Abstract:
|
Item Response Theory (IRT) is often used to estimate participant proficiency and item difficulty in educational testing and has recently been proposed for use in forensic science. In testing situations, it is increasingly common to collect collateral information — such as time spent on each question or perceived item difficulty — alongside responses to each item. There are therefore multiple responses for each participant × item pair that may depend on shared latent variables. For instance, reported difficulty may be related to both proficiency and ‘true’ item difficulty. We present an approach to modeling multiple observed variables using a single latent variable model. This IRT-like model is applied to data in which forensic examiners report both a source conclusion (whether a crime-scene sample matches a reference sample) and a subjective rating of the difficulty of the comparison task. We find that there are some participants who over or under-report difficulty after adjusting for proficiency, item difficulty, and other participants’ ratings. We also discuss the implications of these findings within the criminal justice system and outline future directions of study.
|