Cognitive models of decision-making with identifiable parameters: Diffusion decision models with within-trial noise

Diffusion Decision Models (DDMs) are a widely used class of models that assume an accumulation of evidence during a quick decision. These models are often used as measurement models to assess individual differences in cognitive processes such as evidence accumulation rate and response caution. An underlying assumption of these models is that there is internal noise in the evidence accumulation process. We argue that this internal noise is a relevant psychological construct that is likely to vary over participants and explain differences in cognitive ability. In some cases a change in noise is a more parsimonious explanation of joint changes in speed-accuracy tradeoffs and ability. However, fitting traditional DDMs to behavioral data cannot yield estimates of an individual's evidence accumulation rate, caution, and internal noise at the same time. This is due to an intrinsic unidentifiability of these parameters in DDMs. We explored the practical consequences of this unidentifiability by estimating the Bayesian joint posterior distributions of parameters (and thus joint uncertainty) for simulated data. We also introduce methods of estimating these parameters. Fundamentally, these parameters can be identified in two ways: (1) We can assume that one of the three parameters is fixed to a constant. We show that fixing one parameter, as is typical in fitting DDMs, results in parameter estimates that are ratios of true cognitive parameters including the parameter that is fixed. By fixing another parameter instead of noise, different ratios are estimated, which may be useful for measuring individual differences. (2) Alternatively, we could use additional observed variables that we can reasonably assume to be related to model parameters. Electroencephalographic (EEG) data or single-unit activity from animals can yield candidate measures. We show parameter recovery for models with true (simulated) connections to such additional covariates, as well as some recovery in misspecified models. We evaluate this approach with both single-trial and participant-level additional observed variables. Our findings reveal that with the integration of additional data, it becomes possible to discern individual differences across all parameters, enhancing the utility of DDMs without relying on strong assumptions. However, there are some important caveats with these new modeling approaches, and we provide recommendations for their use. This research paves the way to use the deeper theoretical understanding of sequential sampling models and the new modeling methods to measure individual differences in internal noise during decision-making.