Evaluating the consequences of impaired monitoring of learned behavior in attention-deficit/hyperactivity disorder using a Bayesian hierarchical model of choice response time

Abstract

Objective: Performance monitoring deficits have been proposed as a cognitive marker involved in the development of attention-deficit/hyperactivity disorder (ADHD), but it is unclear whether these deficits cause impairment when established action sequences conflict with environmental demands. The current study applies a novel data-analytic technique to a well-established sequence learning paradigm to investigate reactions to disruption of learned behavior in ADHD.

Method: Children (ages 8-12) with and without ADHD completed a serial reaction time task in which they implicitly learned an 8-item sequence of keypresses over 5 training blocks. The training sequence was replaced with a novel sequence in a transfer block, and returned in 2 subsequent recovery blocks. Response time (RT) data were fit by a Bayesian hierarchical version of the linear ballistic accumulator model, which permitted the dissociation of learning processes from performance monitoring effects on RT.

Results: Sequence-specific learning on the task was reflected in the systematic reduction of the amount of evidence required to initiate a response, and was unimpaired in ADHD. When the novel sequence onset, typically developing children displayed a shift in their attentional state while children with ADHD did not, leading to worse subsequent performance compared to controls.

Conclusions: Children with ADHD are not impaired in learning novel action sequences, but display difficulty monitoring their implementation and engaging top-down control when they become inadequate. These results support theories of ADHD that highlight the interactions between monitoring processes and changing cognitive demands as the cause of self-regulation and information-processing problems in the disorder. (PsycINFO Database Record

Figure 1

Schematic of the Linear Ballistic Accumulator model (Brown & Heathcote, 2008); Evidence for each response accumulates separately until one “winning” accumulator terminates at the response boundary (b). The start point of each accumulator for a given trial is sampled from a uniform distribution bounded at 0 and A, and the rate of evidence accumulation for that trial is sampled from a normal distribution with a mean of v.

Figure 2

Example stimulus moving in 3 steps of a sequence. “ITI” represents the interstimulus interval between each trial.

Figure 3

Plots of actual response time quantiles (dots) compared with those predicted by fitted model parameters (lines) for group-level fits. Black = correct, Gray = errors

Figure 4

Learning effects in mean response times (ms) and accuracy. For all charts: dotted lines = ADHD, solid lines = controls. Error bars represent standard error of the mean.

Figure 5

Learning effects in means of individual LBA parameters. For all charts: dotted lines = ADHD, solid lines = controls. Error bars represent standard error of the mean for parameter values across subjects.