Selective attention reflects a set of cognitive processes that filters incoming information to maintain ongoing cognitive activity (Broadbent, 1954), allowing individuals to focus on what is important while ignoring irrelevant information. The load theory (LT) of selective attention, which has elicited a spirited debate in the literature, proposes that two fundamental mechanisms mediate selective attention. The first mechanism is perceptual, excluding distracters from perception when the level of perceptual load in processing task-relevant stimuli is high enough to exhaust perceptual capacity. This is a passive selection process that occurs because under high perceptual load, limited resources are available to process distracters. By contrast, in situations of low perceptual load, spare capacity not used to process task-pertinent stimuli will “spill over” to the processing of irrelevant distracters (Lavie and Tsal, 1994; Lavie, 1995; Lavie and Cox, 1997; Rees et al., 1997; Lavie et al., 2014). The second mechanism reflects “active” cognitive control that depends on higher cognitive functions, such as the executive control component of working memory (WM) (Lavie et al., 2004; Lavie, 2005). In contrast to perceptual load, distracter interference increases under high WM load. Carrying out the more demanding primary task is associated with augmented competition for the pool of limited executive control resources, resulting in the reduced ability of individuals to actively maintain current processing priorities that distinguish between relevant and irrelevant information (de Fockert et al., 2001; Lavie et al., 2004; Lavie and De Fockert, 2005; Rissman et al., 2009; Burnham, 2010; Kelley and Lavie, 2011).

Studies that have tested the cognitive load aspect of LT have yielded very inconsistent findings. For example, there is behavioral and neurophysiological data to suggest that increasing demands on the executive control component of WM lead to an increase in distracter effects, in keeping with the tenets of LT (Sreenivasan et al., 2007; Konstantinou et al., 2012, 2014; Konstantinou and Lavie, 2013; Roper and Vecera, 2014). In stark contrast, other studies have demonstrated a reduction in the processing of task-irrelevant distracters when demands on executive control are augmented, as indexed by decreased amplitudes of event related potentials (ERPs), such as N1, P3, and reorienting negativity (RON) (Berti and Schroger, 2003; Rose et al., 2005). These findings are not consistent with LT. Similarly, neuroimaging data suggest that increased cognitive demands (e.g., mental subtractions) are associated with a reduction of activity in visual areas related to the processing of distracters (Spinks et al., 2004).

Working memory load has also been investigated in cross-modal studies. It is worth noting that in many real world situations, individuals must contend with multimodal stimulation (e.g., reading a newspaper or driving while ignoring irrelevant sounds). The investigation of the impact of cognitive WM load on cross-modal processing has also produced mixed results. For instance, some studies suggest that cognitive WM load does not influence early processing of task-irrelevant auditory stimuli, as indexed by the amplitude of N1 and mismatch negativity (MMN) components (Otten et al., 2000; SanMiguel et al., 2008). Other investigations have reported that increased WM load is associated with greater amplitude of the N1-P2 complex and MMN in response to task-irrelevant auditory stimuli, as would be predicted by LT (Zhang et al., 2006; Regenbogen et al., 2012). However, there is a body of evidence that has yielded the opposite findings, indicating that increasing WM load reduces the processing of task-irrelevant distracters in the auditory modality. For example, increasing WM load has been associated with a reduction of brainstem evoked responses to task-irrelevant sounds (Sörqvist et al., 2012a) and a decrease in the amplitude of different auditory ERPs, such as the MMN, P3, and RON (Munka and Berti, 2006; SanMiguel et al., 2008; Lv et al., 2010). Similarly, fMRI studies have found that under higher WM load, there is diminished activation in areas associated with processing of irrelevant stimuli presented in a different modality (Klemen et al., 2010; Regenbogen et al., 2012; Sörqvist et al., 2016).

Several hypotheses have been developed that may help account for the variability of findings in investigations of LT. For example, the absence of a clear operational definition of load in various studies could contribute to contradictory results (Benoni and Tsal, 2013). In many cases, increasing task demands may be associated with an augmentation of both perceptual and cognitive load, which according to LT would influence behavior and processing in opposing ways. In addition, because WM is not a unitary process (Smith and Jonides, 1999; Baddeley, 2012), manipulating the load on short-term maintenance operations may have a different impact on the processing of distracters than manipulating the demands on executive control operations, with the former being similar to augmenting perceptual load (Pasternak and Greenlee, 2005; Sreenivasan et al., 2007; Konstantinou and Lavie, 2013; Konstantinou et al., 2014; Roper and Vecera, 2014). Some research suggests that the impact of WM load on selective attention depends on the extent to which the contents of the WM task overlap with targets or distracters in a selective attention task (Yi et al., 2004; Kim et al., 2005; Park et al., 2007). Differences between experimental tasks also may contribute to conflicting evidence for LT. For example, according to Sörqvist et al. (2012b), when WM load is manipulated in a single-task paradigm (e.g., n-back task), processing of irrelevant information decreases under high load. However, processing of irrelevant information increases under high load when WM load is manipulated in a dual-task situation, such as maintaining items in WM while performing an unrelated task (see de Fockert et al., 2001; Lavie et al., 2004).

The current study focused on another potential source of the disparate findings associated with LT by considering the potential impact of individual differences in the executive capacity (EC) of subjects. Most investigations have not measured the WM capacity of subjects, nor accounted for the ways in which this factor may affect the relative difficulty of an experiment under different task loads. There is evidence that the same experimental task-load will be associated with different levels of difficulty depending on a subject’s WM capacity (de Fockert, 2013; Sörqvist and Rönnberg, 2014; Murphy et al., 2016). High WM capacity has been linked with a more steadfast locus of attention and less susceptibility to auditory distraction (Beaman, 2004; Sörqvist et al., 2012a,b; Hughes et al., 2013; Sörqvist and Rönnberg, 2014), a framework that we believe would benefit from further investigation using neurophysiological variables (e.g., ERP, fMRI).

Because predictions derived from LT are very dependent on the extent to which increased load taxes the WM capacity of subjects, thereby limiting available resources to maintain processing priorities, it is critical to attend to each subject’s individual capacity. Most studies in the LT field include small samples, which increases the likelihood of uneven distributions of the WM capacity of subjects across different investigations. For example, if a study were largely composed of subjects with low WM capacity, increasing task load for such individuals may severely strain available resources, making it very difficult for them to exert control over the processing of task-irrelevant distracters. In contrast, if another study mainly included subjects with high WM capacity, the same increase in task load would not deplete their resources. Rather, the more challenging task may further engage the attention of such subjects, resulting in diminished processing of distracters.

Sörqvist and Rönnberg’s recently developed the “neuro cognitive task-engagement/distraction trade-off” (TEDTOFF) model of WM capacity and cross-modal auditory distraction (Sörqvist and Rönnberg, 2014), which can help to scaffold the current study. The model assumes that task difficulty (e.g., load) and WM capacity (i.e., individual differences in the ability to control attention and filter task-irrelevant information) are critical factors that mediate distractibility. Within this framework, higher task difficulty leads to “protection” from distraction by increasing focal-task engagement (i.e., facilitating attention to the attended stimulus) and more actively suppressing the processing of irrelevant stimuli (Sörqvist and Marsh, 2015; Sörqvist et al., 2016). WM capacity is understood as reflecting individual differences in the ability of executive control operations to enhance the focus of attention on task-relevant stimuli and gate the processing of task-irrelevant stimuli. In contrast to LT, the TEDTOFF model considers the difficulty of the task to be more critical than the type of load manipulated (i.e., perceptual or cognitive) (Sörqvist and Rönnberg, 2014).

The aim of the present study was to investigate whether WM load modulates early distracter processing in a cross-modal paradigm. Subjects were engaged in a visual WM task (using an oddball paradigm) while instructed to ignore irrelevant auditory stimuli. Within this experimental context, we addressed some of the methodological limitations in the LT literature on cross-modal selective attention by controlling for the potential confounding effects of task difficulty and WM capacity. First, the EC of subjects, which incorporates indices of WM proficiency, was carefully measured using standard neuropsychological tests. This method has the advantage of relying on well-established norms to characterize the EC of our sample and facilitate comparisons with subjects who participate in future studies. Second, we titrated the level of difficulty of the high load condition so that it would have a relatively uniform impact on all participants. The low load condition was the same for all subjects, but in the high load condition, the number of visual targets was determined individually for each participant to keep task performance consistent across subjects (i.e., accuracy rate of ∼80%). In accordance with our framework, we anticipated that subjects with higher EC would have to be given a larger number of target stimuli to hold in WM to make their performance comparable to that of subjects with lower EC. An important measure of the success of our titration process would be finding that under the high WM load condition, performance on the primary task did not vary as a function of EC or the number of target letters presented to subjects. To the best of our knowledge, no prior study of LT has controlled for variation in subjective difficulty of the primary task across participants. This approach allowed us to address the impact of increased cognitive load of the visual task on the processing of task-irrelevant auditory stimuli in a manner that limited the influence of differences in subjective task difficulty and EC across subjects.

ERPs were used to measure neural activity in response to stimuli presented in both the task relevant modality (visual) and task-irrelevant modality (auditory). ERPs in response to stimuli from the primary task provided neural indices of attentional engagement and resource utilization under different load conditions. The posterior slow wave (SW) served as an index of sustained attention and effort (Ruchkin et al., 1980; Gevins et al., 1996; Rushby et al., 2005; Daffner et al., 2011). We anticipated that in response to visual targets, the posterior SW would increase under the high load condition (Ruchkin et al., 1980; Ruchkin and Sutton, 1983). Also, the P3b component was measured as an index of the process of categorizing an event (Squires et al., 1973; Ford, 1978; Kok, 2001) or updating memory after an event has been categorized (Donchin, 1981; Donchin and Coles, 1988). Consistent with many reports in the literature, we expected a decline in the visual P3 amplitude under the high load condition (Mecklinger et al., 1992; Ruchkin et al., 1992; Gevins et al., 1996). It has been hypothesized that resources under the low load condition had been devoted to decision-making/updating are “reallocated” under the high load condition to the operations of maintenance, manipulation, and sustaining attention, leading to a decline in P3 amplitude (McEvoy et al., 1998; Watter et al., 2001; Daffner et al., 2011). The auditory N1 component, a frontocentral negativity peaking 50–150 ms after stimulus onset, served as a measure of early perceptual processing of task-irrelevant auditory distracters (Woods, 1995). Our primary test of LT involved determining the impact of visual WM load on the amplitude of the auditory N1. If we found that WM load of the primary visual task had no effect on the N1 response to task-irrelevant auditory stimuli, it would be consistent with the idea of separate pools of attentional resources within each sensory modality that do not interfere with each other (Otten et al., 2000). In contrast, finding that visual WM load modulates N1 response in the unattended auditory modality would suggest a shared reservoir of resources, with potential for interference between modalities. Observing that increasing WM load in the primary task is associated with greater processing of distracter stimuli (as indexed by an enhanced auditory N1) would be consistent with the classic LT of cognitive control. In contrast, if we found that augmenting WM load is coupled with an attenuation of the auditory N1, our results would be a challenge to LT. Such findings would be more consistent with the TEDTOFF model that suggests that higher visual WM load increases attentional engagement and “protects” from distraction.

The present study employed a cross-modal paradigm to investigate the impact of WM load on distracter processing, using ERPs as critical dependent variables. To address a potential source of inconsistent findings in the LT literature, the study aimed to control for the confounding effects of task difficulty and EC across subjects. Well-normed, standardized neuropsychological tests rather than experimental tasks were used to define EC, which may allow for greater generalization of results. The WM load manipulation of the primary visual task was effective: relative to the low load condition, the high load condition was associated with lower accuracy and longer RTs. The success of the titration process in controlling the level of task difficulty across subjects was demonstrated by finding that under high WM load, behavioral performance on the primary task was not modulated by EC group. As expected, to achieve this goal, under the more demanding load condition, subjects with higher EC needed to be given a significantly greater number of target stimuli to make their performance comparable to that of subjects with lower EC. Similarly, the ERP indices of processing task-irrelevant stimuli were not modulated by EC group. A trend was noted in the average waveforms suggesting a greater load-related augmentation of posterior SW activity to visual targets for the high than average EC group. This finding may indicate that in response to a similar level of task difficulty, subjects with higher EC are able to mobilize more attentional resources, which would be consistent with the view that higher WM capacity is associated with a more steadfast locus of attention (Sörqvist and Rönnberg, 2014; Sörqvist and Marsh, 2015).

The N1 component served as a marker of early processing of task-irrelevant auditory distracters. The tenants of LT lead to the prediction that augmenting WM load of the primary visual task would result in an increase in the N1 amplitude to auditory distracters because competition for limited EC resources reduce an individual’s ability to sustain current processing priorities. The main findings of the experiment are not consistent with LT. Increasing WM load was linked to a decrease, not an increase, in N1 amplitude in response to task-irrelevant auditory distracters. Our results are consistent with previous findings indicating that augmentation of WM load is associated with a reduction in the magnitude of auditory-evoked brainstem potentials to irrelevant sounds, a marker of the very earliest processing of auditory distracters (Sörqvist et al., 2012b). In line with our findings, other groups have also reported reduced distracter processing under high compared to low WM load in cross-modal paradigms (Munka and Berti, 2006; SanMiguel et al., 2008; Klemen et al., 2010; Regenbogen et al., 2012; Sörqvist et al., 2012a, 2016). Importantly, this pattern of findings was upheld after controlling for the potential influence of differences in subjective task difficulty and EC across participants.

The results of the current study are in line with the TEDTOFF model, which states that augmenting task difficulty (i.e., WM load) is associated with increased attentional engagement and “protection” from distraction (Sörqvist and Rönnberg, 2014; Sörqvist, 2015, Sörqvist et al., 2016). Our ERP data offer additional support for this theory by providing a non-behavioral measure of increased “steadfastness” of the locus of attention. Our results demonstrated that the amplitude of the posterior SW increased under the high load condition, consistent with previous work (Ruchkin et al., 1980; Ruchkin and Sutton, 1983; Daffner et al., 2011). There is evidence that the posterior SW is an index of sustained attention and effort processes that likely underlie the steadfastness of attention (Ruchkin et al., 1980; Gevins et al., 1996; Rushby et al., 2005). Correlation analyses indicated that when WM load is augmented, the greater the increase in attention to the primary task (as measured by the posterior SW), the greater the reduction of the processing of task-irrelevant auditory standards (as measured by the N1). Moreover, as WM load increases, a smaller decline in target accuracy is associated with ERP measures of enhanced attention to the primary task (indexed by the posterior SW) and reduced processing of irrelevant stimuli (indexed by the N1).

One concern is that the load–related increase in posterior SW to visual targets directly influences the measurement of the subsequent N1 to auditory standards, reflecting the same effect that is time-locked to different experimental stimuli. This account could explain why the load–related differences in the average waveforms to auditory standards appear to begin prior to the N1 itself. There are several reasons why we believe this explanation to be unlikely. First, the interstimulus interval (ISI) between visual and auditory events was jittered between 315 and 665 ms, reducing the likelihood that electrophysiologic responses to auditory events were time-locked to visual events (Luck, 2005). Second, there was limited overlap in the spatial distribution of the posterior SW to visual targets and N1 to auditory standards, which would be necessary to directly link the measurement of one component to that of the other. Also, when the SW was re-measured at the Fz cluster (where the N1 was measured) no load effect was observed. Moreover, if the posterior SW and N1 reflected the same experimental effect, one would anticipate a close correlation between the two components within each load, but none was found. Finally, the pattern of results was confirmed using PCA, a method that can parse variance due to the N1 factor from that reflecting electrophysiologic activity that precedes or follows it (Dien, 2012). To further address this issue, future studies should include trials with longer ISIs to eliminate the possibility of temporal overlap between electrophysiologic responses to visual and auditory stimuli.

The dissociation found between the pattern of response for the posterior SW and the P3b component was predicted. In contrast to the SW, the amplitude of the P3b diminished under the high load condition, which is consistent with many other reports (Mecklinger et al., 1992; Ruchkin et al., 1992; Gevins et al., 1996). Under high load, resources that had been allocated to decision-making/updating may be utilized for other cognitive operations like sustaining attention, leading to greater uncertainty about target identity and a smaller P3b amplitude (DeSwart et al., 1981; Johnson, 1986; McEvoy et al., 1998; Watter et al., 2001; Daffner et al., 2011). Another potential contribution to load-related reduction of the size of P3b might be increased trial-to-trial variability in the peak latency of the P3b due to greater task difficulty. We tried to mitigate this potential confounder by measuring mean, and not peak, amplitude for the P3b, as previously recommended (Luck, 2005).

There are limitations to the current study, which focused on controlling for task difficulty and EC. It did not manipulate variables that have been shown to impact distracter processing using within-modality paradigms, such as which component of WM load is varied (e.g., maintenance vs. control), or the degree of overlap between the contents held in WM and the features of target or distracter stimuli (Kim et al., 2005; Park et al., 2007; Konstantinou et al., 2014), both factors that could be addressed in future research. Although our sample size was at least as large as many in the literature, it was nonetheless relatively modest and thus the findings need to be replicated. The inability to find a main effect of EC group or an interaction between EC group and load for our experimental variables could be the result of insufficient power. It is worth noting that almost none of the p-values from these analyses even approached a trend. Our analyses of the early processing of task-irrelevant auditory stimuli concentrated on standard stimuli, which is common to investigations that have relied on the N1 component (Näätänen and Picton, 1987; Munte et al., 1998; Schroder et al., 2014). Additional work is needed to determine whether a similar pattern of response would be observed for more salient events.

It would also be informative for future studies to include more challenging levels of WM load. We strongly suspect that the relationship between task difficulty and protection from distraction may follow an inverted U–shaped curve. If demands are sufficiently augmented, capacity–limited control resources would eventually be depleted, leading to an increase in the processing of task–irrelevant distracters. The theoretical curves of high EC individuals may be ‘shifted to the right’. If so, it would require much more demanding conditions for high EC adults to exhaust resources that mediate top–down control, resulting in an increase in distractibility. If this framework were correct, the TEDTOFF model, which is supported by the current findings, would require further modification. Finally, although we suspect that the correlation between the load-related increase in posterior SW to visual targets and the load-related decrease in N1 to auditory distracters is the result of active, top–down process and not a passive one (i.e., exhaustion of resources), additional research is necessary to confirm this hypothesis.

In summary, an increase in WM load was associated with electrophysiological evidence of an increased steadfastness of attention on the primary visual task and a reduction in the early processing of auditory distracters. These findings occurred within the context of controlling for the confounding effects of task difficulty and WM capacity across subjects, an approach that future studies should take into consideration. The results of this study present a challenge to LT, which may need to be modified.

KD and PH designed the experiment. ET analyzed the data and prepared the figures. KD, PH, ET, and SS interpreted the results of the experiment. SS drafted the manuscript. KD, PH, ET, and SS edited and revised the manuscript. KD, PH, ET, and SS approved the final version of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.