Proactive control of irrelevant task rules during cued task switching

Generalized eta squared is a more conservative estimate of effect size than partial eta squared. Generalized eta squared is additionally reported because some have suggested that it may facilitate a more common framework for effect size that can be compared across experiments of varying types (Bakeman, 2005; Olejnik & Algina, 2003).

Relevant material for this paper can be accessed via Open Science Framework at https://​osf.​io/​fveqb/​. This repository contains the raw data for Experiments 1–3 as well as a summary document detailing all of the analyses conducted by T.S.B. for the manuscript, using the R statistical language, and summarized via R Markdown. This document should provide the code needed to reproduce the relevant results contained within the manuscript for each experiment. Please note that this material has not been peer-reviewed.

In addition to the exceptions noted in the text, for purposes unrelated to the present experiment, we presented half of the trials in the single and mixed-task blocks for both the baseline and incentive blocks in the upper portion of the screen and the other half in the lower portion of the screen. Presentation of stimuli and tasks was perfectly balanced across the two locations (i.e., each location was mostly congruent). We also asked participants to perform a secondary task in which they had to press the “y” key anytime the letter F or the number 5 occurred regardless of what task was cued during the mixed-task blocks. There were 11 trials of this nature. It was stressed that this task was secondary and it was most important for participants to perform the primary task well and try to earn an incentive when possible. Because the magnitude of the TRCE was extremely similar in Experiment 3 and in Experiment 1, which employed neither of these design features, we do not believe they had any influence on the primary results (cf. Kessler & Meiran, 2010, finding that holding in mind irrelevant task rules, such as those of the secondary task, does not affect the TRCE; see also Rubin & Meiran, 2005, finding that holding multiple stimulus-response rules in working memory does not affect mixing cost).

We performed two supplementary analyses to examine whether a speed-accuracy tradeoff (SAT) was responsible for the incentive-driven shifts in performance. Neither the approach of plotting a cumulative accuracy distribution nor the Heitz (2014) approach of examining macro-SAT and conditional accuracy functions supported the assumption that a SAT was driving the primary patterns pertaining to incentive-related changes in performance (i.e., faster responding and reduction in mixing cost for RT on non-incentive trials in incentive compared to baseline block). These analyses were instead consistent with a non-specific increase in errors in the incentive block.

The relatively long ISIs used in the present experiments may also explain why switch costs were relatively small. The relatively long cue-to-stimulus interval provided ample time for preparation of the current task set, while at the same time the relatively long response-to-cue interval allowed for dissipation of the previous set. Both may have served to reduce switch costs.