An ERP investigation of age differences in the negativity bias for self-relevant and non-self-relevant stimuli

Abstract

As we age, we show increased attention and memory for positive versus negative information, and a key event-related potential (ERP) marker of emotion processing, the late positive potential (LPP), is sensitive to these changes. In young adults the emotion effect on the LPP is also quite sensitive to the self-relevance of stimuli. Here we investigated whether the shift toward positive stimuli with age would be magnified by self-relevance. Participants read two-sentence scenarios that were either self-relevant or non-self-relevant with a neutral, positive, or negative critical word in the second sentence. The LPP was largest for self-relevant negative information in young adults, with no significant effects of emotion for non-self-relevant scenarios. In contrast, older adults showed a smaller negativity bias, and the effect of emotion was not modulated by self-relevance. The three-way interaction of age, emotion, and self-relevance suggests that the presence of self-relevant stimuli may reduce or inhibit effects of emotion on the LPP in young adults, but that older adults do not show this effect to the same extent.

1. Introduction

Not all information that we encounter is processed to the same extent, and the factors that influence which stimuli are prioritized can change with age. In this paper, we present a study using event-related potentials (ERP) to examine how emotional valence and self-relevance affect stimulus prioritization during initial processing and how these processes are affected by age. We begin by discussing valence biases in cognitive processing across the lifespan and how these manifest in ERP. We then discuss the relationship of emotional processing to self-relevance and how self-relevance modulates the neural processes elicited by emotional stimuli, before introducing the present work.

1.1. Age and the positivity effect

A significant body of research suggests that younger adults tend to show increased attention to and better memory for negative stimuli versus positive and neutral stimuli (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; Rozin & Royzman, 2001; for important exceptions, see Sedikides & Skowronski, 2020). However, there is a shift in attention and memory toward positive stimuli in older age. This change—referred to as the “age-related positivity effect”—can be due to decreased attention and memory to negative stimuli and/or increased attention and memory to positive stimuli, and it can manifest as a decreased negativity bias or as a positivity bias in older participants (Carstensen & DeLiema, 2018; Reed, Chan, & Mikels, 2014).

There are two general approaches to explaining these changes with age. One approach situates the positivity effect in the context of general cognitive and neural decline with age. For example, Cacioppo, Bernston, Bechara, Tranel, and Hawkley (2011) argue that the positivity effect may result from decline in functioning of the amygdala, which leads to lower arousal experienced in response to negative stimuli (but not positive or neutral stimuli). The other approach, represented by Socioemotional Selectivity Theory (SST; Carstensen, Isaacowitz, & Charles, 1999; English & Carstensen, 2017) attributes the positivity effect to changing motivation with age. This theory proposes that as people age, their perception of decreasing future time horizons leads to different motivational goals. In the earlier portions of life, the acquisition of knowledge and self-improvement are particularly important. Under these goals, negative information may be prioritized because it is more relevant to updating knowledge and adapting behavior for the future (see discussion in Baumeister et al., 2001; Peeters & Czapinski, 1990). As we age and time horizons are reduced, we instead prioritize emotion regulation to maximize meaningful positive experiences with more immediate payoffs. Thus, the positivity effect is thought to reflect age differences in the motivated processing of emotional information, with older adults prioritizing the processing of positive over negative information.

1.2. ERP investigations of the positivity effect

In the event-related potentials (ERP) literature, the most consistent marker of the additional processing afforded to emotional stimuli is the late positive potential (LPP; Citron, 2012; Hajcak, Weinberg, MacNamara, & Foti, 2012). The LPP is centroparietally distributed, begins around 400 ms to words (often earlier to pictures), and usually lasts at least a few hundred ms. It is generally larger to both negative and positive stimuli than to neutral, but it is also very sensitive to context and to the goals of the participant. For example, as discussed below, non-self-relevant emotional stimuli may fail to modulate the LPP when encountered in a context with self-relevant stimuli. Many studies have shown that the pattern observed on the LPP depends on the task given to participants, demonstrating that the LPP is sensitive not just to the properties of stimuli, but also to the goals of participants (Delaney-Busch, Wilkie, & Kuperberg, 2016; Fields & Kuperberg, 2016; Fischler & Bradley, 2006; Naumann, Maier, Diedrich, Becker, & Bartussek, 1997). Perhaps most relevant to Socioemotional Selectivity Theory, several studies have shown that emotion regulation goals modulate LPP amplitude (Speed & Hajcak, 2018). For example, the LPP is reduced to negative pictures when participants are instructed to reduce their emotional reaction (Moser, Hajcak, Bukay, & Simons, 2006) or when they are told to reappraise images as less negative (Hajcak & Nieuwenhuis, 2006). Importantly, the LPP is also sensitive to individual differences; for example, a number of studies show that the LPP to negative stimuli is larger in those with mood disorders (Speed & Hajcak, 2018).

This combination of features suggests that the amplitude of the LPP reflects the relevance of a stimulus with regard to context- and participant-specific motivational goals (Hajcak & Foti, 2020). This makes the LPP a valuable neural marker for examination of the age-related positivity effect. If the positivity effect arises from age differences in controlled processing in the service of changing motivations with age, this would predict age differences in the LPP to positive versus negative stimuli. Young adults should show a larger LPP to negative than positive stimuli while older adults should show a larger LPP to positive stimuli, or should at least show a reduced LPP to negative stimuli compared to younger adults.

Indeed, several studies have shown a pattern consistent with this prediction. Wood and Kisley (2006) reported a negativity bias on the LPP (negative > positive) for young adults while older adults showed equivalent LPP amplitude to positive and negative (see also Kisley, Wood, & Burrows, 2007). Langeslag and van Strien (2009) report a similar pattern, but with some evidence of a positivity bias for older adults (see also Meng et al., 2015; Pehlivanoglu & Verhaeghen, 2019). Mathieu et al. (2014) show a negativity bias for both older and young adults when negative stimuli are highly arousing (compared to moderately arousing positive stimuli), but only young adults showed a negativity bias with less arousing negative stimuli. Taken together, these studies suggest that emotional processing as indexed by the LPP is reduced to negative stimuli as people age (but see Renfroe, Bradley, Sege, & Bowers, 2016).

1.3. Self-relevance, emotion, and the LPP

Although they have generally been studied separately, self-relevance and emotion have clear overlap. We are most likely to have emotional reactions to events and information that are self-relevant (e.g., Brunyé, Ditman, Mahoney, & Taylor, 2011; Grezes et al., 2013), and the mere fact that we care enough about something for it to produce a strong emotional reaction in some sense makes it self-relevant. Indeed, some theories of emotion propose that stimuli must have some degree of self-relevance to be emotional (Lazarus, 1991). Behaviorally, both self-relevant and emotional stimuli are more likely to be attended to and remembered (Compton, 2003; Cunningham & Turk, 2017; Kensinger & Schacter, 2016; Symons & Johnson, 1997). There is evidence that these enhancements for self-relevance and emotion may be due to partially overlapping neural mechanisms (Gutchess & Kensinger, 2018) and that memory for such stimuli may be better preserved with age than memory for neutral, non-self-relevant stimuli (Kensinger, Allard, & Krendl, 2014; Kensinger & Gutchess, 2015).

In the ERP literature, self-relevance often elicits a late positivity that is very similar to the LPP seen to emotional stimuli (reviewed in Knyazev, 2013). For example, a larger posterior positivity is seen to participants’ own name or face (e.g., Tacikowski & Nowicka, 2010), as well as to self-relevant words (Gray, Ambady, Lowenthal, & Deldin, 2004) and objects (Miyakoshi, Nomura, & Ohira, 2007). The impact of self-relevance on the LPP elicited by emotional stimuli has also been examined in a number of studies. These studies have generally shown that the effect of emotion on the LPP is larger for self-relevant stimuli (Fields & Kuperberg, 2016; Herbert, Herbert, Ethofer, & Pauli, 2011; Herbert, Pauli, & Herbert, 2011; Li & Han, 2010; Pinheiro et al., 2016; Schindler, Wegrzyn, Steppacher, & Kissler, 2014; Shestyuk & Deldin, 2010; but see Fields & Kuperberg, 2012). Interestingly, most of these studies fail to show an emotion effect (negative/positive > neutral) for non-self-relevant stimuli, even while using stimuli that have been shown to generate an emotion effect in other studies. This highlights the context sensitivity of the LPP and suggests that non-self-relevant emotional stimuli may not be prioritized for processing when more motivationally relevant self-relevant emotional stimuli are present in the environment (see the Discussion section of this paper and Fields & Kuperberg, 2016).

1.4. The present study

Summarizing the literature reviewed above, the LPP that is commonly observed to emotional stimuli is modulated by self-relevance and may be a marker of a shared mechanism that leads to enhanced processing for both emotional and self-relevant stimuli. This shared mechanism may be relatively well-preserved with age, accounting for the fact that memory declines less for socioaffective stimuli with age. It has also been established that the LPP is sensitive to motivational differences with age that lead to differential processing of positive versus negative stimuli. However, while a number of previous studies have examined the interaction of self-relevance and emotion on the LPP or the interaction of age and emotion on the LPP, no study to date has examined age, self-relevance, and emotion together using ERPs.

Here we employed a paradigm that has previously shown interactions of self-relevance and emotion on the LPP in younger adults (Fields & Kuperberg, 2012, 2016) to compare the pattern of this interaction for younger and older adults. Participants read two-sentence scenarios with a neutral, positive, or negative outcome in the second sentence. Self-relevance was varied by changing the subject of the second sentence from a person’s name to “you”. For example: “A man knocks on Sandra’s/your hotel room door. Sandra/You see(s) that he has a tray/gift/gun in his hand”. This design is based on research showing that sentences in the second person lead to mental models built from the reader’s own perspective (Brunyé, Ditman, Mahoney, Augustyn, & Taylor, 2009; Brunyé et al., 2011; Brunyé, Taylor, Gardony, Ditman, & Giles, 2013). ERPs were recorded to the neutral, positive, or negative critical word (underlined in the preceding example) in the second sentence.

We expected that older and younger adults would show differences in how they processed positive vs. negative words, with younger adults (but not older) showing larger LPP amplitude to negative words than positive words. This would be consistent with a number of previous ERP studies and a large behavioral literature as reviewed above. Our primary question in the present work was whether this pattern would be modulated by self-relevance. Given that the positivity effect is theorized to emerge largely from motivational factors, we might expect that valence differences across age are particularly pronounced for self-relevant stimuli. More specifically, if the negativity bias in young adults serves the goal of knowledge acquisition as proposed by SST, it should be particularly large for self-relevant stimuli, which are the stimuli that it would be most important to learn. A similar effect may be expected for the positivity effect in older adults, with a particular bias toward self-relevant positive stimuli, as these are the stimuli that are most relevant for emotional well-being.

2. Methods

2.1. Participants

Participants were 51 young adults (18–31 years old) and 48 older adults (60–86 years old) recruited from paper fliers and electronic advertisements posted throughout the greater Boston area and by contacting individuals who had asked to be informed of new studies in our laboratory. Of these, data from 8 young adult participants and 8 older adult participants were unusable due to excessive EEG artifact (>25% of trials rejected; see description of artifact procedures below) or technical problems with the EEG recording. This left 43 young (24 female) and 40 older (26 female) participants included in all results reported below. Participants were administered a battery of cognitive tests, the results of which are presented in the Supplementary Materials.

2.2. Stimuli

Stimuli were the same as those used in previous studies that have shown interactions between self-relevance and emotion on the LPP in younger adults (Fields & Kuperberg, 2012, 2016). In brief, stimuli consisted of 222 sets of two-sentence scenarios with Emotion (neutral, positive, and negative) and Self-Relevance (self and other) conditions crossed in a 3 × 2 factorial design. The first sentence introduced a situation involving one or more people, only one of which was specifically named (evenly split between male and female names); this sentence was always neutral or ambiguous in valence. The second sentence continued the scenario and was the same across all emotion conditions except for the critical word, which was pleasant, neutral, or unpleasant. To create the self condition, the named protagonist was changed to “you”. See Table 1 for examples.

Table 1. Examples of two-sentence scenarios in each of the six conditions.

The critical word is underlined (but did not appear underlined in the actual stimulus lists).

Other			Self
Neutral	Positive	Negative	Neutral	Positive	Negative
A man knocks on Sandra’s hotel room door. She sees that he has a tray in his hand.	A man knocks on Sandra’s hotel room door. She sees that he has a gift in his hand.	A man knocks on Sandra’s hotel room door. She sees that he has a gun in his hand.	A man knocks on your hotel room door. You see that he has a tray in his hand.	A man knocks on your hotel room door. You see that he has a gift in his hand.	A man knocks on your hotel room door. You see that he has a gun in his hand.
Fletcher writes a poem for a class. His classmates think it is a very intricate composition.	Fletcher writes a poem for a class. His classmates think it is a very beautiful composition.	Fletcher writes a poem for a class. His classmates think it is a very boring composition.	You write a poem for a class. Your classmates think it is a very intricate composition.	You write a poem for a class. Your classmates think it is a very beautiful composition.	You write a poem for a class. Your classmates think it is a very boring composition.
Vince spends time with his relatives over the vacation. This turns out to be a characteristic experience for him in many ways.	Vince spends time with his relatives over the vacation. This turns out to be a wonderful experience for him in many ways.	Vince spends time with his relatives over the vacation. This turns out to be a disastrous experience for him in many ways.	You spend time with your relatives over the vacation. This turns out to be a characteristic experience for you in many ways.	You spend time with your relatives over the vacation. This turns out to be a wonderful experience for you in many ways.	You spend time with your relatives over the vacation. This turns out to be a disastrous experience for you in many ways.
After dinner, Lydia is involved in a discussion. Many of her remarks surprise people.	After dinner, Lydia is involved in a discussion. Many of her remarks impress people.	After dinner, Lydia is involved in a discussion. Many of her remarks hurt people.	After dinner, you are involved in a discussion. Many of your remarks surprise people.	After dinner, you are involved in a discussion. Many of your remarks impress people.	After dinner, you are involved in a discussion. Many of your remarks hurt people.
Carmelo has been in his current job for over a year. He learns that he is getting a transfer this month.	Carmelo has been in his current job for over a year. He learns that he is getting a bonus this month.	Carmelo has been in his current job for over a year. He learns that he is getting a pay-cut this month.	You have been in your current job for over a year. You learn that you are getting a transfer this month.	You have been in your current job for over a year. You learn that you are getting a bonus this month.	You have been in your current job for over a year. You learn that you are getting a pay-cut this month.

Details of stimulus norms are reported in previous publications (Fields & Kuperberg, 2012, 2015, 2016), and are also summarized in the Supplementary Materials for this manuscript.

2.3. Procedure

2.3.1. Stimulus presentation

Lists were constructed such that each of the 222 scenarios appeared in a different condition in each of six lists (thus appearing in all conditions across lists). Each list had 37 trials in each of the six conditions. Each participant was randomly assigned to one of the six lists, and trial order within a list was fully randomized uniquely for each participant.

Trials were self-paced: each began with the word “READY” until the participant pressed a button to begin the trial. In each trial, the first sentence then appeared in full until the participant pressed a button to advance to the second sentence. The second sentence began with a fixation cross displayed for 500 ms, followed by an interstimulus interval (ISI) of 100 ms, followed by each word presented individually for 500 ms with an ISI of 100 ms (a schematic of stimulus presentation can be seen in Figure S2 in the Supplementary Materials). Participants were asked to refrain from blinking during the second sentence of each scenario (which contained the critical word), but no restrictions were given for other parts of the trial.

Because there is evidence that the positivity effect is strongest when task constraints are minimal (Reed et al., 2014), participants were given no task during initial stimulus presentation other than to silently read each scenario, but they were told that they would later be answering questions about the scenarios. After the EEG session, participants were given a memory test. This paper is not concerned with memory, but methods and results for the memory test are described in the Supplementary Materials to show that participants read and comprehended the scenarios during the EEG recording and that the manipulations of Valence and Self-Relevance produced expected behavioral effects.

2.3.2. Electroencephalographic recording

EEG was continuously recorded during the encoding session. Data was collected using a BioSemi ActiveTwo EEG system and ActiView v6.05 EEG acquisition software (http://www.biosemi.com/). The EEG was recorded from 32 Ag/AgCl electrodes in an elastic cap placed according to the international 10–20 system.¹ In addition, electrodes below and to the left of the left eye and above and to the right of the right eye were recorded to monitor for blinks and eye movements, and electrodes on each mastoid were recorded to serve as the reference. The EEG signal was amplified, filtered online with a low pass 5th order sinc response filter with a half-amplitude cutoff at 104 Hz, and continuously sampled at 512Hz.

2.4. Data processing and analysis

Code used for data processing can be found on the Open Science Framework page for this project: https://doi.org/10.17605/osf.io/egxbc.

EEG and ERP data processing was conducted in EEGLAB v14.1.1 (https://sccn.ucsd.edu/eeglab/index.php; Delorme & Makeig, 2004) and ERPLAB v6.1.4 (https://erpinfo.org/erplab; Lopez-Calderon & Luck, 2014).

The EEG was first referenced to the average of the two mastoid electrodes. Segments of the EEG with more than 10 seconds between event markers (representing breaks in the experiment) were automatically deleted. For each segment of continuous EEG, we removed the DC offset by subtracting the average voltage of the entire segment, then applied a high-pass 2nd-order Butterworth infinite impulse response filter with a half-amplitude cut-off of 0.1 Hz (Kappenman & Luck, 2010; Tanner, Morgan-Short, & Luck, 2015).

For the purposes of artifact correction, we performed independent components analysis (ICA) using the extended infomax algorithm (Lee, Girolami, & Sejnowski, 1999). ICA was performed on the continuous EEG (i.e., prior to epoching). Segments of EEG with significant artifact that was not neural, ocular, or muscular in origin were identified via visual inspection and excluded from the data submitted to the ICA training algorithm.

We extracted segments from 200 ms before to 1100 ms after all events of interest. The previously obtained ICA solution was applied to the segmented data. Components corresponding to ocular activity (blinks and saccades) were identified via visual inspection and removed (0–4 components per participant). We then used spherical spline interpolation as implemented in EEGLAB to replace channels with bad signal for a significant portion of the experiment. Interpolation was employed for 21 (12 young adults, 9 older adults) out of 83 participants with a maximum of two channels (always non-adjacent) interpolated for any given participant.

After independent components were removed and bad channels were replaced, epochs with remaining artifact were identified via artifact detection algorithms implemented in ERPLAB. The parameters of these algorithms (e.g., voltage thresholds) were tailored to each participant via visual inspection of the data, but were consistent across all conditions within each participant. Trials containing a blink or large saccade within the first 200 ms of a trial were rejected even if the artifact was corrected via ICA, as these trials may have a delayed neural response due to the eyes being closed or averted during stimulus presentation. Rejection rates ranged from 0 to 24.8% across participants with an average of 9.9%. Rejection rates did not significantly differ by Self-Relevance, Valence, Group or any interaction of these factors in an ANOVA (all ps > 0.19).

Importantly, all artifact correction and rejection procedures that involved experimenter decisions (i.e., removal of data from the ICA training set, which independent components to remove, which electrodes to interpolate, and the parameters for artifact detection algorithms) were determined by an experimenter blind to group membership.

After artifact correction and rejection was completed, trials not marked for artifact were averaged within conditions of interest to form ERPs.

2.4.1. Statistical analysis of ERP data

Recent simulation work has suggested that mass univariate analysis (see description below) provides the best balance of Type I error control, flexibility, and power for analysis of ERP data (see results and discussion in Fields & Kuperberg, 2020). The mass univariate approach is especially useful for the analysis of ERP components with variable timing. This is true of the LPP (Fischler & Bradley, 2006; Hajcak & Foti, 2020), which has been seen to vary by around 100 ms even within the paradigm employed here depending on the task (Fields & Kuperberg, 2016; see also Holt, Lynn, & Kuperberg, 2009). Also of concern in the present work was the possibility that the timing of effects would differ by age, as has been shown on several ERP components (e.g., Kutas & Iragui, 1998), and as suggested by visual examination of previously reported age effects on the LPP (Langeslag & van Strien, 2009; Meng et al., 2015; Wood & Kisley, 2006). This added uncertainty about timing is yet another reason to prefer a mass univariate approach for within-group analysis (since it allows for a data-driven approach to identifying when effects are present), but it presents a challenge for between-group analyses. Our goal was to test for differences in the amplitude of the LPP across age groups independent of any differences in timing. In a traditional mean amplitude approach, it is possible to simply use different time windows for the two groups, but the mass univariate approach tests at each time point individually and independently, so there is no way to compare groups independent of timing differences.

We therefore took a two-step, hybrid approach. First, we conducted mass univariate analyses to examine the effects of Valence and Self-Relevance within each group. We maximized power in these analyses by using a subset of electrodes and time points where the LPP was likely to appear, but these spatial and temporal choices were broad enough to capture the range of timing and scalp distributions seen in previous work (see below). The results of these mass univariate analyses allowed for a data-driven identification of where and when Valence and Self-Relevance effects were centered within each age group. These results were used to identify ROIs for each group, and the average amplitudes within these ROIs were then used to test for interactions of Valence and/or Self-Relevance with Group.²

2.4.1.1. Within Group Analyses

Statistical analysis of ERPs was conducted via the Mass Univariate Toolbox (Groppe, Urbach, & Kutas, 2011) and Factorial Mass Univariate Toolbox (Fields, 2017). We used a cluster-corrected mass univariate approach (Groppe et al., 2011; Maris & Oostenveld, 2007). Briefly, this approach consists of conducting an ANOVA independently at each time point and electrode of interest. Clusters are identified as adjacent time points/electrodes with effects surpassing a threshold, and all F-values in the cluster are summed to form a cluster mass statistic. A permutation approach is used to estimate the null distribution for this cluster statistic, which is then used to calculate a p-value for each cluster. We used the method of data reduction to construct the permutation test for our factorial design (Welch, 1990). Briefly, for each main effect we first averaged across the levels of the other factor, and for the Self-Relevance × Valence interaction we first calculated self – other difference waves for each level of the Valence factor. In all cases, this reduced the design to a one-way ANOVA, which allows for an exact permutation test (for a more detailed description, see the supplementary material for Fields & Kuperberg, 2020).

For all analyses, the F-value that would give p = .01 in an uncorrected parametric test was used as the threshold for cluster inclusion and electrodes within approximately 7.5 cm of each other (assuming a head circumference of 56 cm) were considered neighbors. 100,000 permutations were performed for each test. Statistical analysis was conducted on data low-pass filtered at 10 Hz (half amplitude cut-off, 2nd-order Butterworth infinite impulse response filter) to reduce the impact of high frequency noise and downsampled to 128 Hz (Groppe et al., 2011; Luck, 2014, Ch. 13).

To maximize power (Fields & Kuperberg, 2020), we conducted mass univariate analysis in a subset of electrodes and time points within which the LPP was likely to appear. These analysis parameters were chosen based on previous literature and examination of the aggregated grand average from trials (AGAT): the simple mean of all trials across all conditions and participants, which is an unbiased method of parameter selection (Brooks, Zoumpoulaki, & Bowman, 2017; Luck & Gaspelin, 2017). All analyses were conducted at a group of midline frontal, central, and parietal electrodes (Fz, FC1, FC2, Cz, C3, C4, CP1, CP2, Pz, P3, P4), reflecting the fact that effects of interest had previously been observed at both frontal and posterior sites. For young adults, a time window of 500–900 ms was used for the LPP. The N400 component that precedes the LPP is known to be delayed with age (Kutas & Iragui, 1998), and visual inspection of previous results suggested that the LPP may also be delayed with age (Langeslag & van Strien, 2009; Meng et al., 2015; Wood & Kisley, 2006). Based on the AGAT and previous results, the LPP was examined in a 550–1000 time window for older adults. We also conducted analyses on the N400 component and exploratory analyses across all time points and electrodes, which are reported in the Supplementary Materials.

2.4.1.2. Between Group Analyses

As noted above, results of these mass univariate analyses within each group were used to determine ROIs for between-group comparisons. The mean amplitude in these ROIs was used to examine the interaction of each effect with Age via a Valence × Self-Relevance × Age Group split plot ANOVA calculated in jamovi (https://www.jamovi.org/).

3. Results

Results of the post-EEG memory test, additional ERP analyses, and additional figures are available in the Supplementary Materials. Full results of all mass univariate analyses can be found on the Open Science Framework page for this project: https://doi.org/10.17605/osf.io/egxbc.

3.1. Young Adults

3.1.1. Visual examination and description of ERP results

As shown in Figure 1, visual examination of the young adult results showed that the LPP was larger to the Self-Negative condition compared to all other conditions at centroparietal sites starting around 400 ms and continuing to the end of the epoch (1100 ms). All other conditions elicited similar LPP amplitude, although there was some evidence of an increased LPP to the Other-Negative condition as well. It should be noted that although there were multiple local peaks within the LPP time window, these are unlikely to represent subcomponents of the LPP: The critical word was always mid-sentence and the following word was presented at 600 ms; thus, the peaks in the LPP time windows are likely the early ERP response to the following word (which was matched across conditions).

Figure 1. ERP Results.

Waveforms show all six conditions for each age group separately at posterior sites where the LPP is usually largest. Scalp maps show the standard deviation of the Self-Other difference wave across the three Valence conditions (i.e., the numerator of the Cohen’s f effect size for the interaction effect) calculated from the mean amplitude in 100 ms time windows. Figures showing main effects in each group and follow-ups for the Self-Relevance × Valence interaction in young adults are available in the Supplementary Materials.

3.1.2. Mass univariate analysis

Results of the mass univariate analysis in the LPP time window are shown in Figure 2. There were significant clusters for the main effect of Self-Relevance from the beginning of the time window to 650 ms (p = .020) and for Valence (p < .001) in a cluster that that spanned the full analysis time window (waveforms and scalp maps for main effects are shown in Figures S3 and S4 in Supplementary Materials).

Figure 2. Mass univariate analysis of ERP data.

Time point/electrode combinations not included in a significant cluster are in gray. For locations included in a cluster, the color represents F-statistic at that location.

There was also a significant cluster for the interaction effect from 604 to 650 ms (p = .024). We followed-up the interaction effect by examining the effect of Valence within the Other and Self conditions separately (see Figures S5 and S6 in the Supplementary Materials). The Valence effect was not significant within the Other condition (all clusters: p > .08), but was significant in a cluster spanning the entire time window for the Self condition (p < .001). Further follow-ups within the Self condition revealed that the LPP was larger to the Negative condition compared to the Positive (p < .001) and Neutral (p < .001) conditions, which did not significantly differ (no clusters found).

3.2. Older adults

3.2.1. Visual examination and description of ERP results

As shown in Figure 1 (and Figures S7 and S8 in the Supplementary Materials), visual inspection of the ERPs showed an effect of Valence with both Positive and Negative scenarios eliciting a larger LPP at centroparietal sites starting around 500 ms and continuing to the end of the epoch. There was also a smaller effect of Self-Relevance with Self scenarios eliciting a larger LPP from around 300 ms to 1000 ms at right central sites. These effects appeared to be largely additive rather than interacting.

3.2.2. Mass univariate analysis

Results of the mass univariate analysis in the LPP time window are shown in Figure 2. There was a significant cluster for the main effect of Valence from the beginning of the time window to 932 ms (p < .001). Follow-up analyses showed that both the Negative (p < .001) and Positive conditions (two clusters: p = .005 and p = .031) elicited a larger LPP than the Neutral condition for most of the time window at a broad range of electrodes. In addition, the Negative condition elicited a larger LPP than the Positive condition from 588 to 807 ms (p = .019). There was also a significant cluster for the main effect of Self-Relevance that spanned the entire time window at a set of midline and right central and parietal electrodes (p = .005).

In contrast to the results seen in young adults, the Self-Relevance × Valence interaction was not significant in older adults (one cluster: p = .299). Follow-up tests of the effect of Valence separately in the self and other conditions confirmed that, unlike the young adults, older adults showed significant effects of valence in both conditions (both ps = .001; see Figure S9 and S10 in the Supplementary Materials).

3.3. Interactions of Self-Relevance and Valence with Age

To examine interactions with Age Group, mean amplitudes were calculated from ROIs determined from the mass univariate analyses within each group (see Methods). These mean amplitudes were submitted to an Age × Self-Relevance × Valence ANOVA. The Greenhouse-Geisser correction was applied to the degrees of freedom for all analyses. Generalized omega squared (ω_G²) is reported as a measure of effect size (Olejnik & Algina, 2003). For the purposes of calculating ω_G², age was not considered a measured factor so that results are comparable to studies examining only one age group.

The main effects of Valence and Self-Relevance on the LPP were largest at CP2 for young adults and C4 for older adults. As these are adjacent electrodes and the scalp distributions were generally similar, we simply averaged these two electrodes for both groups. To account for the delayed effects for older adults, we used a time window of 500–900 ms for the young participants and 600–1000 ms for the older participants. The ANOVA showed that Age did not significantly modulate the effect of Self-Relevance [F(1, 81) = 0.04, p = .839, ω_G² = −.001] or Valence [F(1.94, 157.48) = 0.59, p = .553, ω_G² = −.001].

For young adults, the Self-Relevance × Valence interaction on the LPP was largest at CP1 and CP5 from 604–650 ms. This interaction was not significant for older adults, so we identified the largest non-significant cluster. This cluster was also centered at CP1 and CP5 (see scalp maps in Figure 1) and spanned 549 to 596 ms. We therefore examined the Age Group × Self-Relevance × Valence interaction at CP1 and CP5 in a time window of 550–650 ms.³ These data are shown in Figure 3. The ANOVA showed a significant Age Group × SelfRelevance × Valence interaction [F(1.99, 161.59) = 9.59, p < .001, ω_G² = .012]. This analysis confirmed that the difference in the Self-Relevance × Valence interaction revealed in the within-subject analyses above is statistically significant.

Figure 3. Valence × Self-Relevance × Age interaction on the LPP.

Plotted is the amplitude (µV) averaged across CP1 and CP5 from 550–650 ms (see text). Error bars show the standard error of the mean. Dots show individual participants’ amplitude for each condition.

The within-group analyses reported above serve as one way to break down the pattern driving the three-way interaction. To examine whether the positivity effect differed by Self-Relevance, we also examined the Age × Valence interaction separately for other-relevant and self-relevant scenarios. There was an Age × Valence interaction for the Self-Relevant scenarios [F(1.96, 158.35) = 7.64, p < .001, ω_G² = .017], but not the Other-Relevant scenarios [F(1.96, 158.33) = 2.01, p < .139, ω_G² = .004]. Within the Self-Relevant scenarios, Welch’s t-tests revealed that the Negative - Neutral [t(78.97) = 2.53, p = .014] and Negative – Positive [t(80.72) = 3.76, p < .001] contrasts were significantly larger for young adults, whereas the Positive - Neutral contrast did not differ by Age [t(81) = −1.22, p = .228]. That is, the Age × Valence interaction within Self-Relevant scenarios was driven by the particularly large LPP to Self-Negative stimuli in young adults.

4. Discussion

We conducted a study examining the effect of self-relevance and emotional valence on the processing of social vignettes in younger and older adults. The results showed that the late positive potential (LPP) was particularly large for the self-relevant negative scenarios for young adults. In contrast, older adults showed a smaller negativity bias on the LPP and the effect of valence did not differ by self-relevance. We begin by discussing the findings in young adults in comparison to the previous literature, then we discuss the differences observed in older adults.

4.1. A negativity bias modulated by self-relevance in younger adults

Young adults showed a larger LPP to negative than to positive or neutral scenarios, and this effect was larger for self-relevant stimuli. That self-relevant stimuli would have a larger emotional impact is consistent with previous ERP studies examining the interaction of self-relevance and emotion on the LPP (Fields & Kuperberg, 2016; Herbert, Herbert, et al., 2011; Herbert, Pauli, et al., 2011; Li & Han, 2010; Pinheiro et al., 2016; Schindler et al., 2014; Shestyuk & Deldin, 2010). It is also consistent with the proposal from Socioemotional Selectivity Theory (SST) that valence biases are related to motivational goals and that the negativity bias serves a goal of knowledge acquisition: self-relevant information is surely the most important information for us to learn.

Like most of the previous studies examining self-relevant emotional stimuli in young adults, the effect of emotion on the LPP did not reach significance for the other-relevant scenarios. This contrasts with a large literature showing large effects of emotion on non-self-relevant stimuli when self-relevant stimuli are not included in the study (reviewed in Citron, 2012; Hajcak et al., 2012; Olofsson, Nordin, Sequeira, & Polich, 2008), including a study with stimuli very similar to the other-relevant stimuli used here (Holt et al., 2009). This highlights the context sensitivity of the LPP: apparently, when self-relevant stimuli are common in the experimental context, non-self-relevant stimuli are less likely to draw the additional processing reflected by the LPP, even when they are emotional.

A number of theoretical frameworks propose that enhanced processing for motivationally relevant stimuli, such as those that are emotional and/or self-relevant, is achieved, at least in part, via inhibition of processing of competing stimuli. For example, this is the core idea of Mather and Sutherland’s (2011) arousal-biased competition model, which proposes that emotional arousal serves to enhance processes by which some stimuli are selected for processing at the expense of other stimuli that are inhibited, because neural processing is fundamentally competitive (see also Mather, Clewett, Sakaki, & Harley, 2016). In the ERP literature both the P300, which is often thought to be closely related to the LPP (Fields & Kuperberg, 2016; Hajcak & Foti, 2020), and the LPP itself have been linked to such inhibitory processes (Brown, van Steenbergen, Band, de Rover, & Nieuwenhuis, 2012; Polich, 2007). Inhibition is often invoked to explain effects seen with directly competing stimuli, such as when emotional stimuli draw overt attention at the expense of neutral stimuli (Nummenmaa, Hyona, & Calvo, 2006) or when emotional stimuli are remembered at the expense of neutral backgrounds (Kensinger, Garoff-Eaton, & Schacter, 2007). However, similar effects can be seen across trials: attention and memory can be impaired for neutral stimuli presented before and after emotional stimuli (Schmidt & Schmidt, 2016; Weinberg & Hajcak, 2011), and memory is worse for neutral stimuli when they are presented in lists mixed with emotional stimuli than when they are presented in neutral-only lists (e.g., Barnacle, Tsivilis, Schaefer, & Talmi, 2018; Talmi, Luk, McGarry, & Moscovitch, 2007; Watts, Buratto, Brotherhood, Barnacle, & Schaefer, 2014).

Our results and those of previous studies suggest that the modulation of the emotion effect on the LPP by self-relevance may follow a similar pattern. Although these results have often been framed as self-relevance enhancing the effect of emotion, it seems more accurate to say that the presence of self-relevant scenarios in the experimental context inhibits the effect of emotion in non-self-relevant scenarios (see also discussion in Fields & Kuperberg, 2016).

4.2. A positivity effect with no modulation by self-relevance in older adults?

Older adults also showed the largest LPP to negative stimuli overall, but the difference between negative and positive was smaller in this group and, unlike young adults, older adults additionally showed an increased LPP for positive compared to neutral words. These results appear consistent with the literature on the positivity effect and previous ERP studies examining the effects of age on the LPP (Kisley et al., 2007; Langeslag & van Strien, 2009; Mathieu et al., 2014; Meng et al., 2015; Pehlivanoglu & Verhaeghen, 2019; Wood & Kisley, 2006). However, the interaction of Age and Valence was not statistically significant; instead, only the full Age × Valence × Self-Relevance interaction was significant. Follow-ups suggested two ways of looking at this interaction.

One way of summarizing the pattern driving the three-way interaction is that there was there was an Age × Valence interaction for the self-relevant scenarios, but not the other-relevant scenarios. Within the self-relevant scenarios, this interaction was driven by a particularly large LPP to the negative condition in young adults. This is generally in line with our prediction that the positivity effect (in this case, seen as a reduction in the negativity bias) would be larger for self-relevant stimuli. However, we did not necessarily expect the lack of positivity effect for the other-relevant stimuli, as several previous studies showing the positivity effect on the LPP did not use self-relevant stimuli.

Some insight into the LPP response to the other-relevant scenarios may be gained from the other way of summarizing the three-way interaction: self-relevance did not modulate the effect of valence in older adults, in contrast both to the present results in younger adults and to the results in several previous studies in younger adults as reviewed above. As can be seen in Figures 1 and 3 (see also Figures S9 and S10 in the Supplementary Materials), older adults did show some evidence of an interaction in approximately the same time window and spatial location as the interaction observed for younger adults, and the pattern was consistent with predictions related to the positivity effect: the LPP to positive and neutral scenarios increased with self-relevance while the LPP to negative scenarios decreased with self-relevance. But this modulating effect of self-relevance was clearly smaller for older adults: the interaction of self-relevance and valence did not reach significance for this group, and, in contrast to the young adults, the effect of valence was separately significant for both other and self conditions.

As noted above, a large literature suggests that the other-relevant stimuli would elicit an emotion effect on the LPP in young adults if self-relevant stimuli were not part of the broader context, so it seems that younger adults de-prioritize, or inhibit, non-self-relevant emotional stimuli when more salient self-relevant emotional stimuli are present in the broader context. Following from this interpretation of the young adult data, one explanation of the three-way interaction we observed is that older adults fail to adjust their goals based on the broader experimental context; that is, they simply processed the non-self-relevant scenarios the same way they would if there were not self-relevant scenarios in the stimulus set. This may be because they do not develop such context-dependent goals, or because they struggle to implement them. If the interaction effect seen in young adults reflects inhibition of the LPP effect in non-self-relevant stimuli, an attenuated interaction in older adults would be consistent with research and theory suggesting that many deficits that arise with aging can be explained by reduced capacity for inhibition (Hasher & Zacks, 1988; Lustig, Hasher, & Zacks, 2007; but see Rey-Mermet & Gade, 2018). Support for this idea comes from a recent study where younger and older adults showed similarly sized emotion effects on the LPP when images were task-relevant, but young adults showed greater suppression of the LPP to emotional images when these images were distractions from the main task (Pehlivanoglu & Verhaeghen, 2019).

4.3. Open questions and future directions

On the current evidence, our explanation of the three-way interaction of age, self-relevance, and valence remains speculative. Although several previous studies have reported a pattern similar to what we report for the young adults, this is the first study examining the interaction of self-relevance and valence in older adults. It will therefore be important for future research to examine whether older adults consistently fail to show modulation of the emotion effect on the LPP by self-relevance. One alternative explanation of our results is simply that the self-relevance manipulation was less effective for older adults. However, this possibility seems inconsistent with the full pattern of results: the expected main effects of self-relevance were observed for both the LPP and the memory data (see Supplementary Materials) in older adults, and there was no evidence that these effects of self-relevance differed meaningfully by group (ω_G² < .001 for all Age × Self-Relevance interactions). These findings are consistent with previous literature showing that self-related biases in perception, attention, and memory generally do not decline with age (Gutchess, Kensinger, & Schacter, 2007; Gutchess, Kensinger, Yoon, & Schacter, 2007; Hamami, Serbun, & Gutchess, 2011; Hess, 2014; Mattan, Quinn, Acaster, Jennings, & Rotshtein, 2017).

Another interesting question relates to valence effects in younger adults. While the negativity bias we observed in young adults is consistent with a large literature showing a general bias to negative stimuli in this age group (Baumeister et al., 2001; Carstensen & DeLiema, 2018; Rozin & Royzman, 2001), not all previous ERP studies have shown a larger LPP to negative stimuli. This includes studies examining self-relevant stimuli (Herbert, Herbert, et al., 2011; Pinheiro et al., 2016; Shestyuk & Deldin, 2010). This is likely due in part to variability in the stimuli employed (for comparison of positive and negative pictures from different categories, see Franken, Muris, Nijs, & van Strien, 2008; Weinberg & Hajcak, 2010), but previous studies using the same stimuli as the present work have also shown mixed evidence for the negativity bias (Fields & Kuperberg, 2012, 2016).

It is not entirely clear what accounts for this variability, but the task given to participants may be one important factor. Carstensen and colleagues have argued that age differences in valence biases are most likely to be seen when the experimental task does not impose its own goals that influence which stimuli are motivationally relevant (Carstensen & DeLiema, 2018; Reed & Carstensen, 2012), and a meta-analysis of relevant studies supported this proposition (Reed et al., 2014). It is also well-established that the LPP is quite sensitive to task-imposed goals (Delaney-Busch et al., 2016; Fields & Kuperberg, 2016; Fischler & Bradley, 2006). In contrast to previous studies using this paradigm, the present study gave participants no active task when reading the scenarios, which is consistent with the idea that the negativity bias in young adults is most likely to emerge under such circumstances. However, it is not clear that task demands can explain all findings in the literature (e.g., Herbert, Herbert, et al., 2011 showed a larger LPP to positive stimuli with no task), and further research will be needed to understand when a negativity bias is seen on the LPP in young adults.

With regard to the age-related positivity effect, it may be informative to examine the effect of age in paradigms where young adults have shown a positivity bias on the LPP. If older adults show an even larger positivity bias, this would suggest that these are simply paradigms where the positive stimuli are more salient than the negative stimuli (either inherently, or in relation to the task and context). If, on the other hand, the age-related shift is eliminated in these paradigms, it would suggest additional factors need to be taken into account to fully understand effects of valence and age on the LPP.

4.4. Summary and conclusions

The present work extends previous findings showing that differences in the processing of stimuli of differing valence across the lifespan can be seen on a basic neural marker of prioritized processing for emotional stimuli, the LPP. We observed a negativity bias in young adults that was reduced in older adults, as seen in previous studies. However, main effects of Valence did not interact with age; instead, only the full three-way interaction of Self-Relevance, Valence, and Age was significant. This interaction was driven by a positivity effect for self-relevant scenarios that was not seen on other-relevant scenarios, and by the fact that only the young adults showed a clear modulation of the emotion effect by self-relevance. At present, it’s not clear what accounts for this lack of interaction in the older adults, and it will need to be replicated in future work, but an intriguing possibility is that is has more to do with failure to deprioritize non-self-relevant emotional stimuli than a failure to increase processing for self-relevant emotional stimuli. In any case, these results touch on a number of important topics in cognitive aging research, and they add to a handful of existing studies to show that the LPP, as a neural marker of prioritized processing, can be useful for understanding motivational and cognitive changes in aging.

Highlights.

• Self-relevance modulated previously observed effects of valence on the late positive potential (LPP) of the event-related potential for younger adults, but not for older adults.

• Young adults showed an increased LPP specifically for negative words in self-relevant social vignettes.

• Older adults showed a reduced negativity bias on the LPP, and the effect of valence was not modulated by self-relevance.