Linear discriminant analysis (LDA) is frequently used for classification/prediction problems in physical anthropology, but it is unusual to find examples where researchers consider the statistical limitations and assumptions required for this technique. In these instances, it is difficult to know whether the predictions are reliable. This paper considers a nonparametric alternative to predictive LDA: binary, recursive (or classification) trees. This approach has the advantage that data transformation is unnecessary, cases with missing predictor variables do not require special treatment, prediction success is not dependent on data meeting normality conditions or covariance homogeneity, and variable selection is intrinsic to the methodology. Here I compare the efficacy of classification trees with LDA, using typical morphometric data. With data from modern hominoids, the results show that both techniques perform nearly equally. With complete data sets, LDA may be a better choice, as is shown in this example, but with missing observations, classification trees perform outstandingly well, whereas commercial discriminant analysis programs do not predict classifications for cases with incompletely measured predictor variables and generally are not designed to address the problem of missing data. Testing of data prior to analysis is necessary, and classification trees are recommended either as a replacement for LDA or as a supplement whenever data do not meet relevant assumptions. It is highly recommended as an alternative to LDA whenever the data set contains important cases with missing predictor variables.

Transabdominal and transvaginal ultrasound (US) examinations were performed in 184 asymptomatic postmenopausal volunteers to determine prospectively (a) the frequency and natural history of simple adnexal cysts in healthy postmenopausal women and (b) the relationship between cyst activity and both hormone replacement and length of time since menopause. Eighty-three simple adnexal cysts were found in 52 women. Thirty-two of 184 women (17%) had 37 cysts identified at initial examination; 46 new cysts appeared in 31 women (11 of whom previously had cysts). Forty-nine women with 72 cysts were reevaluated with subsequent US scanning over a period of 3-23 months. Thirty-eight of the 72 cysts (53%) disappeared completely, 20 (28%) remained constant in size, eight (11%) enlarged by 3 mm or more, two (3%) decreased in size by 3 mm or more, and four (6%) both increased and decreased in size on repeated examinations. No statistical relationship was found between presence of cysts or cyst activity with respect to the type of hormone replacement or length of time since menopause.

In previous limited investigations of the human femur/stature ratio we (Feldesman and Lundy: Journal of Human Evolution 17:583-596, 1988; Feldesman et al.: American Journal of Physical Anthropology 79:219-220, 1989) have shown it to be remarkably stable across ethnic and gender boundaries. In this study we evaluate the femur/stature ratio in 51 different "populations" of contemporary humans (n = 13,149) sampled from all over the world. We find that the mean ratio of femur length to stature in these populations is 26.74%, with a very restricted range of variation. When we compare mean femur/stature ratios of males and females, there are no statistically significant differences. ANOVA performed on a naive grouping of samples into "whites," "blacks," and "Asians" indicates that there are significant racial differences (P less than 0.001). When we subject these groups to Tukey's HSD procedure (a post-hoc test), we find that "blacks" are responsible for the significant ANOVA, being significantly (P less than 0.005) different from the other ethnic groups. "Whites" and "Asians" are not significantly different (P = 0.067) under the conditions of this analysis, although all these racial comparisons may be suspect given the small sample sizes. We tested the efficacy of the ratio in three situations: predicting stature of repatriated white Vietnam veterans; predicting stature in a random sample of South African blacks (of known stature), and predicting the stature of a single Akka pygmy. In the first and third cases, the femur/stature ratio does better than the traditionally recommended regression equation, while in the second case the predictions from the femur/stature ratio are less accurate than from the appropriate regression equation. These results encouraged us to apply this ratio to mid- and late-Pleistocene fossil hominids, where the choice of reference population for stature estimates continues to trouble workers. We estimated stature for a sizeable number of Homo erectus (HE), early Neanderthal (EN), Near Eastern Neanderthal (NEN), and early anatomically modern Homo sapiens (EAMHS) by using the simple relationship: stature (cm) = femur length (cm) * 100/26.74. Our results show that HE fossils are slightly taller on average than either EN or NEN samples, which do not differ significantly in stature, while EAMHS fossils are significantly taller than all three earlier groups. While these results are not surprising, our stature estimates for these fossils differ from currently published estimates based on sample-specific regression-based formulae.(ABSTRACT TRUNCATED AT 400 WORDS)

The present study examines the relationship between femur length and stature in children between the ages of 8 and 18 years. In previous investigations, my colleagues and I reported the surprising finding that femur length bears a nearly constant relationship to stature in adult humans regardless of ethnicity or gender. This earlier study revealed that the femur/stature ratio averages 26.74% in adult humans, and that using the ratio to predict stature from femur length yields remarkably accurate estimates. The current study shows that femur/stature ratios of children between the ages of 8 and 11 differ significantly from their older counterparts. Between the ages of 12 and 18, there are no significant differences due to age in the femur/stature ratio; however, there are significant differences in this age group attributable to gender. This study also shows that the worldwide average adult femur/stature ratio does not adequately describe children in this age range. This study strongly documents the adolescent growth spurt in the femur/stature ratios of both males and females at the precise time one would expect to see the spurt occur (10-12 in females; 12-14 in males). This growth follows a nearly identical trajectory in both genders, with relative femur growth dominating before the peak years of the growth spurt, and relative stature growth dominating afterward. This accounts for the ratio's rise to maximum values just before peak growth, and its decline toward the adult ratio thereafter. These findings require us to use separate adolescent femur/stature ratios of 27.16 (females) and 27.44 (males) to estimate the stature of children between the ages of 12 and 18. Preliminary testing shows these ratios to be more accurate in estimating stature than the properly selected Trotter and Gleser adult regression equation. Use of the adolescent male ratio with the Homo erectus juvenile WT 15000 results in a lower stature estimate (157.4 cm) than previously reported. It is suggested that continued testing of the ratio occur, but that the values herein derived may be useful in routine forensic cases involving children in this age range, and with subadult paleontological specimens.

No abstract