for murdered relatives (Utley, 1973:5). Thus, serving with the American army did not necessarily create a moral dilemma for Indian scouts. Hostile Indians’ mobility and familiarity with the terrain made guerilla warfare effective and it sometimes required large numbers of troops to confront relatively small numbers of fighters. In such warfare, Indian scouts provided vital services as guides and interpreters.
After 1866 the army was often assigned the task of confining Indian tribes to defined lands “reserved” for Indians – reservations. Typically this was done by signing a treaty, with a tribe ceding tribal territory in return for the right to a reduced territory and goods to be provided by the federal government. For plains tribes in this era there was continuing pressure on the key resource – the bison herds. By the late 1870s the bison herds had largely been depleted and most plains tribes depended upon food issued by the government and lived in encampments near the agency.
Even after they were defeated and confined to reservations, however, bands of Indians would occasionally leave the reservation and were then subject to capture by the army who would return them. Battles in this period often represented last ditch stands by Indians who did not want to move to a reservation.6 While fighting in the northern plains and Rocky Mountain States was greatly diminished after 1880, fighting continued in the southwestern territories of Arizona, New Mexico, and Texas until 1886 with the surrender of the Apache warriors led by the leader known to whites as Geronimo. General George Crook, perhaps the most able of the military leaders fighting in the West, found it essential to recruit Apaches to fight the hostile members of the same tribe (Utley, 1973: 378). The continued warfare in the Southwest is probably why Arizona is the most common state of origin for scouts in the sample about to be analyzed.
By the 19th century most military in economically advanced countries (including the US) recorded the height of soldiers in order to have a physical description in case of desertion and in order to document that the soldier met the height requirements. Height requirements were imposed inasmuch as short men were at a disadvantage in hand-to-hand combat and exceptionally tall men were not suitable for the cavalry on account of the high center of gravity. In order to estimate the height of Native American men, data on the height of scouts were extracted from the National Archives (N=12,999) (Table 2).7 Information available includes height, age, state of birth, date of enlistment, and occupation prior to enlistment. Indians were eminently suitable as scouts because they knew the local terrain the best. The minimum and maximum height requirement to be eligible to be in the U.S. military also applied to scouts.8 We do not know about other possible requirements, but assume that within the acceptable range of heights the men were a random sample from their respective population. The distribution of adult heights is perfectly normal between the range of 66-75 inches. Outside of this range there does appear to be an obvious shortfall9 (Figure 4).The fact that the distributions are normal enables us to use truncated regression in order to correct for the height restrictions. The use of truncated regression enables us to infer the height of the general population of Indian men from that of the scouts.