Abstract
This paper proposes new measures of individual differences in typing behaviour which provide a means of accurately verifying the identity of the typist. A first study examined the efficacy of a multivariate measure of inter-key latencies and a probabilistic discriminator statistic in conjunction with an individual filtering system which eliminates occasional disfluent keystrokes. The results indicate that, under optimum conditions but with a very small test sample, these measures lead to better typist verification than measures suggested earlier by Umphress and Williams and then Leggett and Williams. A second study validated the improved algorithm under more ecologically valid conditions and showed that when training and test sessions were separated by one week, typist verification using the new algorithm achieved combined false-acceptance and false-rejection rates of 0.9% and 3.8% for test samples of 300 and 50 digraphs respectively.