Individual preferences for task coordination strategies in multitasking: exploring the link between preferred modes of processing and strategies of response organization

In our daily lives, we are often confronted with the demand to multitask—that is, to cope with two or more tasks overlapping in time. To this end, more than one task-set must be maintained concurrently, and ways need to be found to coordinate the concurrent tasks effectively (Koch, Poljac, Müller, & Kiesel, 2018). Our basic assumption of the present study is that elementary challenges of task coordination in multitasking involve two interwoven aspects, which we refer to as task processing and response organization.

Task processing concerns how the internal processing of multiple tasks is organized. It involves, for example, issues such as serial versus overlapping processing, or maintaining and activating different task-sets. For the last 40 years, this aspect of multitasking has been the main focus of interest, involving two different lines of research. These two lines include research on central limitations of parallel task processing based on the psychological-refractory-period (PRP) paradigm (Pashler, 1994a; Telford, 1931), and research on cost effects when switching between tasks based on different variants of the task-switching paradigm (Kiesel et al., 2010). Both paradigms focus on a high control of timing of tasks and a prescribed sequence of tasks, resulting in no or only very few degrees of freedom for own response organization, which makes them particularly suitable to reveal specific constraints involved in task processing. As a result, single mechanisms could be identified, including but not limited to central limitations in response selection (Pashler, 1994a), issues of crosstalk and stimulus–response bindings (Hazeltine, Ruthruff, & Remington, 2006; Hommel, 1998; Navon & Miller, 1987), or cost effects due to a reconfiguration of task-sets when switching between tasks compared to task repetitions (Monsell, 2003). These mechanisms represent constraints of cognitive processing humans have to cope with when performing multiple tasks concurrently.

In contrast, issues of response organization particularly arise in completely self-organized multitasking and involve the aspect of how responses to concurrent tasks are organized over time, reflected in a certain sequence of behavioral responses to the different tasks. Thus, response organization can differ in respect to when and for how long a person works on each of the concurrent tasks. The standard paradigm, reflecting such demands, is represented by the typical concurrent dual-tasking paradigms of the 1970s and 80s (Navon & Gopher, 1979). Here, participants are required to work concurrently on two independent streams of tasks (Wickens, Mountford, & Schreiner, 1981). Usually, they are instructed to maximize their performance by achieving as many trials of both tasks as they can within a given time window, without further instructions or constraints on how to achieve this. That is, participants have full control over how often they repeat one task, switch between tasks, or whether or not they group their responses. Studies using this paradigm have usually addressed performance only on a more holistic level, using relatively global performance measures of multitasking efficiency and task interference (Navon & Gopher, 1979; Wickens, 2002). Thus, they have acquired broad knowledge about which tasks can or cannot be performed concurrently with considerable efficiency, but largely neglected any detailed investigation of the strategies of response organization involved in this performance (see for an exception, Damos & Wickens, 1980). Other paradigms involving issues of response organization in addition to task processing are variants of task-switching or PRP paradigms providing at least some degrees of freedom for individual response organization. Examples are the voluntary task-switching paradigm (Arrington & Logan, 2005), or PRP paradigms, which do not impose a specific response sequence by means of stimulus presentation or instruction, thus providing degrees of freedom for individual choices of how responses are organized (De Jong, 1995; Kübler, Reimer, Strobach, & Schubert, 2018; Pashler, 1994b).

Interestingly, it seems that individuals differ in their preferences of how they process tasks and how they organize responses during multitasking. In this context, preferences can be understood as a stable tendency (comparable to a bias or affinity) how individuals spontaneously deal with multitasking demands. The existence of such preferences is suggested by a variety of incidental observations and anecdotes, which can be found in the literature (e.g., Damos & Wickens, 1980; Jersild, 1927; Kessler, Shencar, & Meiran, 2009; Schumacher et al., 2001). For example, already in the very first experimental series on task switching, Jersild (1927) observed that a subgroup of participants did not show switch costs but switch benefits. Based on this observation, he assumed that these participants had overlapped the processing of both tasks, whereas others tended to process the tasks in a strict serial manner. Similarly, observations of individual preferences can be found in results of PRP research that investigates the effects of different practice schedules on the PRP effect (i.e., the prolongation of the response to the second task after a short stimulus onset asynchrony). The results suggest that even after extensive practice of one of the component tasks or both tasks, not all participants were able to eventually reduce the PRP effect to a negligible minimum (Maquestiaux, Lague-Beauvais, Ruthruff, & Bherer, 2008; Maquestiaux, Ruthruff, Defer, & Ibrahime, 2018; Schumacher et al., 2001, Exp. 3).

Individual differences in the way responses are organized become particularly apparent when experimental paradigms are used that offer at least some degrees of freedom in this respect. An example from PRP research includes the common finding that, if no definite sequence of responses is required, participants do not always organize their responses to the tasks in a strictly serially manner but tend to group their responses together in a considerable number of trials (Ulrich & Miller, 2008). Other observations include individual differences in the so-called repetition bias in voluntary task switching (Kessler et al., 2009; Mittelstädt, Dignath, Schmidt-Ott, & Kiesel, 2018; Mittelstädt, Miller, & Kiesel, 2018), or the observation of different strategies of response scheduling in concurrent task performance, for example, massed, alternating, or simultaneous responding (Damos & Wickens, 1980).

To explore both aspects of individual differences in multitasking in a more systematic manner, two different paradigms have been proposed (Reissland & Manzey, 2016), which also were used in the present study. The first one is the task-switching with preview (TSWP) paradigm, which enables to identify individual preferences with respect to a serial versus overlapping mode of task processing (Reissland & Manzey, 2016). In classical task switching paradigms, only the stimulus of one of two tasks is visible at a time, thus, requesting a strict serial processing of both tasks (Kiesel et al., 2010). In the TSWP paradigm, the stimuli of both tasks are always visible concurrently, with a cue marking the task that has to be performed in the current trial (i.e., the currently relevant task). After each response to the relevant task, only the stimulus of this latter task is updated, while the stimulus of the second task remains. Thus, while working on the marked (relevant) task, the concurrently visible task stimulus of the other task serves as a preview to the stimulus, one has to respond to after the next task switch (Spector & Biederman, 1976). By presenting the task stimuli in this manner, combined with a regular task sequence (e.g., AAABBB), the TSWP principally allows but does not enforce an overlapping processing of the switch stimulus while still performing the currently relevant task. Thus, it provides degrees of freedom for different modes of task processing, but still imposes a certain pattern of response organization (e.g., three times responses to task A followed by three times responses to task B). Using this paradigm, Reissland and Manzey (2016, Experiment 1) found two subgroups of participants differing in whether they made use of the preview to optimize task switches. Thereby, they substantiated previous anecdotal observations of individual biases concerning a more serial versus overlapping mode of task processing (Jersild, 1927) by a more systematic approach. A study by Brüning and Manzey (2018) confirmed these results based on a larger sample size (n = 45), but also found a third subgroup of participants that could not be unambiguously classified as serial or overlapping processors (i.e., so-called semi-overlapping processors). Besides, they showed that at least the individuals using an overlapping processing mode were able to flexibly change to a more serial mode of processing in case of increased risk of crosstalk between tasks. This suggests that the spontaneous preference for a certain mode of processing is not a rigid predisposition, but a kind of cognitive style that can be flexibly adapted to the current environmental constraints and might be at least partly under voluntary control.

To additionally explore individual preferences at the level of response organization strategies, Reissland and Manzey (2016) used what we refer to as free concurrent dual-tasking (FCDT) paradigm. This paradigm directly corresponds to the concurrent dual-tasking paradigms from the 1970s and 80s as described above. Similar as in the TSWP, the stimuli of two streams of tasks are always concurrently visible. However, beyond that, participants in the FCDT are also completely free to decide when and for how long they want to work on a given task. Using this paradigm, combined with applying a detailed post hoc analysis of individual response patterns, Reissland and Manzey (2016) could identify three different strategies of response organization. One subgroup of individuals used a blocking strategy; that is, they focused on one task for a considerable number of trials (i.e., 65 repetitions on average) before switching to the other. Another subgroup of individuals preferred a switching strategy, reflected by switches between tasks about every 4–7 trials. The remaining individuals permanently alternated between both tasks. However, a closer look at the time characteristics of their response patterns showed a remarkable similarity to a response grouping pattern described previously in the studies by Damos and Wickens (1980) and Damos, Smist, and Bittner (1983). Stereotype response patterns reflecting the different strategies are shown in Fig. 1. The blocking strategy seems to be motivated by keeping concurrent tasks as separate as possible, thereby reducing multitasking demands to a possible minimum (i.e., only maintaining two task-sets). In contrast, the other two strategies have in common that individuals strive for some sort of task interleaving as a means of coordinating multiple tasks.

The replicable findings of individual preferences for different modes of task processing in the TSWP paradigm and certain strategies of response organization in the FCDT paradigm raise the question of a possible link between these two aspects. Although it seems plausible that both aspects of multitasking are interwoven, it is still unclear to what extent the way in which an individual cognitively prefers to deal with multitasking requirements directly translates into the temporal organization of responses to the tasks. At least the findings of different styles of response organization in voluntary task-switching and in PRP research with SOA = 0 referred to above suggest that modes of task processing and behavioral response organization do not just represent two sides of the same coin. Instead, they are two aspects of multitasking, which, at least to some extent, could be combined in different ways. For example, in the PRP paradigm, the same serial processing at the stage of response selection can be associated with different sorts of response organization (e.g., serial responding or response grouping, see Pashler, 1994b).

The present study particularly aimed to examine possible links between the two aspects and how they both interact in determining the overall multitasking efficiency. Specifically, we assumed that the way how individuals prefer to process the tasks in the TSWP (serially vs. overlapping) ideally would be predictive for how they organize their responses when performing the FCDT to find the best compromise between maximizing performance while at the same time minimizing possible costs of multitasking.

For example, if individuals strongly prefer a serial task processing mode in the TSWP paradigm, thus striving to shield the processing of the currently relevant task from the processing of the preview stimulus, the frequent imposed switches in this paradigm lead to severe performance decrements by accumulating switch costs (Brüning & Manzey, 2018; Reissland & Manzey, 2016). However, with the possibility for free response organization in the FCDT, serial processors can avoid these costs by choosing a strategy of minimizing switches, that is, by working on longer sequences of one task before switching to the other. Thus, for individuals preferring serial processing in the TSWP paradigm, we expected a corresponding preference for the blocking strategy in the FCDT paradigm.

In contrast, overlapping processing during the fixed sequence of the TSWP paradigm allows benefiting from the stimulus preview by generating considerable time gains at switch trials. Hence, individuals with a preference for overlapping processing should strive to switch between tasks to exploit options to save time. As the FCDT paradigm provides both the opportunity to freely schedule task switches and to preview the task stimulus which they would switch to, it should be especially appealing to overlapping processors identified in the TSWP paradigm to organize their responses in the FCDT paradigm in a way that supports overlapping processing. Thus, we predicted that individuals preferring an overlapping processing mode in the TSWP paradigm preferred interleaving strategies of response organization (i.e., either switching or response grouping) in the FCDT paradigm.

Regarding the semi-overlapping processors, the expectations for according links to response organization strategies were less clear-cut, though. Since they do not exhibit a distinct preference for either mode of processing, both could be feasible for them. Hence, the according links for semi-overlapping processors were investigated in an exploratory manner.

To examine the proposed link between both aspects of task coordination in multitasking, we investigated the performance of the same participants in the TSWP and FCDT paradigm. The same set of simple classification tasks as in the study by Reissland and Manzey (2016) were used. Links between the emerging individual differences on both levels of task coordination probed by the different paradigms were examined.

In addition, we were interested in how the resulting combinations of both task coordination levels affected the overall multitasking efficiency in terms of task throughput in blocks where the two tasks had to be performed according to a prescribed schedule (TSWP) or in a self-organized manner (FCDT), relative to single-task performance. Based on the assumed link between modes of processing and response organization strategies, we expected that individuals preferring the overlapping mode of task processing in TSWP and a corresponding interleaving strategy in FCDT would achieve a higher overall dual-task efficiency, reflected in multitasking benefits than participants preferring serial processing and a blocking strategy. The former group should benefit from the preview during switching, whereas the cautious strategy of the latter group should lead to neither costs nor benefits. Worst overall multitasking efficiency in terms of task throughput (how many of the two tasks can be accurately performed in a given time) was predicted for participants preferring serial processing but applying one of the interleaving strategies of response organization in self-organized multitasking, that is, strategies which would accumulate performance costs due to repeated task switching.

The present study was guided by two objectives. First, the experiment was conducted to investigate the hypothesis that preferred modes of task processing and strategies of response organization were systematically linked; that is, they would reflect interwoven aspects of individual differences in multitasking. As will be discussed in some detail below, the results support this hypothesis to a large extent. The second objective was to investigate the hypothesis that the specific combination of preferred mode of task processing and strategy of response organization would determine the multitasking efficiency. This hypothesis was only partially supported by the results.

To investigate the first objective, the preferences of each individual participant for both aspects of task coordination were probed based on two different paradigms. The two paradigms reflected different degrees of freedom for the participants to self-organize their performance of two tasks. The TSWP only demanded task coordination at the level of task processing, whereas the FCDT put additional demands on the level of response organization. The identification of the task processing mode and the response organization strategy individuals preferred followed the rationales described by Reissland and Manzey (2016) and Brüning and Manzey (2018), enriched by the introduction of two additional statistical criteria (i.e., the PSI and the use of Hartigan’s Dip Test) to further objectify the procedure. The results directly confirm the findings of our previous research. Two subgroups differing in their spontaneous preference for serial versus overlapping processing could be identified based on the use of preview in the TSWP, and again, three different types of strategies of response organization (blocker, switchers, and response groupers) emerged in the FCDT paradigm.

Even more important, the data also supported our assumption that individuals applying serial versus overlapping processing in the TSWP paradigm differed in their preferred strategy for organizing the responses to the same two tasks in the FCDT paradigm. As expected, the vast majority (74%) of participants who processed the tasks in the TSWP in a strict serial manner showed a strong preference for blocking their responses to the two tasks in the FCDT. This allowed them to keep the two tasks as separate as possible even in the multitasking situation. In contrast, and also in accordance with our expectations, 24 out of 26 individuals using the preview option for overlapping processing in the TSWP preferred a sort of task-interleaving strategy, that is, either switching or response grouping, in the FCDT. Both effects provide evidence for our general hypothesis that humans tend to organize their responses to two concurrent tasks in a way that is supported by their preferred mode of task processing. However, there are also a noteworthy, albeit small number of serial and overlapping processors in TSWP who have chosen a response strategy under conditions of self-organized multitasking that was not in accordance with our predictions. This in turn emphasizes the fact that the preferences for the two aspects of task coordination, task processing and response organization, are clearly interwoven, but do not simply represent interchangeable aspects of individual differences in multitasking.

The second aim of the current study was to investigate how different combinations of individually preferred modes of task processing and strategies of response organization would affect the multitasking efficiency of participants. The multitasking efficiency was assessed by comparing the task throughput (i.e., number of correct responses in the given time) achieved by a participant while multitasking compared to single-task performance. Based on the results of our previous research (Brüning & Manzey, 2018; Reissland & Manzey, 2016), we expected overlapping processors to perform more efficiently than serial processors in the TSWP paradigm, because they can realize time gains particularly at task switches. Consequently, we also expected overlapping processors to achieve multitasking benefits in the FCDT when choosing a strategy of interleaving both tasks by either switching frequently between both tasks or by applying response grouping. Both strategies might benefit from overlapping processing. In contrast, serial processors in the TSWP were expected to prefer a blocking strategy of response organization in self-organized multitasking to reduce the number of switches between the tasks and thus minimize time losses due to task switches. As a result of these disadvantages of serial processing when using interleaving strategies of response organization, serial processors were expected to perform worst of all subgroups when choosing one of the task-interleaving strategies.

Our results, however, only partially support these assumptions. As expected, applying overlapping processing using the preview in the TSWP paradigm led to significantly better multitasking performance than serial processing. Furthermore, our data also confirm the assumption that serial processors as identified in the TSWP showed worst multitasking performance in terms of highest multitasking costs when applying a switching or response grouping strategy in the FCDT. This finding is most plausible, since the combination of serial processing with a switching or response grouping strategy involves substantial time losses when switching between task, which in turn accumulate over a high number of switch trials. Accordingly, the fact that serial interleavers perform poorly in terms of multitasking efficiency is not surprising. The fact that some serial processors used an interleaving strategy of response organization, at all, suggests that they had not been aware of their relatively inefficient performance. Indeed, research on metacognition of multitasking in controlled laboratory settings (Finley, Benjamin, & McCarley, 2014) as well as in the field (Horrey, Lesch, & Garabet, 2008, 2009; Lesch & Hancock, 2004) shows that people might not necessarily have metacognitive insight on the extent to which they are personally vulnerable to the risks of multitasking. Similarly, also in the PRP paradigm, characterized by high control of timing of tasks, participants often are apparently not aware of their multitasking costs in dual-task settings (Bryce & Bratzke, 2014, 2017; Corallo, Sackur, Dehaene, & Sigman, 2008). This might also explain why, in the debriefing of the experiment, not only most of the serial blockers and overlapping interleavers, but also most of the serial interleavers reported to have chosen this particular response strategy to perform more efficient. However, since the participants were not requested to predict or assess their own performance efficiency in the course of the present study, further studies are necessary to determine whether individuals differing in their preferences for task coordination differ in their accuracy of introspective knowledge, as well.

In clear contrast to our expectations, the combination of overlapping processing and interleaving response strategies was not more beneficial than the combination of serial processing and the blocking strategy. At least descriptively, the subgroup of serial blockers even showed the highest performance efficiency of the three largest subgroups in the FCDT. They performed the two tasks with about the same task throughput as in the single-task condition. Based on their preference for a serial mode of task processing, we expected that they would not be able to realize any multitasking benefits, but would be prone to considerable cost effects when switching between tasks in self-organized multitasking. However, using a blocking strategy in the FCDT, serial blockers reduced the number of task switches to a minimum to cope with the multitasking demand. As a result, they minimized the risk of time costs when switching, so that overall neither the benefits nor the costs of multitasking arose.

Nonetheless, the overlapping interleavers did not outperform the serial blockers in the FCDT. This effect emerged, although the overlapping processors significantly outperformed the serial processors in the TSWP paradigm and although they applied a similar strategy of using preview and realizing a considerable number of fast switches in the FCDT as in the TSWP paradigm. Indeed, our additional analysis of the time gains and losses of overlapping switchers in the TSWP and FCDT paradigms revealed that their multitasking costs were obviously not caused by time losses at task switches, but by time losses in repetition trials. Furthermore, these latter repetitions costs emerged independent of the length of the next switch response, which rules out that the overlapping processors were less able of overlapping processing in the FCDT than the TSWP paradigm. Therefore, what might have been the cause of these cost effects? One might argue that these repetition costs reflect typical mixing costs, which sometimes have been reported from task-switching studies resulting from the demand to keep two task-sets active in memory (Kiesel et al., 2010). However, the fact that both subgroups did not show similar costs in the TSWP renders this interpretation not very plausible. Furthermore, such costs did not emerge in the FCDT data of our serial blockers, despite the fact that they also were requested to maintain the task-sets of both tasks active in memory. Therefore, the cost effects observed in repetition trials of overlapping switchers must be related to multitasking demands in the FCDT that are specific for the switching strategy.

The most obvious difference between the switching and the blocking strategy in this respect is that switchers had to organize their responses and the scheduling of switches “online” while performing the two tasks concurrently. For this purpose, they must have monitored their performance and have chosen their exact order of responses throughout a dual-task block. In contrast, the blocking strategy can be planned beforehand with only minimal demands on online monitoring. We suppose that the repetition costs visible in the data of the overlapping switchers reflect the additional demands involved in monitoring and decision-making about switches associated with a switching strategy. This would also be in line with results from the voluntary task switching paradigm, suggesting that already the act of choosing the task to be performed incorporates top-down processes that generate time losses separable from switch costs (Arrington & Logan, 2004, 2005). Similar costs can also be found in PRP research showing that the coordination of the processing order of two tasks requires additional control processes (Kübler et al., 2018; Szameitat, Lepsien, Cramon, Sterr, & Schubert, 2006; Szameitat, Schubert, Müller, & Cramon, 2002). In sum, these additional demands on the cognitive control might explain the sustained cost effects found for overlapping switchers under conditions of self-organized multitasking. However, as cognitive control demands were not directly manipulated and investigated in the current study, the data are not conclusive in this regard. Nonetheless, if costs due to self-organization could be overcome, for example, through training, overlapping processing might eventually develop its potential and increase overall multitasking efficiency.

To conclude, the results of the present study again confirm earlier findings of individual differences in multitasking, related to how cognitive processes are organized (serially vs. overlapping) and how responses to the tasks are scheduled (separating vs. interleaving). As a new insight, they also provide strong evidence for a correspondence between the individual preferences for these two aspects of task coordination. The majority of individuals preferring a serial processing mode in the more controlled task-switching with preview paradigm preferred a blocking strategy in the free concurrent dual-tasking paradigm, successfully reducing the dual-task demands. Comparably, most individuals preferring overlapping processing preferred an interleaving response organization strategy. This obvious link suggests that individual preferences emerging in both aspects of task coordination might reflect a common basic difference in the individually preferred style of cognitive control, corresponding to the distinction of flexibility and persistence as different styles of cognitive control proposed by Hommel (2015). The sources of these differences are difficult to derive from the currently available data and, thus, are a matter of speculation at the present time. As Hommel (2015) suggests, they might include environmental influences like cultural differences and/or learning experiences, or even genetic predispositions. However, overlapping processors were not consistently able to transfer their performance advantages in terms of time gains at switches to a self-organized context due to the choice of a non-optimal response scheduling and/or new costs. The latter probably resulted from increased performance monitoring demands in completely self-organized multitasking. Further research should try to identify relevant context variables affecting the relative benefits and costs of different sorts of task coordination in terms of multitasking efficiency.

Another interesting question for further research regards possible links of the preferences found at the different levels of task coordination to other cognitive abilities (e.g., working memory, processing speed, or fluid intelligence) or even personality factors (e.g., resistance to change, polychronicity, or motivational tendencies). Thus far, the role of cognitive abilities, personality characteristics, and different attitudes as possible predictors of multitasking performance has only been investigated with respect to comparatively complex work simulations (e.g., Hambrick, Oswald, Darowski, Rench, & Brou, 2010; König, Bühner, & Mürling, 2005; Mäntylä, 2013; Redick, 2016; Zimmermann, Kubik, Persson, & Mäntylä, 2019). The results suggest that particularly working memory capacity can predict the performance in complex multitasks. However, the performance indicators considered in these studies were usually limited to global performance scores, which do not provide much insight in how the superior multitasking performance is achieved with higher working memory capacity. Identifying links between cognitive abilities, personality characteristics, preferred modes of task processing and strategies of response organization would nicely complement this line of research. For example, first hints were found that the preference for serial versus overlapping processing might be related to working memory capacity (Brüning & Manzey, 2018). In this sense, the emergence of different preferences for specific modes of processing and strategies of response organization would reflect mediating factors explaining how higher working memory capacity can be used to achieve better multitasking performance.