Combining Multiple Complex Span Tasks into a Single Working Memory Measure

Purpose: The adequate measurement of working memory (WM) capacity presents some limitations. One of the main challenges is that WM assessment is confounded by the level of expertise of individuals in the particular domain required to perform the task, such as verbal fluency in the case of the speaking span task. Another drawback is that there is little psychometric evidence to support the use of complex span tasks as measures of WM capacity. Therefore, it is not clear if these assessment tools do in fact measure the theoretical construct they are intended to measure (i.e., WM) or something else. The purpose of this study was to address these shortcomings by developing a new measure that combines the listening and the speaking span tasks and collecting validity evidence for its use through the Rasch model. Methodology: The participants were 290 Japanese high school students who were administered the speaking and the listening span tasks for which I collected validity evidence in Bazan (2020) and Bazan (2021), respectively. Both tasks were performed individually on a face-to-face basis with the stimuli being played on a computer that I operated. Performance was audio-recorded and scored dichotomously (i.e., right or wrong) using the same scoring system as in the two previous studies. That is, a credit was given for each item recalled successively in the order of appearance until memory failure to recall in order. Scores were put together and analyzed as if they belonged to a single test through the Rasch dichotomous model. The analysis involved an evaluation of whether later presented items within a set increased in difficulty as predicted by WM theory, person and item fit to the Rasch model, person and item reliability and separation, and the dimensionality of the combined WM measure. Findings: The Wright map confirmed a hierarchy of item difficulty consistent with the theoretical expectation that the further the item appears within the set, the more difficult it should be. Over 96% of participants and almost all items fit the Rasch model, with person and item reliability indices demonstrating high replicability of the ordering of the persons’ ability and item difficulty across similar samples. Person separation indicated that the measure is sensitive enough to separate participants into three levels of the construct (i.e., high spans, average spans, and low spans) whereas item separation showed that the items can be divided into 9 levels of difficulty, which is excellent according to the guidelines. The examination of dimensionality revealed that the combined measure taps into a single unidimensional construct, namely WM capacity. Unique Contribution to Theory, Practice and Policy: This study provides evidence for the usefulness of the combining approach to mitigate the influence of domain-specific skills on WM measurement.