Purchase this article with an account.
Michelle R Kramer, Patrick H Cox, Stephen R Mitroff, Dwight J Kravitz; A Big Data Approach to Revealing the Nature of Carryover Effects. Journal of Vision 2019;19(10):76a. doi: https://doi.org/10.1167/19.10.76a.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Behavior does not occur in isolation—every cognitive act is influenced by prior experiences and can influence behavior that follows. In experimental design, this manifests as a “carryover effect” (i.e., the influence that a trial has on those that follow). Carryover is challenging to study as each trial must be considered individually. Given the power limitations, most studies focus on effects derived from averaging across trials, and randomize and counterbalance trial order to avoid any systematic effect of carryover. However, this does not change the fact that each individual still experiences an effect of carryover, creating a source of noise, particularly for individual difference experiments. Here, we leverage big data to investigate the nature of carryover, showing that performance is drastically influenced by both the absolute number and relative proportion of prior trials that match or do not match the current trial type, even across an interfering task. The optimization of behavior is proportional to the binomial z-test (R2 = 0.950, p = 1.02*10–23) and is domain-general, with the same pattern found across more than one task and dimension. This precise, scale-free mechanism suggests the implicit optimization of behavior likely occurs in local circuits and involves the striatum and/or local synaptic weight changes. Furthermore, using a novel touch-and-swipe response time measure, we can disentangle the influence of trial history on pre-potent motor responses versus actual task-relevant decisions, with each component showing unique and separable effects. The motor response component is speeded with repetition of any condition regardless of whether or not it matches the current trial condition, whereas the decision component follows a linear pattern, with efficiency defined by the degree to which the accumulated evidence is consistent with the current trial condition. These results reveal the striking and systematic mechanisms by which behavior is optimized to the current context.
This PDF is available to Subscribers Only