Forensic identification with imperfect evidence

We study the problem of forensic identification when the trace evidence from the scene of the crime is imperfect: for example, it might be measured with error, or be partially missing. A general framework for imperfect data is developed, and applied in particular to the following cases, singly and in combination: 'measurement error' in the recorded information at the scene of the crime; 'binning', i.e. discretisation of an originally continuous crime measurement; 'paternity testing', in which the DNA profiles of the child and of the mother provide partial information on the true father's DNA; two types of 'laboratory error', one in which the error is equally distributed among all possible results, and a second in which there is a bias in the error mechanism producing a false match; and 'partial data', as when there is information on the offender's DNA profile, but not on his racial group.