With the aim of investigating the mechanisms of dual-task response-code conflict in young and older adults, we implemented an auditory–manual single-stimulus onset paradigm with one versus two concurrent speeded choice responses (Huestegge & Koch,
2009; Pieczykolan & Huestegge,
2018). In this paradigm, we varied the compatibility between the location implied by the high or low pitch of the auditory stimulus and the high or low location of the motor response (S-R compatibility; cf. Rusconi et al.,
2006). This allowed us to manipulate the congruency between the response codes guiding the two manual responses in dual-response trials (R-R congruency) and elicit response-code conflict through opposing mapping rules (e.g., high pitch—low response combined with a high response). Previous studies have investigated cross-modal response-code conflict in young adults. Here, we assessed the effect of intra-modal (manual only) response-code conflict across the lifespan compared to dual-response conditions with a single, common mapping rule for both responses but with the same types and numbers of stimuli and responses.
Crosstalk in dual-tasking with response-code conflict
In both age groups, we observed crosstalk effects in trials with intra-modal response-code conflict, as reflected in higher dual-task costs in R-R incongruent versus congruent conditions. First of all, this pattern indicates that our approach to facilitating the confusability of task sets by juxtaposing two largely overlapping unimodal S-R sets with opposing mapping rules did indeed lead to substantial mutual interference. Thus, maximizing the dimensional overlap between tasks with opposing S-R mapping rules in a unimodal setting (cf. Kornblum et al.,
1990; Meyer & Kieras,
1997) compares favorably with previous work on crosstalk induced by cross-modal R-R conflicts (Huestegge & Koch,
2009,
2010; Pieczykolan & Huestegge,
2014). Nevertheless, despite the even higher dimensional overlap in our unimodal setting, relative to previous bimodal ones (using very distinct, less automatically synchronized effectors), the average crosstalk effect on RT was somewhat smaller than in cross-modal studies. This is likely due to the differential impact of crosstalk on dual-task costs of S-R compatible versus incompatible responses, as will be discussed below. Overall, we conclude that crosstalk effects in the context of response-code conflict generalize to situations with unimodal mapping selection (e.g., between hands), as compared to bimodal selection (e.g., between hand and eyes).
As expected, dual-task costs on speed and accuracy were not different between S-R compatible and incompatible responses when response codes were congruent to each other. In other words, the performance difference between single- and two-hand responding was not significantly affected by S-R mapping difficulty as long as only a single mapping rule was to be implemented (R-R congruency). This agrees well with the view that such mutually congruent mappings in settings that require two redundant responses at the same time are dealt with via one conjoint response selection process, rather than two independent ones, the result of which is then deployed to both response effectors for execution (Fagot & Pashler,
1992). As such, any dual-response costs in this situation cannot result from interference between two competing response selections but rather reflect general dual-execution costs (Pieczykolan & Huestegge,
2017,
2018). As our paradigm obviates stimulus-related interference, these costs likely arise from demands for two-hand coordination. Consequently, any manipulation that affects response selection, such as S-R compatibility, would be additive to these dual-execution costs.
When, however, two conflicting S-R mapping rules were to be implemented at the same time (R-R incongruent trials), the picture looked somewhat different: S-R compatible responses were slowed down more than S-R incompatible ones when being paired with the opposite mapping, respectively (see full vs. broken lines in all panels of Figs.
2 and
3). Thus, with conflicting response codes, dual-task costs differed between S-R compatible and incompatible responses, similar to what has been reported for cross-modal R-R conflicts in this paradigm (Huestegge & Koch,
2009,
2010; Pieczykolan & Huestegge,
2014). As alluded to above, this situation is markedly different in that it implicates that two different task representations, each associated with a different hand, are represented in parallel. Along the same lines, a recent study using a single visual compound stimulus to trigger uni- or bimanual choice reactions showed that bimanual performance was significantly slowed when the condition implied independent mapping selections, as compared to bimanual conditions with the selection of a single mapping (termed "single task representation"; Schumacher et al.,
2018). Although the authors could not pinpoint the mechanism behind this interference resulting from representing the situation as two tasks (vs. only one), effects located at the perceptual or motor level can be excluded.
At any rate, if the interference observed in our data, ensuing from having to handle two distinct (and incongruent) S-R mapping rules for either hand simultaneously, would solely result from between-task crosstalk, this interference should be symmetrical. That is, the costs of dual-tasking should be about equivalent for both S-R compatible and incompatible mapping selections because from a rational observer’s perspective, mapping confusion should cut both ways similarly (i.e., impose costs additively). This, evidently, was not the case.
As mentioned in the “
Introduction”, such an asymmetry in dual-task costs is also at odds with a simple central capacity sharing model, in which the first task is thought to receive more processing capacity than the second one (e.g., 80%:20%) (Mittelstädt & Miller,
2017; Navon & Miller,
2002; Pieczykolan & Huestegge,
2019; Tombu & Jolicœur,
2003). In our setting, this model would similarly predict symmetrically increased dual-task costs for both S-R compatible and incompatible responses in R-R incongruent trials, as compared to conditions with congruent response codes, or even lower dual-task costs for the (easier) S-R compatible task (Pieczykolan & Huestegge,
2014). Our findings, however, are consistent with an extended, more flexible model of capacity sharing during dual-tasking, which incorporates the notion of a strategic, task-difficulty-based allocation of limited processing capacity. In particular, Pieczykolan and Huestegge (
2014) observed an analogous pattern with conflicting cross-modal response codes and suggested that processing resources are allocated strategically according to perceived task difficulty (such as S-R mapping selection difficulty) for shielding tasks against crosstalk in multiple-task scenarios (Huestegge & Koch,
2010; Pieczykolan & Huestegge,
2014). As a consequence, when a response-code conflict is present (R-R incongruency), the (more demanding) S-R incompatible response mapping would be prioritized and allocated more processing resources. From this, it is predicted that the more difficult mapping is shielded better against crosstalk and, therefore, a relatively larger part of dual-task costs gets conferred onto the de-prioritized (easier) S-R compatible response.
Alternatively, task-set shielding may be biased because top–down cognitive control is already involved in the mapping selection for S-R incompatible responses, aiming to solve the incompatibility-induced response-code conflict (see Rusconi et al.,
2006). This more controlled mode of task processing, relative to the more automatic processing of an S-R compatible mapping, should induce an attentional bias that may protect the ongoing mapping selection against conflicting external information (i.e., crosstalk), akin to resistance against distractor interference. According to this view, the observed bias in the effectiveness of (sub)task shielding is more a side-effect or natural consequence of an existing bias in attentional resource allocation to one of the (sub)tasks, while in the original proposal, it results from a strategic decision to allocate more resources based on perceived task difficulty or, possibly, processing fluency. Whatever will be shown in future research to be the exact mechanism, we, demonstrating this bias in a unimodal setting, can already exclude effects that may be specific to cross-modal conflicts such as those potentially related to the dominance of oculomotor over manual response selection, by which the control demands of saccades are prioritized over the control of manual responses (see Huestegge & Koch,
2013).
The strong interaction between R-R congruency and age showed that older (vs. young) participants suffer from higher dual-task speed and accuracy costs in conditions with response-code conflict (R-R incongruent vs. congruent conditions). This indicates that older adults have specific difficulties in situations where two mutually incongruent response codes need to be selected concurrently from otherwise highly similar task sets, compared to conditions where two redundant response codes (with a common S-R mapping) need to be selected. Since the age-related results’ pattern of dual-task costs was consistent across both speed and accuracy, we can exclude a difference in the speed–accuracy trade-off between age groups as an explanation. Rather, our results agree with Hartley’s (
2001) findings of specific age-related dual-task disadvantages when two similar manual responses have to be executed. Accordingly, the larger response-related crosstalk from opposing mapping rules might be related to an increased response-code confusability, which is further exacerbated in advanced age. The notion of response-code confusability maintains that the concurrent mapping of the same spatial stimulus features (or implications) to a set of motor responses according to different mapping rules for each effector leads to mutual confusion and interference among the spatial codes associated with either effector (Huestegge & Koch,
2009,
2013; Pashler,
1994; Pieczykolan & Huestegge,
2018).
This notion and our findings align with research showing that task-set shielding gets compromised with aging when ambiguity arises from stimuli and response specifications (Mayr,
2001; Mayr & Liebscher,
2001). These age-related difficulties in (consecutive) dual-tasking have been attributed to a potential inhibitory deficit affecting the attentional mechanisms for the processing of one task (i.e., impaired task-set shielding), as well as to a potential deficit in scheduling attention across different task channels, which might cause distraction among parallel processing streams (Hein & Schubert,
2004; Mayr,
2001; Mayr & Liebscher,
2001).
Alternatively, older adults may utilize attentional top–down control to allocate additional processing resources when it is not needed such that performance is not improved but eventually harmed overall. This processing mode has been recently termed the over-reliance on central attention in advanced age and denotes the increased voluntary allocation of attention to central processes as a general strategy to compensate cognitive deficits that come with age (Maquestiaux & Ruthruff,
2021). Accordingly, in trials with response-code conflict, older participants could have overapplied top–down attention to one specific task set, leaving the other one unattended, thereby harming their dual-task performance overall and causing disproportionately higher dual-task costs in the unattended task. For example, older adults might have overapplied attentional resources to the S-R compatible (vs. incompatible) response, although it does not necessarily need it, impeding the automatic activation of response codes eased by the mapping compatibility. Alternatively, it is possible that the elderly overapplied attentional resources to the S-R incompatible response, leaving the compatible one without resources to activate and execute its associated response selection efficiently. As we did not observe a clear age-specific difference in how central attention was distributed between the two mutually incongruent task sets, as reflected in the lack of a three-way interaction, it remains for future studies to investigate in which dual-task crosstalk scenarios different strategies are prioritized by which age group.
To assess whether the detrimental age effect was a domain-specific process, or could instead be explained by generalized slowing, we implemented the RT data transformation proposed by Madden et al. (
1992). We observed the same pattern of significant effects and interactions with the transformed values as the ones obtained with the untransformed performance scores (dual-task costs on RT and ER), with the following exceptions: trivially, due to the transformation of the young participants’ RT values, the main effect of age was not significant anymore. Further, the previously strong interaction between age and R-R congruency now showed only a tendency towards significance with a small to medium effect size. Thus, the evidence is not fully clear-cut to distinguish between a generalized slowing explanation or a task-dependent process of motor response-code conflict. More research would be required to disentangle the mechanisms underlying older adults’ performance impairments when two mappings, due to contrary S-R mappings, need to be selected in parallel and, therefore, shielded from each other (Mayr,
2001; Mayr & Liebscher,
2001).
All in all, our results show that the control of response-related conflict, such as the spatial congruency between response codes, in highly similar task sets might be a source of age-related interference. This corroborates notions that older adults face increased difficulties in shifting and shielding multiple task sets or response selection mappings, leading to an increased response-code confusability.
Alternative explanations for asymmetric dual-task costs with response-code conflict
The observed asymmetry of the increase in dual-task costs for S-R compatible versus incompatible responses in trials with response-code conflict is counterintuitive and at odds with classic models of dual-tasking. In the following, we, therefore, discuss some alternative accounts of this finding.
Response selection bottleneck
One of the most influential accounts of the costs arising from dual-tasking is the response selection bottleneck model, which maintains that the central response selection stage cannot be dealt with in parallel when performing two tasks at once. Accordingly, response selection needs to be finished in Task A before it can start in Task B (Pashler,
1994). From this perspective, the overall higher dual-task costs in R-R incongruent (vs. congruent) trials and their asymmetry according to S-R compatibility would be explained by assuming a separate response selection process for either response, implemented in a serial fashion.
Drawing from the pertinent literature on the PRP effect, the response selection stage in auditory-manual two-choice tasks, for S-R mappings without incompatibility, can be estimated to take at least about 200 ms in young adults and roughly 100 ms more in older ones (Allen et al.,
2014; De Jong,
1995; see also Hein & Schubert,
2004, for comparable estimates from a similar two-choice visual discrimination task). Assuming that the queuing of the subtasks’ response selections is the only (relevant) source of dual-task costs beyond general dual-execution costs (R-R congruent conditions), it follows from the observed asymmetric cost pattern that the S-R incompatible subtask must have entered the response selection bottleneck first in most trials, delaying the S-R compatible response selection. This selection order is actually inconsistent with findings from PRP experiments showing that the less time-consuming (“easy”, e.g., S-R compatible) response selection is preferably processed before the more time-consuming (“difficult”, e.g., S-R incompatible) one (Leonhard et al.,
2011; Ruiz Fernández et al.,
2011). However, let us for now assume this was true in our setting with full temporal task overlap. In this case, the serial-selection model would predict that S-R incompatible responses suffer almost no additional costs from dual-tasking since their response selection would hardly ever be delayed, whereas S-R compatible responses should be strongly delayed by at least 250 or 350 ms in young or older adults, respectively.
5 Obviously, our results look different: On the one hand, there is a larger-than-expected delay for S-R incompatible responses, which amounts to 10 or 100 ms in young or older adults, respectively, as obtained by contrasting RTs of S-R incompatible trials between (serially processed) R-R incongruent conditions and (conjointly processed) congruent conditions (see Table S1). On the other hand, there is a much smaller delay than hypothetically predicted for S-R compatible responses (80 and 190 ms in young and old in our study). Together, this suggests that subtask order was different (i.e., S-R compatible responses were selected first) in a subset of trials (for details, refer to the last section of “Details on the analysis of response grouping” in the Appendix).
Based on the observed dual-task costs for S-R incompatible responses (resulting from trials where these responses were not selected first and therefore postponed, according to the bottleneck model) and the hypothetical costs that would be expected if those responses never came to be selected first, the ratio of trials in which S-R incompatible responses indeed were postponed can be estimated to be about 5 or 33% in young or older adults, respectively (ratio out of 10 or 100 ms observed costs vs. 200 or 300 ms theoretical costs in young or old, respectively). Accordingly, the expected delay of the S-R compatible response selection in the other 95 or 66% of trials would amount to about 140 or 240 ms (i.e., 95 or 66% of 250 or 350 ms) in young and older participants, respectively. These numbers are still way above the observed dual-task costs for S-R compatible responses in R-R incongruent trials (80 and 190 ms). This, as well as the atypical subtask order implied by the observed cost asymmetry, makes it very unlikely that a serial response selection strategy, subject to a structural bottleneck, is the mechanism behind the dual-task costs we observed in the condition with conflicting response codes.
Response grouping
A phenomenon often observed in dual-task paradigms, especially in conditions with two manual-response tasks presented very close together in time, is that participants adopt a strategy termed
response grouping (Pashler,
1994; Pashler & Johnston,
1989; Ulrich & Miller,
2008) or synchronized responding (Fagot & Pashler,
1992). Using this strategy, a participant would select the first response but hold it in, waiting until the second response is also ready to be executed, presumably because it is easier to emit two (manual) responses simultaneously than to emit them in rapid succession.
As our paradigm with easy-to-synchronize bimanual responses to a common stimulus was likely to promote the strategy of response grouping (without inviting it explicitly), it might be argued that the higher dual-task costs for S-R compatible (vs. incompatible) responses in R-R incongruent trials reflect this phenomenon. In particular, promptly executing the S-R compatible response after finishing its easy and fast selection could have been stalled by waiting for the more difficult and slower S-R incompatible selection to finish. To examine this possibility, we followed Miller and Ulrich’s (
2008) procedure to exclude trials with grouped responses and then repeated the statistical analysis. The assumption here was that if the effects obtained with all trials were still present after excluding grouped responses, it would indicate that participants not only engaged in response grouping but, at least in a subset of trials, suffered from asymmetric crosstalk causing differential dual-task interference. The results support this assumption. Thus, it appears that under conditions of response-code conflict, participants implement two different strategies that vary from trial to trial: response grouping and a (possibly strategically) biased resource allocation, both leading to a pattern in which the easier (S-R compatible) response suffers higher dual-task costs. This variability in strategy across trials is in line with findings by Miller and Ulrich (
2008), suggesting that the decision on whether or not to group responses on a given trial is made online. It remains for future research to answer the question for the key factors that influence this decision in settings like ours with full temporal overlap between two R-R incongruent tasks.
Future outlook and relevance
Our study implemented a very specific task combination, focusing on output-specific crosstalk using a single-stimulus onset paradigm. It would need to be tested how well our findings generalize to other more conventional task settings considering input–output modality compatibility effects. We suggest that future studies should assess the relevance of output-related features that interfere with the ability to perform two tasks simultaneously and the changes across age within the context of a content-dependent central interference model. For example, one could test which kinds of response-code conflict increase or decrease the probability of code confusability by manipulating the modalities of effector systems in combination with different single-input systems and controlling for general effector-based processing prioritizations (see Huestegge & Koch,
2013). Furthermore, given the study’s cross-sectional nature, it would be beneficial to use longitudinal designs to analyze developmental and environmental factors that contribute to the age-related dual-task deficits.
The present results also have several implications for product design targeted at older people. In our times, humans regularly use devices that require the simultaneous use of fingers and hands, such as display-control, medical, or navigation devices. As already pointed out by Proctor et al. (
2005), considering the age-related increase in response-code confusability and task-shielding difficulties might reduce errors and improve the usability for older adults, especially so in the context of multitasking. It would be particularly interesting to investigate under which specific intra- and cross-modal dual-task scenarios different strategies within the context of task-shielding and over-reliance on central attention are implemented.
Furthermore, fundamental dual-task crosstalk has been largely ignored in aging-related research, despite its potential relevance for detecting preclinical markers of cognitive decline in advanced age. Studying the interference between simultaneous cognitive and mobility-related processes in older adults has become an emerging research field, especially in the context of gait (Beurskens & Bock,
2012; Brustio et al.,
2017; Liebherr et al.,
2016; Porciuncula et al.,
2016; Smith et al.,
2016), postural control (Boisgontier et al.,
2013; Stelzel et al.,
2017), and upper-arm movement (Toosizadeh et al.,
2019). It seems that performing mobility or postural tasks becomes less automatic with age, negatively affecting secondary cognitive tasks. These observations have been attributed to a possible decrease in the ability to allocate and share attentional resources between mobility-/posture-related and cognitive tasks among the elderly (Boisgontier et al.,
2013; Brustio et al.,
2017). Identifying early changes within clinical settings could possibly predict frailty and disability and allow quick interventions to prevent a speeded disease progression or adverse outcomes (Brustio et al.,
2017; Ceïde et al.,
2018; Liebherr et al.,
2016).