ABSTRACT Stature estimation from long bone lengths yields varying accuracies due to differences in linear regression fit. We evaluated the impact of the correlation between stature and humeral, femoral, and tibial lengths on the performance of ordinary least squares (OLS), major axis (MA), and reduced major axis (RMA) fits. Anatomically reconstructed stature was used as the dependent Y ‐variable, whereas bone length was used as the independent X ‐variable. The sample consists of 938 skeletons from the European Holocene. Variations in the correlation coefficient were analyzed by bootstrapping the sample using the principal component subsampling technique (PCA), which increased the correlation by reducing extremes in the orthogonal distance to the first principal component fit, and Z ‐score subsampling, which decreased the correlation by removing extremes based on Z ‐score distances from the stature mean. The PCA subsampling technique reduced differences in mean absolute percentage errors between the regression models and resulted in convergence of their slopes. The Z ‐score subsampling technique reduced the agreement in stature estimates among the regressions. OLS and RMA exhibited greater agreement in stature estimation than MA. OLS demonstrated better agreement than RMA in samples with increasing correlation coefficients. However, RMA showed greater consistency across samples with varying correlation coefficients and should be preferred when skeletal samples are small and body proportions are unknown. Stature estimates derived from humeral length were more sensitive to differences in correlation coefficients than those based on femoral or tibial length. Consequently, humeral stature estimates should only be considered reliable when applied to large, proportionally homogeneous samples.
Sládek et al. (Sun,) studied this question.