Evidential value of physicochemical data—comparison of methods of glass database creation

Glass databases used in forensic laboratories are commonly generated by collection of samples from a very small area of many different objects (Method 1). These frequently yield univariate data in the form of refractive indices measured using the Glass Refractive Index Measurement (GRIM) technique, or multivariate data in the form of elemental compositions obtained by the SEM‐EDX technique. We investigate whether the within‐ and between‐object variances and covariances estimated from data collected using Method 1 are as reliable as those obtained from databases formed by samples collected from several areas of an object (Method 2). Reliability of the Method 1‐ and Method 2‐estimated parameters is evaluated in terms of the performance of likelihood ratio (LR) models that measure the evidential value of physicochemical data for forensic purposes. Two‐level random effect models assuming that the within‐object distribution is normal, and that the between‐object distribution is either normal or obtained by kernel density estimation (KDE), are applied to comparison of glass fragments of known origin, and the rates of false positive and false negative answers recorded. These rates are similar for Methods 1 and 2 for the refractive index (RI) data, and Method 1 performs better than Method 2 for elemental composition data, suggesting that the current method of database generation is appropriate for the estimation of these sources of variability for glass samples. Copyright © 2010 John Wiley & Sons, Ltd.