Quantifying performance in magnitude comparison tasks using a drift-diffusion model

We investigate the viability of the drift-diffusion framework to account for behaviour on magnitude comparison tasks. Data from both published studies on magnitude comparison and a simulation are analysed to estimate the key drift-diffusion model parameters, using the EZ-diffusion method and HDDM package. All methods resulted in linear mappings between drift rate and difficulty (indexed using 1 - smaller/larger), with an intercept that was consistently close to zero for non-symbolic tasks. The EZ method was rapid and simple to apply, but subject to bias when using aggregate data or when accuracy was very high. In contrast, the HDDM tool produced results that were less biased, but individual differences were under-estimated. We conclude that application of parameter estimation methods, particularly in research on individual differences, requires careful consideration of their limitations.