Abstract
Forensic science evidence is a crucial part of the criminal justice system, and forensic examiners have long been held in high regard due to their ability to link suspects to crimes in an objective and scientifically rigorous way. Recently, however, concerns have been raised that forensic examiners could be susceptible to contextual bias, in which exposure to information not relevant to the examination could unduly influence their conclusions. Contextual bias has now been demonstrated in a wide range of forensic disciplines (e.g., fingerprints, DNA). The field of handwriting analysis, however, has received relatively little attention. Handwriting analysis is likely to be particularly susceptible to bias, because examiners are asked to opine on a product of a human behaviour that varies widely—even across samples written by the same person. In this thesis, we limited our focus to signature examinations, which involve very small samples of handwriting.
In Study 1, we used a signal detection framework to examine the effects of task-irrelevant contextual information on laypeople’s judgements about questioned signatures. In Study 1A, we presented participants with 20 trials comprising one questioned signature and four known signatures. On each trial, participants were asked to indicate whether they thought the questioned signature was genuine or forged. Half of the trials were accompanied by high-biasing contextual information (e.g., stating that the author of the questioned signature had a criminal record); the remaining trials were accompanied by low-biasing information (e.g., stating that the questioned signature was written by a suspect). Contrary to our expectations, the contextual manipulation did not significantly influence participants’ ability to discriminate between genuine signatures, nor did it affect their response bias. In Study 1B, we addressed several limitations of Study 1A that could have prevented us from detecting a context effect. This time, we observed that high-biasing contextual information not only increased participants’ response bias; it also reduced their ability to discriminate between genuine and forged signatures. This latter finding suggested that contextual information might influence the process by which participants examined the signatures.
In Study 2, we therefore focused on the signature examination process using a similar paradigm. We explicitly asked participants either to focus on similarities or differences between the questioned and known signatures. We hypothesised that if high-biasing contextual information led participants to use a positive test strategy, then we should be able to eliminate the context effect by forcing participants to identify similarities between the signatures (i.e., to engage in a negative test strategy). This hypothesis was not supported, however, suggesting that efforts to change the signature examination process per se could be insufficient to mitigate contextual bias. As such, it is important to develop practical and effective ways to either limit exposure to contextual information, or to find ways to present contextual information in a way that is least likely to influence examiner’s opinions.
For this reason, in Study 3, we turned our attention to practitioners. Specifically, we explored the current state of contextual information management in questioned document examination by conducting interviews with 19 international professional document examiners—both from government laboratories and private practice. As well as canvassing practitioners’ views on contextual bias, we sought information about the sources of contextual information that they encounter, the information that is relevant—and not relevant—to their examinations, their methods for reducing the potential for bias, and the perceived barriers to implementing context management strategies. Based on our findings, we provide recommendations for developing and implementing practical and effective context management systems—both for forensic science in general, and handwriting analysis in particular.