The Skull That Broke Race Science

Deep Dive

The Skull That Broke Race Science

Chapter 3 of the tour tells this story in four short beats: Franz Boas measured thousands of European immigrants and their American-born children in the early 1900s, found their head shapes had shifted across a single generation, and used the finding to argue against the idea that head form was a fixed racial marker. Ninety years later, two teams of anthropologists went back to his original numbers with modern statistics and reached opposite conclusions about whether he was right. The tour chapter names that disagreement. This deep-dive works through it — what data Boas actually collected, what specific statistical machinery each side of the 2002–2003 exchange brought to bear on it, and what each side actually found, in their own reported numbers.

This is worth doing carefully because the disagreement is real and still, in a meaningful sense, unresolved. Nothing here is debunked pseudoscience of the kind traced in the "Reading the Face" deep-dive — Boas, Sparks and Jantz, and Gravlee, Bernard, and Leonard are all working in good faith, with defensible methods, on the same publicly available dataset, and they do not agree on what it shows. That is not a flaw in the process. It is what an honest, still-open methodological argument in physical anthropology looks like from the inside.

What Boas actually measured

Between 1908 and 1910, Franz Boas and a team of thirteen assistants took anthropometric measurements on 17,821 European immigrants and their children in New York City, under a commission from the U.S. Immigration Commission.¹ The sample was stratified into seven immigrant groups: East European Hebrews (the largest, at roughly six thousand individuals), Bohemians, Sicilians, and Neapolitans (each represented by about three thousand people), and smaller samples of Poles, Hungarians and Slovaks, and Scots. Boas and his team also aimed to assess general physiological development, but because subjects couldn't be measured unclothed, only stature survived from that part of the plan across the full sample. To capture what he called the "characteristic racial types" of each group, they separately recorded maximum head length, maximum head width, the width of the face between the cheekbones (bizygomatic breadth), and hair, eye, and skin color. Boas excluded eye and skin color from his main analysis over concerns about standardizing them across observers, and gave hair color only brief, secondary treatment.

The core comparison was the cephalic index — head breadth divided by head length, multiplied by 100 — the same ratio nineteenth-century craniometrists had used to sort populations into fixed types. Boas ran three linked analyses to test whether that ratio held steady across a generation raised in a different environment. First, he compared American-born children to foreign-born children within each immigrant group. Second, recognizing that different birth cohorts might simply reflect different waves of immigrants, he split the American-born children further, comparing those born within ten years of their mother's arrival to those born more than ten years after — reasoning that if the U.S. environment was doing the work, its effect should grow stronger with longer exposure. Third, and most directly, he compared children to their own immigrant parents rather than to the group average, on the logic that if American-born children diverged further from their own parents than foreign-born children did from theirs, environment rather than a shift in immigrant stock was the more plausible explanation.

What he concluded — and how big the numbers actually were

Boas reported that the cephalic index shifted measurably between the foreign-born and American-born generations, in every immigrant group he tested, though not by the same amount or even in the same direction. In his summary table of mean differences, American-born Hebrew males ran about two points lower in cephalic index than their foreign-born counterparts, while American-born Sicilian males ran about one to two points higher — the two groups moving toward each other rather than both drifting the same way.¹ He read the convergence as the more striking finding: traits that had been treated as the most stable, most diagnostic markers of "race" in Europe were, in his data, converging toward a common American type within a single generation. In his own words from the preliminary report, "not even those characteristics of a race which have proved to be most permanent in their old home remain the same under our new surroundings."

Boas anticipated that a finding this contrary to expectation would draw fire, and it did — the Immigration Commission itself distanced itself from his conclusions, and European and American colleagues raised objections ranging from measurement error to sampling bias almost as soon as the preliminary report appeared. In 1928, well into a career defined partly by this dispute, Boas published the full, handwritten raw dataset — more than five hundred pages of it — so that other researchers could check his work directly rather than take his summary tables on faith. For most of the twentieth century, few did; the dataset sat largely unexamined by anyone applying inferential statistics that Boas himself did not have available in 1912.

2002: a heritability challenge

In 2002, Corey Sparks and Richard Jantz published a reassessment of Boas's data in the Proceedings of the National Academy of Sciences, using statistical tools that were not available to Boas or, for that matter, to most of the twentieth-century researchers who cited his finding.² Working from more than eight thousand individuals in Boas's dataset, they took two approaches. First, they used the parent-child pedigree information embedded in Boas's records — mothers, fathers, sons, and daughters were all identifiable — to estimate narrow-sense heritability by maximum likelihood, a quantitative-genetics technique for partitioning how much of the variation in a trait is attributable to additive genetic effects versus environment. Second, they ran a large battery of t-tests directly comparing American-born and European-born children within each immigrant group, the same basic comparison Boas had made, but now with modern significance testing applied to it.

Both lines of evidence pointed the same direction, in their reading. The heritability estimates for head length and head breadth both came out above 0.5 — meaning, by their interpretation, that most of the variation in those measurements tracked genetic relatedness rather than which side of the Atlantic someone was born on; facial breadth came out somewhat lower, implying a larger environmental component for that one trait. On the direct group comparisons, only 156 of a possible 448 t-tests could even be run, because many age-and-sex subgroups were too small to test reliably — and of those 156, only 11 showed a statistically significant difference between American-born and European-born children, all of them in the Hebrew sample. Sparks and Jantz concluded, in their own words, that when Boas's data "are subjected to a modern analysis, they do not support his statements about environmental influence on cranial form" — a direct challenge to ninety years of the finding being cited as settled.

The reanalysis that reopened the case began, by Sparks's own account, as a professor handing a book to a student. Sparks was Jantz's master's advisee at the University of Tennessee when Jantz pulled his copy of Boas's data volume off the shelf and said, "Corey, know what you need to do?"³

2003: a different model, a different reading

Clarence Gravlee, H. Russell Bernard, and William Leonard had been working on their own reanalysis of the same dataset independently, unaware of the Sparks and Jantz paper until it appeared in print; their own reanalysis was published a few months later, in the March 2003 issue of American Anthropologist.⁴ Rather than treating birthplace as a simple two-group split, they modeled cephalic index using analysis of covariance (ANCOVA), controlling for age and sex and testing an interaction term between birthplace and immigrant group directly — a design choice motivated by Boas's own observation that the shift ran in different directions for different groups. That interaction term came out strongly significant, meaning the birthplace effect genuinely varied group to group rather than reflecting one uniform pattern. Broken out by group, the American-born versus foreign-born difference in age- and sex-adjusted cephalic index was highly significant for the four largest samples — Hebrews, Sicilians, Central Italians, and Bohemians, each at p < .001 — while the three smallest groups (Poles, Scots, Hungarians and Slovaks) showed weaker or no detectable effect, a pattern the authors attributed to those subsamples simply being too small to reliably detect a real but modest difference.

Gravlee, Bernard, and Leonard also went further than a group-mean comparison, running parent–offspring correlations separately for U.S.-born and foreign-born children. Head form tracked much more closely between foreign-born children and their parents than between U.S.-born children and their parents — the correlation between a child's cephalic index and the average of both parents' cephalic index (the "midparent" value) ran meaningfully higher for foreign-born children than for American-born children in their analysis, and the gap widened further for children born more than ten years after their mother's arrival, consistent with Boas's original claim that the environmental effect accumulated with longer exposure. A separate regression treating time since the mother's arrival as a continuous variable, rather than the two-group split Boas's 1912 computational tools required, did find a statistically significant relationship between time elapsed and cephalic index in the two largest groups — but the authors were candid that the relationship, while statistically real, was modest in size, explaining only a small share of the overall variation in head shape. Taken together, they concluded that "modern analytical methods provide stronger support for Boas's conclusion than did the tools at his disposal" — the opposite verdict from Sparks and Jantz, drawn from the same raw measurements.

Gravlee's case for environment over biology did not stop at skulls. Two years after this reanalysis, he reported that among adults in one southeastern Puerto Rican community, blood pressure tracked how darkly a person was socially classified rather than how dark their skin actually measured — the same logic his cranial reanalysis applied to head shape, run on a different trait.⁵

The rebuttal, and where it landed

Sparks and Jantz did not concede the point. Later in 2003, the journal ran the exchange as a named debate feature, headed "Did Boas Get It Right or Wrong?" — Sparks and Jantz's response appeared in the June issue of American Anthropologist, defending their original approach.⁶ Their core objection was less about any single number and more about what kind of evidence should carry more weight: a heritability estimate derived directly from the parent–offspring pedigree structure in Boas's own data, they argued, was a more direct test of how much of cranial form is genetically transmitted than a set of group-mean comparisons and regression coefficients, however statistically significant those comparisons came out. A birthplace effect can be real and still be small relative to the underlying heritable variation in a trait — and a moderately sized, statistically significant shift, they maintained, is not automatically evidence against a high heritability estimate measured a different way.

Neither side treated the exchange as resolved. Both papers were, by design, working from the identical raw measurements Boas published in 1928; the disagreement traced back to a genuinely contestable modeling choice — whether to center the analysis on heritability estimation from pedigree structure, as Sparks and Jantz did, or on direct, covariate-adjusted comparison of birthplace groups, as Gravlee, Bernard, and Leonard did — and different defensible choices produced different headline conclusions from the same numbers.

Two legitimate readings of one dataset

It is worth being precise about what this disagreement is and is not. It is not a case of a debunked claim being quietly retracted, and it is not a case of a modern team catching an old one in an error of measurement or fraud — the data-entry error rate Gravlee, Bernard, and Leonard found in Boas's handwritten records, checked page by page against the original, came out under half a percent. It is two statistically sophisticated teams applying different, individually defensible analytical frameworks to the same public dataset and arriving at different headline conclusions: Sparks and Jantz's heritability-first framework reads the data as showing head form is substantially genetically determined and only trivially responsive to birthplace; Gravlee, Bernard, and Leonard's covariate-adjusted group-comparison framework reads the same measurements as showing a real, if modest, environmentally driven shift, concentrated in the largest and best-measured samples and more pronounced in children with longer exposure to the American environment.

Both readings survive contact with the raw numbers. Sparks and Jantz are correct that most direct European-versus-American comparisons in the full 156-test battery did not reach significance, and that measured heritabilities for head length and breadth ran above 0.5. Gravlee, Bernard, and Leonard are correct that, once birthplace is modeled with age, sex, and group-specific interaction terms rather than pooled into one blunt comparison, the largest four immigrant samples show highly significant shifts, and that parent–offspring similarity is measurably weaker for American-born children than for their foreign-born counterparts. Neither team has been shown to be using the data wrong. What they disagree about is which statistical lens is the right one to point at a century-old, non-randomly sampled pedigree dataset — and that is a genuinely open methodological question in quantitative genetics and biological anthropology, not a question about whether race is biologically fixed. On that larger point, the point Boas was actually arguing in 1912, both sides of this exchange agree: nothing in either reanalysis revives the idea that head form sorts populations into fixed, heritable racial types the way nineteenth-century craniometry claimed. What remains genuinely contested is a narrower and more technical question — how much of the shift Boas measured in one specific dataset was environment, how much was genetics, and which statistical model best answers that question from data collected before either team's methods existed.

The 2003 exchange was not the last word even for its own participants. In 2010 Jantz returned to Boas's numbers and traced the two largest shifts to specific causes — American-born Hebrew children's head shape changed as families gave up the old infant-cradling practices, American-born Sicilian children's as they grew up under conditions worse than their parents had known in Europe — while still maintaining the overall shifts were smaller than Boas's summary implied.⁷ The dispute deepened rather than closed.

Further reading