Early Selection versus Late Correction: Age-Related Differences in Controlling Working Memory Contents

Abstract

We examined whether a reduced ability to ignore irrelevant information is responsible for the age-related decline of working-memory (WM) functions. By means of event-related brain potentials we will show that filtering is not out of service in older adults but shifted to a later processing stage. Participants performed a visual short-term memory task (change-detection task) in which targets were presented along with distractors. To allow early selection, a cue was presented in advance of each display, indicating where the targets were to appear. Despite this relatively easy selection criterion, older adults’ filtering was delayed as indicated by the amplitude pattern of the contralateral delay activity. Importantly, WM-equated younger adults did not show a delay indicating that the delay is specific to older adults and not a general phenomenon that comes with low WM capacity. Moreover, the analysis of early visual potentials revealed qualitatively different perceptual/attentional processing between the age groups. Young adults exhibited stronger distractor sensitivity that in turn facilitated filtering. Older adults, in contrast, seemed to initially store distractors and to suppress them after the fact. These early-selection versus late-correction modes suggest an age-related shift in the strategy to control the contents of WM.

Performance in tasks that require the temporary storage and manipulation of information declines with age (e.g., Bopp & Verhaeghen, 2005; Cowan, Naveh-Benjamin, Kilb, & Saults, 2006). One possible reason for these changes is that the available storage space in working memory (WM, Baddeley & Hitch, 1974) diminishes with age. However, another possibility is that, due to age differences in executive processes that control access to WM (e.g., Kane & Engle, 2002), the available storage space is less efficiently used. In particular, older adults may be less able to prevent irrelevant information from consuming capacity (e.g., Hasher & Zacks, 1988; Hasher, Zacks, & May, 1999).

One of the most compelling pieces of evidence in favor of such a filtering/inhibition account of the age-related WM decline comes from an fMRI study by Gazzaley, Cooney, Rissman and D'Esposito (2005). In a delayed recognition task, in which a sequence of targets and distractors was presented, older adults showed a specific deficit in preventing distracting information from being processed. Whereas young adults were able to differentially suppress neural activity in those sensory cortical regions that are related to the processing of the currently irrelevant information, older adults were not capable of doing so.

While the Gazzaley et al. approach allows the assessment of top-down modulations of perceptual processing, it provides no information about the fate of relevant and irrelevant information across the delay interval. In the past decade, the so-called contralateral delay activity (CDA; Vogel & Machizawa, 2004) of the EEG has been proven to be a useful tool in this regard (e.g., Vogel, McCollough, & Machizawa, 2005; Jost, Bryck, Vogel, & Mayr, 2011; Jost & Mayr, 2015 online first; Sander, Werkle-Bergner, & Lindenberger, 2011a; Spronk, Vogel, & Jonkman, 2012; Störmer, Li, Heekeren, & Lindenberger, 2013). The CDA is a sustained negative wave measured over the posterior cortex. Its amplitude increases as the number of items represented in visual WM increases, reaching an asymptotic limit at each individual's WM capacity (Vogel & Machizawa, 2004). This provides evidence that the CDA is a measure of the number of objects represented in visual WM. Moreover, Vogel, McCollough et al. (2005) showed that the response of the CDA to the presence of irrelevant distractors could be used as an indicator of filtering efficiency that served as a strong predictor of individual WM capacity.

Jost et al. (2011) measured the CDA while old and young adults had to store red items and ignore blue and green ones. Generally, the amplitude of the CDA not only increased with the number of targets, but also when distractors were presented, indicating that filtering was not perfect and that distractors were stored to some extent. Moreover, older adults showed smaller filtering scores than young adults, but only early in the retention interval indicating an early filtering deficit (for related results, see Gazzaley et al., 2008) that completely disappeared for the second half of the retention interval. In contrast, age-independent individual differences in filtering were found throughout the delay interval.

One possible reason for this observed pattern of age differences is that old adults’ problem is not so much with filtering per se. Rather, it may simply take them longer to distinguish between relevant and irrelevant information. In this case, the observed filtering delay should disappear when an easy selection criterion is used. Therefore, one goal of the current study was to examine to what degree the age-related delay in filtering can be replicated with an easier selection criterion. We, therefore, used a design similar to Jost et al. (2011), but targets and distracting items were now presented in different locations (for a similar procedure, see Experiment 2 in Vogel, McCollough et al., 2005). In comparison with other selection attributes, location seems to provide a more basic-level, earlier selection criterion (e.g., Anllo-Vento & Hillyard, 1996). To allow location-based selection, a cue was presented in advance of each display and indicated where the targets were to appear, allowing individuals to shift attention to the target location before targets and distractors were actually presented. There is evidence from other tasks showing that age differences in interference from distracting information are largely reduced when distracting and target information is presented in separate locations (Hartley, 1993) or when the position of the target is known in advance (Plude & Hoyer, 1986). This suggests that distracting information from a non-attended location should be much easier to ignore even by older adults than a distracting feature within an attended location.

If a deficit in distinguishing between targets and distractors was responsible for the age-related delay in filtering observed by Jost et al. (2011), then older adults should benefit from an easier selection criterion and age-related differences in the time course of filtering should disappear. However, should we continue to find evidence for different time courses, the age-related filtering delay can be regarded as a more general phenomenon that must be due to reasons other than a difficult selection criterion.

Another potential explanation of the observation by Jost et al. (2011) that age effects in filtering were restricted to the first half of the retention interval and diminished at the end is that older adults used a different, later filter mechanism (as suggested by Gazzaley et al., 2008, see also Gazzaley, 2013) to remove distractors from WM. Specifically, older adults may rely on “late correction” instead of “early selection” (Braver, 2012). Thus, they initially encode distractors in WM and suppress them (according to the task rules) after the fact during the retention interval. Younger adults, in contrast, use the task goals during stimulus encoding to optimally bias processing towards task-relevant features (e.g., Desimone & Duncan, 1995). In the present study, we explore the possibility that these functionally different modes to control the contents of WM are responsible for the age-related variations in the time courses of filtering as observed in the study by Jost et al. (2011). If older adults do not rely on early selection but on a later and thus functionally different control process, then they should differ from younger adults already during early processing of targets and distractors.

How can we identify potential age differences in early selection? Different from the Jost et al. (2011) paradigm, the location-based selection criterion yields highly salient distractors, which, as will be shown below, modulate the amplitudes of the N1 and the N2pc. The N1 is associated with the early perceptual processing of visual stimuli (e.g., Heinze, Luck, Mangun, & Hillyard, 1990; Luck, 2014). Moreover, it was found that the N1 is sensitive to the allocation of attention (Mangun, 1995; for a review) as well as discrimination between different types of stimuli (Mangun & Hillyard, 1991; Vogel & Luck, 2000). The second component, the N2pc (posterior contralateral N2), is assumed to reflect the deployment of spatial attention (e.g., Luck, 2014; Luck & Hillyard, 1994). It is usually observed as a more negative voltage at contralateral than at ipsilateral scalp sites relative to the attended position of items in a bilateral display. More recently it has been suggested that it also constitutes an index of item selection and individuation within the attended hemifield (for review, see Anderson, Vogel, & Awh, 2014; Mazza & Caramazza, 2015). However, evidence in the context of WM tasks is rare, since WM-CDA studies by and large neglected N2pc effects, because of an overlap with the CDA. Typically, this also holds for the studies that used the filtering paradigm (see e.g., Jost et al., 2011; Vogel, McCollough et al., 2005; but see Störmer et al., 2013).

If both younger and older adults rely on early selection, but older adults are just less efficient or take longer, then age-related differences in these two components should emerge in terms of quantitative differences such as attenuated amplitudes or latency effects. However, if older adults do not rely on early selection but on late correction, we would expect to find qualitative differences between young and old adults with less differentiation between targets and distractors in old adults’ early ERP components (N1 and N2pc) than in those of young adults.

Method

Participants

Twenty-nine students (age-range: 19-27 years) and thirty-three older adults (age-range: 63-78 years) took part in the study. Participants were paid for their participation or received course credits. All participants reported having normal or corrected to normal vision. Younger and older participants were comparable with regard to their educational level. Younger participants were students and the older participants had a professional education or higher educational qualifications. The older participants were screened for mild cognitive impairment or dementia using the DemTect (Kalbe et al., 2004). All older adults reached values that were normal for their age (M = 17.42 points, SD = 1.06, range = 15-18 points, maximum value of the test: 18 points).

Four younger and nine older participants had to be excluded from further analysis because of excessive eye movements (in more than 50% of the trials). The effective sample comprised twenty-five younger (M age = 22.40 years, SD = 2.38, 16 female) and twenty-four older adults (M age = 68.21 years, SD = 3.19, 12 female).

Stimuli, Task, and Procedure

WM and filtering performance were measured by means of the change-detection task (Phillips, 1974; Luck & Vogel, 1997), i.e., a short-term memory task in which a varying number of visual stimuli is presented for a brief interval of 200 ms. After a retention interval of about 1 sec, memory for the stored items is tested. Given that the capacity of visual WM is on average about three to four items (Luck & Vogel, 1997), increasing the number of items in the memory display will lead to an overload and hence a decrease of WM performance.

The experiment consisted of two parts. In the first part participants performed a standard behavioral version of the change-detection task to estimate visual WM capacity. The second part served to measure filtering efficiency. Participants performed a filter version of the change-detection task (Vogel, McCollough et al., 2005) where they had to filter out irrelevant information on basis of the spatial location, meanwhile ERPs were recorded.

Estimation of Visual Working-Memory Capacity

On each trial a varying number of colored squares was presented for 200 ms. The task was to maintain the colors of the squares over a retention interval of 900 ms. Subsequently a single probe test array (i.e., only one item was tested) was presented and the participants had to decide whether this item has changed in color or not compared to the item at the same position in the memory array. The test array remained until the participant responded. Responses were unspeeded and accuracy was stressed.

Number of squares in the memory array (set size) was either 2, 4, 6, or 8. The squares had a size of 0.77° × 0.77° and appeared within a centered area of 7.17° × 7.17° visual angle (viewing distance of 70 cm). They were presented on a gray background with a fixation cross in the middle of the screen. Colors were randomly selected (either red, blue, green, black, yellow, or purple) with the constraint that the same color could appear at most twice in an array. In 50% of the trials the color changed between the memory and test arrays and the new color value was selected randomly from all of the other values. Participants were instructed to press the right button of a controller when the color had changed and the left button when the color stayed the same.

Trials were presented in 3 blocks of 80 trials each. In each block the number of trials for each set size was the same and half of the trials in each set size were change and the other half no-change trials (i.e., 10 trials).

Filtering Task

As in the Jost et al. (2011) study stimuli were red bars that differed in orientation (0°, 45°, 90° and 135°). Target and distractors were presented at separate spatial locations. The target items were presented either in the upper or lower half of the display; the relevant location was indicated with an arrow presented in advance allowing for early selection (see Figure 1A). In trials where distracting items were presented additionally, they were presented in the half above or below the target items. Otherwise this half remained empty.

Figure 1.

A. Stimulus sequence of a trial. Shown is an example of a distractor-present trial. As indicated by the arrow presented in advance, items presented in the upper right location should be stored. The two distracting items in the lower right location are to be ignored. To keep visual stimulation constant, the same amount of information was presented in the left hemifield. Note that in the test array the orientation of the target item has changed. B. Examples of memory arrays for the experimental conditions, when the upper right location is relevant. Number of targets (either one or three) and distractor presence were manipulated orthogonally. C. Contralateral and ipsilateral waveforms time locked to the onset of the memory array exemplarily for the set size 3 condition (for all conditions see Supplementary Figure 1). Negative voltage is plotted upward.

We manipulated the number of targets, which could be either one (set size 1) or three (set size 3) and we manipulated whether the targets were presented alone or with two distracting items. This results in four different conditions, i.e., set size 1, set size 3, set size 1 + distractors, and set size 3 + distractors (similar to Jost et al., 2011) and an equal number of non-distractor and distractor trials.

The memory array was bilateral, that means on both sides of the fixation cross a whole memory array was presented, but participants had to attend only to one hemifield. This bilateral presentation allows isolating activation related to encoding and maintenance from unspecific activation (see Gratton, 1998, for a description of this rationale, see also the Supplementary Material). This is important especially for the CDA (see Vogel & Machizawa, 2004).

Therefore, stimuli (with a size of 0.39° × 1.33° visual angle) appeared in four different spatial locations around the fixation cross (top right, bottom right, top left, and bottom left quadrant). Each location had a size of 3.45° × 3.45°, a 1.23° distance from the x-axis, and a 1.72° distance from the y-axis (thus, with a viewing distance of 70 cm, the complete display was 10.34° × 9.36°). The arrow in advance indicated not only the relevant location (top or bottom) but also the relevant hemifield (left or right). The arrow had a length of 1.23° and was presented radially with a 0.74° distance from the fixation cross, so that it pointed toward the relevant location (see Figure 1). Orientation of the bars were randomly selected with the restriction that the orientations within a quadrant always differed from each other. This holds for the memory as well as for the test arrays.

At the beginning of each trial a fixation cross was presented on a gray background for 500 ms followed by the arrow that indicated the target position for 200 ms. After a variable interval of 200 to 400 ms the bilateral memory array was presented for 200 ms. The retention interval was 900 ms. Memory for targets was tested with a full-probe test array (see Figure 1A). In 50% of the trials, the orientation of one of the targets has changed and participants were asked to indicate this change with a button press (right button for change and left button for non-change). Note that the distracting items always stayed the same. The test array remained visible until the response. The inter-trial interval was 2000 ms.

Participants performed 14 blocks of 65 trials each. The first trial of each block was a filler trial and discarded from analysis. Within each block the number of trials was the same (i.e., 2) for all possible combinations of experimental conditions, change versus no-change trials, attended hemifield, and target location. Every block was followed by a break, which the participant could finish with a button press. Participants were instructed to ignore the distracting information as good as possible and to respond as accurate as possible. Moreover, to minimize eye movements, participants were requested to fixate the fixation cross until the end of a trial.

EEG Recording and Preprocessing

ERPs were recorded with Ag/AgCl electrodes inserted in an elastic cap (EASYCAP, Munich, Germany) with positions extrapolated from the International 10-20 system. Horizontal eye movements were measured with electrodes on the outer canthi of the eyes and blinks with one electrode below the right eye. An electrode on the nose tip served as reference and an electrode on the right mastoid was the ground. The impedances for all electrodes were kept below 10kΩ. Signals were recorded with two 32-channel DC amplifiers (Brain Amps, Brain Products, Munich, Germany), sampled at 500 Hz, and low-pass filtered with 250 Hz.

Data preprocessing and ERP averaging were run with Brain Vision Analyzer software. Signals were filtered offline with a band-pass of 0.1-30 Hz (24 dB/oct) and a 50 Hz notch-filter. Trials with eye movements, blinks and other artifacts (e.g., muscle activities) were excluded from further analyses. Mean number of remaining trials per condition was 149 for younger adults (min = 85 trials per condition, max = 196 trials per condition) and 125 trials for the older adults (min = 49 trials per condition, max = 187 trials per condition). The EEG signal was segmented into 1200-ms epochs starting 100 ms before the onset of the memory array and lasting until the end of the retention interval (i.e., consisting of 100 ms baseline, 200 ms memory array and 900 ms retention interval). Segments were averaged across trials for each individual and condition. Baseline correction was carried out with a 100-ms-prestimulus baseline.

Data Analysis

Figure 1C shows the contralateral and ipsilateral waveforms time locked to the onset of the memory array. As can be seen, the presentation of the memory array elicits distinct P1 and N1 components between 100 and 200 ms (both at contralateral and ipsilateral sites) followed by two relative negativities at contralateral sites: a transient one between 200 and 300 ms, i.e., the N2pc, and another, more sustained one that lasts until the end of the retention interval, i.e., the contralateral delay activity (CDA). As in the study by Jost et al. (2011) we analyzed the amplitude modulation of the CDA to investigate whether distractors are encoded in WM. In addition, we also investigated the N1 and the N2pc (both preceding the CDA) as these are regarded to reflect initial perceptual and attentional processes.

The occipital N1 is a distinct negative component that peaks around 180 ms poststimulus. N1 amplitude differences between the conditions were measured at three, for the N1 typical posterior electrodes contralateral to the attended hemifield (O1/O2, PO3/PO4, and PO7/PO8). (Footnote 1) For the statistical analysis, mean amplitudes from 20-ms time windows centered around the grand-average peak of each condition were extracted (see Figure 2 for the means of the extracted values). Since peak latencies differed across the conditions and between the age groups (varying between 172 and 222 ms, see Supplementary Figure 2), the time windows were defined separately for each condition and group.

Figure 2.

A. Performance in the change-detection task measured as percent correct responses. Performance decreases with set size, but also with distractor presence. This distractor effect is the same for young and older adults. Error bars indicate standard errors of the mean. B. Amplitudes of the posterior N1. The amplitude of the younger adults primarily reflects a distractor effect whereas the amplitude of the elderly is sensitive to both distractor presence and increasing set size. C. N2pc amplitudes. For both younger and older adults the amplitude is larger for set size 3 than for set size 1 reflecting item individuation and selection. Younger adults also show an increase for the distractor conditions, which is not present for the elderly.

Both the N2pc and the CDA are ERP phenomena that are contralateral to the attended hemifield and are obtained by subtracting ipsilateral from contralateral posterior activity. In the present study, the resulting difference waves (see Figure 3) were averaged across seven posterior electrodes (O1/O2, PO7/PO8, PO3/PO4, P7/P8, P5/P6, P3/P4, P1/P2, detailed information regarding the topographies are provided in the Supplementary Material) and hemispheres. As can be seen in Figure 3A, N2pc and CDA are pronounced negativities that are distinguishable with regard to onset and time course – the N2pc between 200 and 300 ms precedes the more sustained CDA – and as will be shown below also with regard to their functional significance.

Figure 3.

A. Grand average contralateral minus ipsilateral difference waves for younger and older adults. Negative voltage is plotted upward. The negative wave around 200 and 300 ms reflects the N2pc and the more sustained negativity the CDA. Note that the N1 was not larger contralateral than ipsilateral and is, therefore, not apparent in the difference waves. B+C. Difference wave set size 3 minus set size 1 + distractors (smoothed with a 5Hz low-pass filter) reflecting the time course of CDA filtering scores. Vertical lines indicate the time point when 50% of the maximum amplitude is reached. Significant differences in this 50% mark (B) suggest that older adults exhibit a delay in filtering. This also holds when older adults are compared with younger adults with low working-memory capacity K (C).

For the statistical analyses of the N2pc and the CDA, amplitudes for the different conditions were exported for time windows of 25 ms and smoothed using a moving average (n = 3) to reduce noise (for a similar procedure see Jost et al., 2011). Since the timing of the N2pc differed between the age groups, the statistics was adjusted such that it optimally fits the time course in each group. More precisely, we determined for each of the conditions and age groups the 25-ms time window, in which the amplitude was largest. The resulting values are shown in Figure 2C.

For the CDA, visual inspection of the data revealed that for both younger and older adults it started at around 350 ms after the onset of the memory array. We therefore analyzed the CDA amplitude differences with 25- ms time windows (see above) between 350 and 1100 ms.

For the N1 and the N2pc all four conditions were included in the analyses starting with a superordinate 2×2×2 analysis of variance (ANOVA) with the repeated-measure factors set size and distractor presence as well as the group factor age. For the CDA, the critical conditions for investigating filtering efficiency are the set size 1, the set size 3, and the set size 1 + distractors conditions (see also Jost et al., 2011). The number of targets in this distractor condition is equal to the number of targets in the set size 1 condition, whereas the total number of items in the distractor condition is equal to the number of targets in the set size 3 condition. Given that the CDA amplitude increases with more items stored, filtering efficiency can be measured by comparing the CDA amplitudes with the following rationale: If an individual is highly efficient in filtering distractors and stores only the target, then the CDA amplitude in the set size 1 + distractors condition will be similar to the amplitude in the set size 1 condition. If, however, filtering is inefficient and the target plus the two distractors are stored, then the amplitude will be similar to that in the set size 3 condition. Consequently, filtering efficiency is indexed by the amplitude difference set size 3 minus set size 1 + distractors. The resulting values multiplied with −1 indicate how efficiently distractors were prevented from being stored, with higher values reflecting better filtering (see Jost et al., 2011). Note that a similar measure of filtering efficiency cannot be calculated for the set size 3 + distractors condition, because the design does not include a no-distractor condition with the same total number of items (i.e., set size 5).

ERP results will be presented in chronological order (i.e., N1, N2pc, and CDA). Additional information with respect to analyses, results, and functional significance of the ERP components is provided in the Supplementary Material.

Results and Discussion

Working Memory Capacity

WM capacity K was estimated from the behavioral change-detection task by means of a standard formula (Cowan, 2001; Pashler, 1988): K = set size × (hit rate - false alarm rate). K was calculated for set sizes 4, 6, and 8 and averaged over the resulting measures to receive a single K score for each individual. Mean capacity for older adults was 2.12 items (SD = 0.64) and for younger adults 2.74 items (SD = 0.94). This difference was significant with t(47) = 2.69, p = .010, d = 0.77, and indicates an age-related decline of WM capacity.

Performance in the Filtering Task

Reduced WM capacity of older adults was also present in performance in the filtering task (see Figure 2A). Performance decreased with increasing set size and this effect was larger for older than for younger adults. A 2×2×2 ANOVA with the repeated-measure factors set size and distractor presence as well as the group factor age revealed a main effect of Set Size, F(1, 47) = 255.15, p < .001, η_p² = .84, that interacted with age, F(1, 47) = 17.29, p < .001, η_p² = .27. In contrast, distractor presence had only a numerically very small effect, F(1, 47) = 10.66, p = .002, η_p² = .19, that did not differ for younger and older adults (0.8% vs 1.0%), F < 1. Given that this task was specifically designed to test the dynamics of filtering with an easy selection criterion, the small distractor effect and the absence of an Age × Distractor interaction is not surprising.

N1 and N2pc: Early Perceptual Processing and Attentional Selection

N1 amplitudes are depicted in Figure 2B (see Supplementary Figure 2 for the waveforms). As can be seen, the pattern of effects differed for younger and older adults. This was also confirmed by a 2×2×2 ANOVA with the repeated-measure factors set size and distractor presence as well as the group factor age. In both groups, the amplitudes were larger when distractors were present, but this effect was more pronounced in younger than in older adults. Both the main effect Distractor Presence, F(1, 47) = 102.89, p < .001, η_p² = .69, and the interaction Distractor Presence × Age, F(1, 47) = 4.23, p = .045, η_p² = .08, were significant. Set size had different effects in young and old adults, as indicated by the interaction Set Size × Age, F(1, 47) = 14.13, p < .001, η_p² = .23 (all other effects p > .191).

Separate analyses for the two groups confirmed this pattern. Young adults showed a large amplitude increase with distractor presence, main effect Distractor Presence, F(1, 24) = 82.45, p < .001, η_p² = .78, and a small amplitude decrease with increasing set size, F(1, 24) = 7.37, p = .012, η_p² = .24. For older adults the N1 amplitude significantly increased with both distractor presence, F(1, 23) = 29.62, p < .001, η_p² = .56, and set size, F(1, 23) = 7.07, p = .014, η_p² = .24. Thus, whereas for younger adults the N1 amplitude solely increased with distractors (set size even had a reversed effect), older adults also showed sensitivity to the number of targets. In other studies (e.g., Kursawe & Zimmer, 2015; Libertus, Woldorff, & Brannon, 2007; Mazza, Pagano, & Caramazza, 2013) it has been observed that the N1 amplitude increased with the number of elements presented in the visual field (without distinguishing between targets and non-targets e.g., Nan, Knösche, & Luo, 2006). The amplitude pattern of the younger adults observed in our study suggests that they have treated objects in one quadrant as a single item. This indicates that they were more biased towards the discrimination of the different locations where targets and distractors were presented, and hence towards the attribute that differentiated targets from distractors (i.e., location). In contrast, older adults also processed other qualities of the display such as the increase of the number of targets within the relevant location.

N2pc amplitudes are depicted in Figure 2C (see Figure 3 for the waveforms). As for the N1, also the amplitude pattern in the N2pc differed for the two age groups. Whereas younger adults showed an amplitude increase with set size for the no-distractor condition and an additional increase when distractors were presented, older adults only showed the set-size effect for the no-distractor conditions. Consistent with this observation the superordinate ANOVA revealed a significant Set Size × Distractor Presence interaction, F(1, 47) = 6.79, p = .012, η_p² = .13, as well as a Distractor Presence × Age interaction, F(1, 47) = 13.17, p = .001, η_p² = .22.

With regard to the set-size effect, a further analysis revealed a significant difference for the no-distractor trials (set size 1 vs. set size 3), F(1, 47) = 5.31, p = .026, η_p² = .10, that did not interact with age (F < 1). This pattern fits the often-observed set-size effect in the N2pc and presumably reflects item individuation and selection (see Anderson et al., 2014; Mazza & Caramazza, 2015) before the information is stored in WM. Interestingly, this set-size effect did not differ for younger and older adults.

With regard to the distractor effect, only young adults showed a significant amplitude increase when distractors where presented. The follow-up ANOVA for young adults revealed a significant distractor presence effect, F(1, 24) = 18.45, p = .000, η_p² = .44 (for older adults F < 1). This difference between distractor and no-distractor trials was further validated by directly comparing the amplitudes of the distractor condition with one target and two distractors and the set size 3 condition and thus holding the total number of items constant, t(24) = 2.61, p = .016, d = 0.35. This pattern of results suggests that in young adults, distractor presence triggers an additional process – presumably a kind of “selection-for-suppression” process (for more details on this argument, see Supplementary Material). For older adults there was, if anything, the reverse tendency observed, that is, the amplitude in the distractor condition was even slightly smaller than in the set size 3 condition (p = .604).

To summarize these results: Both for the N1 and the N2pc, qualitatively different amplitude patterns were observed for young and older adults. In both components younger adults exhibited stronger distractor sensitivity than older adults. Moreover, whereas young adults seemed to select distractors for suppression before information was actually encoded in WM (as signaled by the N2pc effects), older adults exhibited no evidence of discriminating distractors from targets.

CDA: Encoding and Maintenance of Targets and Distractors

The CDA as an indicator of the actual fate of distractors in WM is depicted in Figure 3A. Except for generally larger amplitudes for younger than for older adults, the broad characteristics were similar in the two groups. Consistent with previous findings (e.g., Vogel & Machizawa, 2004) the CDA increased with set size, that is, it increased with the number of targets presented in the memory array (see Supplementary Material for detail). Most importantly, the CDA amplitudes also reflect that filtering was relatively easy both for young and older adults. The amplitude of the set size 1 + distractors condition was much closer to the amplitude of the set size 1 than to the amplitude of the set size 3 condition.

To look more closely at filtering efficiency, filtering scores were calculated as difference waves set size 3 minus set size 1 + distractors (multiplied with −1 to receive positive filtering scores). As can be seen in Figure 3B, filtering efficiency was lower in older than in younger adults, but only early in the retention interval. This age effect was significant in the time windows between 325 and 450 ms, with t(47) values between 1.75 (p = .044, one-tailed, d = 0.51) and 2.63 (p = .013, d = 0.74). Later in the retention interval (i.e., in the second half), filtering scores were the same for older and younger adults (all ts < 1.08). (Footnote 2)

A closer look reveals that this age effect resulted from a latency shift of the difference wave. In order to test for this delay, we determined for each group the time point when 50% of the maximal amplitude was reached by applying the jackknife method (Miller, Patterson, & Ulrich, 1998; Ulrich & Miller, 2001). This method basically is a “leave-one-out” ERP-averaging procedure (often used to measure the onset latency of the lateralized readiness potential) that has the advantage of increasing statistical power (for details, see Luck, 2014). Individual CDAs are often noisy and the true 50% marks are thus difficult to determine. The jackknife approach seems to be a good method to deal with this problem, because it increases the signal-to-noise ratio of the waveforms.

A direct comparison of the determined time points revealed that older adults reached the 50% of the maximal filtering score significantly later than younger adults (M = 458 ms vs. M = 351 ms), t(47) = 9.20, p < .001 (Footnote 3). Note that we came to the same result without using the jackknife procedure. Thus, the age-related delay in filtering is not restricted to more difficult selection criteria, but also occurs when selection is easy. Moreover, a closer look at the CDA filtering scores also suggest that older adults might “make up” the early filter deficits during the maintenance phase: At the end of the retention interval age effects disappeared, which is also in line with previous findings (Jost et al., 2011) and the behavioral data.

To investigate whether the observed delay in filtering is age-specific or whether it accompanies reduced WM capacity in general, filtering efficiency of younger adults with low WM capacity was examined. We selected younger individuals (n = 13) with individual WM capacities smaller than or equal to the median of the entire young-adults group. Their mean WM capacity was with 1.99 (SD = 0.43) comparable to the older adults’ WM capacity (M = 2.12, SD = 0.64), t < 1. Despite comparable capacity, filtering scores during the early time windows differed between these two groups (see Figure 3C), i.e., between 300 and 375 ms with t(35) values from 1.68 (p = .051, one-tailed, d = 0.59) to 2.42 (p = .021, d = 0.83). Also, the time point when 50% of the maximal filtering score was achieved, was with a mean of 329 ms earlier for the young adults with low WM capacity than for the group of older adults (M = 458 ms), t(35) = 4.76, p < .001. Moreover, low capacity and high capacity young adults did not differ (t < 1) and the 50% mark even was reached (numerically) earlier by the low capacity group. These results indicate that the delay in filtering is specific to older adults and not a general consequence of lower WM capacity.

General Discussion

The aim of the present study was to investigate age-related differences in visual WM capacity and its relation to filtering efficiency. Particularly, we were interested in whether the age-related delay in filtering that was found in previous work (Jost et al., 2011) reflects a selection difficulty effect or, alternatively a functional difference between old and young adults in the way WM content is controlled. In the latter case we expected to find that the delay in older adults’ filtering generalizes to an easy selection criterion. In addition, older and younger adults’ ERP components associated with early processing of targets and distractors were expected to respond differently to the filtering demands.

Easy versus Difficult Selection Criteria

The analysis of the CDA filtering scores revealed an age-related delay in filtering similar to the delay that was observed in the previous study with a different selection criterion (Jost et al., 2011). This is remarkable given that our procedure promoted optimal attentional focusing on the targets: Targets and distractors were presented in different locations and information about where the targets occurred was provided in advance (which both proved to reduce interference from distractors, e.g., Fox, 1995; Yantis & Johnston, 1990). Thus, focused attention without distractor processing was in principle possible. Selection on the basis of color as in the study by Jost et al. (2011), in contrast, requires that at least this feature needs to be processed for both targets and distractors before the decision can be made which objects are relevant and which ones are not. Thus, potential age-related impairments in perceptual processing might require longer data accumulation and/or delayed filtering. In the present study, however, this influence should be minimized, because focusing attention is possible for both groups given that deployment of spatial attention is not impaired in older adults (e.g., Hartley, 1993; for a review on attention and aging, see Kramer & Kray, 2006). Moreover, the number of objects that have to be processed was relatively small – the critical condition in which the filtering delay in the CDA was observed only has one target (plus two distractors). Thus, it is unlikely that limited perceptual processing capacity is (exclusively) responsible for the age-related filtering delay. Rather the delay seems to be a more general phenomenon in the context of WM tasks with filtering demands that is not restricted to situations when target selection/distractor filtering is relatively hard. Moreover, the observation that young adults with low WM capacity did not exhibit a similar delay indicates that the filtering delay is not just a byproduct of reduced WM capacity but is specific to older adults.

Processing Differences in Discrimination and Selection

The analysis of the early visual potentials revealed that older and younger adults differed already during initial perceptual/attentional processing. These effects were not just in terms of attenuated or delayed components in older adults, but rather indicated qualitative differences between old and young adults. Younger adults exhibited a clear N1 amplitude increase from the no-distractor (set size 1, set size 3) to the distractor conditions (set size 1 + distractors, set size 3 + distractors) and, accordingly, with the number of locations in which the items were presented. Older adults showed the same amplitude increase in the distractor condition, but in addition, were sensitive to the number of targets, which had no, or even reverse effects in young adults. Although, the exact mechanisms that are reflected by the N1 are not fully understood, the pattern of effects observed in the N1 seems to be predictive of the efficiency of early selection processes. A stronger bias towards the attribute that discriminates targets and distractors (i.e., location) seems to promote distractor filtering. In contrast, increased early processing of other qualities of the display such as the number of targets, seems to go along with delayed filtering (as in older adults).

The N2pc amplitude pattern also nicely fits this interpretation. A pronounced distractor effect was observed in the younger adults. This modulation, however, was not observed for the elderly. Assuming that the N2pc reflects item selection and individuation (e.g., Anderson et al., 2014; Mazza & Caramazza, 2015), the larger amplitude in the distractor conditions for young adults and the absence of such an effect in the elderly suggests that younger adults select the distractors for suppression. Note that older adults do not seem to be generally impaired in item selection and individuation, because the set-size effect was very similar for young and older adults. This finding is in accordance with other reports of preserved set-size effects in the N2pc in older adults (although in these studies a general N2pc amplitude reduction was observed for older adults, Pagano, Fait, Monti, Brignani, & Mazza, 2015; Störmer et al., 2013). In our study, the only difference between the age groups in the N2pc was seen in the distractor effect, indicating that only distractor handling was different rather than the general process of item selection and individuation. From other studies, there is evidence from the N2pc that even distractor suppression is preserved with advancing age. Lien, Gemperle and Ruthruff (2011), for instance, reported that resistance to attentional capture by salient but irrelevant stimuli was the same for old and young adults. This finding not necessarily contradicts our observations, but indicates that some aspects of attention seem to be spared in older adults, while others change with age.

Störmer et al. (2013) reported age differences in the N1 and N2pc in a similar paradigm. In their study, targets and distractors differed in shape. Older adults exhibited an early sensitivity to the number of items regardless of stimulus type (evidence comes from the analysis of the N1), which is in accordance with the present findings that early processing in older adults is less biased towards the target and distractor discriminating feature. Moreover, the N2pc effects in the Störmer et al. study also indicate that attentional selection is attenuated in older adults. However, object-based selection used in this study produced very subtle perceptual differences between targets and distractors. As a consequence, filtering was relatively hard even for the young adults, so that differential effects on the actual fate of the distractors (in the CDA) were not found.

In the present study, younger adults exhibited stronger distractor sensitivity in both the N1 and the N2pc and, as indicated in the CDA effects, they also reached maximum filtering efficiency earlier than the older adults. These findings indicate functionally different processing between the age groups throughout the processing stream. It is also important to note that the observed N1 and the N2pc pattern for the elderly were not present for low capacity young adults. Moreover, the specific pattern in each age group was observed even for subgroups of low and high WM capacity (see Supplementary Material for details).

The present findings also indicate that under certain conditions older adults can fully compensate for their initial delay in filtering (although in other studies distractors in the change-detection task affected older adults’ memories more than that of the younger adults, e.g., Jost et al., 2011; Sander, Werkle-Bergner, & Lindenberger, 2011b). As in the study by Jost et al. (2011) ERP filtering scores at the end of the retention interval were the same for older and younger adults. In contrast to the previous study, however, the performance decrease when distractors were presented in the memory array also was the same for older and younger adults. This even holds when the total number of presented items exceeds WM capacity (as in the condition with 3 targets + 2 distractors) – a situation in which filtering is essential. Thus, the delay in filtering does not always lead to an observable drop in performance. This observation suggests that whatever older adults do within the retention interval, after distractors were initially encoded, helps to preserve capacity for targets. Filtering, is thus not completely impaired in older adults, but rather seems to be shifted to a later processing stage (see also Gazzaley et al., 2008).

Early Selection versus Late Correction: Age-Related Differences in Controlling the Contents of Working Memory?

The results so far indicate that older and younger adults differ in the temporal dynamics of distractor processing or (to stick with the terminology from the attention literature) with regard to the locus of selection. Even during early perceptual and attentional processing, younger adults exhibited a stronger sensitivity to the feature that allows discriminating targets and distractors than older adults. As a consequence, younger adults filtered distractors more efficiently than older adults, who exhibited lower distractor sensitivity and initially encoded distractors in WM. However, later in the retention interval (or in performance) older adults did not show signs of stronger distractor load effects than young adults indicating that there are other mechanisms besides early selection that allow protecting against distractors. Older adults seem to suppress distractors after the fact, relying on a form of late selection or “late correction”.

One possibility mentioned during the review process is that older adults might be forced into a later selection mode because of a reduced capacity or bandwidth when it comes to processing objects on the screen in parallel. We believe, such an interpretation would be plausible in a situation with a larger number of objects on the screen and a more difficult filtering criterion (as in the study by Jost et al., 2011). However, with the small number of relevant objects (only one target in the critical filtering condition) and the very simple, location-based filtering criterion the demands on parallel processing should have been relatively small and therefore are unlikely to have enforced the qualitative differences in WM filtering between old and young adults. Instead, we believe these results suggest that there is a strategic difference in the way young and older adults control the contents of WM and that both the early ERP effects and the later, CDA effects are a consequence of these age differences.

This interpretation is broadly consistent with theories that assume a flexible application of early versus late selection (e.g., Johnston & Heinz, 1978; Lavie, 1995; Vogel, Woodman, & Luck, 2005; Yantis & Johnston, 1990) and in particular with the notion of an age-related shift to late-selection processes (see Velanova, Lustig, Jacoby, & Buckner, 2007) or “reactive” control (Braver, 2012). According to the load-shift model of cognitive aging proposed by Velanova and colleagues, executive functions diminish with age. Older adults are less effective at implementing early-selection processes and, as a compensatory mechanism, shift to late selection/allocating (frontal control) resources to late processing stages. Although inspired by age-effects in long-term memory retrieval, load shift is proposed to apply to different functional domains. Based on the more general idea that (frontal) top-down signals can bias perceptual processing (Desimone & Duncan, 1995; Miller & Cohen, 2001), early selection can be achieved when task goals are activated in advance. According to Velanova et al. younger adults expend considerable resources to early selection which, when applied to the task of the present study, will result in highly efficient filtering. Older adults, in contrast, fail to appropriately filter incoming information and this increases the burden on later processing stages.

The load-shift model (Velanova et al., 2007) specifically proposes late selection as compensatory mechanism. The dual mechanisms of cognitive control (DMC) framework proposed by Braver and colleagues (e.g., Braver, 2012; Braver, Gray, & Burgess, 2007; Braver, Paxton, Locke, & Barch, 2009) in contrast, is more neutral in this regard. It assumes two distinct mechanisms of cognitive control that both have merits. According to this framework, cognitive control can be asserted in a proactive or a reactive manner. The model is conceptualized as a general framework to explain variations in cognitive control across situations and individuals. Importantly, it makes the specific predicition that aging goes along with a general shift to reactive control where control is mobilized after a high interference event is detected (for review, see Braver, 2012).

Our main finding that the temporal dynamics of filtering differ between young and older adults and the fact that older adults operate in a kind of “late correction” mode even under easy selection conditions is in accordance with the above proposed age effects. Younger adults in the present study proved to optimally bias attention towards features that discriminate between targets and distractors which in turn facilitates early target selection – corresponding to proactive control in the DMC terminology. In contrast, older adults were less biased during early processing which allows distractors gaining access to WM. The retention interval then seems to be used not only for storing the targets, but also for a late correction (i.e., reactive suppression of the distractors).

From the present findings it is difficult to evaluate why older adults rely on late correction. According to Braver (2012) the reactive control strategy has the advantage “of being computationally efficient”, because it frees up resources – in our case for encoding – that otherwise would have been consumed for forming and activating the settings to filter distractors. Similarly, Lindenberger and Mayr (2014; see also Mayr, Spieler, & Hutcheon, 2015) have argued that older adults may adopt a strategy of relying on environmental input rather than on internal representations. Favoring one type of control over the other may also be related to the characteristics of the task situation. When distractors occur in only 50 percent of the trials, as in the present study, late correction might be as adequate as early selection. However, when distractor probability increases or motivational incentives for filtering distractors are high, proactively controlling what is encoded into WM, might be the better choice. Whether older adults are capable of adjusting their mode of controlling the contents of WM to situational factors is an interesting question for future research.

Conclusion

The present study provided evidence that older and younger adults differ in the way they control the contents of WM. While young adults initiated processing with relatively high sensitivity to those features of the display that discriminate between targets and distractors, older adults initially stored distractors and suppressed them in a “just in time” manner during the course of maintenance. Although these results highlight the importance of filtering for WM in general and for the age-related decline in particular (see Hasher & Zacks, 1988; Hasher et al., 1999), they also indicate that filtering is not out of service in older adults, but shifted to a later processing stage. Importantly, younger adults with low WM capacity did not show a similar delay in filtering. Thus, our findings support the claim that older adults are not just like low WM young adults (Jost et al., 2011) and adds direct evidence of age differences in the way WM is controlled.