Event-Related Brain Potentials Reveal Strategy Selection in Younger and Older Adults

Abstract

It is well-established that younger adults prioritize information accrued during different stages of stimulus evaluation (“early” versus “late”) to optimize performance. The extent to which older adults flexibly adjust their processing strategies, however, is largely unexplored. Twenty-four younger and twenty-four older participants completed a cued flanker task in which one of three cues, indicating the probability that a congruent array would appear (75%, 50%, or 25%), was presented on each trial. Behavioral and ERP (CNV, LRP, N2, and P3b) analyses allowed us to infer cue-driven changes in strategy selection. Results indicate that when both younger and older adults expected an incongruent array, they prioritized late, target information, resulting in a decreased susceptibility to the performance-impairing effect of distractors, extending the conclusions of Gratton and colleagues (1992) to older adults and supporting the claim that strategic control remains largely intact during healthy aging.

Introduction

Cognitive control encompasses a set of mechanisms that, based on current task goals and contextual information, adjust attentional biases (or attentional weights allocated to different stimuli or stimulus features; Gratton, Cooper, Fabiani, Carter, & Karayanidis, 2018) in order to support goal-directed behavior. However, as individuals age, research suggests that certain aspects of control may suffer, with evidence demonstrating that older adults process information more slowly, exhibit increased susceptibility to the deleterious effects of distracting stimuli, and have lower performance on working memory and inhibitory tasks, compared to younger adults (Fabiani, 2012). What remains unclear, however, is the extent to which older adults can utilize task-relevant, contextual information to modify their expectations (and, consequently, their attentional biases) during performance on control tasks. In other words, can older adults dynamically alter their attentional biases in a strategic (i.e., context-dependent) manner similarly to younger adults? To answer this question, a cued-flanker task was administered to participants while EEG was recorded. As a follow-up to Gratton, Coles, and Donchin’s (1992) ERP study of strategic control in younger adults, here we show that older adults also flexibly adjust their attentional weights in a trial-by-trial manner, as evidenced by corresponding changes in behavior and electrophysiological indices of response preparation. In addition, at least amongst younger adults, we demonstrate that the N2 component of the ERP, a putative index of the detection of response conflict (Botvinick et al., 2001; Folstein & Van Petten, 2008; Larson et al., 2014), is also susceptible to cue-driven modulation.

The congruency (or interference) effect (CE) for a given experimental condition is typically measured as the difference in mean reaction times (RT) between incongruent and congruent trials in conflict tasks such as the Eriksen flanker (Eriksen & Eriksen, 1974), Stroop (Stroop, 1935), and Simon (Simon, 1969) tasks. In the original view of Eriksen and Eriksen (1974), the size of the CE is typically taken to reflect the mutual inhibition associated with the activation of competing response channels (see also Gratton, Coles, Sirevaag, Eriksen, & Donchin, 1988), which occurs in conflict conditions, but not in no-conflict conditions. According to this view, activation of response channels should occur automatically when the related information is presented. Specifically, incorrect responses could be automatically activated, at least at some level, whenever the stimulus contains information associated with them. Furthermore, in a series of experiments, Gratton et al. (1992) demonstrated that the size of the CE also varies when the expected relationship between target and noise information – and therefore the expected probability of conflict between response channels – is manipulated. This leads to the idea that, by manipulating participants’ expectations for the upcoming “target-noise” conflict level (e.g., by manipulating the proportion of congruent items within a block or the previous trial’s conflict level; see Braem et al., 2019 for a review of dynamic control manipulations), one can modulate the degree of activation of incorrect response channels that is produced by otherwise identical stimulus information (i.e., the weight given to information provided by the stimulus to activate responses). In turn, this leads to a moderation or enhancement of the conflict between alternative response channels, and ultimately of the CE.

Particularly clear evidence of the role of conflict expectations in modulating the CE comes from a variant of the conflict paradigm in which symbolic cues are used to indicate the relative probabilities of upcoming congruent (low conflict) or incongruent (high conflict) stimuli (Bugg & Smallwood, 2016; Chao, 2011; Correa et al., 2014; Ghinescu et al., 2010; Lamers & Roelofs, 2011). In their original paper, Gratton et al. (1992), using a flanker task with younger adults, first observed this cue-driven modulation of the CE in two experiments (3a and 3b). Participants appeared to adjust the attentional weights allocated to peripheral (distractors) vs. central (target) stimuli as a function of the information conveyed by explicit, probabilistic cues that varied on a trial-by-trial basis. When a cue indicated that a congruent array was more likely to appear, participants weighted peripheral information extracted during the early, incomplete stage of stimulus evaluation more heavily, resulting in larger CEs and greater initial activation of the incorrect response on incongruent trials (as assessed via electromyograms and the lateralized readiness potential, LRP, an ERP measure of the relative activation of competing responses; Gratton et al., 1988). In contrast, when a cue indicated that an incongruent array was more likely, participants weighted central information extracted during the late, complete stage of stimulus evaluation more heavily, resulting in smaller CEs and lesser initial activation of the incorrect response on incongruent trials. Cues indicating that congruent and incongruent arrays were equiprobable generally resulted in intermediate CEs and intermediate levels of incorrect motor preparation.

While strategy-driven cognitive control is well established in younger adults, evidence for this phenomenon in older adults remains sparse and inconclusive. Some studies using cue-based control tasks suggest age-related deficits in the updating and maintenance of task-relevant context representations (Braver et al., 2005; Paxton et al., 2008) and, consequently, in the strategic adjustments of control. Other studies, utilizing proportion-congruent manipulations instead of conflict cuing, indicate that strategic control remains intact or is only partially hindered during normal aging (Bélanger et al., 2010; Mutter et al., 2005; West & Moore, 2005; Xiang et al., 2016). Importantly, however, the former findings, which hint at age-related declines in strategic control, utilized the AX-CPT task (Rosvold et al., 1956; Servan-Schreiber et al., 1996). In the AX-CPT paradigm, given that participants are instructed to respond solely to the probe, X (versus Y), when it is immediately preceded by the appropriate cue, A (versus B), the cue-probe contingency is particularly taxing for working memory, a domain where younger adults hold a decided advantage over older adults (Park, 2000; Salthouse, 1990). And since the maintenance of cue information is essential for optimal task performance (a feature absent from other cued conflict paradigms), these heightened working memory demands could exacerbate age-related disparities in performance. This may prove particularly problematic when the cue-probe interval is prolonged (e.g., to 5 seconds; Braver et al., 2005).

However, evidence also suggests that interindividual variability in cognitive ability increases with age (at least on measures of memory, processing speed, and visuospatial function; Christensen et al., 1999; Hultsch et al., 2002; Wilson et al., 2002). This increased heterogeneity among older adults may inadvertently lead to larger sampling errors, such that, for example, high performing older adults are overrepresented in some samples and underrepresented in others. These disparities in sample characteristics may partially account for discrepant findings on strategic control and aging. Nevertheless, it appears that a significant gap in the literature exists regarding older adults’ capacity to utilize cues in preparation for conflict detection and resolution, in conditions in which working memory load is minimized, thus distinguishing between proactive control per se and working memory abilities.

Electrophysiological studies on the relationship between aging and strategic adjustments of control have typically relied upon conflict adaptation, or the congruency sequence effect, as a window into the neural mechanisms supporting dynamic cognitive control. Unfortunately, this effect’s interpretation is often confounded by repetition priming effects and/or feature integration biases (Hommel et al., 2004; Mayr et al., 2003). Nevertheless, a recent, large ERP study (Larson et al., 2016) demonstrated that younger and older adults exhibit similar behavioral conflict adaptation effects. Both groups also displayed N2 conflict adaptation effects, such that the N2 amplitude congruency effect (i.e., the difference in N2 amplitude between incongruent and congruent trials) was significantly smaller following incongruent than congruent trials. However, for P3b amplitude, Larson and colleagues observed a significant conflict adaptation effect for younger adults only. To our knowledge, no studies have utilized the lateralized readiness potential (LRP) to examine strategic control in the context of aging, in a manner akin to Gratton and colleagues (1992). However, while Wild-Wall and colleagues (2008) did not assess strategic control per se, they administered a standard, non-cued flanker task to younger and older adults and found that incongruent trials elicited an initial “positive dip” in the target-locked LRP, an index of incorrect response activation, in both age groups. This suggests that, like the N2, the LRP indexes (in part) response conflict not only in younger adults, but in older adults as well, and therefore may provide a window into potential age-related differences in strategic control.

In the present study, we administered a cued flanker task to younger and older adults while EEG was recorded, to replicate and extend the conclusions on strategic selection from Gratton and colleagues (1992) to older adults. Specifically, in addition to behavioral measures, we examined ERPs time-locked to cue and target (flanker) onset, including the contingent negative variation (CNV), LRP, N2, and P3b components. In an S1–S2 paradigm such as the current one, the cue-locked CNV primarily indexes motor preparation for an upcoming imperative stimulus, and, to a lesser extent, stimulus expectancies and working memory (Leuthold et al., 2004; Walter et al., 1964). Therefore, we expected cues predicting congruent arrays to elicit larger CNV amplitudes than cues predicting incongruent arrays, reflecting enhanced preparedness (or reduced hesitancy) to respond. We anticipated that the probabilistic cues would modulate the level of initial response activation (i.e., after imperative stimulus presentation), as reflected in the target-locked LRP, such that cues predicting a congruent array would elicit larger CEs than equiprobable cues or cues predicting an incongruent array, thereby revealing the relative preference for information accrued during the early versus late stage of stimulus evaluation, and vice versa. For the conflict N2 component, similarly to the LRP, we expected that cues predicting a congruent array would elicit larger N2 CEs than equiprobable cues or cues predicting an incongruent array. However, given the dearth of ERP research on strategic control and aging, we remained agnostic as to whether age would interact with both cue and congruency. Lastly, given the well-known age-related decline in processing speed, we predicted that the latency of the P3b component in older adults would be significantly prolonged compared to younger adults, reflecting the increased time needed to categorize the central target stimulus. For both age groups, we expected a prolonged P3b latency on incongruent compared to congruent trials, due to interference from incompatible flankers.

Method

Participants

Twenty-five younger adults (13 females, age range: 18–29 years) and thirty older adults (18 females, age range: 65–80 years) gave informed written consent to participate in the study, which was approved by the Institutional Review Board of the University of Illinois at Urbana-Champaign. None of the participants reported a history of psychiatric or neurological illness or exhibited any signs of dementia (scores ≥ 51 on the modified Mini-Mental Status Examination; Teng & Chui, 1987). Neither younger adults (assessed via Beck’s Depression Inventory-Second Edition; Beck et al., 1996) nor older adults (assessed via the Geriatric Depression Scale; Yesavage et al., 1982) displayed symptoms of moderate to severe depression. All participants were right-handed, native English speakers and received monetary compensation for their time. Six older adults (4 females) were excluded after flanker task practice due to exceedingly low task performance (accuracy ≤ 60% in all cases). One younger adult (male) was excluded for the same reason. This yielded a final sample consisting of 24 younger adults (age range = 18–29, 13 females) and 24 older adults (age range = 65–80, 14 females).

As commonly found in aging studies¹, older adults reported more years of education, had higher age-adjusted IQs, and displayed superior vocabulary knowledge. An overview of the sample’s descriptive characteristics is provided in Table 1.

Table 1:

Descriptive characteristics of the sample – Mean (standard deviations)

	Younger Adults	Older Adults	p-value
N	24 (13 females)	24 (14 females)
Age (years)	21.67 (3.07)	71.38 (4.19)	< .001
Education (years)	15.13 (1.83)	17.46 (2.78)	.001
IQ (age-adjusted) *	111.75 (14.28)	124.08 (14.81)	.005
Shipley’s Vocabulary Scale	31.33 (3.61)	36.42 (2.92)	< .001

^*Kaufman Brief Intelligence Test – 2^nd Edition

Task, Stimuli, and Procedure

The data reported here come from a larger study that also included blocks of flanker trials in which the cues were uninformative (i.e., not associated with different conflict probabilities) emotional images from the from the International Affective Picture System database (Lang et al., 2008). These data are not analyzed in this report in which we focus solely on the blocks in which the cues were informative (i.e., probabilistically associated with different conflict levels) emotionally neutral images.

A cued flanker task was used to assess cognitive control and its sensitivity to strategic modulation on a trial-by-trial basis (see Figure 1 for a trial schematic). In a flanker paradigm, a central target stimulus is flanked by irrelevant distractors that are either congruent (e.g., >>>>> or <<<<<) or incongruent with the target (e.g., <<><< or >><>>). The participant’s task is to ignore the flankers and respond based on the direction of the central arrow. Individuals who struggle to filter out the task-irrelevant distractors (i.e., flankers) perform poorly, which is reflected in large CEs. As implicated by Gratton and colleagues (1992), probabilistic cues presented prior to the imperative stimulus may modulate this CE by regulating the amount of attentional resources allocated to distractors at different stages of stimulus evaluation. Furthermore, cues predicting that a congruent array will appear should cause participants to expect a congruent array, thereby increasing the attentional weights allocated to distractors during the early stage of stimulus evaluation, leading to increased processing of peripheral distractors and impairing performance on incongruent trials (resulting in larger CEs on predict-congruent, or PC, trials). Conversely, cues predicting that an incongruent array will appear should cause participants to expect an incongruent array, thereby decreasing the attentional weights granted to distractors during the early stage of stimulus evaluation, leading to decreased processing of peripheral distractors and improving performance on incongruent trials (resulting in smaller CEs on predict-incongruent, or PI, trials). Cues indicating that congruent and incongruent arrays are equiprobable (i.e., predict-equiprobable cues, PE), should, in theory, lead to intermediate performance and CEs. We consider these probabilistic cue-based differences in the congruency effect to be manifestations of strategic control.

Figure 1:

Trial schematic. Participants were instructed to press a button on one of two keypads that corresponded to the direction of the target (central) arrow (i.e., a left-pointing arrow required a left button press and vice versa). Participants were given nearly 2 seconds to respond, extending from the onset of the imperative stimulus to the beginning of the next trial. The International Affective Picture System (IAPS) catalog numbers for the images used as cues are 7018, 7100, and 7705 for the screw, fire hydrant, and dresser, respectively (a representative image is shown here: “Yellow Fire Hydrant” by Lee Edwin Coursey, licensed under CC BY 2.0). The presented cue varied randomly on a trial-by-trial basis.

Three neutral, low-arousing images of inanimate objects (fire hydrant, dresser, and screw from IAPS) served as cues that preceded the imperative stimulus array. The images respectively represented a 75% (predict-congruent; PC), 50% (predict-equiprobable; PE), and 25% (predict-incongruent; PI) probability of a congruent stimulus array — that is, each cue validly predicted the probability of a congruent array subsequently appearing, and the presented cue varied randomly on a trial-by-trial basis. The three cue types were equiprobable and participants were explicitly told the congruency probability represented by each cue prior to commencing the task. PC and PI cue images were counterbalanced across subjects. The imperative stimulus consisted of five horizontally oriented arrows that were either congruent (<<<<< or >>>>>) or incongruent (<<><< or >><>>) on any given trial.

Subjects sat in a dimly lit, sound-attenuated and electrically shielded booth, approximately 100 cm in front a computer monitor. Participants were instructed to indicate, as quickly and accurately as possible, the direction (left or right) of the central target arrow, by pressing a button on one of two keypads located on either side of the participant. Stimulus-response mapping remained constant across all participants (i.e., a left-pointing target stimulus always required a left-button press, and vice versa) to eliminate the confounding Simon effect in some participants (Simon, 1969). There were three blocks of 288 trials each, yielding a total of 864 trials.²

Each trial began with a 499 ms cue, followed by a 999 ms fixation. Then, the imperative stimulus appeared for 149 ms and was followed by 1848 ms of fixation before the onset of the next trial (the target and flankers were presented simultaneously). The response window began with the onset of the imperative stimulus and continued until the onset of the next cue (i.e., the next trial). The global probability of a congruent trial within each block was 50%. The imperative stimulus arrays were presented in white typeface on a black computer screen and subtended 2.23° × 0.46° of visual angle. Each cue overlaid a gray background with uniform dimensions such that each composite image subtended 6.98° × 5.35° of visual angle. All stimuli were presented on a monitor (19-in. CRT, refresh rate 60 Hz, screen resolution 1280 × 960; Dell Computer, Round Rock TX) using the E-Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA).

Accuracy feedback was displayed on-screen at the end of each block. Participants were informed of the percentage of correct trials and were told to either “respond more slowly and more accurately”, “respond more quickly”, or “continue to respond as quickly and accurately as you can” if they scored below 75%, above 95%, or in the 75%–95% range, respectively. The feedback was designed to encourage participants to prioritize speeded responses and elicit a reasonable number of errors, a requirement for accurately assessing speed of processing. Participants could take breaks between blocks, as needed.

Before the flanker task, younger adults completed 96 practice trials at the experimental speed. Older adults completed two sets of practice trials. Additional practice was added for older adults to offset difficulties (apparent on preliminary data) for them to complete the task at the experimental speed. As such, we added a slower-paced practice block to familiarize this group with the task. In the first set (48 trials), the inter-stimulus interval (ISI) was increased by 30%, but the cue and imperative stimulus presentation times remained at experimental speed. In the second set (96 trials), each trial ran at the experimental speed.³

EEG Recording and Preprocessing

EEG was recorded from 64 channels using an actiCAP electrode system with Ag-AgCL electrodes and a BrainAmp DC amplifier using BrainVision Recorder software (Brain Products, Gilching, Germany) at a sampling rate of 500 Hz. Electrodes were arranged in accordance with the International 10–20 system (Jasper, 1958), with subsequent division of inter-electrode spaces by a factor of two (5–10 system). The left mastoid was used as the online reference, and electrode AFz was used as the ground. Electrodes were also placed above and below the left eye and at the outer canthi of both eyes to monitor eye movements and blinks (i.e., ocular artifacts). Impedances were kept below 10 kΩ. An online bandpass filter of 0.1–250 Hz was used. To minimize the problem of multiple comparisons, only data from selected active electrodes (FCz, Cz, Pz, C3 and C4), at which previous studies indicated the largest effects for the relevant conditions examined in this study (for reviews, see Fabiani et al., 2007; Folstein & Van Petten, 2008; Polich, 2007; Smulders & Miller, 2012), were submitted to statistical analyses and are reported in the current paper.

Offline, data were preprocessed using custom scripts in MATLAB and EEGLAB (Delorme & Makeig, 2004). Electrodes were re-referenced to the average of the left and right mastoids Luck, 2014). Prior to CNV, N2, and P3b analyses, a 30 Hz low-pass filter was applied, and prior to analysis of target-locked LRPs, a 10 Hz low-pass filter was applied. For preparatory interval analyses (i.e., CNV), epochs lasting 1,698 ms, time-locked to cue onset and terminating with the onset of the imperative stimulus (−200 to 1498 ms), were extracted from the continuous EEG; baseline-correction was applied using the mean amplitude from −200 to 0 ms relative to cue onset. For post-target interval analyses (i.e., LRP, N2, and P3b), epochs lasting 2,197 ms, time-locked to imperative stimulus onset and terminating with the onset of the next cue (−200 to 1,997 ms), were extracted from the continuous EEG; baseline-correction was applied using the mean amplitude from −200 to 0 ms relative to target onset. Epochs with channels showing A/D saturation for greater than 500 ms were removed before correction of ocular artifacts (Gratton et al., 1983). Additionally, epochs showing artifacts greater than 1000 μV in absolute magnitude were excluded. After correction of ocular artifacts (Gratton, Coles, & Donchin, 1983), any other artifacts with an absolute magnitude greater than 200 μV within a 600 ms window were detected and discarded using a 100 ms moving window. For all analyses, only epochs in which the participant had responded correctly were included.

Behavioral Analyses

Incorrect trials and all trials with reaction times less than or equal to 200 ms (i.e., fast guesses) were discarded before statistical analysis. After collapsing across hand (i.e., target arrow direction), the following six conditions were created (cue type, flanker congruency): predict-congruent, congruent (PCC); predict-congruent, incongruent (PCI); predict-equiprobable, congruent (PEC); predict-equiprobable, incongruent (PEI); predict-incongruent, congruent (PIC); predict-incongruent, incongruent (PII).

Three-way mixed ANOVAs were used to analyze the RT and accuracy data, with cue (PC/PE/PI) and congruency (congruent/incongruent) as within-subject factors and age (young/old) as the between-subjects factor. Additionally, the inverse efficiency score (IES; Bruyer & Brysbaert, 2011; Townsend & Ashby, 1978, 1983), an integrated measure of RT and accuracy, which is computed by dividing the mean RT of correct responses by the proportion of correct responses for each condition (IES = RT/proportion correct), was also analyzed using a three-way mixed ANOVA. IES is relatively insensitive to speed-accuracy tradeoffs and is used as a measure of RT that is not biased by fast decisions, thereby providing an index of processing speed that estimates the “true” processing speed when the effects of speed-accuracy tradeoffs are minimized. Data were analyzed using IBM SPSS Statistics for Windows, Version 26. An alpha level of .05 was used for all statistical tests, and all statistics reported are Greenhouse-Geisser-corrected. For those dependent measures that failed Levene’s test for homogeneity of variances, we briefly summarize the results of analyses of log-transformed data in the footnotes, in addition to analyses of the original data in the main text. Lastly, as we were particularly interested in examining the evidentiary strength in favor of or against an age-related effect on strategic control, we report the inclusion Bayes factor (BF_incl), which indicates the likelihood of the observed data occurring under all models with a particular effect — in this case, the three-way interaction effect (Age × Cue × Congruency) — relative to all models without that particular effect. BF_incl > 1 indicates evidence in favor of an effect’s inclusion, whereas BF_incl < 1 indicates evidence against an effect’s inclusion — values greater than 3 or less than 1/3 indicate “substantial” evidence in favor of or against an effect, respectively (Wetzels et al., 2011). JASP, an open-source statistical package, was used to compute the inclusion Bayes factor (Bergh et al., 2020; JASP Team, 2020; Rouder et al., 2012).

ERP Analyses

All electrophysiological measurements were obtained via ERPlab (Lopez-Calderon & Luck, 2014), and all measurements were conducted on single-subject averaged waveforms. However, as one of our dependent variables contained negative values (i.e., LRP CE amplitude) and the logarithm is only defined for positive values, for these data, a robust mixed ANOVA was performed on the 10% trimmed means (i.e., the sample means after excluding the 10% largest and 10% smallest values), which is summarized in the footnotes. This test, which is less sensitive to outliers than regular ANOVA, was performed using the ‘bwtrim’ function (from the WRS2 package; Mair & Wilcox, 2020) in RStudio (RStudio Team, 2020)

Cue-locked ERPs

To examine the effects of cue-induced expectations on preparatory processing, we quantified the mean amplitude of the late CNV in the last several hundred milliseconds of the preparatory interval (1200–1498 ms), immediately prior to imperative stimulus onset. Electrode Cz was chosen for analyses, as prior studies have demonstrated that CNV amplitudes are largest at the vertex (for a review, see Donchin et al., 1978). The late CNV has been described as a pre-motor potential reflecting anticipation/expectancy or a participant’s readiness to respond, but its functional significance is still somewhat unclear (Fabiani et al., 2007). For analyses of early sensory-related components, including the cue P2 (P2_Cue) and frontal negativity, please see the supplementary materials (see Supplementary Figure S1 for grand averaged, cue-locked waveforms at multiple midline and occipital electrodes).

Target-locked LRPs

To assess the extent to which motor preparation is impacted by response conflict and whether this effect is modulated by cue-induced expectations, we examined the LRP based on ERP data from lateral electrodes C3 and C4, roughly overlapping the hand regions of the motor cortex. As a reminder, the LRP is a pre-motor potential that indexes response preparation, or preparation for voluntary movements of one side of the body (e.g., a hand or foot). The LRP was computed by subtracting the average activity for correct response trials only, at each time point, for electrodes ipsilateral to the responding hand from the average activity, at each time point, for electrodes contralateral to the responding hand (Gratton et al., 1988). To mitigate confusion over which set of waveforms were analyzed, the grand-averaged pre-subtraction LRP waveforms are presented in Figure 2. Instead, we analyzed and report the congruency effect, difference waveforms in the results section. These waveforms are very similar to those reported in Gratton et al. (1988; 1992), indicating the presence of a clear difference in the LRP waveform for congruent and incongruent trials, and specifically indicating the presence of a dip in the direction of the incorrect side for the incongruent, but not the congruent trials, as well as a delay in the latency of the deflection of the LRP waveform in the direction of the correct response for incongruent relative to congruent trials. Since these main effects of congruency on the LRP waveform are well-established and are not the focus of the current study, they were not further analyzed here.

Figure 2:

Grand average lateralized readiness potential (LRP) waveforms time-locked to target (flanker array) onset. Solid and dashed lines indicate the congruent and incongruent conditions, respectively, while red, green, and blue lines indicate the predict-congruent (PC), predict-equiprobable (PE), and predict-incongruent (PI) conditions, respectively.

Target-locked LRP Congruency Effect.

In order to highlight and examine in detail the putative Cue × Congruency interaction effect on the relative activation of correct and incorrect responses, LRP difference waveforms were computed such that, for every cue, the (averaged) incongruent LRP waveform was subtracted from the (averaged) congruent LRP waveform for every subject. This yielded three waveforms (PC, PE, and PI) per individual, each indicating the difference in the relative activation over time of correct and incorrect responses on congruent relative to incongruent trials, for each cue condition. Peak amplitude and onset latency measurements were then taken in a time window spanning 200–500 ms (this time window encompasses the visually identifiable peaks based on the grand averages for both age groups across all conditions, and is consistent with prior literature; Smulders & Miller, 2012) and were submitted to a two-way mixed ANOVA, with cue (PC/PE/PI) as the within-subjects factor and age (young/old) as the between-subjects factor. Onset latency was defined as the latency at which the waveform reached 50% of its (most positive) peak voltage (based on the local peak amplitude). When justified by the occurrence of significant main effects or interactions, follow-up analyses were conducted.

Target-locked ERPs

In order to examine the effect of conflict on the N2 component and the extent to which this effect is modulated by cue-induced expectations, we measured the negative, local peak amplitude from 200 to 500 ms (this time window encompasses the visually identifiable peaks based on the grand averages for both age groups across all conditions, and is consistent with prior literature; Folstein & Van Petten, 2008) at electrode FCz. These measurements were then submitted to a three-way mixed analysis of variance (ANOVA) with cue (PC/PE/PI) and congruency (congruent/incongruent) as the within-subject factors and age (young/old) as the between-subjects factor. To discern the effect of age on component latency, N2 local peak latency was also measured from 200–500 ms at electrode FCz and submitted to a three-way mixed ANOVA. For P3b analyses, measurements of positive, local peak amplitude and latency were taken 300–700 ms post-target (this time window encompasses the visually identifiable peaks based on the grand averages for both age groups across all conditions, and is consistent with prior literature; Polich, 2007) onset at electrode Pz. Both P3b measurements were then submitted to three-way mixed ANOVAs, with cue and congruency as the within-subjects factors, and age as the between-subjects factor. When justified by the occurrence of significant main effects or interactions, follow-up analyses were conducted.

Results

Behavior

Average behavioral effects are presented in Figures 3 and 4. As a statistical adjustment for multiple comparisons, we report the Holm-Bonferroni-corrected p-values (p_Holm) for all pairwise comparisons.

Figure 3:

From top to bottom: Mean response time (RT), accuracy (ACC), and inverse efficiency score (IES = RT ÷ ACC) for each of the three cue types (PC, PE, PI) by age (younger — light colors, older — dark-colors) and congruency (congruent — solid, incongruent — striped). YA = younger adults; OA = older adults; Con = congruent; Inc = incongruent; PC = predict-congruent; PE = predict-equiprobable; PI = predict-incongruent.

Figure 4:

From top to bottom: Mean response time (RT), accuracy (ACC), and inverse efficiency score (IES = RT ÷ ACC) congruency effects (CE; incongruent minus congruent) for younger (light) and older adults (dark). YA = younger adults; OA = older adults; PC = predict-congruent; PE = predict-equiprobable; PI = predict-incongruent.

Response Time

Analyses revealed a statistically significant main effect of age [F(1,46) = 26.24, p < .001, η_p² = .363], indicating that younger adults responded faster than older adults. Additionally, the main effect of congruency was significant [F(1,46) = 198.384, p < .001, η_p² = .812], indicating that most participants responded quicker on congruent trials than on incongruent trials. The Age × Congruency interaction was significant [F(1,46) = 4.588, p = .038, η_p² = .091], revealing that older adults exhibited a larger congruency effect than younger adults, which was driven predominantly by prolonged RTs for older adults on incongruent trials (see Figures 3 and 4).⁴ The Cue × Congruency interaction was also significant [F(1.559, 71.699) = 5.738, p = .009, η_p² = .111], demonstrating that the PC_CE was larger than the PI_CE [t(47) = 2.860, p_Holm = .019]. However, while the PC_CE was larger than the PE_CE [t(47) = 1.749, p_Holm = .087] and the PE_CE was larger than the PI_CE [t(47) = 2.153, p_Holm = .073], these contrasts did not survive an adjustment for multiple comparisons. The aforementioned results and the absence of evidence for a three-way interaction with age [F(1.559, 71.699) = 1.037, p = .344, η_p²= .022, BF_incl = .001] are consistent with the proposal that PC and PI cue-induced expectations, in both younger and older adults, led participants to favor information extracted during early (more noise-sensitive) and late (less noise-sensitive) stages of stimulus evaluation, respectively (see Gratton et al., 1992). Lastly, the main effect of cue failed to reach statistical significance [F(1.756, 80.780) = 1.202, p = 0.302, η_p² = .025] and did not interact with age [F(1.756, 80.780) = 0.059, p = 0.924, η_p² = .001].

Accuracy

Analysis of accuracy rates (i.e., proportion of trials correct; Figure 3) yielded a significant main effect of congruency [F(1, 46) = 41.181, p < .001, η_p² = .472], with participants performing more poorly on incongruent trials than congruent trials. Additionally, the main effect of cue was significant [F(1.891, 86.976) = 3.878, p = .026, η_p² = .078], and pairwise comparisons revealed that accuracy was significantly higher following PI cues than PC cues [t(47) = 2.643, p_Holm = .033]. Accuracy was also higher following PI than PE cues, but this result was not statistically reliable [t(47) = 1.826, p_Holm = .148], and PE and PC cues did not differ significantly with respect to their effect on accuracy [t(47) = 1.094, p_Holm = .279]. The Cue × Congruency interaction, however, was not significant [F(1.96, 90.17) = 2.493, p = .089, η_p² = .051]. The main effect of age was not significant [F(1, 46) = 1.184, p = .282, η_p² = .025] and did not interact with cue [F(1.89, 86.98) = 1.173, p = .312, η_p² = .025] or congruency [F(1, 46) = 0.037, p = .849, η_p² = .001]. Lastly, the Age × Cue × Congruency interaction failed to reach the significance threshold, and Bayesian analysis indicates substantial evidence against its inclusion [F(1.96, 90.17) = 1.595, p = .209, η_p² = .034, BF_incl = 4.99 × 10⁻⁵].⁵

Inverse Efficiency Score

A significant main effect of age [F(1, 46) = 20.340, p < .001, η_p² = .307] indicated that younger adults exhibited faster accuracy-adjusted RTs than older adults (Figure 3). And, as expected, a significant main effect of congruency [F(1, 46) = 82.60, p < .001, η_p² = .642] demonstrated that accuracy-adjusted RTs were slower on incongruent trials than congruent trials. Additionally, the Cue × Congruency interaction [F(1.413, 64.996) = 10.179, p = .001, η_p² = .181] revealed, once again, that the PC_CE was significantly larger than the PI_CE [t(47) = 3.519, p_Holm = .002] and the PE_CE was significantly larger than the PI_CE [t(47) = 3.673, p_Holm = .002]. However, while the PC_CE was larger than the PE_CE (Figure 4), this result did not reach the significance threshold [t(47) = 1.880, p_Holm = .066]. The main effect of cue approached but did not reach significance [F(1.315, 60.491) = 2.894, p = .083, η_p² = .059] and did not interact with age [F(1.315, 60.491) = 1.380, p = 0.253, η_p² = .029]. Finally, the Age × Congruency [F(1, 46) = 2.802, p = 0.101, η_p² = .057] and the Age × Cue × Congruency interactions [F(1.413, 64.996) = 2.963, p = 0.076, η_p² = .061, BF_incl = .002] did not exceed the significance threshold.⁶

Event-Related Potentials

As a statistical adjustment for multiple comparisons, we report the Holm-Bonferroni-corrected p-values (p_Holm) for all pairwise comparisons.

Contingent Negative Variation (CNV)

Neither the main effect of cue [F(1.923, 88.453) = 1.136, p = .324, η_p² = .024] nor its interaction with age [F(1.923, 88.453) = .008, p = .990, η_p² = 1.78 × 10⁻⁴] exerted significant effects on the amplitude of the CNV. However, a significant main effect of age was observed [F(1, 46) = 5.344, p = .025, η_p² = .104], with older adults displaying larger (more negative) CNV amplitudes (M = −4.05 μV) than younger adults (M = −2.16 μV). While these results suggest a lack of differential cue-based preparation at this stage, the main effect of age suggests increased effortful, preparatory processing prior to the upcoming imperative stimulus for older adults relative to younger adults, perhaps to compensate for age-related deficits in working memory and proactive control (Figure 5).

Figure 5:

Top: Grand average ERP waveforms time-locked to cue onset for younger (YA; solid lines) and older adults (OA; dashed lines). Red, green, and blue lines represent ERP waveforms elicited by predict-congruent (PC), predict-equiprobable (PE), and predict-incongruent (PI) cues, respectively. The interval encompassed by the black rectangle indicates the time window (1200–1498 ms) used to measure the mean amplitude of the contingent negative variation (CNV). Bottom: Topographic plots of mean CNV amplitude from 1200–1498 ms for younger (left column) and older adults (right column).

LRP Congruency Effect

The average difference between the LRP for congruent and incongruent correct trials is presented in Figure 6 (and Supplementary Figure S2), separately for each cue condition and for younger and older adults. The waveforms show an early positive deflection, reflecting activation of the correct response for congruent trials and of the incorrect response for incongruent trials, at latencies < 350 ms in younger adults and < 500 ms in older adults. This deflection is consistent with the proposal that relatively early processing stages favor the response indexed by the majority of features present in the stimulus (see Gratton et al., 1992). The waveforms suggest, though, that this effect of early stimulus analysis on response activation is greater for PC cues than for the other types of cues. The waveforms also suggest that the peak latency of the effect of early stimulus analysis on response activation is affected by age, but not by cue type.

Figure 6:

Grand average lateralized readiness potential (LRP) congruency effect waveforms (CE; incongruent minus congruent) time-locked to target (flanker array) onset. Red, green, and blue lines indicate the predict-congruent (PC), predict-equiprobable (PE), and predict-incongruent (PI) conditions, respectively. For both younger and older adults, the early, positive-going (downward) deflection occurs when the difference in response activation levels on congruent trials (correct activation > incorrect activation) is larger (i.e., more negative) than the difference in response activation levels on incongruent trials (incorrect activation > correct activation). Rectangles indicate the time windows during which local peak amplitude measurements were taken (200–500 ms).

Amplitude.

Consistent with these visual impressions, we observed a significant main effect of cue on the amplitude of the LRP congruency effect [F(1.962, 90.236) = 4.962, p = .009, η_p² = .097] such that PC cues elicited larger LRP CEs than both PE (t(47) = 2.870, p_Holm = .018) and PI cues (t(47) = 2.308, p_Holm = .051), although the latter did not survive correction for multiple comparisons. There was not a statistically reliable difference (see Table 2 and Figure 7 for the relevant descriptive statistics and bar graphs) in the LRP CE between PE and PI cues (t(47) = −.651, p_Holm = .518). The main effect of age was not significant [F(1, 46) = 2.274, p = .138, η_p² = .047] and did not interact with cue [F(1.962, 90.236) = .772, p = .463, η_p² = .017, BF_incl = 1.193] — however, an inclusion Bayes factor of 1.193 provides weak evidence in favor of the two-way interaction effect.⁷ The main effect of cue indicates that PC cues induce greater activation of the incorrect response on incongruent trials than PI and PE cues, suggesting that participants utilize contextual information to modify their processing strategies in service of response activation; the absence of a significant cue by age interaction (with an F < 1) suggests that this phenomenon occurs to an extent that is relatively similar in younger and older adults.

Table 2:

Mean onset latencies (50% fractional peak latencies) in ms (standard errors) for LRP congruency effect waveforms

Cue Type	Younger Adults	Older Adults
Predict-Congruent	223.25 (6.04)	305.58 (14.32)
Predict-Equiprobable	229.75 (6.58)	299 (12.70)
Predict-Incongruent	216 (8.98)	306.67 (12.85)

Figure 7:

(a) The mean lateralized readiness potential (LRP) amplitude congruency effect (CE; difference in amplitude between incongruent and congruent trials) for younger (light) and older adults (dark). (b) Average local peak N2 amplitudes and (c) average local peak P3b amplitudes for each of the three cue types (PC, PE, PI) by age (younger — light colors, older — dark-colors) and congruency (congruent — solid, incongruent — striped). YA = younger adults; OA = older adults; Con = congruent; Inc = incongruent; PC = predict-congruent; PE = predict-equiprobable; PI = predict-incongruent.

Latency.

A significant main effect of age was observed [F(1, 46) = 44.651, p < .001, η_p² = .493], indicating an age-related lengthening of the processes leading to activation. In addition, neither the main effect of cue [F(1.672, 76.917) = .099, p = .873, η_p² = .002] nor its interaction with age were significant [F(1.672, 76.917) = .919, p = .388, η_p² = .020, BF_incl = .066]. The absence of a within-subject effect (i.e., cue) in the presence of a between-subject effect (i.e., age) suggests that the effect of cue on the speed of the processes leading up to response activation based on early stimulus analysis must be relatively minor, if anything.⁸

N2 Component

Stimulus-related ERP activity at FCz (the electrode at which conflict-related N2 is most commonly observed) is presented in Figure 8 for younger and older adults. This figure clearly reveals a negative deflection, peaking at around 300 ms latency, which we interpret as a “conflict” N2 (see Figure 9 for topographical maps of N2 amplitude by condition and age group at the time of their respective peak latencies, as indicated in Table 3). The N2 component of the ERP, which typically exhibits a frontocentral scalp topography, is sensitive to response conflict such that N2’s are larger (more negative) on incongruent (high conflict) than congruent (low conflict) trials.

Figure 8:

Younger (solid lines) and older adults’ (dashed lines) grand average event-related potential (ERP) waveforms time-locked to target (flanker array) onset for congruent and incongruent trials (red and black lines), following predict-congruent (PC), predict-equiprobable (PE), and predict-incongruent (PI) cues at electrodes FCz (left column) and Pz (right column). Rectangles indicate the time windows during which local peak amplitude measurements were taken (200–500 ms and 300–700 ms at FCz and Pz, respectively).

Figure 9:

Topographic plots of N2 amplitude for each of the six conditions at their respective peak latencies, as listed under each plot, for younger adults (top) and older adults (bottom). PC = predict-congruent; PE = predict-equiprobable; PI = predict-incongruent.

Table 3:

Mean local peak latencies in ms (standard errors) for the N2 and P3b ERP component

Component	Cue Type	Congruency	Younger Adults	Older Adults
N2	Predict-Congruent	Congruent	314 (11.49)	288 (9.48)
		Incongruent	316.83 (7.01)	300.08 (9.41)
	Predict-Equiprobable	Congruent	304.50 (10.18)	294.50 (10.97)
		Incongruent	314.17 (6.74)	297.92 (10.40)
	Predict-Incongruent	Congruent	312.50 (10.87)	289.58 (10.25)
		Incongruent	316.08 (7.21)	301.83 (13.61)
P3b	Predict-Congruent	Congruent	418.92 (12.69)	527.50 (9.40)
		Incongruent	489.08 (14.43)	546.08 (22.67)
	Predict-Equiprobable	Congruent	444.83 (18.43)	528.92 (12.22)
		Incongruent	466.50 (16.65)	557.75 (20.56)
	Predict-Incongruent	Congruent	444.92 (20.20)	531.58 (12.95)
		Incongruent	477 (15.57)	545.58 (24.21)

Amplitude.

The main effect of congruency was significant, indicating that N2 amplitude was more negative on incongruent than congruent trials [F(1, 46) = 18.022, p < .001, η_p² = .281], thereby replicating the conflict N2 effect (see Figure 7 for bar graphs and Figure 8 for younger and older adults’ waveforms). We also observed a significant main effect of age [F(1, 46) = 17.869, p < .001, η_p² = .280] such that younger adults’ N2 amplitudes were significantly more negative than older adults’. However, this effect seems to emerge around the earlier P2 window, 150–250 ms (for these analyses, see Supplementary Material). However, the Age × Congruency interaction was not statistically significant [F(1, 46) = 4.028, p = .051, η_p² = .081]. A significant Cue × Congruency interaction [F(1.818, 83.627) = 6.581, p =.003, η_p²= .125] was observed, but the three-way interaction with age was not significant [F(1.818, 83.627) = 2.727, p = .076, η_p² = .056, BF_incl = .216]. To resolve the two-way interaction, for each subject, we computed the congruency effect for each cue and conducted pairwise comparisons. These post-hoc analyses confirmed that the PC_CE was larger than the PI_CE [t(47) = −3.264, p_Holm = .006] and PE_CE [t(47) = −2.107, p_Holm = .081], clearly demonstrating the effect of cue-induced expectations on N2 amplitude and suggesting that the amount of conflict experienced can be modified by prior contextual information (although the latter comparison did not survive an adjustment for multiple comparisons). While the PE_CE was larger than the PI_CE, this difference was not significant [t(47) = −1.509, p_Holm = .138]. Lastly, the main effect of cue was not statistically significant [F(1.966, 90.451) = 0.828, p = .438, η_p² = .018] and did not interact with age [F(1.966, 90.451) = 1.125, p = .328, η_p² = .024].

Latency.

Analysis of N2 peak latency did not reveal any significant main effects or interactions (see Table 3 for descriptive statistics). The main effect of congruency [F(1, 46) = 1.996, p = .164, η_p² = .042] was not significant. The main effect of age was not significant [F(1, 46) =, p = .127, η_p² = .050], and the Age × Congruency interaction failed to reach statistical significance [F(1, 46) = .141, p = .709, η_p² = .003]. The main effect of cue also failed to reach significance [F(1.916, 88.158) = .185, p = .823, η_p² = .004] and did not interact with age [F(1.916, 62.921) = .549, p = .572, η_p² = .012]. Lastly, neither the Cue × Congruency interaction [F(1.946, 89.512) = .015, p = .984, η_p² < .001] nor the Age × Cue × Congruency interaction [F(1.946, 89.512) = .589, p = .553, η_p² = .013, BF_incl = 3.09 × 10⁻⁵] were statistically significant.⁹

P3b Component

The average waveforms at Pz (the electrode most commonly used to examine this component) are presented in Figure 8. The waveforms shows a large positive deflection peaking at a latency of approximately 400–500 ms, which we interpret as the P3b (see Figure 10 for topographical maps of P3b amplitude by condition and age group at the time of their respective peak latencies, as indicated in Table 3). The P3b component of the ERP, which typically exhibits a posterior-parietal scalp topography, indexes updating of working memory contents and is sensitive to manipulations of stimulus probability.

Figure 10:

Topographic plots of P3b amplitude for each of the six conditions at their respective peak latencies, as listed under each plot, for younger adults (top) and older adults (bottom). PC = predict-congruent; PE = predict-equiprobable; PI = predict-incongruent.

Amplitude.

P3b amplitude was significantly larger on congruent than incongruent trials [F(1, 46) = 35.678, p < .001, η_p² = .437]. The Age × Congruency interaction was not statistically significant [F(1, 46) = 3.824, p = .057, η_p² = .077], but a significant main effect of age was observed [F(1, 46) = 4.301, p = .044, η_p² = .086], with older adults exhibiting larger P3bs than younger adults (see Figures 7 and 8 for bar graphs and waveforms, respectively). In addition, while we did observe a significant Cue × Congruency interaction [F(1.664, 76.526) = 5.751, p = .007, η_p² = .111], this interaction was qualified by a significant Age × Cue × Congruency interaction [F(1.664, 76.526) = 4.889, p = .014, η_p² = .096, BF_incl = .016], with follow-up analyses showing a significant Cue × Congruency interaction for older adults [F(1.627, 37.421) = 10.104, p = .001, η_p² = .305], but not younger adults [F(1.634, 37.586) = .935, p = .384, η_p² = .039]. Amongst older adults, post-hoc comparisons showed that the PI_CE was significantly larger than the PC_CE (t(23) = −3.691, p_Holm = .004) and PE_CE (t(23) = −2.821, p_Holm = .019). Additionally, the PE_CE was significantly larger than the PC_CE (t(23) = −2.326, p_Holm = .029). However, an inclusion Bayes factor of .016 provides strong evidence against the three-way interaction effect, so these post-hoc results should be interpreted with caution. Lastly, neither the main effect of cue [F(1.887, 86.787) = 1.541, p = .221, η_p² = .032] nor its interaction with age [F(1.887, 86.787) = .419, p = .647, η_p² = .009] were significant.

Latency.

For P3b peak latency, the main effect of congruency was statistically significant [F(1, 46) = 6.196, p = .016, η_p² = .119], with the latency of the incongruent P3b delayed compared to the congruent P3b, presumably reflecting prolongation of the stimulus evaluation process (see Table 3 for descriptive statistics). The P3b peaked significantly later for older than younger adults [F(1, 46) = 23.205, p < .001, η_p² = .335], presumably reflecting reductions in processing speed for older adults, but age did not interact with congruency [F(1, 46) = .705, p = .406, η_p² = .015]. Moreover, the main effect of cue was not significant [F(1.818, 83.644) = .216, p = .785, η_p² = .005] and did not interact with age [F(1.818, 83.644) = .226, p = .777, η_p² = .005] or congruency [F(1.817, 83.564) = 1.288, p = .280, η_p² = .027]. Finally, the Age × Cue × Congruency interaction did not reach the significance threshold [F(1.817, 83.564) = 2.032, p = .142, η_p² = .042, BF_incl = 5.81 × 10⁻⁴].¹⁰

Discussion

The present study sought to conceptually replicate and extend the findings on strategy selection in conflict tasks from Gratton and colleagues (1992, Experiments 3a and 3b) to older adults. We were particularly interested in discovering whether older adults would adopt cue-dependent processing strategies, like younger adults. The presence of significant cue by congruency interactions for RT, IES, LRP, and N2 suggests that both younger and older adults adjust their processing settings in a cue-driven manner, such that PC and PI cues lead participants to favor information extracted during the early and late stages of stimulus evaluation, respectively. Moreover, these cue-driven attentional adjustments have clear behavioral and electrophysiological manifestations, such that PC cues lead participants to make a more extensive use of information extracted during the early, incomplete stage of stimulus processing, resulting in an increase of the RT, IES, LRP, and N2 congruency effects (CEs). Conversely, PI cues cause participants to prioritize information extracted during the late, complete stage of stimulus processing, resulting in a diminution of the RT, IES, LRP, and N2 CEs. The peak latency of the P3b component for both younger and older adults was significantly prolonged on incongruent trials compared to congruent trials, presumably reflecting the perceptual interference from incongruent flankers. The P3b peak latency for older adults was significantly prolonged compared to younger adults, which is consistent with age-related slowing. Lastly, while we did not observe a significant cue effect on CNV amplitude, the age effect was significant, perhaps indicating greater cognitive/motor preparation among older adults prior to imperative stimulus onset.

For RT, IES and LRP amplitude, the fact that age did not interact with the cue-dependent congruency effect (i.e., the three-way interaction was absent) is somewhat surprising. Based on the dual mechanisms of control (DMC) account (Braver, 2012; Braver & West, 2008) and older adults’ diminished working memory capacities (which are essential for proactive control), one might expect them to struggle with updating and/or maintaining and subsequently utilizing cue information. Therefore, for older adults, the cue by congruency interaction should either be absent or significantly attenuated. The fact that the older adults in our study effectively adjusted their strategies as a function of cue is inconsistent with DMC predictions. However, the cue-flanker interval in our study was relatively short at only 999 ms, and therefore it was not particularly taxing for working memory maintenance. It is possible that if this interval were prolonged (e.g., to 5000 ms), or an additional memory-load was imposed, older adults would experience more problems maintaining cue-derived context representations in working memory, leading to performance that is modulated less by the cue. Alternatively, one could argue that our cue-flanker interval was sufficiently long to allow older adults to benefit from the cues, and that they would only exhibit proactive control deficits at relatively shorter intervals (e.g., 500 ms), perhaps due to inefficient updating of working memory contents and/or slower processing of cue information. Interestingly, though, this finding suggests that not all phenomena that can be attributed to proactive control are necessarily affected by aging. Perhaps, only conditions in which proactive control is limited by processing speed or working memory abilities, such as when updating or maintenance are significantly challenged, exhibit an age-related effect.

Analysis of the conflict-related N2 (Kopp et al., 1996; Van Veen & Carter, 2002; Yeung et al., 2004) revealed an enhanced N2 CE when a congruent array was expected (PC), which was driven by an increase in amplitude on incongruent trials and a decrease in amplitude on congruent trials. Conversely, we observed a relatively attenuated N2 CE when an incongruent array was expected (PI) or when congruent and incongruent arrays were equiprobable (PE). In contrast to these findings, employing a blocked probability (proportion-congruent) manipulation, Bartholow and colleagues (2005) observed a more complex pattern of results in which, at electrode Fz, the N2 CE was largest and only significant in the expect-incompatible blocks. However, at electrode Cz, the N2 CE was significant across all block types (i.e., expect-compatible, expect-neutral, and expect-incompatible), but the magnitude of the N2 CE did not differ between any of the three conditions. So, whereas Bartholow and colleagues tentatively concluded that the N2 solely indexed response conflict, our results suggest that the N2 indexes not only response conflict, but also conflict expectations. Furthermore, our findings dovetail nicely with other ERP studies of cognitive control in aging reporting that conflict N2 amplitude decreased with age (Hsieh et al., 2020; Wascher et al., 2011; Wild-Wall et al., 2008) but remained amenable to strategic, trial-to-trial adjustments of control (Larson et al., 2016).

Surprisingly, we did not observe the typical age-related reduction in P3b amplitude (Kropotov et al., 2016; Polich, 2007; Porcaro et al., 2019; Van der Lubbe & Verleger, 2002). Instead, similarly to CNV amplitudes, P3b amplitudes were larger for older adults than younger adults, which may reflect some underlying neural compensation, thereby allowing these older adults to perform similarly to younger adults. Although in line with several extant theories (Cabeza, 2002; Reuter-Lorenz & Cappell, 2008), this explanation is somewhat speculative in the context of this study. Additionally, P3bs elicited by congruent trials were significantly larger than those elicited by incongruent trials, which, in part, may be the result of increased latency jitter (i.e., single trial variability) on incongruent trials. Based on the grand average waveforms (Figure 8), this appears to be the case especially for older adults, who lack a discernible P3b peak for all incongruent conditions. However, among older adults, conflict expectations seemingly modulated P3b amplitude in an anticipated manner, with larger P3b CEs observed following PI cues than PE or PC cues. Moreover, if P3b amplitude partly indexes the perceived probability of an event occurring (Donchin, 1981), then one would expect a larger P3b on congruent trials preceded by PI than PC cues, as they are, contextually, less probable. Conversely, on incongruent trials, PC cues should elicit larger P3bs than PI cues, as they are, contextually, less probable. Once again, though, the results of these follow-up analyses should be treated with skepticism, as the inclusion Bayes factor strongly suggests that the P3b amplitude data are more likely to be observed under models without the three-way interaction effect than models with it. Finally, in alignment with previous studies, P3b latency was significantly prolonged for incongruent trials (compared to congruent trials) and older adults (compared to younger adults), presumably reflecting an extension of the stimulus categorization process and an age-related decrease in processing speed, respectively.

If we assume that the primary determinant of late CNV amplitude is the degree of motor preparation for an upcoming imperative stimulus (with anticipation/expectancy and working memory as lesser contributors; Leuthold et al., 2004), the absence of a cue-related effect on CNV may be noteworthy. Some theories of the conflict adaptation effect invoke the idea of a “hot hand” (Mayr et al., 2003) or re-instantiation of motor programs based on memory cues (Spapé et al., 2011). Both of these would predict that the existence of a Cue × Congruency interaction implies differences in motor preparation as a function of the cue. Specifically, following PC cues and in anticipation of an “easy”, low-conflict stimulus, participants might exhibit greater activity over motor cortex, reflecting a heightened readiness to respond. In contrast, following PI cues and in anticipation of a “hard”, high-conflict stimulus, participants might exhibit lesser activity over motor cortex, reflecting a reduced readiness to respond (or increased reluctance/hesitancy). Theoretically, pre-motor activity following PN cues, as indexed by late CNV amplitude, would reflect intermediate levels of motor preparation. However, the results of the current study do not support these predictions, suggesting that motor preparation was largely equivalent after the different types of cues, and favoring a cognitive control explanation for the Cue × Congruency interaction.

Another theory about CNV is that its amplitude may also reflect the maintenance of contextual information in working memory (Kray et al., 2005). Therefore, within this framework, we could say that the proportion of attentional resources allocated for maintenance of contextual information did not differ as a function of cue. In other words, the working memory load imposed by each cue type was roughly equivalent.

On a methodological note, it is also possible that our online high-pass filter (0.1 Hz) was too stringent, thereby attenuating ultra-low-frequency modulations of the late CNV. However, this is unlikely to have affected the results in a significant manner, since with the type of filter used in the current study, the CNV should not be attenuated more than a few percentage points.

As for the main effect of age in which CNV amplitudes were larger for older adults than younger adults, several other studies have reported similar findings (Kropotov et al., 2016; Miyamoto et al., 1998; Wild-wall et al., 2007), which the authors attributed to enhanced effortful, preparatory processing, perhaps to compensate for age-related decreases in neural efficiency. It should be noted that this age-related pattern is not universal, with some studies reporting larger CNVs for younger adults or no group differences (Bennett et al., 2004; Golob et al., 2005). Further research is needed to disentangle the extent to which motoric and various non-motoric processes, including stimulus expectancies and working memory, contribute to the CNV.¹¹

Several limitations of the current study should be noted. A possible problem is that, as is often the case with aging research conducted in college towns, our sample of older adults possessed higher-than-average levels of educational attainment, IQ and vocabulary performance, suggesting that they may not be representative of the broader population of older adults in the United States. The fact that our older adults were generally high performers may have contributed to the absence (in all analyses besides that of P3b amplitude) of an age-related effect on strategic control (i.e., the Age × Cue × Congruency interaction). Moreover, relatively well-preserved cognitive abilities — but especially working memory — may have allowed our older adults to perform similarly to their younger counterparts, thereby providing tentative evidence against the claim that strategic control declines with age. Given this, our findings may be limited in their generalizability to the broader community of older adults, as well as lower performing, non-college-bound younger adults. In addition, as our sample size was relatively modest (N = 48), we conducted a post-hoc power analysis (MorePower 6.0 statistical calculator; Campbell & Thompson, 2012) to reveal if our sample was sufficiently large to detect three-way interactions of a small-to-medium effect size (i.e., η_p² = .01–.06; Cohen, 1988). Given the average, observed effect size for all three-way interactions in the current study (η_p² = .046), results indicated that an N of approximately 104 would be required to attain the recommended statistical power of .80 at a nominal alpha of .05 to detect such effects. Therefore, while we can reasonably dismiss the presence of medium-high to large effects of age on proactive control in the present study (i.e., η_p² = .10–.14 — however, see P3b amplitude results), we cannot rule out the possibility of small-to-medium effects. Third, as we only sampled older adults aged 65 to 80, we cannot extend our conclusions to the oldest elderly individuals (80+). Lastly, the use of a cross-sectional research design, as opposed to a longitudinal design, reduces the strength of any age-related causal inferences one may draw from the current findings.

In conclusion, the current findings suggest that, like younger adults, healthy older adults can dynamically adjust their attentional biases in a trial-by-trial, cue-driven manner. The claim that strategic modulation of cognitive control remains at least partially intact during aging is supported by behavioral and electrophysiological results. These results clearly demonstrated that the congruency effect was larger following cues that predicted a congruent array as opposed to an incongruent array, with uninformative cues (i.e., 50/50) generally resulting in intermediate values. We believe that this graded congruency effect is the consequence of cue-dependent, strategic adjustments to the preference for information accumulated during the early, incomplete stage of stimulus evaluation versus the late, complete stage of stimulus evaluation. Therefore, the results of this study suggest that older adults may possess a relatively intact ability to proactively control their processing settings, at least when working memory is not overly taxed. These findings partially contradict research that suggests that proactive control, or the ability to maintain and utilize contextual representations to modulate attentional control, declines with age. Future research, with more widely representative samples, is needed to determine whether this apparent contradiction is due to the fact the paradigm used in the current study has low working memory requirements.

Highlights.

• We examined to what extent older adults flexibly adjust their processing strategies

• Response preparation differed as a function of expectation and conflict

• Contrary to prior research, older adults exhibited fairly intact proactive control

• Not all phenomena attributed to proactive control are necessarily affected by aging

• ERPs suggested differential processing of informative versus uninformative cues