Do all subjects fit the same recognition memory model? Comparisons of continuous, discrete, and hybrid models using extended multinomial processing trees.

Share

Abstract

In Gutkin et al. (2024), we studied traditional models of recognition memory, such as the two-high-threshold model and signal detection theory, using the benefits of multinomial processing tree models to quantify the effects of both ordinal and continuous variables. Simulations found that including relevant variables, such as response times (RTs) and confidence levels (CLs), significantly improved the estimation accuracy of model parameters and the power of model selection in fitting data from a recognition memory experiment. However, when applying these extended models, we found that no single model was suitable for all subjects. Thus, in the present study, we explored whether recognition memory is best represented as a continuous or discrete process by proposing a hybrid model that extends the original Atkinson–Juola (AJ) model (Atkinson & Juola, 1973, 1974; Juola et al., 1971). Data were sourced from a recognition memory experiment conducted by Juola et al. (2019), involving manipulations of relative target frequencies and measuring RTs and CLs. Our findings indicate that the extended AJ model provides a superior fit for most participants compared to purely discrete or continuous models. Additionally, the patterns of CLs and RTs align with the AJ model’s assumptions about their distributions. Despite these insights, there remains notable variability in cognitive processing and decision-making strategies across individuals. Therefore, there is still a need for a more general model, as suggested in the present article, that could accommodate the diverse strategies and individual differences influencing search, decision making, and response processes underlying recognition memory.