Keywords: Expert testimony, forensics, likelihood ratio, score-based inference
The correct analyses and interpretation of evidence play a fundamental role in the fair administration of justice. The United States incarcerates more people than almost any other country in the world, in many cases based on erroneous "science", on expert testimony that exceeds what science can support, or both. Many of the more popular forensic tools lack a statistical foundation that might permit estimation of error rates or other measures of uncertainty when the expert testifies that a sample collected at the crime scene has the same source as a sample collected from the suspect. Statisticians have proposed the use of likelihood ratios to determine whether the evidence supports the prosecutor or the defense hypotheses in a trial. But even if we knew how to compute a likelihood ratio for, e.g., fingerprints, research has shown that jurors do not know how to interpret many types of statistical statements. We present some of the research currently underway at CSAFE (Center for Statistics and Applications in Forensic Evidence) in areas including ballistics, and shoeprint analysis.