DNA fingerprinting for forensic identification: some statistical issues.

In DNA fingerprinting work, it is well known that, although fragment lengths of a sample are uncorrelated, measurement errors of fragment lengths are correlated. Unless the fragment length is very large, the measurement errors are directly proportional to fragment lengths. In the currently practiced matching procedures for forensic identification, these facts are ignored or inadequately used. Some of these issues are addressed in this study. Based on available empirical observations, an appropriate statistical model and techniques are proposed. These techniques lead to the definition and construction of elliptical match windows, instead of independent interval (linear) windows currently used. Simulation experiments indicate that the performance of the proposed procedure is superior to the current procedures. Further, the proposed procedure does not make use of some assumptions underlying the current procedures. Extensions of the procedure to utilize replicate measurements of crime-scene and/or suspect's samples are indicated.