Statistics Seminars: The Evaluation of Evidence for Autocorrelated Data with an Example using Traces of Cocaine on Banknotes
25 November 2013 14:00 in CM221
Much research in recent years for evidence evaluation in forensic science has focussed on methods for determining the likelihood ratio where the data have been generated by various random phenomena. The likelihood of the evidence is calculated under each of two proposi- tions, that proposed by the prosecution and that proposed by the defence. The value of the evidence is given by the ratio of the likelihoods associated with these two propositions. One form of evidence evaluation is related to discrimination in which the problem is one of source identity. The two propositions are that the source is or is not associated with criminal activity. The aim of this research is to evaluate this likelihood ratio under two explanations, one an extension of the other, for the random phenomena by which the data have been generated. The first is when the evidence consists of continuous autocorrelated data. The second is when the observed data are also believed to be driven by an underlying latent Markov chain. Four models have been developed to take these attributes into account: an autoregressive model of order one, a hidden Markov model with autocorrelation of lag one and a nonparametric model with two different bandwidth selection methods. Application of these methods is illustrated with an example where the data relate to traces of cocaine on banknotes. The likelihood ratios using these four models and one based on an assumption of independence are calculated for these data, and the results compared.
The research is supported by an EPSRC CASE award, voucher number 009002219.
Contact firstname.lastname@example.org for more information