This version is available at https://doi.org/10.14279/depositonce-10980
Copyright applies. A non-exclusive, non-transferable and limited
right to use is granted. This document is intended solely for
personal, non-commercial use.
Terms of Use
Brüning, J., & Manzey, D. (2017). Flexibility of individual multitasking strategies in task-switching with
preview: are preferences for serial versus overlapping task processing dependent on between-task
conflict? Psychological Research, 82(1), 92–108. https://doi.org/10.1007/s00426-017-0924-0
This is a post-peer-review, pre-copyedit version of an article published in Psychological Research. The
final authenticated version is available online at: http://dx.doi.org/10.1007/s00426-017-0924-0.
Jovita Brüning, Dietrich Manzey
Flexibility of individual multitasking
strategies in task-switching with preview:
are preferences for serial versus
overlapping task processing dependent on
between-task conflict?
Accepted manuscript (Postprint)Journal article |
1
2
3
Flexibility of individual multitasking strategies in task-switching with preview: Are 4
preferences for serial versus overlapping task processing dependent on between-task conflict? 5
6
Jovita Brüninga, b & Dietrich Manzeya 7
8
This is a post-peer-review, pre-copyedit version of an article published in Psychological Research. 9
10
11
12
aTechnische Universitaet Berlin 13
Work, Engineering and Organizational Psychology 14
Marchstrasse 12, F7 15
10587 Berlin 16
Germany 17
email: dietrich.manzey@tu-berlin.de 18
jovita.bruening@tu-berlin.de 19
bCorresponding author. Please correspond to: jovita.bruening@tu-berlin.de. 20
21
Acknowledgements 22
This research was supported by grant DFG 3759/4-1 provided from the Deutsche Forschungsgemeinschaft to the 23
second author. Thanks are due to Jessika Reissland for helpful comments to an earlier draft of this article, to 24
Marie Mückstein for collecting the data, and to Marcus Bleil for his technical support. 25
26
2
Abstract 1
The prevalence and the efficiency of serial and parallel processing under multiple task demands are highly 2
debated. In the present study, we investigated whether individual preferences for serial or overlapping (parallel) 3
processing represent a permanent predisposition or depend on the risk of crosstalk between tasks. Two groups 4
(n = 91) of participants were tested. One group performed a classical task switching paradigm, enforcing a strict 5
serial processing of tasks. The second group of participants performed the same tasks in a task-switching-with-6
preview paradigm, recently introduced by Reissland and Manzey (2016), which in principle allows for 7
overlapping processing of both tasks in order to compensate for switch costs. In one condition, the tasks included 8
univalent task stimuli, whereas in the other bivalent stimuli were used, increasing risk of crosstalk and task 9
confusion in case of overlapping processing. The general distinction of voluntarily occurring preferences for 10
serial or overlapping processing when performing task switching with preview could be confirmed. Tracking 11
possible processing mode adjustments between low- and high-crosstalk condition showed that individuals 12
identified as serial processors in the low-crosstalk condition persisted in their processing mode. In contrast, 13
overlapping processors split up in a majority adjusting to a serial processing mode and a minority persisting in 14
overlapping processing, when working with bivalent stimuli. Thus, the voluntarily occurring preferences for 15
serial or overlapping processing seem to depend at least partially on the risk of crosstalk between tasks. 16
Strikingly, in both crosstalk conditions the individual performance efficiency was the higher, the more they 17
processed in parallel. 18
19
Keywords: individual differences, task processing, task switching, preview, parallel processing, serial processing 20
21
3
Flexibility of individual multitasking strategies in task-switching with preview: Are preferences for serial versus 1
overlapping task processing dependent on between-task conflict? 2
3
The tendency to aim for efficient multitasking performance, that is coping with two or more concurrent 4
tasks at the same time or in close succession (Salvucci, 2013), can frequently be observed in our everyday life. 5
For example, we switch rapidly between writing an e-mail and answering a phone call or we listen to the news 6
while driving a car. Comparably, also many multimedia devices demand that we efficiently organize multiple 7
tasks at once or in close succession. But is this task-organization realized by actually performing processing steps 8
of the tasks simultaneously or by rapidly switching between the tasks? And does the task-processing mode 9
represent an individual characteristic or is it flexibly adjusted to the kind of tasks we perform? While the first 10
question involves the basic issue of serial versus parallel processing, the second question addresses the issue of 11
rigidity versus flexibility of individual preferences for these two modes of task-processing. 12
Thus far, the issues of serial versus parallel processing in multitasking has primarily been addressed in 13
the context of the psychological refractory period (PRP) paradigm (for a review, see Fischer & Plessow, 2015). 14
The PRP effect refers to the observation that in dual-tasking the response to the second task is usually the more 15
prolonged the more both tasks overlap in time. The most popular model explaining this effect is the response-16
selection-bottleneck (RSB) model (Pashler, 1994). According to this model, peripheral processes of perception 17
and response execution can actually be processed in parallel to other processes. However, the central stage of 18
information processing reflecting response selection can handle just one task at a time and, thus, requires a strict 19
serial processing of tasks at this stage if performed in close temporal proximity (Pashler, 1994). Alternative 20
models do not assume such strict structural central limitation of human information processing, but a limitation 21
of the response selection stage in terms of a limited amount of processing resources, which can be flexibly 22
allocated and shared among different tasks (Navon & Miller, 2002; Tombu & Jolicoeur, 2002, 2003; Wickens, 23
1984, 2002). These latter models allow for parallel as well as serial task processing even at the central stage of 24
response selection, depending on the chosen strategy of resource allocation. Research findings supporting this 25
idea include findings of backward crosstalk (i.e., facilitated performance in a first task, when a subsequent task 26
requires compatible elements; Fischer, Miller, & Schubert, 2007; Hommel, 1998; Janczyk, Pfister, Hommel, & 27
Kunde, 2014; Miller, 2006; Navon & Miller, 1987, 2002; Schubert, Fischer, & Stelzel, 2008) which are at least 28
difficult to explain in terms of a structural limitation. Other findings suggest that serial and parallel processing 29
represent two modes of task-processing, which can be applied with some flexibility, depending on instructions 30
4
(Lehle & Hübner, 2009), on task characteristics (Fischer, Gottschalk, & Dreisbach, 2014; Luria & Meiran, 1
2005), or on the stress state of an individual (Plessow, Schade, Kirschbaum, & Goschke, 2012). 2
With respect to the efficiency of the two modes of processing there is some consensus that a serial mode 3
of central processing might be of general advantage for multitasking, at least in situations where participants 4
have no degrees of freedom to choose the order of responses, either because of the specific instruction or a 5
sufficiently long stimulus-onset asynchrony (Logan & Gordon, 2001; Miller, Ulrich, & Rolke, 2009). One reason 6
for this superiority of serial processing is the reduction of possible crosstalk effects between the two tasks, 7
because stimulus-response mappings are less prone to confusion (Navon & Miller, 1987). 8
An attempt to broaden the view on serial and parallel processing in multitasking by addressing it outside 9
of the PRP paradigm has recently been made by Reissland and Manzey (2016, Experiment 1). They extended the 10
issue of serial versus parallel processing to the task switching domain. In the usual task switching paradigms (for 11
a review, see Kiesel et al., 2010) the participants have no option to use a parallel processing strategy and, thus, 12
have to work on two tasks in a strict serial order. In this case the response time (RT) to the first task after a 13
switch is significantly prolonged indicating switch costs. These are usually attributed to a time-consuming 14
process of task-set reconfiguration which partly can only be performed, when the stimulus of the new task is 15
available (Monsell, 2003) and/or effects of task-inertia, that is residual effects of the task performed before 16
(Allport, Styles, & Hsieh, 1994), which add to the time usually needed to respond to the new task (cf. Meiran, 17
2010). In their task-switching-with-preview (TSWP) paradigm, Reissland and Manzey (2016, Experiment 1) 18
extended this approach by providing options for parallel processing of the two tasks. For this purpose, they 19
adopted the often used alternating runs paradigm of task switching (Rogers & Monsell, 1995), comprising two 20
classification tasks (A and B), which had to be performed in an AAABBBAAA scheme. As a novel feature, this 21
paradigm was combined with an option to preview the stimulus of the task one has to switch to (i.e., the next 22
stimulus of task B is already displayed while the participant is still working on the AAA sequence; see for an 23
early example of a similar approach also Spector & Biedermann, 1976). Providing such preview gives 24
participants the opportunity to at least partly start processing of the task they have to switch to next, while still 25
working on the currently relevant task. Thus, it combines the requirement of task switching with the option of 26
overlapping (parallel) processing usually studied in the PRP paradigm1. It was expected that such overlapping 27
processing, if used, would reduce the switch costs compared to the switch costs occurring in an alternating runs 28
1 Note that, in the PRP literature the term “parallel” processing is used to describe concurrent processing exactly at the central stage of
response selection. We prefer to use the term “overlapping” processing to refer to any partial or full concurrent processing of two tasks,
which can include all processing stages.
5
paradigm without preview, or even turn them into switch benefits. The latter would depend on how far the stages 1
of processing the preview stimulus (perception, response selection, response execution) were executed while the 2
ongoing task was performed. 3
Their results provide evidence that the preview option led indeed to a reduction of switch costs 4
compared to a control group (i.e. a non-preview group), working on the same tasks in a classical alternating runs 5
paradigm of task switching, which enforces serial processing. However, not all participants contributed to this 6
preview effect. Only about one half of the individuals seemed to repeatedly make use of the stimulus preview. 7
This led to negative switch costs (i.e. benefits) in this subgroup in a considerable number of switch trials which 8
reduced their mean switch costs to about 50 ms only, about 200 ms less than in the control group. Remarkably, 9
the RT to task repetitions before the switches was not compensatory prolonged compared to the control group. 10
This suggests that the participants of this subgroup indeed started to process the task they had to switch to, 11
already, while still being involved in processing the currently relevant task. In contrast, the remaining 12
participants of the preview group retained switch costs similar to those observed in the control group. Obviously, 13
they did not use the provided stimulus preview and processed both tasks entirely serially. Note that these 14
individual differences in task-processing modes were not induced by any specific instruction. 15
This finding confirmed anecdotal observations from the very first set of task switching studies (Jersild, 16
1927). By displaying series of mixed tasks on printed lists, Jersild (1927) incidentally included an option of a full 17
preview of the next stimulus participants had to respond to. While the majority of participants showed the usual 18
switch costs, others even benefited from task alternations by achieving modest sized switch benefits instead of 19
costs. Jersild assumed that these latter participants already started processing the next task while still executing 20
the response to the other, thus using the incidentally provided preview for overlapping task processing. Also in 21
the PRP literature hints to systematic individual differences with respect to serial versus parallel processing 22
modes can be found. Schumacher et al. (2001) trained eleven participants with a PRP task by presenting both 23
tasks simultaneously without prescribing a certain order of responses (Experiment 3). Due to this training, the 24
mean PRP effect was considerably reduced to about 80 ms. However, this reduction was only due to a subgroup 25
of five participants for whom the mean PRP effect diminished to a negligible degree (14 ms), pointing to almost 26
perfect dual-task performance. The remaining participants still showed a substantial PRP effect even after 27
training. Schumacher et al. assumed that this difference reflects personal preferences for either a “daring 28
strategy” of task processing, characterized by a “great deal of processing overlap” (p. 107), or a “cautious 29
strategy” reflecting a more serial processing mode. 30
6
Altogether these findings add a new perspective on the issue of serial and overlapping processing in 1
multitasking. Findings from the PRP paradigm already show that serial and parallel processing on central 2
processing stages in multitasking are not mutually exclusive but seem to reflect strategic choices. Beyond that, 3
the available data from the TSWP paradigm suggests that individuals differ principally in their “natural” 4
preference for a serial or overlapping processing of entire tasks. Moreover, the results from the TSWP paradigm 5
suggest that overlapping processing is not necessarily less efficient than serial task processing in multitasking. 6
Conversely, its results show that the overlapping processing mode, when used in the TSWP paradigm, is 7
significantly more efficient than serial processing in reducing time-losses at task switches. 8
However, the study by Reissland & Manzey (2016) does not yet allow to infer, whether these individual 9
preferences are a permanent predisposition or not. A preference for the serial task-processing mode might be 10
explained by a strong tendency for task shielding, which helps avoiding possible issues of crosstalk between two 11
tasks, especially, when the stimuli of the tasks are concurrently available. Nevertheless, in the study by Reissland 12
and Manzey serial processing was used by a considerable percentage of participants although the risk of 13
crosstalk between two tasks was relatively low due to the use of well-distinguishable univalent task stimuli and 14
unambiguous stimulus-response mappings. Thus, it seems to be a sort of conservative multitasking strategy that 15
prioritizes risk avoidance and a clear structure of task-processing over performance maximization. This would 16
correspond to the “cautious” strategy described by Schumacher et al. (2001). However, regarding the preference 17
for overlapping processing the answer is less clear-cut. Here, two alternative explanations might apply: (1) The 18
participants might have intentionally chosen this processing mode in order to optimize their multitasking 19
performance. In this case, they would have accepted the comparatively low risk of possible crosstalk between the 20
two tasks to organize the task switching process as fast as possible. Such a processing mode would correspond to 21
the “daring” strategy of task-processing in PRP tasks described by Schumacher et al. (2) Alternatively, the 22
participants might have simply been unable to shield the two tasks against each other, thus, processed the 23
information provided by the preview involuntarily. For example, it has been suggested that in case of a 24
comparatively low load of perceptual information both, task relevant as well as task irrelevant information will 25
be processed within the bounds of an individual’s perceptual capacity (Lavie, 1995, 2010). Thus, participants 26
possessing a higher capacity might naturally be more prone to distractor effects by concurrent task stimuli than 27
others. In case of minimized crosstalk between the tasks this incidental side-effect might have led to performance 28
advantages in terms of time savings at switches, which turned the usual switch costs actually in switch benefits. 29
Based on that, the aim of the current study is to shed light on the issue to what extent preferences for 30
serial versus overlapping processing reflect a permanent and somewhat rigid predisposition versus an 31
7
individual’s strategic choice in the face of different task characteristics. Based on the TSWP paradigm, we 1
compared two conditions varying in their degree of risk of crosstalk between the two tasks and accordingly in 2
their degree of suitability for either of the processing modes. In order to manipulate the risk of crosstalk we 3
varied the “valence” of the stimulus material. While in one part of the experiment univalent stimuli were used for 4
the two classification tasks to be performed according to the TSWP paradigm, bivalent stimulus material was 5
applied for both tasks in another part. Since bivalent stimuli are task-unspecific, they bear increased risk of 6
crosstalk in terms of confusion of stimulus-response mappings, when processed in an overlapping manner. 7
Accordingly, we expected all participants applying the serial processing mode with univalent stimuli to apply 8
this mode also in case of bivalent task stimuli, which even more require efficient task-shielding due to increased 9
risk of crosstalk. As a consequence of the task-shielding, they should not exhibit any mixing costs, that is, the 10
phenomenon that RTs are higher in task repetitions in mixed blocks than in single task blocks (see, e.g., Los, 11
1996). However, due to the fact that they do not make use of the preview option, the serial processors should still 12
show considerable switch costs directly comparable to those, which usually arise in classical task switching 13
paradigms with univalent and bivalent stimuli. This also should result in a low overall multitasking efficiency, 14
reflected in less correct responses in alternating compared to single-task performance as measured by the overall 15
dual-task performance efficiency (ODTPE) measure (see the appendix for a detailed description). Regarding the 16
overlapping processors, the considerations mentioned above lead to two clearly distinct expectations. According 17
to the hypothesis of a flexible adaptation of the task-processing mode, we hypothesized that a considerable 18
number of individuals applying overlapping processing in case of less risk of crosstalk (univalent task stimuli) 19
would change to a more serial processing mode allowing for better task-shielding in response to bivalent stimuli. 20
As a consequence, this subgroup should show a better multitasking efficiency in terms of lower switch costs (i.e. 21
benefits due to a preview effect) and higher ODTPE scores than serial processors in the condition with less risk 22
of crosstalk, but essentially the same multitasking efficiency as serial processors in the condition with high risk 23
of crosstalk. Only individuals who are able to practice overlapping processing effectively even in the face of 24
considerable risks of crosstalk should stay with this mode independent of the stimulus valence and, thus, 25
outperform serial processors with respect to multitasking efficiency in both conditions. However, if the observed 26
preference for an overlapping processing mode is a fixed predisposition leading to an involuntary processing of 27
the preview stimulus, the overlapping processors should not be able to adjust their task processing to a more 28
serial mode in case of higher risks of crosstalk. In this case, performance of the overlapping processors should 29
decline with increased risk of crosstalk, reflected not only in high switch costs but also in considerable mixing 30
costs, and an overall low multitasking efficiency. 31
8
Another question not yet addressed in previous research regards possible individual preconditions 1
associated with preferences for either of the two processing modes. Individual differences in the extent of 2
working memory capacity (WMC) could be one apparent candidate as it has been shown to affect different 3
aspects of executive functions and attentional control (Engle, 2002; Engle & Kane, 2003; Kane, Conway, 4
Hambrick, & Engle, 2007). Since overlapping processors need to hold and manipulate an additional task 5
stimulus in their working memory while still working on another task, the extent of using the preview option for 6
overlapping processing might depend on the WMC of an individual. If this is the case, a smaller WMC would 7
also severely limit the adaptation of processing mode to different risks of crosstalk. Thus, we expected that 8
individuals with comparatively high WMC make more extensive use of the preview and show more flexibility in 9
their choice of processing mode. 10
Method 11
Participants 12
A total of 96 right-handed volunteers took part in this study. The dataset of four participants were 13
excluded from the analysis due to high error rates (ER; > 15%) in at least one of the performed tasks. The dataset 14
of one further participant was excluded, because his mean RT in task pure repetition trials was higher than four 15
standard deviations (SD) above the according group mean. Thus, the final sample included 91 participants (53 16
female, mean age = 25.2 years, SD = 2.9 years, range = 18 – 30 years). All volunteers were native German 17
speakers with normal or corrected-to-normal vision. They received 7 Euro/hour or course credit and an 18
additionally monetary bonus for each correctly answered stimulus. 19
Tasks 20
Two pairs of tasks were used for the experiment. The first task pair included a digit categorization task 21
and a letter categorization task. Thus, this pair comprised univalent task stimuli (digits vs. letters) involving a 22
comparatively low risk of crosstalk when performed in mixed blocks. In the digit categorization task, participants 23
were required to decide whether a presented digit was lower or higher than five (1, 2, 3, 4 vs. 6, 7, 8, 9). In the 24
letter categorization tasks presented consonants had to be classified as to whether their co-occurring vowel 25
contained an "e" or not. Target letters were the consonants D, P, T, W2. Non-target letters in this task were the 26
consonants H, K, J and Q. 27
2 Note that the letters ‘D’, ‘P’, ‘T’, and ‘W’ are pronounced [deː], [peː], [teː], and [veː] in German, while the letters ‘H’, ‘K’, ‘J’,
and ‘Q’ are pronounced [haː], [kaː], [jɔt], and [kuː].
9
The second pair of tasks included two categorization tasks based on the same set of letters (A, B, C, E, 1
O, U, X, Z) and, thus, comprised bivalent stimuli, which provide a comparatively high risk of crosstalk when 2
performed in mixed blocks. The first task of this pair required the categorization of a presented letter according 3
to its position in the alphabet (first half: A, B, C, E vs. second half: O, U, X, Z). The second task was to 4
categorize letters into vowels (A, E, O, U) and consonants (B, C, X, Z). 5
The individual WMC of our participants was assessed by means of the automated operation span task 6
(Ospan, Turner & Engle, 1989; Unsworth, Heitz, Schrock, & Engle, 2005). The task requires participants to 7
solve a series of simple math operations while trying to remember a set of letters (set sizes 3–7), which needed to 8
be recalled at the end of each test trial. 9
Stimuli and Apparatus 10
The experimental stimuli were displayed in light grey (RGB = 245, 245, 245; font size = 24 px) on dark 11
grey background (RGB = 90, 90, 90) on an Acer LCD screen (1280 x 1024 px, sampling with 60 Hz). Stimulus 12
presentation and response recording was controlled by a custom-made JAVA software running on an Intel 13
Pentium (2.9 GHz, 8 GB RAM; Windows 7 Pro). Participants responded by pressing predefined letters on a 14
standard keyboard, which were marked with color points for easier recognition. The keys ‘S’ and ‘A’ were used 15
with the index and middle finger of the left hand to respond to one task and the keys ‘K’ and ‘L’ with the index 16
and middle finger of the right hand to respond to the other task. The task-hand assignment was counterbalanced 17
between participants. The Ospan task was presented on the same screen as the experiment using the Psychology 18
Experiment Building Language (Mueller, 2012; Mueller & Piper, 2014). 19
Procedure 20
One to four participants were simultaneously tested at independent PC workstations, separated by 21
opaque screens. Each participant performed the Ospan task before the actual experiment was started. 22
Within two experimental groups (non-preview vs. preview) two crosstalk conditions were performed in 23
counterbalanced order. Instructions for each condition were presented on the computer screen and could be read 24
self-paced. The tasks were structured in fixed time periods and participants were instructed to maximize their 25
amount of correct responses in the given time. First, the according pair of single-tasks was explained including a 26
30-s block of task familiarization. Subsequently, the participants practiced each single-task for further 60 s to 27
account for initial practice effects. Finally, the respective task switching procedure in mixed blocks was 28
described and practiced for 120 s. The data collection phase then included three runs per condition, each 29
comprising one 120-s mixed block followed by two single-task blocks lasting 60 s, each. The single-task blocks 30
were included in counterbalanced order to control for the stability of single-task performance. 31
10
For mixed blocks an alternating run schema (Rogers & Monsell, 1995) of task presentation was used, 1
with both tasks (A and B) of the respective task pair presented in a predictable AAABBB task sequence. With 2
this form of task presentation, four different sorts of trials could be distinguished: single-task trials, pure 3
repetition trials (trials in mixed block which were neither preceded nor followed by a task switch: AAABBB), 4
pre-switch trials (trials in mixed blocks, not preceded but followed by a task switch: AAABBB) and switch trials 5
(trials in mixed blocks following a task switch directly: AAABBB). The distinction of pure repetition trials and 6
pre-switch trials allowed for an unbiased estimation of mixing and switching costs based on comparisons of RTs 7
for single-task and pure repetitions, and pure repetitions and switch trials, respectively. In addition, the separate 8
consideration of pure repetitions and pre-switch trials enabled for the identification and location of possible 9
crosstalk effects between tasks in mixed blocks with preview. 10
The scheme of the stimulus presentations in the mixed blocks of the two experimental groups (non-11
preview/preview) and the two crosstalk conditions is shown in Figure 1. 12
Fig. 1 Schematic of the classical task switching paradigm (non-preview), and the task-switching-with-preview 13
paradigm (preview) including stimuli inducing low or high risk of crosstalk 14
15
The non-preview group always encountered the stimulus of the currently relevant task, only. In order to 16
make the task switches visible also with the bivalent task stimuli in the high-crosstalk condition, the locations of 17
stimulus presentation for the two tasks were separated vertically (distance = 16 px). All task stimuli were shown 18
until a response was recorded. Upon a registered response, the next stimulus appeared immediately (response–19
stimulus interval = 0 ms). 20
11
For the preview group, each mixed block started with the simultaneous presentation of stimuli for both 1
tasks. Corresponding to the non-preview condition the stimuli were presented in vertical arrangement with close 2
spatial proximity, making concurrent perception of the two stimuli possible without eye movements. An external 3
cue (white arrow) supported the participants in following the predictable task switching schedule. The arrow 4
appeared on the left side of the stimulus of the task, which had to be performed first (task A). Upon a response, 5
the next stimulus of this task appeared immediately while the stimulus of the other task (B) remained. Upon the 6
response to the third trial of the currently relevant task (A) in a row, the arrow switched to the stimulus of the 7
other task indicating a task switch. Then the participant performed three trials of the now relevant task B before 8
another switch of the arrow indicated to return to task A and so on. Thus, the participants in the preview group 9
always saw two stimuli, with the stimulus without an arrow providing a preview of the stimulus of the task, 10
which had to be switched to next. This arrangement of stimulus presentation provided the participants of the 11
preview group with the opportunity to apply either a serial or overlapping processing mode of task processing, 12
depending on whether or not they processed the preview-stimulus in an overlapping manner. However, no 13
specific instruction was given concerning these two possible processing modes in the preview group. 14
The task stimuli of each block were randomly drawn from the stimulus sets of the respective tasks with 15
the constraint that no stimulus would be directly repeated and that the two possible responses per task were 16
equally distributed. At the end of each block the participants were provided with feedback on the number of 17
processed trials and the number and the percentage of correct responses of both tasks for five seconds. Short 18
breaks of one and two minutes were included between the experimental runs of one condition and the two 19
crosstalk conditions, respectively. Altogether the experiment lasted about one and a half hour per participant. 20
Design 21
The experiment entailed a 2 (preview vs. non-preview) x 2 (low vs. high crosstalk) x 4 (trial type) 22
design. The first factor represented a between-subjects factor. One half of the sample (preview group, n=45) 23
performed the mixed blocks with the TSWP paradigm, while the second group (non-preview group, n=46) 24
performed the experiment without preview, analog to the classical task switching paradigm. The factor crosstalk 25
was manipulated within subjects. All participants performed the two task pairs varying in the degree of risk of 26
crosstalk. The factor trial type also represented an independent within-subjects factor, including four trial types: 27
single-task, pure repetition, pre-switch, and switch trials. 28
Data Analyses 29
Impact of preview and risk of crosstalk on task switching performance at group level. For each 30
single-task block and the different trial types in mixed blocks the mean RT and ER for each participant were 31
12
calculated. RTs were in principle determined by the period between stimulus presentation and response. 1
However, in the preview condition the stimulus of a switch trial was always visible before actually requiring a 2
response. Therefore, the RT for switch trials in this group was defined as the time between the change of the cue 3
indicating a task switch and the respective response to the stimulus of the new task. ERs were calculated as the 4
rate of false responses per block. Based on these RT and ER data we further calculated the mean switch and 5
mixing costs for each participant and experimental condition. Switch costs were derived by subtracting the 6
means of RT/ER of pure repetition trials from the means of RT/ER of switch trials. The pure repetition trials 7
were used as reference to prevent influences of a preceding or succeeding switch. Mixing costs were calculated 8
as the difference between means of RT/ER of pure repetition trials in mixed blocks and means of RT/ER of 9
single-task trials. The data of each participant in the different experimental conditions were then collapsed across 10
tasks and experimental runs, yielding 245 single-task trials (SD = 32.9) and 215 mixed block trials (SD = 35.4) 11
per task and participant on average. Trials with an RT slower than two SD from the participant’s mean RT in the 12
according block were discarded (5.04% of trials (SD = 0.51%) per participant on average). Only correct 13
responses were considered in the analyses of RTs and measures of efficiency. Based on this data, the first set of 14
analyses addressed the impact of preview and risk of crosstalk on performances at group level. For this purpose, 15
the performance of the preview versus non-preview group in single, repetition, pre-switch and switch trials and 16
both crosstalk conditions were compared. In case of violations of the sphericity assumption, degrees of freedom 17
were corrected according to the Greenhouse-Geisser procedure in all analyses. 18
Impact of risk of crosstalk on individual preferences for serial versus overlapping processing in 19
the preview group. The data of the participants of the preview group were further analyzed with respect to 20
differences in the applied mode of task processing, which represented the main target of our research. For this 21
purpose, we performed fine-grained analyses of their overtly observable response patterns, inspecting for specific 22
cues for overlapping processing. Two separate analyses were performed for the two crosstalk conditions. Note 23
that the same analyses were also performed for the non-preview group to obtain reference values and that 24
participants of the non-preview group were not further classified whatsoever. 25
The fine-grained analyses were based on the following logic: Switch trials can be considered to reflect 26
the time needed to process the given task stimulus plus additional time needed for a task-set reconfiguration 27
and/or overcoming possible task-inertia effects from the preceding task (Allport et al., 1994; Monsell, 2003). In 28
case of serial processing, RTs in switch trials are not expected to become as fast as RTs in single-task trials but 29
to reflect some switch costs. In contrast, if we observe switch RTs in the preview condition that are even faster 30
than the typical RTs in single-task trials along with no increase of mixing costs in the repetition and pre-switch 31
13
trials, this can be taken as clear evidence that at least some processing of the previewed switch stimulus must 1
have taken place already before the switch and, thus, reduced the switch costs to a considerable degree. Based on 2
this rationale, we compared each participant’s switch RT on a single-trial basis with the distribution of RTs in the 3
respective single-task condition of this participant. If the RT of a switch trial was in the range of the 25% 4
quickest responses in the single-task or even shorter, the switch was classified as a fast switch. Thus, these fast 5
switches are considerably shorter than the mean processing time needed for a single-task response, providing a 6
relatively conservative indicator of overlapping processing cases for each individual. 7
However, fast switches could also occur by chance, even though a participant works serially on both 8
tasks. In order to estimate this rate of incidentally occurring fast switches the data of the non-preview group was 9
used. Participants of this group had no option for overlapping processing, thus, any fast switches must have 10
occurred incidentally. The classification of the individuals of the preview group into subgroups of serial versus 11
overlapping processors was based on a comparison of their individual rate of fast switches with the rates of 12
randomly occurring fast switches derived from the non-preview group. If the mean rate of fast switches of a 13
participant of the preview group fits the range of the random distribution derived from the data of the non-14
preview group, it cannot be excluded that this participant had worked on the two tasks in a serial manner. Thus, 15
all participants who showed a rate of fast switches that was below the rate corresponding to the grand mean plus 16
one SD of the random distribution, were classified as serial processors constituting the serial subgroup. 17
Following the same rationale, only participants of the preview group, who showed a rate of fast switches that 18
was higher than the one corresponding to three SDs above the grand mean of the non-preview group, were 19
considered to have worked at least partially in parallel on the two tasks and were classified as overlapping 20
processors representing the overlapping subgroup. All remaining participants in the preview group were 21
considered to show too many fast switches to occur by chance, but too few to indicate a manifest and clear 22
preference for an overlapping processing mode. Thus, we classified them as semi-overlapping processors 23
constituting the semi-overlapping subgroup. 24
Each participant of the preview group was assigned to one of the three subgroups for both the low 25
crosstalk and high crosstalk condition. This procedure enabled to test for systematic changes of the task-26
processing modes dependent on these two conditions. 27
Analyses of performance efficiency at subgroup level. Two additional sets of analyses were 28
performed in order to investigate the impact of different modes of processing on performance efficiency in 29
mixed blocks. The first one included comparisons of mean switch costs and mixing costs between the non-30
preview group and all subgroups of the preview group in both crosstalk conditions. This was done to reveal what 31
14
sorts of costs or benefits occurred depending on the mode of processing and the given risk of crosstalk. In 1
addition, we compared the multitasking efficiency of the three preview subgroups and the non-preview group in 2
both crosstalk conditions by comparing their overall performance in mixed blocks relative to single-task 3
performance. For this purpose, the ODTPE measure proposed by Reissland and Manzey (2016) was used. It 4
represents a straightforward throughput measure based on the number of correct responses the participants could 5
perform in the given time. Thereby, it allows to describe the overall net efficiency participants achieve in mixed 6
blocks relative to their single-task performance, taking speed and accuracy of responses equally into account. 7
The ODTPE is positively correlated with performance, that is, the higher the ODTPE value, the better the 8
performance in mixed blocks compared to single-task blocks (net benefits: ODTPE > 0). Whereas lower ODTPE 9
values reflect poorer performance in mixed blocks compared to single-task blocks (net costs: ODTPE < 0). A 10
detailed description of this measure is provided in the appendix. 11
Impact of working memory capacity on preferring different modes of task-processing. A final data 12
analysis addressed to what extent the preferences for serial versus overlapping processing in the low- and high-13
crosstalk condition were dependent on the individual WMC. For this purpose, all participants of the preview 14
group were categorized as low versus high WMC individuals. This was done based on a median split of the 15
distribution of partial scores achieved by the participants in the Ospan. Since their internal consistency is often 16
reported to be higher compared to the absolute scores, partial scores have been recommended in the literature as 17
the more proper performance measure (Conway et al., 2005; Friedman & Miyake, 2005; Redick et al., 2012). 18
Then, the rate of fast switches as indicators of overlapping processing were compared dependent on WMC. 19
Results 20
Effects at Group Level 21
Table 1 shows the mean RT and ER for the different types of trials, separated for the two experimental 22
groups (non-preview vs. preview) and the two crosstalk conditions (low vs. high crosstalk). We ran a 2 (preview 23
vs. non-preview) x 2 (low vs. high crosstalk) x 4 (trial type) repeated measures ANOVA on RTs. The ANOVA 24
revealed significant main effects of trial type, F(1.1,102) = 328.25, p < .001, ƞp = .79, and risk of crosstalk, 25
F(1,89) = 197.89, p < .001, ƞp = .69, as well as a significant interaction between trial type and risk of crosstalk, 26
F(1.3,111) = 164.99, p < .001, ƞp = .65. Regarding the trial type, Sidak corrected post-hoc pairwise comparisons 27
revealed significant differences between the single and repetition trials (p = .03) with shorter RTs in repetitions 28
compared to single trials. In line with this difference, mean mixing costs in repetition and pre-switch trials were 29
very low and negative in almost all conditions. The pairwise comparisons revealed also a significant difference 30
between the switch and all other trials (all p < .001; all remaining p > .24), indicating that especially the RTs on 31
15
switch trials compared to the other trial types were considerably prolonged. The latter difference was more 1
pronounced in the high-crosstalk than in the low-crosstalk condition. However, neither a significant main effect 2
of preview, F(1,89) = 0.05, p = .83, nor significant interaction effects with crosstalk, F(1,89) = 0.48, p = .49, or 3
trial type, F(1.1,102) = 0.6, p = .46, or both, F(1.3,111) = 1.24, p = .28, were found. 4
ERs were generally low. The corresponding ANOVA for ERs showed a significant main effect of trial 5
type, F(3,267) = 21.44, p < .001, ƞp = .19, crosstalk, F(1,89) = 18.25, p < .001, ƞp = .17, and a significant 6
interaction between these factors, F(2.7,237) = 7.73, p < .001, ƞp = .08. ERs were higher in the high- than in the 7
low-crosstalk condition. For the main effect of trial type Sidak corrected post-hoc comparisons revealed 8
significant differences between the single and all other trials (all p < .001; all other p > .22), showing that ERs 9
decreased from single-task blocks to mixed blocks. This effect was somewhat more marked in the low- than the 10
high-crosstalk condition. Again, no significant main effect of preview, F(1,89) = 1.34, p = .25, and no significant 11
interaction effects with crosstalk, F(1,89) = 0.87, p = .35, or trial type, F(3,267) = 0.87, p = .46, or both, 12
F(2.7,237) = 0.25, p = .84 emerged. 13
Table 1. Performance, mixing and switch costs by crosstalk condition for the non-preview and preview group: reactions
times (RT) in ms, and error rates (ER) in percent.
Trial type
Low-crosstalk
High-crosstalk
Non-preview
Preview
Non-preview
Preview
RT
ER
RT
ER
RT
ER
RT
ER
Single-task
659.8
2.6
648.2
3.0
681.5
2.7
670.6
3.3
Repetition
630.2
1.9
630.4
1.9
676.8
2.0
681.2
2.5
Pre-Switch
623.5
1.8
632.9
1.8
683.2
1.9
688.8
2.2
Switch
825.5
0.8
789.8
1.4
1057.2
2.1
1057.5
2.6
Mixing costs
- 29.6 -0.8
-17.8 -1.1
-4.7 -0.6
10.7 -0.8
Switch costs
195.3
-1.0
159.4
-0.5
380.3
0.0
376.3
0.2
14
Identification of Individual Modes of Task-Processing in the Preview Group 15
The distributions of percentages of fast switches for the non-preview and preview group in the low-16
crosstalk condition are shown in Figure 2. Note that these distributions show the rates of fast switches for each 17
participant of the two groups in ascending order. 18
Fig. 2 Distribution of individual rates of fast switches of the non-preview and preview group in the low-crosstalk 19
condition arranged from the lowest to the highest rates. Individuals of the preview group were classified into the 20
three types of processors based on the standard deviation of rates of fast switches in the non-preview group as 21
indicated. Data of 20 individuals showing a rate of exactly 0% fast switches are aggregated as “1...20”. SD = 22
standard deviation. n = number of individuals 23
16
1
As becomes evident, these distributions differed markedly with respect to their mean, albeit with some 2
overlap in range. The rates of fast switches in the non-preview group ranged from 0% to 20.7% (mean (M ) = 3
2.52%, SD = 4.03%), while the distribution of fast switches in the preview group showed a considerable larger 4
range from 0% to 56.28% (M = 14.09%, SD = 15.54%). Participants of the preview group who showed a mean 5
fast switch rate not higher than one SD above the grand mean of the non-preview group (6.55%) were considered 6
as individuals who did not make any use of the preview and, thus were classified as serial processors. In 7
contrast, those participants in the preview group, who exhibited a fast switch rate of at least 14.61%, that is a rate 8
higher than three SDs above the grand mean of the non-preview group’s distribution of fast switches, were 9
categorized as distinct overlapping processors. Their rate of fast switches appears clearly too high to be 10
explained just by a random occurrence in case of purely serial processing. Instead they obviously had made 11
repeatedly use of the preview option to start processing the switch task stimulus before the actual switch took 12
place. The remaining individuals occasionally exhibited fast switches reflected in continuous values between 13
6.55% and 14.61%. While they exhibited too many fast switches to occur just by chance, they did not seem to 14
develop a coherent task-processing mode. Therefore, these individuals were considered as a third distinct 15
subgroup, labeled as semi-overlapping processors. Overall, this categorization procedure resulted in three 16
subgroups for the low-crosstalk condition: 18 serial processors (fast switches: M = 1.97%, SD = 2.53%), 14 17
semi-overlapping processors (fast switches: M = 10.09%, SD = 2.33%), and 13 overlapping processors (fast 18
switches: M = 35.16%, SD = 12.28%). 19
Figure 3 depicts the corresponding non-preview and preview group’s distributions and the according 20
classification of individuals in the preview group in the high-crosstalk condition. 21
Fig. 3 Distribution of individual rates of fast switches of the non-preview and preview group in the high-crosstalk 22
condition arranged from the lowest to the highest rates. Individuals of the preview group were classified into the 23
three types of processors based on the standard deviation of rates of fast switches in the non-preview group and a 24
17
marked discontinuation in their own distribution (“gap”) as indicated. Data of 55 individuals showing a rate of 1
exactly 0% fast switches are aggregated as “1...55”. SD = standard deviation. n = number of individuals 2
3
Note that both of the latter distributions were more leptokurtic than in the low-crosstalk condition, with a 4
comparatively small range of 0% to 3.34% (M = 0.38%, SD = 0.76%) of fast switches in the non-preview group, 5
and a somewhat broader range of 0% to 28.8% of fast switches in the preview group (M = 3.28%, SD = 6.58%). 6
This was expected because the opportunity of identifying overlapping processing based on the rate of fast 7
switches according to our criterion is much more limited in case of bivalent than univalent stimuli. The reason is 8
that participants of both groups (non-preview and preview) can be expected to need more time for task switches 9
in case of bivalent compared to univalent task stimuli, primarily due to task-set inertia adding to the basic costs 10
of task-set reconfiguration (Meiran, 2010). By comparison, the single-task RTs, with which the switch RTs are 11
compared in order to identify fast switches as defined above, are not affected by the valence of the stimuli and, 12
therefore, they remain quick. As a consequence, the a priori probability of observing fast switches is inherently 13
smaller in case of bivalent compared to univalent stimuli, and, thus, the distributions of rates of fast switches 14
must become relatively more leptokurtic for the high-crosstalk than the low-crosstalk condition. Based on these 15
distributions the participants’ mode of processing was initially classified using the same rationale as in the low-16
crosstalk condition. Thus, individuals showing a rate of fast switches within the range of one SD above the grand 17
mean of the non-preview group, that is less than 1.14%, were classified as serial processors. Applying the same 18
criteria as in the low-crosstalk condition would then have meant to classify all participants with rates of fast 19
switches higher than three SDs above the grand mean of the non-preview group as overlapping processors, and 20
the remaining participants as semi-overlapping processors. However, this classification based on the criteria 21
derived from the very leptokurtic distribution of fast switches in the non-preview condition with a grand mean 22
18
close to zero would have been somewhat biased, as it would have resulted in a markedly narrow range of the 1
rates of fast switches from 1.14% to 2.66% for the classification of semi-overlapping processors and an overly 2
broadened range of 2.66% to 28.8% fast switches for the classification of overlapping processors. Furthermore, a 3
visual inspection of the whole range of distribution of participants classified as (semi-overlapping or overlapping 4
processors) shows a remarkable discontinuation, characterized by almost a doubling of the rate of fast switches 5
from 8.8% to 16.3%. This clearly visible gap in the distribution of fast switches suggests, that the individuals 6
showing rates of fast switches of 16.3% and higher differ qualitatively from the individuals showing rates of fast 7
switches of 1.14% to 8.8%. In order to take this gap into account, we considered a classification of semi-8
overlapping and overlapping processors in accordance to this discontinuation as more adequate, instead of just 9
considering the third SD from the grand mean of the distribution of fast switches in the non-preview group. This 10
resulted in the following subgroup sizes in the high-crosstalk condition: 28 serial processors (fast switches: M = 11
0.16%, SD = 0.31%), 13 semi-overlapping processors (fast switches: M = 4.14%, SD = 2.72%), and four 12
overlapping processors (fast switches: M = 22.3%, SD = 5.3%). Note that the number of participants classified as 13
serial processors was not affected by this procedure. 14
Changes of task-processing modes between low-crosstalk and high-crosstalk condition. Table 2 shows a 15
cross-table of the frequencies of serial, semi-overlapping and overlapping processors in the low- and high-16
crosstalk condition. All individuals (n = 18) classified as serial processors in the low-crosstalk condition also 17
used this mode of processing in the high-crosstalk condition. On the contrary, participants working on univalent 18
stimuli in a more overlapping manner, tended to shift to a more serial processing mode in the high-crosstalk 19
condition. Out of the group of 14 semi-overlapping processors in the low-crosstalk condition, only six persisted 20
in their mode, while eight exhibited a serial processing mode of processing in the high-crosstalk condition. Out 21
of the 13 individuals preferring an overlapping processing mode in the low-crosstalk condition, seven were 22
classified as semi-overlapping processors in the high-crosstalk condition, and two turned out to use a distinct 23
serial processing mode, when working with bivalent stimuli. Remarkably, four out of the overlapping subgroup 24
in the low-crosstalk condition also applied this mode of processing in case of bivalent stimuli, despite the raised 25
risk of crosstalk in this latter condition. An extended McNemar’s chi-square test revealed a significant difference 26
in the distribution of the used processing modes between both conditions (χ2 (3, n = 45) = 17, p < .001). 27
28
29
30
Table 2. Number of participants per subgroup in each crosstalk condition.
19
High-crosstalk
Processors
Serial
Semi-overlapping
Overlapping
total
Low-
crosstalk
Serial
18
-
-
18
Semi-overlapping
8
6
-
14
Overlapping
2
7
4
13
Total
28
13
4
45
1
Impact of Modes of Task-Processing on Performance Measures and Multitasking Efficiency 2
Switch costs. The mean RT/ER switch costs for the non-preview group and the three subgroups of the 3
preview group in both crosstalk conditions are shown in Table 3 (upper panel). Switch costs based on RT in the 4
low-crosstalk condition were analyzed by a one-way ANOVA, using the group categorization (non-preview, 5
preview-serial, preview-semi-overlapping and preview-overlapping) as between-subject factor. As Levene’s test 6
was significant, F(3,87) = 2.85, p = .042, for the RT switch costs in the low-crosstalk condition, we used a 7
Welch correction for the one-way ANOVA and Tamhane corrected post-hoc pairwise comparisons to account 8
for unequal variances between the subgroups. The Welch corrected one-way ANOVA revealed a significant 9
main effect, F(3,35.64) = 36.5, p < .001, np = .50. Whereas the serial processors of the preview group and the 10
participants of the non-preview group exhibited considerable switch costs, the overlapping processors of the 11
preview group were able to reduce these costs to an amount as low as 33.9 ms. The mean switch costs of the 12
semi-overlapping processors were between these extremes. The Tamhane corrected post-hoc comparisons 13
revealed all pairwise differences between the different (sub-)groups as significant (all p < .004). 14
Because of the highly different group sizes, the data of the high-crosstalk condition were analyzed by a 15
Kruskal-Wallis test followed by Bonferroni-Holm corrected Mann-Whitney U test. Again the overall effect 16
(H(3) = 34.53, p < .001), as well as most pairwise comparisons became significant (all p < .003) indicating 17
essentially the same pattern of results as for the low-crosstalk condition. Only the pairwise comparison between 18
the serial subgroup and the non-preview group just failed to become significant (p = .052). A closer inspection of 19
the data suggested that the lower RT switch costs in the (semi-) overlapping subgroups were primarily caused by 20
the participants’ ability to even generate considerable mean switch benefits, in terms of switch times lower than 21
repetition times, with their fast switches (semi-overlapping subgroup: -124.7 ms in both conditions; overlapping 22
subgroup in low-crosstalk: -116.9 ms, in high-crosstalk: -89.4 ms). Switch costs reflected in ERs were low in 23
both crosstalk conditions (-1.2% to 0.8%) and did not differ significantly from zero in any group (all p > .11). 24
Mixing costs. The mean RT/ER mixing costs for the different (sub-)groups and the two crosstalk 25
conditions are shown in Table 3 (lower panel). Mixing costs in both crosstalk conditions were analyzed for all 26
subgroups separately by one-sample t-tests. None of the mixing costs observed differed significantly from zero 27
20
(all p > .28). Therefore, we refrained from testing for differences between the subgroups of the preview group 1
and the non-preview group. Moreover, also the mean RT to trials immediately preceding a fast switch was not 2
significantly prolonged compared to the mean single-task RTs for neither of the different (sub-)groups in both 3
conditions (all p > .09). Similarly, also mixing costs reflected in ER were all below zero, actually indicating 4
small mixing benefits. 5
Table 3. Performance and switch costs by crosstalk condition for the non-preview and preview group: reactions times (RT)
in ms, and error rates (ER) in percent.
Costs
Group
Low-crosstalk
High-crosstalk
RT
ER
RT
ER
Switch
Non-preview
195.3
-1.01
380.3
0.04
Serial
281.1
-0.19
491.1
0.06
Semi-overlapping
119.5
-0.66
230.5
0.77
Overlapping
33.9
-0.60
46.1
-1.16
Mixing
Non-preview
-29.6
-0.78
-4.7
-0.62
Serial
-28.0
-1.14
10.0
-0.34
Semi-overlapping
-4.2
-1.23
13.0
-1.79
Overlapping
-18.3
-1.02
8.2
-0.60
6
Overall Dual-Task Performance Efficiency. The results of the analyses of the overall multitasking 7
efficiency based on the ODTPE measure are shown in Figure 4, separately for the low-crosstalk (left) and high-8
crosstalk (right) condition. As becomes evident from the figure, a relatively similar picture emerges in the low- 9
and high-crosstalk condition with the overlapping subgroup outperforming the other subgroups and the non-10
preview group. Remarkably, the overlapping subgroup even achieved an ODTPE value above zero (0.15%) in the 11
low-crosstalk condition. Thus, with a low risk of crosstalk between task stimuli, the overlapping task-processing 12
mode did not result in multitasking costs, but in small multitasking benefits compared to single-task performance. 13
These differences in ODTPE scores were confirmed by a one-way ANOVA in the low-crosstalk condition, F(3,87) 14
= 7.07, p < .001, np = .20, and by a Kruskal-Wallis test in the high-crosstalk condition (H(3) = 14.51, p = .002). In 15
the low-crosstalk condition, differences were further substantiated by planned Tukey-HSD corrected post-hoc 16
comparisons, which revealed significant differences between the overlapping subgroup and the other two 17
subgroups of the preview condition (all p < .018) and between the serial subgroup and the non-preview group (p 18
= .008; all other: p > .21). In the high-crosstalk condition, Bonferroni-Holm corrected Mann-Whitney U tests 19
indicated that the overlapping subgroup differed significantly from the non-preview group (p = .007) and from the 20
serial subgroup (p = .003). All other differences were not statistically significant (all p > .017). 21
Fig. 4 Mean overall dual-task performance efficiency (ODTPE) scores for the non-preview group and each preview 22
subgroup in the low (left panel) and the high (right panel) crosstalk condition. Error bars represent ± two standard 23
errors. * p < .05. ** p < .01. *** p < .001. ns = not statistically significant 24
21
1
Task-Processing Mode and Working Memory Capacity (Ospan) 2
The median split of the distribution of the partial score in the preview group (md = 65) resulted in a 3
subgroup “low working memory capacity” of 23 individuals (M = 53, SD = 9.5) and a subgroup “high working 4
memory capacity” of 22 individuals (M = 71, SD = 4.2). A 2 (low vs. high WMC) x 2 (low vs. high crosstalk) 5
repeated measures ANOVA on the rate of fast switches revealed significant main effects of WMC, F(1,43) = 6
4.73, p = .035, ƞp = .10, and risk of crosstalk, F(1,43) = 47.72, p < .001, ƞp = .53, as well as a significant 7
interaction between WMC and risk of crosstalk, F(1,43) = 7.31, p = .01, ƞp = .15. While in the low-crosstalk 8
condition the mean rate of fast switches was considerably lower in the low WMC subgroup (M = 8.8%, SD = 9
3.1%) than in the high WMC subgroup (M = 19.6%, SD = 3.1%), this difference diminished noticeably in the 10
high-crosstalk condition (low WMC: M = 2.2%, SD = 1.4%, high WMC: M = 4.4%, SD = 1.4%). 11
Discussion 12
The current research addressed the question to what extent individual preferences for serial versus 13
overlapping processing in the TSWP paradigm are determined by a permanent and rigid predisposition for one of 14
these task-processing modes, or by an individual’s strategic choice. For this purpose, we compared performances 15
when switching between two simple classification tasks with either univalent (low risk of crosstalk) or bivalent 16
(high risk of crosstalk) task stimuli under conditions of classical task switching demanding a serial processing of 17
the two tasks versus conditions of TSWP providing options for overlapping processing. However, it was an 18
essential precondition to replicate the finding of previous research (Reissland & Manzey, 2016) that the preview 19
option provided by the TSWP paradigm was actually used for overlapping processing at least by a considerable 20
22
number of individuals. Therefore, our first set of analyses focused on whether we would already find indications 1
of use of preview in the TSWP conditions, compared to classical task switching, at an overall group level. Such 2
an effect, reflected in reduced or even absent switch costs, was reported at least for conditions with low risk of 3
crosstalk (univalent task stimuli) by Spector and Biedermann (1976) and Reissland and Manzey (2016). 4
Nevertheless, the overall group analysis only revealed a strong effect of risk of crosstalk on switch costs. 5
Replicating an effect well-known from classical task switching studies, which has often been confirmed since 6
then (Allport et al., 1994; Rogers & Monsell, 1995), switch costs were higher in the conditions with the higher 7
risk of crosstalk between tasks. However, no overall benefit of preview was found in either condition. Regarding 8
the preview effect, a little reduction of switch costs was just observed in the low-crosstalk condition, but this 9
effect was too small to become statistically significant at the overall group level. 10
The obvious reason for the weak preview effect at the overall group level is that, as expected, the 11
participants performing the TSWP paradigm did not represent a homogenous group but included participants 12
who differed in their use of preview. This was revealed by the main set of analyses, including fine-grained 13
analyses of the individually chosen task-processing modes in the two crosstalk conditions. The rationale behind 14
the classification of subgroups exhibiting different modes of task-processing in the preview condition was based 15
on a statistical approach taking a comparison of the distribution of so called fast switches that were detected on a 16
single-trial basis in the preview-group and the non-preview group into account. Along with unchanged RTs for 17
trials preceding a task switch, fast switches were taken as an indicator of overlapping processing. In contrast to 18
Reissland and Manzey (2016), who found a clear dichotomy of serial and overlapping processors, indicated by a 19
discontinuation of the distribution of fast switches in their preview group, the current data does not support a 20
clear-cut classification. Instead a more gradual shift from serial processing towards clear overlapping processing 21
was observed when a task-preview was available. One reason for this difference is the considerable larger sample 22
included in the current research which represents a better basis for analyzing the nature of individual differences. 23
As we aimed for a comparison between participants, who processed unambiguously in either a serial or 24
overlapping manner, we built extreme groups. For this purpose, we defined those individuals as overlapping 25
processors, who, based on the comparison of the fast switch rates’ distributions, were highly unlikely to belong 26
to the same population as the participants of the non-preview control group, which had to work serially on the 27
tasks. In an analog way, all participants whose number of fast switches did not differ from the distribution found 28
for the non-preview group were considered as distinct serial processors. All other participants constituted the 29
remaining semi-overlapping subgroup, which exhibited a mix of serial and overlapping processing. This method, 30
based on a strict definition of fast switches and a consideration of possible trade-offs in terms of prolonged 31
23
response times for trials preceding a fast switch, represents a rather conservative approach to identify 1
overlapping processing, as only incidents are considered where alternative explanations can be excluded. Thus, 2
the rates of fast switches used for classification of the subgroups of semi-overlapping and overlapping processors 3
can be regarded as an estimation of the minimum number of incidents of overlapping processing. 4
Turning now to the performance consequences of the preferences for serial versus overlapping 5
processing, we first consider the effects arising for the different subgroups in the condition with low risk of 6
crosstalk before we turn to the question to what extent the preferred mode of processing and its efficiency was 7
affected when the risk of crosstalk increased and what individual predispositions could be leading to differences 8
in the preferred mode of processing. 9
Individual Preferences for Serial or Overlapping Processing Mode in the Low-Crosstalk Condition 10
The replication of the basic finding of participants, who preferred to process the tasks in a serial manner 11
even when preview was available and others who used the preview for overlapping processing, albeit to a 12
different degree, confirms anecdotal observations reported from one of the earliest task switching studies 13
(Jersild, 1927), and is also in line with similar conclusions derived from research with other multitasking 14
paradigms (e.g., Schumacher et al., 2001, Experiment 3). Inspecting the switch costs for the different subgroups 15
in the condition with preview shows that the participants identified as overlapping processors were able to reduce 16
their mean switch costs to negligible 34 ms. In contrast, the mean switch costs of the semi-overlapping 17
processors were still substantial (120 ms) and those of the serial processors were higher (285 ms) than the mean 18
costs in the non-preview control group (195 ms). Regarding the multitasking efficiency of the three subgroups, it 19
turned out, that the overlapping processors did not only show very small mean switch costs, but were even able 20
to perform slightly more tasks in the given time when working in the mixed condition compared to the single-21
task blocks. Thus, they did not show any multitasking performance decrements but even a small performance 22
gain. This multitasking gain was mainly due to the fact, that these participants actually realized considerable 23
switch benefits instead of costs with their fast switches. The fact that these benefits did not lead to even higher 24
multitasking gains seems to be related to the fact that even the participants classified as overlapping processors 25
were not able to practice overlapping processing with each switch but only with a certain percentage of switches. 26
Dependence of Individual Preference for Serial or Overlapping Processing on Risk of Crosstalk Between 27
Tasks 28
Based on the replication of the individual preferences for different processing modes, the major 29
question of the present study was to what extend individuals would adopt their preferred mode of processing in 30
case of higher risks of crosstalk between tasks? As expected, all participants preferring a serial processing mode 31
24
in the condition with a comparatively low risk of crosstalk between the tasks used this mode also in the condition 1
with high risk of crosstalk. However, the participants of the other two subgroups showed more flexibility and 2
adapted their mode of processing to the increased risk of crosstalk. Working with bivalent task stimuli involving 3
higher risks of crosstalk (Navon & Miller, 1987) or task confusions (e.g., due to stimulus-response bindings, 4
Logan & Gordon, 2001), the majority of these participants adopted a more serial mode of processing, which 5
allowed for better shielding of the tasks against each other. This shift of task-processing mode provides two 6
important insights. First, it provides evidence that, at least for these two subgroups, the chosen modes of task-7
processing in multitasking situations are flexible and adaptive with respect to the nature of tasks. This confirms 8
similar conclusions derived from recent PRP research (Fischer et al., 2014; Lehle & Hübner, 2009; Lehle, 9
Steinhauser, & Hübner, 2009; see for a review Fischer & Plessow, 2015). Second, it rules out that benefits of 10
overlapping processing observed with univalent stimuli reflect a (positive) side-effect of an individual’s inability 11
to shield the processing threads of the tasks effectively against each other. This had been considered as a possible 12
alternative explanation, based on the assumption that individuals differ in their proneness to effects of visual 13
distractor stimuli (flanker), depending on their attentional capacity (Forster & Lavie, 2007; Lavie & Cox, 1997). 14
However, there were also participants who did not adopt a more serial mode of processing in case of 15
higher risk of crosstalk. This held true for four participants, who used an overlapping processing mode in both 16
the low- and high-crosstalk condition, and six out of the 14 participants, who showed some indication of 17
overlapping processing, albeit to a limited extent, in both conditions. Inspecting their RTs, resulting switch costs 18
and multitasking efficiency revealed that these participants obviously persisted in their mode of task processing 19
not because they could not do otherwise, but because it was still efficient. Comparing the mean switch costs and 20
multitasking efficiency of the three subgroups in the low- and high-crosstalk condition provided essentially the 21
same pattern of effects for both conditions. Notably, the highest overall multitasking efficiency emerged for the 22
four participants classified as overlapping processors also in the high-crosstalk condition. 23
The fact that individuals classified as distinct overlapping processor outperformed serial processors in 24
both crosstalk conditions shows that overlapping processing can efficiently be used to optimize task switching 25
processes, when preview of the stimuli of the task one has to switch to next is available. This is in contrast to 26
findings from PRP research, which suggest that a serial mode of processing represents a more advantageous 27
processing mode in multitasking settings, particularly when there are less degrees of freedom to choose the order 28
of responses (e.g., Logan & Gordon, 2001; Miller et al., 2009). However, whereas the finding from PRP research 29
just accounts for a serial or parallel mode of processing on the central stage of response selection our results 30
regard the efficiency of serial or overlapping processing of whole tasks. This reflects an important difference 31
25
between the theoretical basis of our research and the dual-task research based on the PRP paradigm. In contrast 1
to the bottleneck model underlying PRP research which assumes that options of overlapping processing provided 2
by processing stages before (i.e. perception) and after (i.e. motor execution) the presumed bottleneck are 3
commonly used by individuals to optimize multitasking performance (Pashler, 1994), the current data show that 4
this holds true only for a subgroup of participants who have a preference for overlapping processing. In contrast, 5
the data of the subgroup of serial processors suggests that they prefer to shield whole tasks entirely, and do not 6
even use possibilities of parallel processing that should be easily available. The fact that such complete shielding 7
of tasks has not been observed in the PRP paradigm might be related to the fact that, in this paradigm, 8
overlapping processing at least of stages before the bottleneck is enforced by presenting the task stimuli in very 9
close succession for a very short period of time only. 10
The comparison with PRP research and the basic assumptions of the classical bottleneck model also 11
provides some clues what sorts of overlapping processing might have contributed to performance gains of 12
overlapping processors. First, based on the fact that perceptual processes can run in parallel the encoding of the 13
preview-stimulus might have already been completed during the performance of the other task. Second, the 14
overlapping processors might have started top-down processes of task-set reconfiguration (Monsell, 2003) 15
already while still executing the response to the other task. However, the current data are certainly not conclusive 16
in this respect and more research will be needed to clarify this aspect further. 17
Possible Factors Influencing the Choice of Processing Mode 18
Another question addressed in the current study concerns how individuals, who voluntarily prefer an 19
(semi-)overlapping mode of processing when risk of between-task crosstalk is low, but flexibly change to a more 20
serial mode of processing in case this risk increases, differ in their preconditions from those principally 21
preferring a serial mode of processing independent of task characteristics. One candidate that we assumed to be 22
relevant in this respect was the WMC. Our data provide significant support for this assumption. Individual 23
differences in WMC, as assessed by the Ospan task (Turner & Engle, 1989), were found to be one determinant of 24
the use of mode of processing. Individuals with a comparatively high WMC were more likely to use the stimulus 25
preview for overlapping processing than individuals with a low WMC. This is in line with other results pointing 26
to a link between WMC and the efficiency of executive function and attentional control (Kane et al., 2007). Even 27
more importantly, WMC interacted significantly with the extent of overlapping processing under different risks 28
of crosstalk. Individuals with comparatively low WMC processed the two tasks always in a serial manner, 29
independent of whether the risk of crosstalk was high or low. In contrast, the majority of participants with 30
comparatively high WMC adapted their mode of processing to the respective risk of crosstalk that is changed 31
26
from a mode of (semi-)overlapping processing in the condition with low risk of crosstalk to a mode of serial 1
processing, if the crosstalk was high. Out of the four participants who applied an overlapping mode of processing 2
even in the high-crosstalk condition, three belonged to the higher WMC group. Overall, this suggests that WMC 3
is a factor of relevance not only for the preference of a more serial or overlapping mode of processing but also 4
for the capability to adapt these two modes to task characteristics. 5
Conclusions and Limitation 6
Overall, the results of the present study provide several new insights which are relevant for our 7
understanding of serial versus overlapping (parallel) processing in multitasking. Especially they suggest that 8
individuals differ concerning a preference for serial and overlapping processing, that these modes of processing 9
can be regarded as poles of a performance dimension describing the degree to what extend individuals make use 10
of overlapping processing when dealing with different tasks, and that the use of overlapping processing can 11
result in performance advantages compared to serial processing even in case of possible risks of crosstalk. 12
Furthermore, at least individuals with high WMC seem to be flexible in adopting their mode of processing to the 13
task context with practicing overlapping processing when risks of crosstalk are comparatively low and task 14
shielding if these risks are high. However, this latter flexibility of adopting the mode of processing to risks of 15
crosstalk between tasks was only shown for a shift from a more overlapping to a more serial mode of processing, 16
if risks of crosstalk between tasks increase, as in the high-crosstalk condition of the present research. A limitation 17
of the present study regards the use of highly similar tasks even in the low-crosstalk condition, as both tasks 18
included visually presented verbal stimuli, which both demanded manual responses. According to the multiple 19
resources model of Wickens (2002) such task similarity hinders instead of supports overlapping processing. 20
Accordingly, this might have contributed to the fact that a considerable percentage of participants preferred to 21
process even the tasks in our low-crosstalk condition in a serial manner. Follow-up research should investigate to 22
what extent these serial processors would keep their mode of processing, if very diverse tasks in terms of their 23
resource demands are used. The TSWP paradigm combined with fine-grained analyses of responses provides a 24
promising tool for such research. 25
Compliance with Ethical Standards 26
Funding: This study was funded by Deutsche Forschungsgemeinschaft (DFG 3759/4-1). 27
Conflict of Interest: Dietrich Manzey declares that he has no conflict of interest. Jovita Bruening declares that 28
she has no conflict of interest. 29
Ethical approval: All procedures performed in studies involving human participants were in accordance with the 30
ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration 31
27
and its later amendments or comparable ethical standards. This article does not contain any studies with animals 1
performed by any of the authors. 2
Informed consent: Informed consent was obtained from all individual participants included in the study. 3
4
28
References 1
Allport, D. A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set: Exploring the dynamic control of tasks. 2
In C. Umiltà & M. Moscovitch (Eds.), Attention and performance XV (pp. 421–452). Cambridge, MA: 3
MIT Press. 4
Conway, A. R., Kane, M. J., Bunting, M. F., Hambrick, D. Z., Wilhelm, O., & Engle, R. W. (2005). Working 5
memory span tasks: A methodological review and user’s guide. Psychonomic Bulletin & Review, 12(5), 6
769–786. 7
Engle, R.W. (2002). Working memory capacity as executive attention. Current Directions in Psychological 8
Science, 11, 19–23. 9
Engle, R. W., & Kane, M. J. (2003). Executive attention, working memory capacity, and a two-factor theory of 10
cognitive control. Psychology of Learning and Motivation, 44, 145–199. 11
Fischer, R., & Plessow, F. (2015). Efficient multitasking: parallel versus serial processing of multiple tasks. 12
Frontiers in Psychology, 6: 1366. doi:10.3389/fpsyg.2015.01366 13
Fischer, R., Gottschalk, C., & Dreisbach, G. (2014). Context-sensitive adjustment of cognitive control in dual-14
task performance. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40(2), 399–15
416. doi:10.1037/a0034310 16
Fischer, R., Miller, J., & Schubert, T. (2007). Evidence for parallel semantic memory retrieval in dual tasks. 17
Memory & Cognition, 35(7), 1685–1699. 18
Forster, S., & Lavie, N. (2007). High perceptual load makes everybody equal: Eliminating individual differences 19
in distractibility with load. Psychological Science, 18, 377–382. 20
Friedman, N. P., & Miyake, A. (2005). Comparison of four scoring methods for the reading span test. Behavior 21
Research Methods, 37, 581–590. 22
Hommel, B. (1998). Automatic stimulus-response translation in dual-task performance. Journal of Experimental 23
Psychology: Human Perception and Performance, 24, 1368–1384. 24
Janczyk, M., Pfister, R., Hommel, B., & Kunde, W. (2014). Who is talking in backward crosstalk? Disentangling 25
response- from goal-conflict in dual-task performance. Cognition, 132, 30–43. 26
Jersild, A. T. (1927). Mental set and shift. Archives of Psychology, 89. 27
Kane, M. J., Conway, A. R., Hambrick, D. Z., & Engle, R. W. (2007). Variation in working memory capacity as 28
variation in executive attention and control. Variation in Working Memory, 1, 21–48. 29
29
Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp, A. M., & Koch, I. (2010). Control and 1
interference in task switching—A review. Psychological Bulletin, 136(5), 849–874. 2
doi:10.1037/a0019842 3
Lavie, N. (1995). Perceptual load as a necessary condition for selective attention. Journal of Experimental 4
Psychology: Human Perception and Performance, 21(3), 451–468. 5
Lavie, N. (2010). Attention, distraction, and cognitive control under load. Current Directions in Psychological 6
Science, 19(3), 143–148. 7
Lavie, N., & Cox, S. (1997). On the efficiency of visual selective attention: Efficient visual search leads to 8
inefficient distractor rejection. Psychological Science, 8(5), 395–396. 9
Lehle, C., & Hübner, R. (2009). Strategic capacity sharing between two tasks: Evidence from tasks with the same 10
and with different task sets. Psychological Research, 73, 707–726. 11
Lehle, C., Steinhauser, M., & Hübner, R. (2009). Serial or parallel processing in dual tasks: What is more 12
effortful? Psychophysiology, 46, 502–509. 13
Logan, G. D., & Gordon, R. D. (2001). Executive control of visual attention in dual-task situations. 14
Psychological Review, 108, 393–434. 15
Los, S. A. (1996). On the origin of mixing costs: Exploring information processing in pure and mixed blocks of 16
trials. Acta Psychologica, 94(2), 145–188. 17
Luria, R., & Meiran, N. (2005). Increased control demand results in serial processing evidence from dual-task 18
performance. Psychological Science, 16(10), 833–840. 19
Meiran, N. (2010). Task switching: Mechanisms underlying rigid vs. flexible self control. In R.R. Hassin, K.N. 20
Ochsner, & Y. Trope (Eds.), Self control in society, mind and brain (pp. 202–220). New York, NY: 21
Oxford University Press. 22
Miller, J. (2006). Backward crosstalk effects in psychological refractory period paradigms: Effects of second-23
task response types on first-task response latencies. Psychological Research, 70, 484–493. 24
Miller, J., Ulrich, R., & Rolke, B. (2009). On the optimality of serial and parallel processing in the psychological 25
refractory period paradigm: Effects of the distribution of stimulus onset asynchronies. Cognitive 26
Psychology, 58(3), 273–310. doi:10.1016/j.cogpsych.2006.08.003 27
Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140. doi:10.1016/S1364-28
6613(03)00028-7 29
Mueller, S. T. (2012). The Psychology Experiment Building Language, Version 0.13. Retrieved from 30
http://pebl.sourceforge.net. 31
30
Mueller, S. T., & Piper, B. J. (2014). The Psychology Experiment Building Language (PEBL) and PEBL test 1
battery. Journal of Neuroscience Methods, 222, 250–259. 2
Navon, D., & Miller, J. (1987). Role of outcome conflict in dual-task interference. Journal of Experimental 3
Psychology: Human Perception and Performance, 13, 435–448. 4
Navon, D., & Miller, J. (2002). Queing or sharing? A critical evaluation of the single-bottleneck notion. 5
Cognitive Psychology, 44, 193–251. doi:10.1006/cogp.2001.0767 6
Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116, 220–7
244. 8
Plessow, F., Schade, S., Kirschbaum, C., & Fischer, R. (2012). Better not to deal with two tasks at the same time 9
when stressed? Acute psychosocial stress reduces task shielding in dual-task performance. Cognitive, 10
Affective, & Behavioral Neuroscience, 12(3), 557–570. 11
Redick, T. S., Broadway, J. M., Meier, M. E., Kuriakose, P. S., Unsworth, N., Kane, M. J., & Engle, R. W. 12
(2012). Measuring working memory capacity with automated complex span tasks. European Journal of 13
Psychological Assessment, 28, 164–171. 14
Reissland, J., & Manzey, D. (2016). Serial or overlapping processing in multitasking as individual preference: 15
Effects of stimulus preview on task switching and concurrent dual-task performance. Acta 16
Psychologica, 168, 27–40. 17
Rogers, R. D., & Monsell, S. (1995). Costs of a predictible switch between simple cognitive tasks. Journal of 18
Experimental Psychology: General, 124(2), 207–231. doi:10.1037/0096-3445.124.2.207 19
Salvucci, D. D. (2013). Multitasking. In J. D. Lee & A. Kirlik (Eds.), The oxford handbook of cognitive engineering 20
(pp. 57–65). New York, NY: Oxford University Press. 21
Schubert, T., Fischer, R., & Stelzel, C. (2008). Response activation in overlapping tasks and the response-22
selection bottleneck. Journal of Experimental Psychology: Human Perception and Performance, 34, 23
376–397. 24
Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D.E., Lauber, E.J., Kieras, D.E., & Meyer, D.E. 25
(2001). Virtually perfect time sharing in dual-task performance: uncorking the central cognitive 26
bottleneck. Psychological Science, 12(2), 101–108. doi:10.1111/1467-9280.00318 27
Spector, A., & Biederman, I. (1976). Mental set and mental shift revisited. The American Journal of Psychology, 28
89(4), 669–679. doi:10.2307/1421465 29
Tombu, M., & Jolicoeur, P. (2002). All-or-none bottleneck versus capacity sharing accounts of the psychological 30
refractory period phenomenon. Psychological Research, 66, 274–286. 31
31
Tombu, M., & Jolicoeur, P. (2003). A central capacity sharing model of dualtask performance. Journal of 1
Experimental Psychology: Human Perception and Performance, 29, 3–18. 2
Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent?. Journal of Memory and 3
Language, 28(2), 127–154. 4
Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span 5
task. Behavior Research Methods, 37, 498–505. 6
Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman & R. Davies (Eds.), Varieties of 7
attention (pp. 63–101). New York, NY: Academic Press. 8
Wickens, C. D. (2002). Multiple resources and performance prediction. Theoretical Issues in Ergonomics 9
Science, 3(2), 159–177. doi:10.1080/14639220210123806 10
11
32
Appendix: Definition of Overall Dual-Tasking Performance Efficiency (ODTPE) 1
The ODTPE measure is a straightforward throughput measure. It describes how many of two tasks can be 2
performed correctly in a given time when an individual has to cope with these two tasks concurrently or in close 3
succession compared to a situation where the same tasks can be performed under single-task conditions3. We 4
refer to overall benefits of dual-tasking if the overall performance of two tasks performed under dual-task 5
conditions is better, that is more tasks are performed correctly, than what would theoretically be expected in case 6
of strictly separated processing of the two component tasks without considering any dual-tasking costs. Costs of 7
dual-tasking are assumed if overall dual-task performance is worse, that is less tasks are performed correctly, 8
than what would be expected from strictly separated processing. This logic can be illustrated by the following 9
Gedankenexperiment: 10
Assume an individual can correctly solve 90 trials of a letter classification task (task A), and 80 trials of a digit 11
classification task (task B), both in a single-task block of one minute each. If this individual then has to work on 12
both tasks concurrently and/or in close succession for another one-minute block, we would theoretically expect 45 13
correct responses to the letter classification task and 40 correct responses to the digit classification task, given that 14
(1) this individual works on the two tasks in a strict serial processing mode, (2) this individual is able to perform 15
the different tasks with the same speed as under single-task conditions, and (3) neither general costs or benefits of 16
dual-tasking arise. Considering the performance of the component tasks separately and comparing it to the 17
respective single-task performance, this might suggest a performance decrement of 50% in the dual-task condition. 18
However, considering the overall performance for both component tasks together, and the fact that the time 19
available per task was cut by 50% in the dual-task condition, neither a loss nor a benefit in performance was 20
produced. Thus, the throughput of tasks has actually remained the same for both conditions. Dual-task benefits 21
would be reflected in any higher throughput, that is a total number of correctly performed tasks (summed across 22
both component tasks) higher than 85. This, for example, might be possible if some overlapping processing of the 23
two tasks takes place. In contrast, overall dual-task costs would be reflected in the fact that an individual would 24
achieve a fewer overall number of correct responses than could have been expected from the single-task 25
performance in the single-task blocks (< 85). This could be due to, for example, costs of task switching or costs 26
related to outcome conflicts. 27
3 Note that we use the term “dual-tasking” for any situation where an individual has to cope with two tasks, including situations of
real concurrent performance like in PRP-tasks, situations of classical task switching where individuals have to switch between two tasks, or
situations of task switching which provide options of overlapping processing like in our TSWP paradigm.
33
In our experiments, we worked with single-task blocks of one minute and dual-task trials of two minutes. If a 1
participant is working strictly serially (without considering any switch and mixing costs) we, thus, would expect 2
the same number of correct responses in single-task and mixed-trials or dual-task blocks, respectively. Following 3
this reasoning, we defined ODTPE formally as follows: 4
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂=100 ∗��𝑛𝑛𝑛𝑛A_dual + 𝑛𝑛𝑛𝑛B_dual�
�𝑛𝑛𝑛𝑛A_single + 𝑛𝑛𝑛𝑛𝐵𝐵_single��−100 5
With nCA_single and nCB_single defined as number of correct responses in the respective tasks under single-task conditions, 6
and nCA_dual and nCB_dual defined as the corresponding performance in dual-task conditions. Based on this measure, 7
performance benefits of multitasking as described above are reflected in values of ODTPE > 0. In contrast, costs 8
of multitasking are reflected in values of ODTPE < 0. Note that the consideration of the number of correct 9
responses generates an overall efficiency measure that represents costs and benefits reflected in both, response 10
times and error rates. 11
12
