Variance (un)explained: Experimental conditions and temporal dependencies explain similarly small proportions of reaction time variability in linear models of perceptual and cognitive tasks

Abstract

Any series of sensorimotor actions shows fluctuations in speed and accuracy from repetition to repetition, even when the sensory input and the motor output requirements remain identical over time. Such fluctuations are particularly prominent in reaction time (RT) series from laboratory neurocognitive tasks. Despite their omnipresent nature, trial-to-trial fluctuations remain poorly understood. Here, we systematically analysed RT series from various neurocognitive tasks, quantifying how much of the total trial-to-trial RT variance can be explained by three sources of variability that are frequently investigated in behavioural and neuroscientific research: 1) the experimental conditions, which are commonly employed to induce systematic patterns in variability, 2) short-term temporal dependencies such as the autocorrelation between subsequent trials, and 3) long-term temporal trends over entire experimental blocks and sessions. Furthermore, we examined to what extent the explained variances by these sources are shared or unique. In total we analysed 1913 unique RT series from 30 different cognitive control and perception-based tasks. On average, the three sources together explained only about 8-17% of the total variance. The experimental conditions explained on average about 2.5-3.5%, but did not share explained variance with temporal dependencies. Thus, the largest part of the trial-to-trial fluctuations in RT remained unexplained by the three variability sources that are typically analysed here. These may be attributable to observable endogenous factors, such as ongoing fluctuations in brain activity and bodily states, though some extent of randomness may be a feature of behaviour rather than just nuisance.

Boundary updating as a source of history effect on decision uncertainty

“…It is calculated by multiplying the sign of the regression coefficient with the uniquely explained variance of the variable. The uniquely explained variance is obtained through variance partitioning analysis 85,86 , which involves comparing the R-squared values of the full regression model that includes all variables and the reduced regression model that excludes the variable of interest. For instance, to calculate the signed R-squared of variable x , we first calculate the R-squared of the full model that includes variables x, y , and z .…”