?
Новые источники информации в компьютерном тестировании
Traditionally, psychometrics is concerned with theory-based information about human behavior – indicators of the targeted construct, like item responses, performance assessment products, etc. However, over the past forty years, advances in psychometric modeling and the development of information technologies allowed for the analysis of the so-called collateral information. This information is not theory-based and easy to collect in computerized testing. However, most importantly, collateral information is intended solely to increase the reliability of measurements preserving the construct’s original interpretation. This article distinguishes between target and collateral information gathered during computerized testing. A carefully crafted measurement model is required to properly process collateral information along with target information. Social scientists usually choose Item Response Theory (IRT) models as such measurement models due to their clear interpretation, facilitating the discussion of the results of measurements in terms of social sciences. Since the choice of the correct IRT-model is crucial for preserving the original interpretation of the parameter estimates, it is possible to use the classification of such models to describe sources of collateral information systematically. This article introduces a classification of sources of collateral information based on the type of data they describe: (i) collateral information about respondents, (ii) collateral information about items, (ii) collateral information about interactions between respondents and items. The latter type of collateral information is particularly intriguing. Typically, it includes such types of data as item response times, response strategies, actions log data, gaze data, and other types of process data. Additionally to IRT modeling, examples of process mining and sequence pattern mining are also provided as examples of collateral information. The article illustrates the use of collateral information in educational psychometrics with a recent literature review. We describe cases where the measurement model’s choice changes the interpretation of the IRT parameter estimates, which causes the breaking of the conditions defining collateral information. There is large- and small-scale educational and psychological research among cases. We also highlight the most illustrative cases of using collateral information in modern psychometric practice with regard to its source and the IRT-model used to process it. Moreover, we demonstrate that using the new sources of information in computerized testing contributes to developing evidence-based pedagogical practices and makes their application more manageable. The directions for future research in the area of collateral information in psychometrics are provided.