Proposed Method for Predicting Pair Matching of Skeletal Elements Allows Too Many False Rejections

Abstract

Byrd proposes a method for predicting pair matches in commingled remains to reduce visual comparison. The method compares differences between left and right postcranial element measurements in commingled samples with differences in known pairs from a reference sample using a t-score approach. We duplicated his protocol using six elements from two samples of known paired elements (n = 854 to 1063) and calculated the number of pairs correctly predicted. Time commitment was estimated by mathematically attempting matches with all left and right elements in these samples. Although the results show an 86% reduction in the number of potential pairs requiring visual matching, we do not recommend the method because (i) the normality assumption for use of a t-score approach is violated, (ii) no account is made for bilateral asymmetry, and (iii) the high rate of false rejections (up to 22%) undermines its ability to show true incompatibilities for potential pair matches.

The Power of Exclusion using Automated Osteometric Sorting: Pair‐Matching,

“…It has been argued that Byrd and LeGarde's method violates the normality assumption of a t ‐test . This conclusion was based on the lack of normality in the D ‐values of the reference sample as determined by Shapiro–Wilk tests .…”

“…It has been argued that Byrd and LeGarde's method violates the normality assumption of a t ‐test . This conclusion was based on the lack of normality in the D ‐values of the reference sample as determined by Shapiro–Wilk tests . However, using a Shapiro–Wilk test on very large sample sizes will often result in a significance value indicating non‐normality in the majority of scenarios, even if the sample is normally distributed.…”

Univariate and Linear Composite Asymmetry Statistics for the “Pair‐Matching” of Bone Antimeres

“…The value of D should consequently be −55, but in both the 2008 and the 2014 Tables the listed value is 55. This led Vickers et al (:103) to suggest that in both Tables what was actually used was the “ absolute value of sum difference,” or:…”

“…This is performed by estimating an exponential power distribution for the linear composite. The “absolute value of sum difference” suggested by Vickers and co‐workers (:103) can then be shown to follow a folded exponential power distribution. For the “summed absolute value of differences” as described by Lynch and co‐workers (:2), it was necessary to fit a skew exponential power distribution .…”

“…To make matches on the basis of measurements, it is necessary to first characterize the directional and fluctuating asymmetry for paired elements. Previous methods that rely on multiple measurements have generally used a linear composite, such as the sum of differences between left and right side measurements . After finding that this linear composite did not follow a normal distribution, Vickers et al suggested using the sum of absolute differences between sides.…”