30 volunteers aged 60–75 years participated in the study (17 female, 13 male; 29 right-handed, 1 left-handed; mean age 67.8 years, SD 4.1 years). All participants’ behavioral data were analyzed. Due to substantial artifacts in the EEG, two participants’ data were excluded from ERP data analysis (both participants were female; mean age of the remaining sample: 68.1 years, SD 4.1 years). The study was approved by the Ethics Committee of the University of Oldenburg. According to the Declaration of Helsinki, each participant gave written informed consent prior to the beginning of the experiment after all procedures had been explained. Participants received a modest financial compensation (8 €/h) for their participation.

Sounds were created with Matlab (R2012b) and the stimulus delivery was controlled using the Psychophysics Toolbox extension for Matlab (Psychtoolbox 3.0.10). Instructions, visual cues during the training phase and the movie were presented on a wall-mounted TFT monitor. A Soundblaster X-Fi Audio interface was used to generate the audio signals. It was connected to a Tucker-Davis attenuator in bypass mode, which in turn was connected to a Denon PMA 510AE amplifier. Sounds were delivered via a pair of Cambridge Audio S30 speakers, positioned approximately 1.5 m away from the participant on both sides of the TFT monitor in the experimental room. Participants sat comfortably inside an electrically and acoustically shielded chamber while performing the experimental tasks.

Before the main experiment started, participants completed a four-level training procedure with increasing difficulty. In level 1, only the task-relevant stream A was presented, and visual support was given (a white square indicating the occurrence of a deviant tone). In level 2, stream A was presented alone without visual aid. In level 3, both streams (A and B) were presented with visual support for the deviant tone in stream A. In level 4, both streams were presented without visual aid; the procedure was thus identical to the experimental blocks. The training blocks lasted 1 min each and could be repeated if necessary (level 4 was always repeated at least once). Training was finished when performance reached a stable level and the participant had notably understood the task. After EEG preparation, one additional training block (level 4) was presented to refresh the knowledge of the task.

The main experiment consisted of two parts with EEG recordings throughout. First, participants completed the behavioral experiment (stream segregation part). Second, they were presented with the stimuli of the regularity extraction part while watching the silent documentary. After removing the EEG cap, the pure-tone audiogram (via Siemens Unity II audiometer and Sennheiser HAD-200 headphones) was measured at octave frequencies between 125 Hz and 8 kHz for both ears. To measure speech-in-noise comprehension, the Oldenburg Sentence Test (OLSA,¹ Wagener et al., 1999b) was administered, using the adaptive procedure at a noise level of 65 dB SPL (presented with calibrated Siemens CD 310 F free field speakers). In addition, participants filled questionnaires on demographic variables, and they completed the Mehrfachwahl-Wortschatz-Intelligenztest (MWT-B, Lehrl, 1977) as a short screening for verbal intelligence.

The whole experimental session lasted between 2.5 and 3.5 h, including all tests and tasks, electrode application and removal as well as breaks for the participants.

EEG data were continuously recorded using a BrainAmp amplifier system (BrainProducts, Gilching, Germany) with passive Ag/AgCl electrodes from 96 scalp positions using an infracerebral electrode cap with an equidistant electrode layout (Easycap, Herrsching, Germany). The horizontal electrooculogram (EOG) was measured with electrodes placed at the outer canthi of the left and right eye. The vertical EOG was obtained from separate electrodes placed below the left and right eye and from two electrodes above the eyes that were inserted in the electrode cap. The reference electrode was placed at the tip of the nose. EEG and EOG signals were amplified and recorded with a sampling rate of 500 Hz, applying an analog filter with 250 Hz low pass and 0.0159 Hz high pass (time constant 10 s).

Average hearing loss (AHL) values across all frequencies ranged from 4.64 to 33.93 dB (Mean = 19.88 dB, SD = 9.14 dB), indicating wide variance from participants with almost normal thresholds to participants with considerable peripheral hearing loss. A significant correlation between AHL and age was observed (r = 0.43, p = 0.02, N = 30; see Figure 2A), indicating that peripheral hearing ability decreased with increasing age within our sample spanning 15 years of age. To exclude a confounding effect of hearing status on task performance, AHL in the frequency range in which the experimental stimuli were presented (370–554 Hz) was specifically examined. As a proxy, AHL values for 250 and 500 Hz were averaged, yielding a mean AHL of 5.42 dB (SD = 6.83 dB, range −7.5 to 20 dB). This indicates relatively preserved hearing at the frequencies of the auditory stimuli.

OLSA results ranged from −6.6 to −2.5 dB SNR (Mean = −5.36 dB SNR, SD = 0.93 dB SNR). This suggests a mild to moderate impairment in understanding speech in noise relative to the expected average threshold of young normal-hearing subjects at −7.1 dB SNR (Wagener et al., 1999a). Note that numerically lower—more negative—values indicate better performance. Speech-in-noise thresholds were significantly worse (i.e., less negative) with increasing age (r = 0.52, p < 0.01, N = 30; Figure 2B). Moreover, AHL correlated significantly with speech-in-noise comprehension (r = 0.49, p < 0.01, N = 30; Figure 2C).

All participants scored at or well above the population average on the verbal intelligence test MWT-B (verbal IQ range 97–145, mean IQ: 127), which rules out occurrences of major cognitive decline in the sample.

Performance as evaluated by the sensitivity index d’ scores ranged from −0.10 to 4.62 in the predictable condition (Mean = 2.69, SD = 1.52) and from 0 to 4.48 in the random condition (Mean = 2.35, SD = 1.33). The variation across participants was large (see Figure 3A), with some participants performing very well (note that performance without misses and false alarms would lead to a d’ maximum of 6.01) and other participants with d’ near zero failing to handle the target detection task above chance level. On average, performance in both conditions was significantly above chance [predictable condition: t(29) = 9.69, p < 0.001; random condition: t(29) = 9.66, p < 0.001].

Performance in the predictable condition was significantly better than in the random condition, t(29) = 3.39, p < 0.01. This implies that at group level, participants can benefit from the spectrotemporal regularity in the task-irrelevant background for performing the foreground task. At single-subject level, not all participants showed the same amount of benefit (or a benefit at all): Δd’ values ranged from −0.62 to 2.08 (Mean = 0.34, SD = 0.54).

Mean reaction times (see Figure 3B) to detected targets in the task-relevant stream did not differ between conditions [t(29) = −1.25, p = 0.22; predictable condition: Mean = 0.49 s, SD = 0.08 s; random condition: Mean = 0.50 s, SD = 0.08 s].

FDR-corrected running t-tests of the difference wave (deviant minus standard) at the frontocentral electrode position E02 show significant negative and positive deflections in several time ranges, of which the longest one was taken for further statistical analysis (see Figure 4A). It consists in a pronounced negativity in a late time range, starting from 428 ms after the onset of pattern-violating second tone (low-mid-mid instead of low-low-mid). The topography of this late component (measured from 428 to 496 ms) shows a frontocentral negativity with inversed polarity at the mastoids (Figures 4B,C), which is consistent with a generator of this component in auditory cortex (Näätänen et al., 2007). The negativity was identified as an MMN (see below for discussion). Individual MMN amplitudes at the frontocentral electrode location E02 in the time window 428–496 ms varied widely from −2.62 to 1.15 μV (Mean = −0.77 μV, SD = 0.87 μV; mean amplitude was significantly negative at group level: t(27) = −4.65, p < 0.001). No significant correlation between frontocentral MMN amplitude and age was found (r = −0.32, p = 0.10, N = 28, black dots in Figure 2D). Individual MMN amplitudes at common mastoids (CM) varied in the chosen time window from −1.03 to 2.40 μV (Mean = 0.47 μV, SD = 0.66 μV; mean amplitude was significantly positive at group level: t(27) = 3.78, p < 0.001). MMN amplitudes at common mastoids were found to significantly correlate with age (r = −0.54, p < 0.01, N = 28, green dots in Figure 2D), with less positive MMN amplitudes associated with increasing age. The correlations between MMN amplitude and age at E02 and CM do not qualitatively change when controlling for AHL.

Correlation analyses were carried out in search of underlying factors for the wide variation in the predictability-based performance benefit (Δd’). No significant correlation was observed between Δd’ (as a measure of predictability-based performance benefit) and MMN amplitude (as a measure of deviance detection capacities) at E02, r = 0.23, p = 0.25, N = 28 (Figure 5A, black dots) and CM, r = 0.22, p = 0.26, N = 28 (Figure 5A, green dots). There was also no correlation between benefit and AHL (r = −0.10, p = 0.61, N = 30, Figure 5B) nor between benefit and speech-in-noise comprehension (r = 0.08, p = 0.68, N = 30, Figure 5C).

Examining the correlation of benefit and age suggests that there could be a trend toward lower benefit with increasing age (r = −0.33, p = 0.07; Figure 5D), though it does not meet the conventional alpha level of 5% (and much less a Bonferroni-corrected alpha level of 1% to compensate for the five correlation coefficients computed for benefit). This numerical association of benefit and age could reflect a spurious trend, or it could indicate a real effect measured with too low power given the relatively small sample size (N = 30). To separate between these two possibilities, we retrieved the data of all 16 participants (mean age = 65.9 years, SD = 4.0 years) from the elderly group of Experiment 2 by Rimmele et al. (2012a). The current experimental design is highly similar to Experiment 2 of Rimmele et al. (2012a), thus their data were re-analyzed in terms of Δd’ benefit as in the current study by subtracting d’ in the random condition from d’ in their isochronous condition (corresponding to the predictable condition here). A joint correlation analysis based on data from both studies (Figure 6) revealed no significant correlation between age and benefit (r = −0.14, p = 0.34, N = 46), which suggests that the trend in the current experimental data was indeed spurious.

Importantly, Rimmele et al. (2012a) had found no significant benefit for elderly participants in the isochronous condition (Mean d’ = 2.55, SD = 0.70) relative to the random condition (Mean d’ = 2.41, SD = 0.74). In order to identify reasons for the apparent discrepancy with the current result, a post hoc independent-samples t-test was performed to compare the Δd’ results from Rimmele et al. (2012a) with the current data. Δd’ was numerically higher in the current study (Mean = 0.34, SD = 0.54) than in the previous study (Mean = 0.13, SD = 0.32), but this difference was not statistically significant [t(44) = 1.38, p = 0.17]. Again, a joint analysis of both datasets showed that Δd’ is significantly different from zero across both groups of elderly participants [Mean = 0.27, SD = 0.48, t(45) = 3.71, p < 0.001].

The lack of significant correlation of behavioral benefit (Δd’) with any other measured variable (Figure 5) led us to examine the robustness of Δd’ measurement, to rule out that measurement error underlies the absence of correlation. We assessed the robustness of the measurement by calculating Δd’ separately for each consecutive pair of blocks (i.e., subtracting d’ in the first random block from d’ in the first predictable block, etc.)² and determining Cronbach’s alpha of the three Δd’ estimates. Across our 30 listeners, Cronbach’s alpha was modest (0.31). Examination of the individual data showed that one single listener exhibited excessive variation between blocks³ whereas all others’ data were much more consistent. Excluding this one listener’s data increased Cronbach’s alpha to a moderate level of 0.58. Given that Δd’ is a difference score, 0.58 is an acceptable value, and it is considerably higher than the observed correlation of Δd’ with any other measured variable.

To study whether elderly listeners can extract spectrotemporal patterns, we used MMN as an indirect measure of regularity extraction (Schröger, 2005). The “low-mid-mid” deviations from the “low-low-mid” pattern indeed elicited a significant ERP component whose topography is consistent with the auditory MMN component (Näätänen et al., 2001, 2005). The time-range of the MMN component was relatively late, with its peak at 462 ms relative to the onset of the deviant event (second tone in the triplet). MMN usually occurs at about 100–250 ms after deviation onset (Näätänen et al., 2007; Garrido et al., 2009). One possible reason for the late MMN is that the three stimuli of the tone pattern are perceptually bound into an auditory object and that the comparison of the sensory input with the expected template (whose mismatch leads to MMN) takes place after the last tone of the triplet. Relative to this third tone, the observed peak latency is 179 ms, which is well in line with the usual MMN latency (Näätänen et al., 2007; Garrido et al., 2009). Alternatively, an embedded regularity may be encoded in different ways (Horváth et al., 2001), not only in the form of a global pattern regularity (“low-low-mid is continuously repeated”) but also in the form of a local transitional regularity (“a mid-tone is always followed by a low tone”). This transitional regularity is not violated until the third tone of the triplet, which explains the occurrence of a late MMN as well. Other alternatives are that the late negativity reflects the typical prolongation of MMN latency in elderly listeners (Cooper et al., 2006; Näätänen et al., 2011; Getzmann and Näätänen, 2015), or that this negativity is not the typical MMN but the late discriminative negativity (LDN) following MMN, which has been described mainly for children (e.g., Cheour et al., 2001), but also for adults when abstract regularities are employed (Zachau et al., 2005; Bendixen et al., 2012). Altogether, we conclude that the negativity peaking at 462 ms reflects an automatic violation detection process, in which the deviant triplet violates the formed prediction about the sensory input and the predictive model needs to be updated (Winkler and Czigler, 1998).

One may object that comparing ERPs to standard (low-low-mid) and deviant triplets (low-mid-mid) is not ideal due to the physical difference in the second tone (low vs. mid tone). In fact, a significant early difference between standard and deviant ERP traces emerged in the latency range of the N1 component at about 120 ms after onset of the physically different tone (see Figure 4A). To differentiate whether this N1 enhancement for the deviant triplet was due to the physical difference or reflects an early sign of “true” deviance detection (Näätänen et al., 2005), it would have been more advantageous to swap the role of deviant and standard triplets half-way during the passive listening condition. Yet although this has been a standard recommendation in auditory MMN research for many years, recent work on the so-called “primacy bias” (Todd et al., 2011, 2013, 2014; Fitzgerald and Todd, 2018) has established that swapping standard and deviant comes with significant reductions in the overall size of the MMN component. This is because a “lasting first impression” of stimulus probabilities and significance has been formed, reducing the MMN elicited by the deviant that was first experienced as a standard (Todd et al., 2011, 2013, 2014; Fitzgerald and Todd, 2018). Such higher-order effects (signifying predictive coding at different timescales) would reduce overall MMN amplitude and might introduce another source of inter-individual variability. This might have interacted with the aim to quantify pattern MMN on an individual basis in the current study. Importantly, even if the early (120 ms) negativity elicited by deviant relative to standard triplets was confounded by physical difference, it is highly unlikely that this translated to the negativity at 462 ms analyzed here, because the third tone in standard and deviant triplets was physically identical and the early modulation was much smaller than the late MMN.

To sum up, the ERP data from passive listening suggest that—at group level—elderly listeners extracted the pattern regularity and detected deviations from it. Finding this ability at group level is consistent with prior work on pattern regularity extraction in elderly listeners (Alain and Woods, 1999; Näätänen et al., 2011; Rimmele et al., 2012b; Getzmann and Näätänen, 2015). The respective abilities of the individual listeners (as quantified by MMN amplitude) showed a high amount of variation, which was later used to address the question of a possible direct relation with the ability of regularity-based stream segregation.

Note that evidence in favor of pattern regularity extraction in elderly listeners does not imply that this ability is fully preserved in elderly listeners: this conclusion would require comparison with a young-listener control group. Such a control group was not included in the current study because our focus was on examining differences within a group of elderly listeners (60–75 years), not on drawing comparisons across wide age ranges (∼25 years vs. ∼65 years, as in many other studies). Previous studies comparing widely different age groups have consistently found smaller frontocentral MMN amplitudes in elderly as compared to young listeners (Alain and Woods, 1999; Näätänen et al., 2011; Rimmele et al., 2012b; Getzmann and Näätänen, 2015). The fact that we did not find a correlation between frontocentral MMN amplitude and age in the current study (Figure 2D, black dots) does not contradict those prior observations, as one would expect the chronological age to play a more minor role in a sample spanning 15 years (60–75 years) than in samples spanning 40 years of age and more.

Indeed, the correlation between polarity-reversed MMN amplitude at the common mastoids and age was significant in our sample (Figure 2D, green dots). This might indicate that elderly listeners show impairments in the supratemporal but not frontal MMN generators for deviations of our pattern regularity (Giard et al., 1990; Opitz et al., 2002; Näätänen et al., 2007), or it may reflect a shift in the orientation of the superior temporal gyrus dipole (Deouell, 2007). Since our frontocentral MMN amplitude numerically shows an almost parallel trajectory with age as the mastoid MMN amplitude (Figure 2D: both tend to become less positive/more negative with age), the latter explanation seems more likely (see Baldeweg et al., 1999, for a similar observation where frontal MMN increase is paralleled by mastoid MMN decrease, though not in the context of aging). More specifically, it seems plausible that a change in orientation of the superior temporal gyrus dipole reduces MMN amplitudes at mastoid electrodes and increases MMN amplitudes at frontocentral electrodes at the same time. In contrast, a change in strength of the superior temporal gyrus dipole would reduce MMN amplitude at both electrode sites, and one would have to assume a parallel increase in strength of the frontal MMN source to account for the observation that MMN decreases with age at the mastoids while it tends to increase with age at frontocentral electrodes. While this would be a less parsimonious explanation, the current data should not be over-interpreted regarding this issue, especially since the association of age and frontocentral MMN amplitude was not significant by conventional criteria. It is likewise difficult to relate this observation to other MMN studies investigating aging effects by comparing young and elderly listeners, because many studies report frontocentral MMN amplitudes that are directly referenced to the mastoid electrodes, or do not report mastoid data at all. Further studies examining possible dissociations between temporal and frontal MMN generators with age are required, especially in the context of pattern regularities.

In the active-listening part of the current study, elderly listeners showed better foreground task performance when a task-irrelevant background stream carried a spectrotemporal regularity than when it did not. This is in line with the predictive-coding account of auditory stream segregation (Winkler et al., 2009, 2012). It is consistent with prior work on regularity-based stream segregation in young listeners (Andreou et al., 2011; Rimmele et al., 2012a), and informative with respect to previously conflicting results on whether this translates to elderly listeners: While de Kerangal et al. (2021) showed that the ability to track sources in an acoustic scene based on their regularities is largely preserved in elderly listeners, Rimmele et al. (2012a) had found an impairment of elderly listeners in regularity-based stream segregation. Since the current study closely followed the task and design of Rimmele et al. (2012a), we can now exclude task differences to underlie the different findings. The discrepancy of the current results with those of Rimmele et al. (2012a) is most likely due to issues of statistical power, with only 16 elderly listeners taking part in the previous study, compared to 30 elderly listeners in the current one. Alternatively, a procedural difference lies in the more comprehensive task training in the current study, from which the elderly may have benefitted. Post hoc analyses showed that the size of the behavioral benefit from regularity-based stream segregation in the two studies was not significantly different (though numerically higher in the current study), and that the behavioral benefit was significant when jointly analyzing both datasets. This underlines the necessity of replication studies with sufficient statistical power (Maxwell et al., 2015).

The fact that regularity-based stream segregation (i.e., behavioral benefit) does not correlate with peripheral hearing status (Figure 5B) nor with speech-in-noise comprehension (Figure 5C) is consistent with the findings from de Kerangal et al. (2021) who likewise found no such relations. This similar pattern of results in both studies is important on a theoretical level because it implies that the investigated regularity-based processing of complex acoustic scenes cannot trivially be explained by physical confounds in the stimulus setup (which might, e.g., put listeners with better peripheral hearing at an advantage). Instead, regularity-based processing of complex acoustic scenes appears to capture an independent ability, which is important to further investigate. The predictive-coding framework (Friston, 2005; Friston and Kiebel, 2009; Kanai et al., 2015; Denham and Winkler, 2017; Heilbron and Chait, 2018) provides an important theoretical basis for characterizing this ability, and in turn for developing a full understanding of auditory scene analysis and mitigating possible deficits thereof.

We conclude that elderly listeners can benefit from regularity-based stream segregation, but it is not known yet whether they can benefit to the same extent as young listeners. Answering the latter question would require taking measurements from a group of young listeners for comparison, which—as discussed for pattern regularity extraction above—was not in the focus of the current study. In view of the modified conclusion about the general ability of elderly listeners to perform regularity-based stream segregation, it seems warranted to verify the age group difference in this ability reported by Rimmele et al. (2012a) in an independent replication study. Similar to the ability for pattern regularity extraction, the ability for regularity-based stream segregation did not correlate with chronological age (Figures 5D, 6). Again, this does not rule out the existence of an age effect when comparing groups with a wider age range, as one would expect chronological age to dominate the results more when spanning a wider range. In any case, the observed strong inter-individual differences in the capacity to benefit from background regularity (both in the current and in the previous study) warrant further examination.

The key hypothesis of the current study was that listeners whose auditory system detects deviants more readily (as evidenced by MMN during passive listening) can also use predictability more easily for segregating sound sources from one another (as evidenced by regularity benefits in the active-listening task). This was expected to show up as a correlation between MMN amplitude and behavioral benefit. However, no such correlation was observed (Figure 5A). Since the inter-individual variation in both measures was high (see Figure 5A), the failure to find a correlation cannot be explained by floor or ceiling effects, or lack of variance. We must consider the possibility that part of the variation in the single-participant values reflects measurement error rather than true differences in the underlying ability. However, regarding MMN amplitudes, excessive measurement error seems unlikely given the significant correlation of individual mastoid MMN with age. Regarding behavioral benefit, robustness of the measurement was quantified by a Cronbach’s alpha of 0.58, which is considerably higher than the observed correlation of Δd’ benefit with any other measured variable.

To put potential concerns about measurement error further into perspective, it is important to note that well-established findings on auditory aging were replicated in the current study, even within the narrow age range (spanning 15 years) relative to studies comparing participants across different age groups (young vs. elderly). Specifically, we found that higher chronological age was accompanied by a decline in peripheral hearing ability (Figure 2A), which is consistent with prior work (e.g., Gates and Mills, 2005; Glyde et al., 2013). Increasing hearing thresholds challenge locating, detecting, discriminating, and comprehending sounds especially in complex acoustic environments, resulting in impaired speech comprehension (for a review see Pichora-Fuller and Souza, 2003; Martin and Jerger, 2005). In accordance with this, we confirmed that higher chronological age is accompanied by worse speech-in-noise comprehension (Figure 2B). The relation between peripheral hearing status and speech-in-noise comprehension, though significant, is far from being deterministic (Figure 2C). This underlines the partial independence of peripheral and central auditory processes (Plomp, 1978; Pichora-Fuller and Souza, 2003; Alain et al., 2006; Peelle and Wingfield, 2016). In any case, replication of these established findings in the current study rules out the possibility that severe measurement error masked inter-individual differences of relevant auditory abilities.

Other factors may help to understand the absence of a significant correlation between MMN amplitude and behavioral benefit. Before drawing premature conclusions on a putative dissociation between regularity extraction and regularity-based stream segregation, we should consider whether the two variables are valid indicators of the processes they are assumed to reflect. MMN is an indirect measure of regularity extraction (Schröger, 2005), though a dissociation between deviance detection and regularity extraction is unlikely under all current theoretical and modeling frameworks (Näätänen et al., 2007; Garrido et al., 2009; Fishman, 2014; Fitzgerald and Todd, 2020). A more imminent question is whether MMN amplitude gives a valid assessment of the strength or probability of deviance detection, or whether other processes may confound this measurement. Specifically, since we measured regularity extraction via MMN in a separate, passive-listening condition, MMN amplitude differences between participants may reflect different levels of attention to the sound stream while watching the silent documentary. Though attention effects on the MMN itself are small (Sussman, 2007), overlapping components such as the N2b (Novak et al., 1990) may confound the measurement and thereby artificially enhance the measured MMN amplitude. In fact, those participants who do not successfully disregard the irrelevant sound stream while focusing their attention on the video, might actually be those who also have trouble disregarding the task-irrelevant low-low-mid sound stream while focusing their attention on the task-relevant (A) stream. It may be the case that their negativity in the MMN latency range is counter-intuitively larger than those of many others because they fail to ignore the sound stream. This would be consistent with findings from cognitive aging in a wider sense, showing that the ability to inhibit task-irrelevant stimuli decreases with age across modalities and task types (e.g., Weeks and Hasher, 2018; Gaál et al., 2020). Interindividual differences in the ability to ignore task-irrelevant information (“resist interference”) would be one explanation why no consistent relation between MMN amplitude and benefit was found at group level. Future studies should mitigate this concern by measuring regularity extraction and deviant detection ability while participants’ auditory attention is more strongly controlled. It would also be advantageous to measure both abilities with the presence of the “A” sound stream.

Similarly, it could be questioned whether the behavioral benefit is a valid indicator of regularity-based stream segregation in every individual case. The underlying assumption is that more success in segregating the streams automatically translates into higher task performance. Yet this neglects the possibility that some participants may have trouble focusing their attention on the correct stream even when they succeed in segregating the streams (Gaeta et al., 2001; Healey et al., 2008; Horváth et al., 2009; Passow et al., 2012). Denham and Winkler (2017) summarize similarly conflicting results in their review article, stemming from the fact “that predictable sequences attract attention while also being easier to suppress.” They go on to conclude that “the presence of both tendencies may allow the influence of predictability to be easily modulated according to intrinsic preferences, attentional set and task demands” (Denham and Winkler, 2017). This consideration addresses contradictory findings across different studies, but it is well conceivable that it also applies within one study: In the current case, intrinsic preferences of individual participants may have obscured a systematic relation at group level. Therefore, an independent measurement of participants’ ability to focus on a given stream, and also to perform the task without the existence of a background stream, is needed to yield further insights into the involved processes. This would also tap more into related cognitive processes such as resistance toward interference, going beyond the very general cognitive screening in terms of verbal intelligence in the current study.