Multitasking in Complex Environments: An Attempt to Generalize Individual Differences in Multitasking to a Realistic Task Setting

Introduction

Multitasking, that is, the requirement to perform several tasks at the same time, or one after another (Koch et al., 2018), has become part of everyday life. Typically, multitasking requires either simultaneous processing, where two or more tasks are performed concurrently or sequential processing (also called task-switching), where one switches between multiple tasks in close succession (Kiesel et al., 2010; Koch et al., 2018).

Multitasking research can be categorized into two main strands. Cognitive psychologists have primarily investigated the underlying cognitive processes and processing limitations by comparing multitasking costs to single-task situations (Miller & Durst, 2014). For this purpose, simple tasks like letter or digit classification were used as a first strand to draw clear conclusions about cognitive limitations. In contrast, more applied research focuses on the overall performance consequences of realistic multitasking situations (Lam et al., 2022). Therefore, the second strand explores complex tasks that reflect real-world multitasking scenarios, incorporating additional task demands such as task complexity, environmental factors, and time pressure¹, which are considered in the current study. Despite these differences, multitasking paradigms of both strands share fundamental similarities with respect to the cognitive mechanisms involved (e.g., shifting attention or inhibiting distractors; Koch et al., 2018).

Recent studies show that when providing participants with some degree of freedom in their task processing, individual preferences for task processing can be identified (e.g., Brüning & Manzey, 2018; Brüning et al., 2020; Reissland & Manzey, 2016). Using the task-switching with preview (TSWP) paradigm, the authors identified serial processors (i.e., participants who ignored the preview option), parallel processors (i.e., participants who effectively used the preview option to prepare for the next task switch), and semi-parallel processors (i.e., a third subset of participants who used the preview, but unsystematically). These processing styles have also been found to impact multitasking efficiency, with a higher number of correct answers per time observed for a parallel processing style (Brüning et al., 2020), which further was associated with a higher working memory capacity (WMC; Brüning & Manzey, 2018). In contrast, studies using complex paradigms have focused holistically on the effects of variables such as WMC on multitasking performance (Hambrick et al., 2010), without considering the influence of specific processing strategies.

HS and JBi share first authorship. JBr and TR share last authorship.

The main goals of the present study are, therefore, twofold. First, we aim to establish a link between overall performance in TSWP and a more complex multitasking paradigm. Second, we aim to investigate potential performance benefits for serial/parallel processors in dual-task setups that should be beneficial for the individual processing style in the more complex paradigm.

To this end, we conducted an online experiment with two sessions, where participants worked on an online adaptation of the Synthetic Work Task (SynWin; Elsmore, 1994) in the first and on the TSWP in the second session. We selected the SynWin task as our more complex paradigm because it is an established semi-ecologically valid multitasking paradigm (see, e.g., Balkin et al., 2000), for which WMC has also been shown to be associated with better multitasking performance (Hambrick et al., 2010). This task permits interleaving and switching between four distinct subtasks, each with different levels of urgency, and therefore exhibits a higher level of task complexity, time pressure, and distractions than the TSWP paradigm while having qualitatively similar demands on cognitive mechanisms such as WMC and task-shifting.

We hypothesized that performance efficiency (i.e., correct answers within a time frame) correlates positively between the two paradigms. Moreover, to explicitly test performance differences between serial and parallel processors, we used two different dual-task combinations in the SynWin paradigm, one of which should be particularly beneficial for parallel processors. We expected that parallel processors would have a higher multitasking efficiency than serial processors in an arithmetic- and memory task combination, but no differences between serials and parallels in an arithmetic and visual monitoring task combination, as the latter requires less switching and has lower WMC demands than the former.

Methods

Preregistration including experimental setup, hypotheses, and statistical methods (https://tinyurl.com/25z8aeby), and data and scripts (https://tinyurl.com/ycxy34d4) are available at OSF. The study included two experimental sessions with a minimum of one day and a maximum of one week between both sessions.

SynWin Paradigm

In the first session, after giving informed consent, participants performed in an online version of the SynWin paradigm (Elsmore, 1994). We used the Gorilla Experiment Builder (www.gorilla.sc) to create and host the experiment (Anwyl-Irvine et al., 2020). Four distinct subtasks (Figure 1) involving memory, arithmetic processing, visual and auditory monitoring were presented in separate windows of the screen.

OPEN IN VIEWER

Figure 1.

Screenshot of the SIMON multitasking environment.

In the memory task (located in the top left), participants were instructed to memorize a letter sequence and then determine if a random letter, presented about every 8 s, was part of that sequence. The arithmetic task (top right) was constantly available and required solving arithmetic additions. In the visual monitoring task (bottom left), the participants were asked to click on the task window as soon as a needle reached the lower range of a fuel gauge, which took about 30 s for the gauge to go from full to empty. The auditory monitoring task (bottom-right) involved the presentation of two different tones, with the instruction to hit an alarm button when hearing the target tone. Tones were played approximately every 10 s.

In the arithmetic task, participants received 20 points for correct responses and had 10 points deducted for incorrect responses. Similarly, in the memory and auditory monitoring tasks, correct responses were awarded 10 points and missing or incorrect responses resulted in a 10-point deduction. The point to be earned in the visual monitoring task ranged from 0 to 10, based on the level of the tank, and 10 points were subtracted for every second that the tank was empty. The primary goal was to maximize the total score in each experimental condition.

After a short practice session, they first performed all four tasks separately for 1 min each (single-task (ST) environment). This was followed by two different dual-task environments. A combination of the arithmetic and visual monitoring task (DT1) was intended to enforce a more serial task processing, because participants must pause the arithmetic task intermittently to visually monitor the fuel gauge. In contrast, the arithmetic task combined with the memory task (DT2) should foster parallel task processing, because the letter sequence must be continuously held in working memory while solving the arithmetic additions.

Again, after practicing both task combinations, participants worked on both DTs for 2 min each in counterbalanced order. Finally, they first practiced and then performed all four tasks simultaneously for 4 min. The procedure of completing each of the ST, DT, and MT environments for 1 min, 2 min, and 4 min respectively was repeated in the same order (excluding the time spent on the practice sessions). As a result, the participants spent equal amounts of time in each of the experimental conditions (ST, DT, and MT).

TSWP Paradigm

Participants were then invited to the second experimental session to perform the TSWP. We used jspsych (de Leeuw, 2015) to program and a JATOS server (Lange et al., 2015) to host the experiment. The participants were given two classification tasks, first separately and then combined in a task-switching environment. For this purpose, we presented white letter and digit stimuli on a grey screen (Figure 2).

OPEN IN VIEWER

Figure 2.

Trial procedure of the TSWP environment, adapted from Brüning et al. (2021, Figure 2).

In the letter classification task, they had to classify letters as vowels or consonants and press either the ‘A’ or ‘S’ key on the keyboard using their middle and index finger, respectively. In the digit classification task, they were supposed to categorize digits as even or odd and press either the ‘K’ or ‘L’ key on the keyboard with their index finger or middle finger, respectively. Hand-task assignment was counterbalanced across participants.

First, all task conditions were separately demonstrated to the participants followed by a short practice session. Then, subjects worked in the task-switching environment for two experimental blocks that lasted 2 min each. The instruction was to work on three stimuli of one task category (e.g., digits), then switch to the other category (e.g., letters) for the following three stimuli as indicated by an arrow cue. While performing in one task category, the upcoming stimulus of the other task category could be previewed (see Figure 2). Afterwards, participants performed the letter and digit classfication tasks in the single-task environment for 1 min each in counterbalanced order. This procedure of experimental blocks was repeated three times in total. The participants' were instructed to generate as many correct responses as possible and to minimize reaction times.

Results

To test the hypothesis of whether there is a relationship between individual performance in both paradigms, we computed a product-moment correlation between the multitasking efficiency scores of the TSWP task and the SynWin task. The correlation was not significant (r = .16, p = .29), indicating that, at the level of the whole sample, the overall performance efficiency achieved in the more controlled task-switching paradigm is unrelated to the efficiency observed in the more complex paradigm. Descriptively (Figure 3), one reason for this could be that the relationship between the efficiency scores varies depending on whether a parallel, semi-parallel, or serial processing mode is preferred. We conducted exploratory follow-up analyses to examine these tendencies. While there seems to be a descriptively positive relationship for the serial processing mode (n = 10, r = .33), it is negative for semi-parallel processors (n = 12, r = -.68). No identifiable relationship occurs within the group of parallel processors (n = 23, r = -.09). This observation tentatively suggests that the two task environments were more similar for serial processors than for (semi-)parallel processors, and therefore the underlying processing mechanisms required may overlap more between the two environments for serial processors than for (semi-)parallel processors.

OPEN IN VIEWER

Figure 3.

Scatterplot showing the relationship between TSWP and SynWin multitasking efficiency for different processing modes.

In order to analyze whether parallel processors show performance advantages over serial processors in the dual-task condition of the SynWin paradigm, semi-parallel processors were excluded from this analysis. We ran a 2x2 ANOVA with processing mode (i.e., serial vs. parallel) as a between-subjects factor and SynWin DT condition (i.e., maths-fuel vs. maths-memory) as a within-subject factor (Figure 4). The respective SynWin DT efficiency scores served as outcome measures. We found a main effect for dual-task type (F(1, 31) = 54.32, p < .001, η_p² = 0.64) with better performance in DT1 (i.e., arithmetic and visual monitoring task) than in DT2 (i.e., arithmetic and working memory task). However, any potential differences due to processing styles may have been masked in the SynWin task by its complex task structure. The analysis revealed no main effect for processing mode (F(1, 31) = 0.23, p = .64, η_p² = 0.007) and no interaction effect between DT type and processing mode (F(1, 31) = 2.40, p = .13, η_p² = 0.07).

OPEN IN VIEWER

Figure 4.

Barplot showing the mean efficiency score of the respective SynWin DT for different processing modes.

Discussion

In the present study, we (1) aimed to investigate the relationship between individuals’ multitasking performance in a more controlled paradigm and their performance in a more complex, semi-ecologically valid task environment. The study (2) also aimed to explore whether parallel processors may have performance advantages in a complex dual-task environment that was intended to demand this processing mode.

To achieve these aims, we (1) calculated efficiency scores representing the relation of multitask performance to single-task performance for both paradigms and then correlated these scores. However, no significant correlation was found between individuals' scores in the two paradigms.

In addition (2), three out of four subtasks of the SynWin paradigm were selected to form a dual-task that was assumed to primarily involve parallel processing and a second one, which primarily requires serial processing. It was expected that individuals categorized as parallel processors in the TSWP paradigm would show better performance as serial processors in the former dual-task environment. However, the respective analysis did not reveal significant performance differences. Instead, the type of dual-task was found to be a significant determinant of the performance outcome, with participants achieving significantly higher efficiency scores in the maths-fuel dual-task compared to the maths-memory dual-task.

Beyond that, the majority of the SynWin efficiency scores being positive creates additional challenges when interpreting the data. There is a consensus in the literature that multitasking is commonly regarded as being more costly than beneficial in comparison to single-tasking (for an integrative review, see Koch et al., 2018). In our study, this applies to the TSWP paradigm as switching from one task category (e.g., digits) to the other (e.g., letters) is accompanied by multitasking costs for most participants. However, the subtasks of the SynWin paradigm are likely to be combined in a way that makes efficient interleaving of tasks more feasible. For instance, the SynWin fuel task is relatively easy to monitor, so participants can earn points on the other tasks in the interim. Thus, if subjects are supposed to manage all four tasks simultaneously, they likely do not lose points; instead, they gain some.

To summarize, there seem to be constraints with respect to the experimental conditions necessary to generalize findings from a controlled and frequently used multitasking paradigm to a more complex scenario. One reason for this may be that multitasking has different task demands in realistic compared to controlled paradigms due to varying complexity levels of the task environment. That is, the nature of the tasks themselves may contribute to differences in performance. In the SynWin paradigm, all subtasks are permanently available, and participants can choose a strategy to manage all tasks simultaneously. Again, efficient task interleaving can be achieved by taking advantage of natural waiting times occurring for some subtasks as they do not require continuous attention (for a similar approach, see, Mittelstädt et al., 2018). The TSWP, however, is relatively inflexible as the stimuli have to be processed one after another in a predetermined order without the possibility of disregarding any stimuli.

Possibly, the SynWin task involves multitasking costs or benefits associated with planning and managing multiple tasks simultaneously. Participants can (a) schedule themselves when they want to switch between tasks and (b) permanently adapt their strategies to maximize their total scores while reducing mental load. In contrast, the TSWP also provides a preview of upcoming stimulus category to facilitate task switches, but subjects do not have a free choice of when to perform these switches. In combination with different degrees of stimulus complexity and number of presented tasks, too many parameters may differ between the paradigms, making it difficult to establish an identifiable relationship between individual performance levels.

Future research should therefore examine the relationship between one controlled and another more complex paradigm that are more closely aligned to each other. Specifically, an adaption of the SynWin paradigm where the decision when to switch to another subtask is not self-determined but externally scheduled, as in the TSWP, while also providing a preview could be promising. Further, using a complex multitasking paradigm that is more difficult to manage and provokes multitasking costs rather than benefits could enhance similarities with the TSWP paradigm. The Multi-Attribute Task Battery (MATB, Comstock & Arnegard, 1992) is frequently utilized and may serve as a suitable alternative paradigm. In the MATB, all tasks require constant processing, and more frequent task switching is required than in the SynWin—two key similarities to the TSWP that perhaps allow for a closer link between the paradigms. As a result, multitasking is generally rather costly than beneficial (Lam et al., 2022) in the MATB, another similarity to the TSWP. Ultimately, implementing a more realistic task environment other than the SynWin paradigm that can classify different processing modes could be promising for conducting a deeper investigation into the impact of individual processing modes in more complex situations.

Conclusion

Our experiment revealed that a high-level individual multitasking efficiency in the TSWP paradigm does not predict high-level multitasking efficiency in the SynWin task.

In addition, the individual processing mode identified by the TSWP paradigm was not accompanied with performance benefits in the SynWin dual-task combination that was intended to address the respective processing mode.

In summary, the present study reveals a gap between highly controlled multitasking studies and more complex tasks, a phenomenon that is often found in psychological research. To gain deeper insight into how people multitask in more ecologically valid settings and the underlying cognitive processes involved, future research should prioritize efforts in this direction. Fundamentally, interdisciplinary collaboration between cognitive science and applied disciplines that deal with working environments and other day-to-day scenarios is essential to understand the impact of basic cognitive mechanisms on real-world behavior.