Efficiency of evidence accumulation as a formal model-based measure of task-general executive functioning in adolescents

Abstract

Difficulties with executive functioning are implicated in various forms of psychopathology. However, executive functioning task performance frequently demonstrates poor test-retest reliability, questionable convergent validity, and unstable associations with clinical measures. Model-based approaches may improve measurement by providing richer information about mechanisms underlying performance. The present study systematically compared a model-based measure of task-general executive functioning, efficiency of evidence accumulation (EEA), with traditional summary metrics extracted from the same tasks in a longitudinal study of adolescents (N = 637, age = 7-19). EEA demonstrated reasonable stability across development and strong cross-task reliability. Reflecting traditional metrics, EEA related to self-reported effortful control and parent-reported attention, externalizing and total problems. EEA and one traditional metric (go/no-go standard deviation of reaction time) correlated with inhibition-related brain activation in the anterior cingulate cortex and the right superior temporal gyrus. These findings highlight the potential of EEA as a task-general, stable, biologically plausible measure of executive functioning in adolescents. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Figure 1. Efficiency of Evidence Accumulation

Note. This figure depicts how Efficiency of Evidence Accumulation (EEA) is derived from several executive functioning task conditions. As pictured on the left, each condition is modeled with the Diffusion Decision Model (DDM), which assumes that evidence driving a decision (i.e. choosing a response) is gradually accumulated over time until a critical response threshold is reached (Ratcliff & Rouder, 1998). The DDM model parameter of interest is drift rate (v, pictured here in black), which describes the rate of evidence accumulation. Also pictured here are boundary separation (a), which captures response caution, and non-decision time (T_er), which includes encoding and output time. EEA is calculated via a latent factor explaining shared variance in v across various tasks and conditions, thus describing the general rate at which an individual gathers goal relevant evidence among noise to make adaptive choices across tasks and conditions (Weigard & Sripada, 2021). This metric can be calculated in this way from many choice tasks; the three conditions used for this study (Go/No-Go “go”, Go/No-Go “no-go”, and Stop Signal “go”) are pictured here. Data pictured here are simulated to depict a similar v value for the three conditions.

Figure 2

Note. This figure depicts developmental stability (A) and individual-level change from Time 1 to Time 2 (B) for executive functioning scores, as measured by Efficiency of Evidence Accumulation (EEA), Go/No-Go Standard Deviation of Reaction Time (SDRT), Go/No-Go False Alarm Rate, Stop Signal Standard Deviation of Reaction Time, and Stop Signal Reaction Time (SSRT). In (B), scores are plotted against child age in years at time of testing. Each line represents one child. In general, children improved with age on all metrics. This pattern is more apparent for task-general EEA and go/no-go metrics, which may be driven by a higher N for these tasks, as the Stop Signal task was added to the Time 1 protocol partway through data collection (N=297 for EEA, 278 for Go/No-Go, 99 for Stop Signal).

Figure 3. Unique and overlapping regions in which better task-general EEA (red) and Go/No-Go SDRT (blue) correlated with increased inhibition-related brain activation

Note. Slice location of this image is (−3, 44, 11). Image depicted in neurological space. This image depicts substantial overlap in regions correlating with executive functioning performance as measured by task-general EEA (red) and Go/No-Go Standard Deviation of Reaction Time (SDRT; blue) within the anterior cingulate cortex and right superior temporal gyrus. Additive overlap between images is indicated with color blending; therefore, locations indicated in purple represent areas of overlap. None of the other traditional metrics (Go/No-Go False Alarm Rate, Stop Signal Reaction Time, Stop Signal SDRT) correlated with inhibition-related activation. Image generated from activation maps cluster-thresholded with 3dttest++ (Cox, 1996; Cox et al., 2017a, 2017b) in SPM (Wellcome Centre for Human Neuroimaging) and rendered via MRIcroGL (Rorden & Brett, 2000).