The method was developed and refined over decades and has been widely used in forensic anthropology for identifying the population affinity of unidentified remains in legal contexts. It is a legitimate technique with genuine predictive value when applied to populations that are well-represented in reference databases.
The problems arise in specific contexts. The reference databases are built from known populations — mostly populations with well-documented histories and large enough sample sizes to characterize their morphological variation. Early prehistoric populations are not in these databases, because we do not have large samples of well-preserved skeletal remains from them. When a specimen is compared to modern populations, the analysis can identify which modern population it falls closest to, but it cannot directly compare it to the prehistoric population it actually came from.
This matters because populations change over time. The morphological profile of a population — its distribution of skull shapes and facial features — can shift substantially over thousands of years, through natural selection, through genetic drift in small populations, and through admixture with other groups. Modern Native American skulls reflect the morphology of populations as they exist today, after 9,000 years of history that includes all the processes that shape population morphology. They are not necessarily the best comparison for an individual who lived at the beginning of that 9,000-year period.
The statistical resemblance of Kennewick Man’s skull to Ainu and Polynesian populations was real in the sense that the measurements fell closer to those modern populations than to modern Native Americans in the reference database. But interpreting that finding as evidence that Kennewick Man was not ancestrally related to Native Americans required assuming that modern Native American skull morphology is representative of what early Native Americans looked like — an assumption that may not be warranted.
