Assessing model-based inferences in decision making with single-trial response time decomposition

The latent psychological mechanisms involved in decision-making are often studied with quantitative models based on evidence accumulation processes. The most prolific example is arguably the drift-diffusion model (DDM). This framework has frequently shown good to very good quantitative fits, which has prompted its wide endorsement. However, fit quality alone does not establish the validity of a model's interpretation. Here, we formally assess the model's validity with a novel cross-validation approach based on the recording of muscular activities, which directly relate to the standard interpretation of various model parameters. Specifically, we recorded electromyographic activity along with response times (RTs), and used it to decompose every RT into 2 components: a premotor time (PMT) and motor time (MT). The latter interval, MT, can be directly linked to motor processes and hence to the nondecision parameter of DDM. In two canonical perceptual decision tasks, we manipulated stimulus strength, speed-accuracy trade-off, and response force and quantified their effects on PMT, MT, and RT. All 3 factors consistently affected MT. The DDM parameter for nondecision processes recovered the MT effects in most situations, with the exception of the fastest responses. The extent of the good fits and the scope of the mis-estimations that we observed allow drawing new limits of the interpretability of model parameters. (PsycInfo Database Record (c) 2021 APA, all rights reserved).