Likelihood Ratios in Forensic Statistics: When or When Not to Use Them

There has been significant controversy about the use of likelihood ratios in forensic statistics.  For analysis of single source DNA samples the likelihood ratio is well accepted.  Applying the approach more broadly has proved difficult because the likelihood ratio requires assumptions about the probability distributions of observed evidence under different hypotheses.  For some complex evidence types, such as latent prints or questioned documents, developing appropriate distributions is difficult. In these cases alternative approaches that resemble the likelihood ratio have been developed.  This talk reviews the argument for the likelihood ratio, discusses the current state of affairs for several evidence types, and addresses remaining challenges to widespread use.