Can time-based decay explain temporal distinctiveness effects in task switching?

Abstract

In task switching, extending the response–cue interval (RCI) reduces the switch cost—the detriment to performance when switching compared to repeating tasks. This reduction has been used as evidence for the existence of task-set decay processes. Recently, this has been challenged by the observation of sequential dependencies on the RCI effect: switch cost is only reduced at longer RCIs when the previous trial had a short RCI. This trial-wise variation of RCI is thought to affect the temporal distinctiveness (TD) of a previous task's episodic trace, affecting the probability of its automatic retrieval on the current trial; importantly, TD is thought to be independent of the current trial's RCI. The present study highlights a dependency between the current RCI and TD, and demonstrates that a decay model can reproduce some patterns of data attributed to TD. Further, the decay account makes a strong prediction when TD is held constant: repetition response times should slow as the RCI increases, and switch response times should be facilitated. This prediction was tested via re-analysis of extant data and three experiments. The re-analysis provided some evidence for the decay account, but Experiments 1 and 2 report slowing for task repetition and switch trials, which cannot be explained by a task-set decay process. Experiment 3, which utilized tasks requiring perceptual judgements, showed small evidence for decay. We conclude that the data are largely consistent with the TD account and that the evidence for decay of higher-level task-sets is not convincing.

Keywords:

• Task switching
• Decay
• Temporal distinctiveness

Part of this work was presented at the Experimental Psychology Society conferences at Lancaster University, UK (April 2013), and Bangor University, UK (July 2013). Many thanks are due to Himeh Horoufchin and Iring Koch for providing the raw data from Horoufchin, Philipp, and Koch (2011a) and to George Houghton for fruitful discussion about the modelling presented here. We thank Iring Koch, Andrea Philipp, and five anonymous reviewers for their detailed and constructive comments on previous drafts.

Notes

¹Note that there are other scenarios in Table 1 where the current trial's RCI differs but the overall TD is constant. For example, 100–1000 and 200–2000 both have an RCI-ratio of 0.1, but the current RCIs are 1000 ms and 2000 ms, respectively. Thus, this scenario also allows one to examine decay whilst controlling for TD. However, the number of exemplars of fixed RCI-ratio and varying RCI is greatest for RCI-ratio of 1 (i.e., unity), and therefore it is exclusively this scenario which is examined in this paper.

²We thank Iring Koch for raising this possibility.

³We are extremely grateful to an anonymous reviewer for encouraging us to explore this issue.

⁴It should be noted that only one set of temporal “ages” were calculated for each RCI ratio, even though multiple ages are evident for some RCI ratios. For example, an RCI ratio of 1 arises from sequential RCIs of 100–100, 200–200, 1000–1000, or 2000–2000. Thus, the value for in this example ranges from 100 ms to 2000 ms. But, all that matters is the ratio of the RCIs; for example, the verbal theory (and the SIMPLE instantiation) predicts the same retrieval probability for all of the different conditions that give rise to an RCI ratio of 1. Thus, to avoid redundancy, the model only used one exemplar of temporal age for each RCI ratio, without loss of generality.