Seventy-three volunteers participated in the study. The data set of one participant had to be excluded due to high error rates (ERs; > 15%) in both tasks. Two further participants were excluded, because they showed no coherent response strategy. They blocked their responses for about 70% of the time, but started to group their responses in the middle of the experiment. For the remaining 70 participants (46 females), a clear preference could be found for both levels of task coordination. Their ages ranged from 18 to 33 years (mean age = 25.5 years, standard deviation, SD 3.5 years); they had normal or corrected-to-normal vision, were either right-handed or ambidextrous, and were able to speak German at native language level. All participants received 11.25 Euro or course credit, and an additional monetary bonus for each correctly answered stimulus, which could accumulate up to five Euro.
Task-switching with preview
In contrast to the common task switching paradigms (Kiesel et al., 2010), TSWP provides participants additionally with a preview to the task stimulus they have to respond to after the next task switch. That is, the stimulus of the task that will become relevant after the next task switch is already visible, while participants still work on the currently relevant task. A scheme of an exemplary stimulus presentation is illustrated in Fig. 2 (left). As becomes evident from this figure, each block of the TSWP starts with the concurrent presentation of one stimulus of each task, A and B, the participant will have to perform, for example, a letter and a digit classification task, as shown in Fig. 2. The participant is required to work on the two tasks in a predictable AAABBB sequence, corresponding to an alternating runs scheme in classic task-switching research (Rogers & Monsell, 1995). To guide the participants through this sequence, an additional arrow cue is provided, which marks the task that needs to be performed in the current trial (see Fig. 2). Responses to the two tasks (yes, no) have to be provided with the index and middle fingers of different hands (e.g., left hand for task A and right hand for task B). Upon each given response to the marked task, the stimulus of this task is immediately updated in the subsequent trial, that is, without any response–stimulus interval. In the meantime, the stimulus of the other task remains visible throughout all three trials of the currently marked task. After the third trial, a shift of the arrow to the other task indicates that the participant finally has to switch to exactly this other task stimulus, and to start the work on the three trials of this task. Thus, the remaining stimulus of the respective other task always provides a “preview” of what the specific task will be after the next switch. Participants are instructed to work as quickly and accurately as possible through the sequence of the two tasks for a predetermined time. However, no instructions are provided whatsoever about the use of the information provided by the preview stimulus. Thus, the participants are completely free with respect to process or to ignore this information. Accordingly, it was up to them either to use the preview stimulus to prepare the switch while still working on the currently relevant task, or to ignore that stimulus and process it after the switch. A demonstration of the paradigmFootnote 1 is provided on the open science framework platform: https://osf.io/sb6wq/.
Free concurrent dual-tasking (FCDT)
The general features of the presentation of task stimuli and the response recording correspond to the one in the TSWP paradigm. However, in contrast to the TSWP, the participants are not requested to work on both tasks in a predefined sequence, but are also completely free in respect to when and how long they want to work on a given task. They are only instructed to maximize their throughput of both tasks within the given time without neglecting one task in favor of the other, but do not get any instructions whatsoever how to organize the coordination of tasks in order to achieve this goal. Thus, the participants can freely organize their responses to the two tasks with the only constraint that they should perform the two tasks with the same priority. A typical scheme of stimulus presentation is presented in Fig. 2 (right). Like in the TSWP, each FCDT block starts with the presentation of one stimulus of each task. However, no additional cue is provided to mark the currently relevant task. Instead, it is up to the participants to decide which task to start with and after how many trials to switch between tasks. Upon each response to one of the tasks, only the task stimulus of the task just responded to is updated, whereas the task stimulus of the other task remains visible until the participants finally decide to switch to that task. Given that the responses to the two tasks are provided by different hands, the specific sequence of responses to the two tasks can then be derived post hoc from the timeline of response recordings. This, in turn, provides the basis for the identification of different sorts of response organization in terms of blocking, switching, or response grouping. Note that, like in the TSWP paradigm, the participants always have the opportunity for overlapping processing of the stimuli of both tasks, which might be used or not to optimize task-switching performance. Thus, the paradigm is suitable for analyzing the extent to which the preferred mode of cognitive task processing, as identified in the more controlled TSWP paradigm, also transfers to a situation of free dual-tasking and how this mode relates to the pattern of self-organized response scheduling. A demonstration of the paradigm is provided on the open science framework platform: https://osf.io/e6wgx/.
The experiment contained two simple classification tasks, which were used in both, the TSWP and the FCDT paradigm. The digit classification task consisted of a set of digits, which had to be classified according to their parity (2, 4, 6, or 8 vs. 3, 5, 7, or 9). In the letter classification task, participants had to decide whether a presented letter was a vowel or a consonant (A, E, I, U vs. G, K, M, or R).
Stimuli and apparatus
The experimental stimuli were displayed in light grey (RGB = 245, 245, 245; font size = 24 px) on dark grey background (RGB = 90, 90, 90) on an Acer LCD screen (1280 × 1024 px, sampling with 60 Hz). In single-task blocks, the stimuli were presented in the center of the screen. All dual-taskFootnote 2 blocks started with the simultaneous presentation of stimuli for both tasks. The stimuli were then presented vertically with close spatial proximity (distance = 16 px), allowing concurrent perception of the two stimuli without eye movements. Stimulus presentation and response recording were controlled by a custom-made JAVA software running on an Intel Pentium (2.9 GHz, 8 GB RAM; Windows 7 Pro). Participants responded by pressing predefined letter keys on a standard keyboard. The keys ‘K’ and ‘L’ were used with the index and middle finger of the right hand to respond to one task and the keys ‘S’ and ‘A’ with the index and middle finger of the left hand to respond to the other task. The task-hand assignment was counterbalanced between participants. The keys were marked with color points for easier recognition.
One to three participants were tested simultaneously at independent PC workstations, separated by opaque screens. The experiment was structured into three parts: (1) introduction and practice of the single tasks, (2) introduction and practice of the FCDT paradigm followed by the experimental phase, and (3) introduction and practice of the TSWP paradigm followed by the experimental phase. Note that all participants received the same order of the two paradigms to avoid that their voluntarily exhibited response sequence pattern in the FCDT would be influenced by the prescribed response sequence in the TSWP paradigm.
Throughout training and experimental phase, the single-task and dual-task trials had to be performed for a specified time rather than a specified number of trials. Performance assessments were based on task throughput, which is the number of trials correctly performed within the given time, reflecting both speed as well as accuracy. Accordingly, participants were instructed to maximize their performance in terms of number of correct responses to one or both tasks within the given time.
Instructions for the single tasks and the two paradigms were presented on the computer screen and could be read self-paced. The training of both single tasks comprised a 30-s block for task familiarization and a subsequent 60-s block for further practice, respectively. The subsequent training and experimental phase of both paradigms followed the same general block structure. For both paradigms, the initial practice phase included a 60-s dual-task block for task familiarization and an additional 120-s dual-task block for further practice, respectively. Following this practice phase, the experimental phase for each paradigm included three runs. Every run comprised two 120-s dual-task blocks followed by one 60-s block of each single task. All runs began with the dual-task condition to prevent the influence of practice effects in the two single tasks biasing the identification of processing modes. The single-task blocks were included to control for stability of single-task performance. The order of single tasks was counterbalanced across these blocks.
The task stimuli of each block were randomly drawn from the stimulus sets of the respective tasks with the constraint that no stimulus would be directly repeated and that the two possible responses per task were equally distributed. During single- and dual-task blocks, task stimuli were shown until a response was recorded. Upon a recorded response, the next stimulus appeared immediately (response–stimulus interval = 0 ms).
After every block, participants were provided with feedback on the number of performed trials and the number and the percentage of correct responses of each task for 5 s. Feedback was provided primarily to maintain the participants’ motivation and to inform them about their error rates to ensure that participants not only speed up their responses, but also try to keep their error rates within reasonable limits. Throughout the experimental procedure, short breaks of 1 and 2 min were included between the experimental runs and between the two paradigms, respectively. Altogether, the experiment lasted approximately one and a half hour.
All participants performed both paradigms. Participants were then post hoc categorized into subgroups, separately for each dual-task paradigm. Based on their performance in the TSWP, we classified participants regarding their mode of task processing depending on the degree to which they used the preview (i.e., serial, semi-overlapping, or overlapping). In the FCDT, we classified participants regarding their strategy of response organization based on their response pattern (i.e., blocking, switching, or response groupers). The resulting 3 (categorization in FCDT) × 3 (categorization in TSWP) contingency table then was used for analysis of the proposed correspondence between both levels of task coordination.
For both paradigms, three different trial types were considered: single-task trials, repetition trials, and switch trials. Repetition trials included all trials in which participants performed the same task as on the previous trial. Switch trials were defined as trials in which participants performed a different task compared to the previous trial. For each single-task block and the different trial types in dual-task blocks, the mean inter-response intervals (IRIs), defined as the time interval between two subsequent responses, and error rates (ERs), defined as the number of incorrect responses compared to the total number of responses given, were calculated for each participant. IRIs had to be used instead of reaction times that are usually calculated as the interval between stimulus and response. This was necessary, because in the FCDT paradigm, one or more responses to the other task could occur between the onset of the preview of the switch stimulus and the response to this switch stimulus. Hence, IRIs were more appropriate to assess the time needed for responding to a task. Only correct responses were considered in the analyses of both, IRIs, and measures of efficiency. In both paradigms, the data of the different trial types of each participant were aggregated across tasks and experimental runs. Averaged across tasks and participants, this yielded 266 single-task trials (SD 31) and 488 dual-task trials (SD 84.6) for the TSWP paradigm. For the FCDT paradigm, 254 single-task trials (SD 30) and 473 dual-task trials (SD 79.5) averaged across tasks and participants were available. Regarding the identification of outliers at single trial level, first, all trials with an IRI longer than 5 s were discarded. Subsequently, trials slower than two SD from the participant’s mean IRI in the according trial type were excluded. This yielded in excluding 4.5% of trials (SD 0.8%) in the FCDT and 4.4% of trials (SD 0.7%) in the TSWP paradigm per participant on average.
Identification of individually preferred modes of task processing in the TSWP
The identification of individually preferred modes of task processing was based on a fine-grained analysis of the overtly observable response time patterns, following the rational and criteria described in our previous study (Brüning & Manzey, 2018). In a first step, the switch-trial data were inspected for specific cues, the so-called fast switches that provide a first indicator of possible overlapping processing. For this purpose, we compared each switch-trial IRI with the first quartile of the distribution of IRIs in the respective single-task block of a participant. Only switch trials with an IRI at least as fast as the 25% quickest responses in the respective single-task trials were classified as fast switches. Since such fast switches are considerably shorter than the mean processing time needed for a single-task response, it can be assumed that not only task-set reconfiguration and/or overcoming possible task-inertia effects from the preceding task (Allport, Styles, & Hsieh, 1994; Monsell, 2003), but also some stimulus processing must have been carried out prior to that fast switch IRI. Based on the outlined analysis, the fast switch rate of an individual was calculated by relating the number of fast switches to the number of all switches performed by the individual.
However, before considering the fast switch rate as indicator of the degree of overlapping processing for a single participant, two alternative sources of fast switches, a compensational prolongation in trials directly preceding the switch (i.e., in the pre-switch interval) or a production just by chance (e.g., due to unrelated muscular pre-activation), had to be excluded. To rule out the former, we tested whether participants showed longer mean responses in the interval before a fast switch than before all non-fast switches. The underlying logic is that the pre-switch interval (PSI) preceding a non-fast switch could entail mixing costs resulting from the performance of two tasks rather than one, but should not lead to other time losses. In contrast, the PSI preceding a fast switch could include a combination of mixing costs and potential compensational prolongations in case of interleaving, but still serial processing. Only if no compensational prolongation can be found in the PSI of a fast switch compared to non-fast switches, overlapping processing must have been applied. Accordingly, the comparison of PSIs preceding fast and non-fast switches allows for identifying such compensational prolongations without confounding them with potential mixing costs. The comparisons are schematically depicted in Fig. 3. Finally, only those participants whose mean PSI of fast switches was equal or shorter than their according mean PSI of all non-fast switches, were considered to be overlapping processors.
The second alternative, that is, that fast switches might have occurred randomly, was addressed by comparing the individual’s fast switch rate found in the present TSWP paradigm with a distribution of fast switch rates produced by chance. This latter distribution was obtained from a sample performing task switching without a preview option, thus only allowing for serial processing. In the present study, we used data acquired in our previous study (Brüning & Manzey, 2018), which included a sample of N = 46 participants performing an alternating runs task-switching paradigm using similar simple classification tasks as in the current study but without providing a preview. Without the option to use a preview, the participants of this control group were forced to work serially on the two tasks. Only those individuals of the current study whose fast switch rate was three SDs above the grand mean of the random distribution derived from the control group (14.61%) were finally classified as overlapping processors. In contrast, individuals showing a fast switch rate in the present TSWP paradigm, which fitted to the distribution of incidental fast switches by being lower than the grand mean plus one SD (6.55%) were still defined as serial processors. All remaining participants were considered to show too many fast switches to occur by chance, but too few to indicate a manifest and clear preference for an overlapping processing mode, and, thus were classified as semi-overlapping processors. Note that our criteria, combined with the procedural aspect that single-task blocks always followed dual-task blocks, altogether represent a relatively conservative indicator of overlapping processing.
Identification of individually preferred strategies of response organization in the FCDT
The identification of preferred strategies of response organization was based on post hoc analyses of the response sequences over time produced by the individual participants in the FCDT. This included how often participants repeated a given task and how they organized task switches. Capitalizing on the approach and criteria introduced by Reissland and Manzey (2016), we inspected the switch rates of participants to distinguish between blockers and switchers, but also considered the distribution of IRIs in switch trials to distinguish response groupers from switchers, as well. The switch rate was defined as the number of switches related to the maximum number of switches that would have been possible given the number of each task performed in the given time. Participants who showed a switch rate below 10% and, thus, minimized the number of task switches while multitasking, were classified as blockers as they obviously preferred to separate the performance of both tasks as much as possible. In contrast, all individuals with a switch rate higher than 10% and an unimodal distribution of IRIs in switch trials were classified as switchers. In this case, most of the switch IRIs fluctuate evenly around the mean IRI of switch trials. Finally, individuals who performed a high number of switches (i.e., switch rate > 50%) along with a bimodal distribution of switch IRIs were classified as response groupers. They typically produced prolonged switch IRIs while processing the stimuli of both tasks internally, followed by a very short response when they finally have processed both tasks and executed the according response in close succession. To distinguish response groupers from switchers, we tested whether the distribution of switch IRIs deviated from a unimodal distribution (i.e., most likely bimodal) by means of Hartigan’s dip test (Hartigan & Hartigan, 1985). Since such tests are highly sensitive for signs that contradict unimodal distributions, we considered a p value of p < 0.001 as critical to confirm bimodality. However, all tests for bimodality were also confirmed by visual inspection. For an illustration of the strategies, compare the stereotype sequences of responses to both tasks as shown in Fig. 1 in Sect. 1.
Analyses of multitasking efficiency
The multitasking efficiency achieved by individual participants in TSWP and FCDT was assessed by the overall dual-task performance efficiency (ODTPE) measure proposed by Reissland and Manzey (2016) and refined by Brüning and Manzey (2018). It describes how many trials of the digit and letter classification tasks which a participant can perform correctly in the 2-min dual-task blocks (TSWP or FCDT), relative to the overall number of correct trials achieved in the two 1-min single-task blocks. Thereby, it represents a straightforward throughput measure, considering speed and accuracy of responses equally. Positive ODTPE scores indicate that the task throughput in dual-task blocks is higher than in single-task blocks, thus reflecting multitasking benefits. Negative ODTPE scores indicate multitasking costs, and ODTPE scores = 0 indicate that the task throughput in dual-task blocks is the same as in single-task blocks. A detailed description of this measure can be found in the appendix of our previous study (Brüning & Manzey, 2018).