After more than a century of experimental research on human multitasking abilities, it seems fair to conclude that they are limited (for reviews see Koch et al., 2018
; Pashler, 1994
). One of the standard experimental paradigms that has emerged from this research tradition is the Psychological Refractory Period (PRP) paradigm (e.g., Pashler, 1994
), in which participants perform two choice–reaction time tasks (Task 1 and Task 2) under different conditions of temporal overlap (i.e., different stimulus onset asynchronies, SOAs). The typical observation in this paradigm is a selective response slowing in Task 2 with increasing temporal overlap of the two tasks (i.e., the PRP effect).
Only about a decade ago, Corallo and colleagues demonstrated that introspection in the PRP paradigm is blind to the PRP effect (Corallo et al., 2008
). Specifically, they observed that trialwise estimates of RT2 (introspective reaction times, IRTs) were unaffected by SOA. This basic observation has now been confirmed several times (Bratzke & Bryce, 2016
; Bratzke & Janczyk, 2021
; Bryce & Bratzke, 2014
; Marti et al., 2010
). This apparent unawareness of multitasking costs has been regarded as evidence for a central processing bottleneck that encompasses response selection and conscious perception (Corallo et al., 2008
; Marti et al., 2010
). However, results from a recent introspective PRP study, which used a timeline method to assess the full subjective time course of a PRP trial, raise doubts about this interpretation (Bryce & Bratzke, 2022
). In this study, unawareness of the PRP effect seemed to depend on the stimulus modality order of the two tasks, with an unawareness in auditory-visual and an awareness in visual–auditory dual tasks. This observation is clearly inconsistent with the idea of an amodal conscious perception bottleneck and hints at a memory-related source of the introspective blind spot.
To our knowledge, all previous studies on introspection about dual-task performance using the “standard” visual analog scale (VAS) method assessed introspective RTs either for both tasks separately or only for Task 2. None of these studies, however, asked participants to estimate the total processing time of a PRP trial.1
Previous models of task organization in the multitasking context have considered the objective time demands of to-be-scheduled tasks as an important factor in task organization (e.g., Salvucci, 2005
). If a dual task needs to be scheduled within a larger overarching multitasking context, the total time demands to complete both tasks may be just as important as, if not more important than, the time demands of each separate task. Furthermore, when it comes to task scheduling, our subjective representation of the time taken may be more influential than the objective time required and prior research has demonstrated that these can differ considerably from each other.
One could argue that a subjective estimate of total processing time can be reconstructed from the IRTs and the objective or estimated SOA and indeed Marti et al. (2010
) did take such an approach. The results of Marti et al.’s (2010
) reconstruction of the subjective phenomenology of PRP trials suggest that participants can track the effect of SOA on total trial duration; that is, that total trial duration increased with increasing SOA. However, it is not clear whether such a reconstructed estimate equals the actual subjective experience of total trial duration. In a previous study, we observed that participants were much better at reconstructing the temporal course of a PRP trial when they indicated the trial events (i.e., stimulus and response onsets in the two tasks) on a common timeline instead of providing separate RT estimates (Bryce & Bratzke, 2017
). This suggests that it may be especially difficult to extract the time intervals of interest (i.e., RT1, RT2, SOA) from the subjective time course of a PRP trial. In the present study, we therefore examined whether participants can accurately introspect about the total time demands of a dual task when they are asked to provide a direct and explicit estimate of total trial time in the PRP paradigm with the standard visual–auditory modality order. As in previous studies, we also manipulated the perceptual difficulty of Task 2 to test whether participants are sensitive to the rather small effects of Task 2 difficulty on total trial time as well as on RT2.
In the present study, we aimed to uncover whether people can introspect about their total processing time in a standard dual-task situation, namely the Psychological Refractory Period paradigm. The data show that participants were able to report the effects of SOA and Task 2 difficulty on total trial time despite being unaware of the PRP effect when RTs are estimated for each task separately. These results confirm in a more direct way the implication of previous results by Marti et al. (2010
) that the subjective phenomenology of dual-task trials tracks the effects of SOA on total trial time.
The present results again replicated the typical introspective blind spot of the PRP effect that is observed when participants process an auditory Task 1 and visual Task 2 and then provide their IRTs via a VAS (for a recent report of notably spared introspection in the visual–auditory task order using a timeline method, see Bryce & Bratzke, 2022
). Nevertheless, the present IRT results also showed some deviations from the typical pattern; that is, both IRTs increased with increasing SOA. Importantly, none of the objective RTs showed a corresponding result pattern. Since both the objective and introspective total trial times did show such an effect, we conducted a follow-up analysis including introspective task order. This analysis revealed that SOA effects on each IRT were modulated by task order, again without corresponding effects on objective RTs. This suggests that experience with the estimation of total trial time affected the IRTs participants provided. Participants who performed the estimation of total trial time in the first half of the experiment showed a tendency to report a similar SOA pattern also in their IRTs during the second half of the experiment. Since ITTs and IRTs were never assessed within the same trial (and also not within the same half of the experiment), this carry-over effect must reflect a long-term effect rather than a short-lived response bias (e.g., a within-trial anchoring effect). We therefore interpret it as an example of cue utilization in introspection about multitasking performance. According to this view, participants use various cues to infer their RTs, for example, the feeling of difficulty or other temporal intervals present in the trial (see Bratzke & Bryce, 2016
; Bratzke et al., 2014
; Bryce & Bratzke, 2014
). Importantly, some of these cues may be invalid. In the present experiment, we posit that participants wrongly inferred longer RTs in long than short SOA trials based on the insight they previously gained when estimating total trial times (i.e., longer total trial times in long than short SOA trials). While the results of the order analysis need to be interpreted with caution due to the halved sample size in the order analysis, this finding constitutes further evidence that IRTs collected in this manner are unstable and particularly vulnerable to bias.
In contrast, the total trial time estimates appear more accurate. Importantly, this was confirmed when the two time intervals (RT and TT) and measures (subjective and objective) were analyzed in a combined ANOVA. But how accurate is introspection about total trial time? First, while the predicted under-additive interaction between SOA and Task 2 difficulty on RT2 as well as on ITT was not significant, estimates of total trial time showed the same qualitative result pattern as objective total trial times. That is, they both showed an increase with increasing SOA, a Task 2 difficulty effect and no two-way interaction. Second, correlational and regression analyses showed that the relationship between subjective and objective measures was much stronger for total trial time than for RT2. While one could argue that participants overestimated total trial time at short SOA (see lower part of Fig. 1
), we caution against interpretations based on absolute ITT values, since these are greatly determined by the rather arbitrary labels of the VASs. In sum, it seems fair to us to conclude that estimates of total trial time rather accurately (even though not perfectly) tracked the effects of SOA and Task 2 difficulty on objective total trial time.
A potential limitation of the experimental design deserves some consideration, namely the fact that in ITT blocks participants provided only one estimate whereas in IRT blocks they provided two estimates. One could question whether the reduced cognitive load in the ITT blocks is responsible for the superior introspective accuracy observed. There is indeed evidence from time perception literature that timing of multiple intervals can affect timing precision and accuracy (Brown & West, 1990
; Bryce & Bratzke, 2015b
; van Rijn & Taatgen, 2008
). Previous introspective PRP studies, however, have demonstrated that dual timing is not the cause of the introspective blind spot in the PRP paradigm. For example, in Bryce and Bratzke (2014
), the same null effect was observed although IRT2 and IRT1 were assessed in separate halves of the experiment. In a study using the method of constant stimuli, in which participants only compared RT2 to presented intervals, the same unawareness of the PRP effect was observed (Bratzke & Bryce, 2016
); in Experiment 1B of Bryce and Bratzke (2022
) participants only reported the events related to Task 2 on the timeline and also did not report a PRP effect. Furthermore, the fact that selectively the PRP effect is not reflected in IRT2, while other effects of Task 2 manipulations are usually reflected, further argues against the dual timing explanation.
Given the present dissociation in the accuracy of different introspective estimates in the multitasking context, it seems pertinent to ask whether introspections about the time demands of task processing could actually impact on people’s behavior in terms of strategic task scheduling. A related theoretical framework is the optimization account by Miller and colleagues, which assumes that both parallel and serial processing are possible in dual-task processing, and that the processing mode is chosen to optimize the total sum of RTs (Miller et al., 2009
). With respect to broader multitasking contexts, computational cognitive architectures like ACT-R (e.g., Anderson et al., 2004
) and EPIC (e.g., Meyer & Kieras, 1999
) have been used to model task organization. As in the optimization account, in these modeling frameworks the time demands of to-be-scheduled tasks play an important role in task organization (e.g., Salvucci, 2005
). However, usually the objective rather than the subjective time demands are considered in these models. We believe that it is important to consider also the subjective time demands of a task, as they can considerably differ from the objective ones. More specifically, according to the present results, subjective task demands of the total trial time may be a reliable source of information, whereas subjective time demands of each separate sub-task may not.
There is a rich literature on the relationship between metacognitive monitoring and cognitive control, suggesting an important functional role of metacognitive experience in behavioral control in higher-level cognitive tasks like studying for an exam or reasoning (e.g., Metcalfe, 2009
, Thompson et al., 2011
). To our knowledge, however, only very few studies have investigated how subjective experience, which might differ from objective performance as in the PRP paradigm, impacts on behavioral adjustments at the micro-level of standard experimental multitasking paradigms. A related study by Desender and colleagues (2014
) using a masked priming paradigm provided evidence that subjective rather than objective conflict (i.e., incongruence between prime and target) triggered conflict adaptation in subsequent trials (but see Abrahamse & Braem, 2015
, for criticism of Desender et al.’s interpretation and Foerster et al., 2017
, for a replication failure). Such sequential modulations of task performance are not unique to single-task performance and have also been observed in dual tasks (Janczyk, 2016
; Olszanowski et al., 2015
; Strobach et al., 2021
). The investigation of sequential modulations in dual-task paradigms may thus provide a starting point for further research regarding the role of introspection in behavioral adaptations in this context (see also Bratzke & Janczyk, 2021
Another related paradigm at the other pole of the multitasking continuum (i.e., switching between tasks without temporal overlap) is the voluntary task-switching paradigm, in which participants can freely choose which of two tasks to perform. Usually a strong tendency to repeat a task instead of switch between tasks is observed (e.g., Arrington & Logan, 2005
). Two previous observations suggest that introspection might play an important role in this repetition bias. Namely, people seem to be aware of their switch costs (Bratzke & Bryce, 2019
), and they tend to switch to another task when the SOA between the new and the old task corresponds to their switch costs (e.g., Mittelstädt et al., 2018
). These observations clearly show that task-switching behavior depends on switch costs; however, it is difficult to distinguish between the role of introspection and objective time demands in voluntary task-switching as introspective and objective RTs are highly correlated in task-switching (see Bratzke & Bryce, 2019
In the present study, introspection about total trial time biased introspection of RT2, but only if attention was drawn toward this time interval by instruction (i.e., asking participants for estimates of total trial time). Thus, introspection about total trial time is probably not what participants do spontaneously in this context, at least when they are asked to introspect about RT1 and/or RT2 at the same time. This raises the questions of what kind of information or processes participants actually introspect about when engaging dual-task processing, and how overarching performance goals (e.g., optimization of task-scheduling) would affect their introspective ‘focus’. Furthermore, the subjective time demands of task processing are certainly only one aspect of introspective and metacognitive experiences during such a task. Which of these experiences arise under which conditions and how they are functional for cognitive and behavioral control remain important questions for future research.
In conclusion, the present study investigated participants’ ability to introspect about the total time taken to process two tasks in the PRP paradigm. The results showed that estimates of total trial time rather accurately tracked the effects of SOA and Task 2 difficulty on objective total trial time, although estimates of each RT reflected unawareness of the PRP effect. The present results thus provide some good news for bad introspection in dual-tasking. Furthermore, the present carry-over effects from estimation of total trial times to RT estimation provide further evidence that introspective RTs are oftentimes biased by (or even based on) potentially invalid cues.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.