When people must respond to two stimuli presented in rapid succession, the response to the second stimulus (Task 2) is delayed, and this delay increases as the interval between the two stimuli (the stimulus onset asynchrony, or SOA) decreases. This PRP effect, first described by Telford (1931) and extensively studied by Welford (1952) and Pashler (1994), is one of the most reliable phenomena in experimental psychology and has provided crucial evidence about the architecture of human information processing.
The Central Bottleneck Model
The dominant account of the PRP is the central bottleneck model, which proposes that while perceptual processing and motor execution can proceed in parallel for two tasks, a central stage — typically identified with response selection — can only process one task at a time. When Task 2's central stage is ready to begin but Task 1's central stage is still in progress, Task 2 must wait, creating the PRP delay.
where A = pre-bottleneck stage, B = bottleneck stage, C = post-bottleneck stage
Slack = max(0, B₁ − SOA − A₂)
The model predicts that at short SOAs, a "slack" period absorbs any time added to Task 2's pre-bottleneck stage (Stage A₂). This leads to the locus of slack logic: if a factor affects Task 2's pre-bottleneck processing, it will have a reduced effect on RT₂ at short SOAs (because the added time is absorbed into the slack) but a full effect at long SOAs. Factors affecting the bottleneck or post-bottleneck stages will have additive effects with SOA.
Empirical Evidence
The locus of slack logic has been applied extensively to determine where various factors exert their effects. Stimulus intensity and word frequency — which affect perceptual processing — show underadditive interactions with SOA, confirming their pre-bottleneck locus. Response-response compatibility effects are additive with SOA, consistent with a bottleneck or post-bottleneck locus. These findings provide converging evidence that response selection is the processing bottleneck.
A major debate concerns whether the central bottleneck is a hard-wired structural limitation or a strategic policy that can be overcome with practice. Meyer and Kieras (1997) proposed EPIC, a model in which the bottleneck reflects a cautious scheduling strategy rather than a capacity limit. Supporting this view, Schumacher et al. (2001) showed that with extensive practice, PRP effects can be virtually eliminated for certain task combinations, suggesting that at least some aspects of the bottleneck are strategic.
Formal Models
Beyond the basic bottleneck model, several formal architectures have been proposed. The central capacity sharing model (Tombu & Jolicoeur, 2003) allows central processing to be shared between tasks rather than strictly serialized, with the sharing proportion as a free parameter. Diffusion model analyses of dual-task performance (Ulrich et al., 2006) have shown that the PRP manifests as a reduction in drift rate for Task 2 at short SOAs, suggesting that central capacity is divided rather than completely withheld.
The PRP paradigm remains one of the most informative methods for studying the architecture of cognition, and the bottleneck model provides a simple but powerful mathematical framework for making precise, testable predictions about dual-task interference.