Task-general efficiency of evidence accumulation as a computationally-defined neurocognitive trait: Implications for clinical neuroscience

Abstract

Quantifying individual differences in higher-order cognitive functions is a foundational area of cognitive science that also has profound implications for research on psychopathology. For the last two decades, the dominant approach in these fields has been to attempt to fractionate higher-order functions into hypothesized components (e.g., "inhibition", "updating") through a combination of experimental manipulation and factor analysis. However, the putative constructs obtained through this paradigm have recently been met with substantial criticism on both theoretical and empirical grounds. Concurrently, an alternative approach has emerged focusing on parameters of formal computational models of cognition that have been developed in mathematical psychology. These models posit biologically plausible and experimentally validated explanations of the data-generating process for cognitive tasks, allowing them to be used to measure the latent mechanisms that underlie performance. One of the primary insights provided by recent applications of such models is that individual and clinical differences in performance on a wide variety of cognitive tasks, ranging from simple choice tasks to complex executive paradigms, are largely driven by efficiency of evidence accumulation (EEA), a computational mechanism defined by sequential sampling models. This review assembles evidence for the hypothesis that EEA is a central individual difference dimension that explains neurocognitive deficits in multiple clinical disorders and identifies ways in which in this insight can advance clinical neuroscience research. We propose that recognition of EEA as a major driver of neurocognitive differences will allow the field to make clearer inferences about cognitive abnormalities in psychopathology and their links to neurobiology.

Keywords: cognitive control; diffusion model; executive function; linear ballistic accumulator; mathematical psychology; transdiagnostic risk.

Figure 1

Schematics of the (A) linear ballistic accumulator (LBA) and (B) diffusion decision model (DDM), which are commonly applied sequential sampling models in the accumulator-type class and random-walk class, respectively. In both illustrations, the models describe a task in which an individual must decide whether a presented arrow is pointing to the left or the right. The LBA assumes that accumulators for the correct choice (right, in green) and incorrect choice (left, in red) start at a level drawn from a uniform distribution between 0 and parameter A and proceed to gather evidence at linear and deterministic rates over time as they race toward an upper response threshold, set at parameter b. The rates of evidence accumulation on individual trials, represented by the light green and light red traces, are drawn from normal distributions with a mean of v (represented by the green, v_right, and red, v_left, arrows) and a standard deviation of sv. The DDM instead assumes a single decision variable that represents the relative amount of evidence for each of the two possible choices (e.g., evidence for right vs. left; these models are typically applied to two-choice decisions). This variable begins at parameter z and drifts over time between boundaries for each possible response, set at 0 (for left) and parameter a (for right). The drift process on individual trials, represented by the light blue traces, is stochastic and moves toward the boundary for the correct choice at an average rate of v (represented by the blue arrow, v_right–left). Efficiency of evidence accumulation, defined as the rate at which an individual is able to gather relevant evidence from the environment to make accurate choices, can be measured in the LBA by subtracting the average accumulation rate for the incorrect choice (v_left) from that of the correct choice (v_right). Efficiency of evidence accumulation is also measured by the DDM’s single average drift rate parameter (v_right–left). Individuals’ level of response caution (i.e., speed/accuracy trade-off) can be indexed by parameters that represent the distance evidence accumulators must travel to trigger a response in both the LBA (parameter b) and DDM (parameter a). Both models also include parameters for time taken up by perceptual and motor processes peripheral to the decision, t0 and Ter, respectively.

Figure 2

Simulated data that illustrate the behavioral manifestations of differences in efficiency of evidence accumulation (EEA). Response time (RT) data from 10,000 trials were simulated with the diffusion decision model implemented in the R package rtdists (149) while varying drift rate (v = 2, 1, 0.5) and holding other diffusion decision model parameters constant (a = 1, z = 0.5, Ter = 0.300). Blue histograms represent simulated correct RTs, while red histograms represent simulated error RTs. As EEA (v) decreases, accuracy rates are reduced and both the mean and standard deviation of RT increase. However, analysis of RTs with the ex-Gaussian distribution, a statistical model that allows Gaussian and exponential components of RT distributions to be indexed separately, reveals that the mean (μ) and Gaussian variability (σ) stay relatively constant, while exponential RT variability (τ; positive skew) substantially increases at lower levels of EEA. Therefore, as demonstrated in previous large-scale simulation studies (91), EEA primarily affects RT distributions’ level of exponential RT variability, with larger τ estimates (i.e., greater levels of positive skew) providing a behavioral hallmark for reduced EEA.

Figure 3

Hypothesized determinants of efficiency of evidence accumulation (EEA) manifested on specific cognitive tasks for a given individual.

Figure 4

Diagrams contrasting the general assumptions of two different approaches to studying neurocognitive contributions to psychopathology: the dominant fractionation paradigm (top) and the alternative efficiency of evidence accumulation (EEA)–based paradigm we highlight in this review (bottom).