Effort expenditure modulates feedback evaluations involving self–other agreement: evidence from brain potentials and neural oscillations

Abstract

The influence of effort expenditure on the subjective value in feedback involving material reward has been the focus of previous research. However, little is known about the impact of effort expenditure on subjective value evaluations when feedback involves reward that is produced in the context of social interaction (e.g. self–other agreement). Moreover, how effort expenditure influences confidence (second-order subjective value) in feedback evaluations remains unclear. Using electroencephalography, this study aimed to address these questions. Event-related potentials showed that, after exerting high effort, participants exhibited increased reward positivity difference in response to self–other (dis)agreement feedback. After exerting low effort, participants reported high confidence, and the self–other disagreement feedback evoked a larger P3a. Time–frequency analysis showed that the high-effort task evoked increased frontal midline theta power. In the low (vs. high)-effort task, the frontal midline delta power for self–other disagreement feedback was enhanced. These findings suggest that, at the early feedback evaluation stage, after exerting high effort, individuals exhibit an increased sensitivity of subjective value evaluation in response to self–other agreement feedback. At the later feedback evaluation stage, after completing the low-effort task, the self–other disagreement feedback violates the individuals’high confidence and leads to a metacognitive mismatch.

Introduction

Imagine how you would feel if, after having just taken a mathematical examination that required you to exert high effort (i.e. a hard exam), you immediately compared your answers with your classmates, and you were informed that your answer disagreed with the other’s answer. Conversely, after an examination in which you exerted low effort (i.e. an easy exam), think how you would feel if you knew your answer disagreed with another person’s answer. This ubiquitous scenario inspired us to explore how people process (or evaluate) feedback suggesting that their judgments agree or disagree with other individuals after the low-effort and high-effort tasks.

Effort refers to mental or physical activity required to achieve a specific goal (Eisenberger 1992; Inzlicht et al. 2018; Pan et al. 2023). Numerous studies have explored how effort expenditure influences feedback evaluations (Inzlicht et al. 2018; Umemoto et al. 2019; Harmon-Jones, Clarke, et al. 2020; Bowyer et al. 2021; Pan et al. 2023; Umemoto et al. 2023). Specifically, several studies have shown that, in cognitive tasks, exerting more effort can increase the subjective value associated with the reward outcome/feedback (Kelley et al. 2019; Harmon-Jones, Willoughby, et al. 2020; Bogdanov et al. 2022; Pan et al. 2023). On the contrary, other studies have suggested that the subjective value of reward is inversely related to the amount of effort required to obtain it, referred to as “effort discounting” (Botvinick et al. 2009; Ghods-Sharifi and Floresco 2010; Wu and Zheng 2023). These inconsistent findings regarding the effects of effort on subsequent reward evaluations are known as the “effort paradox.” Most studies have focused on the relationship between effort expenditure and subjective value attached to rewards (being presented in the form of feedback). In particular, many researchers have manipulated the “material reward,” such as monetary compensation (Zhang K et al. 2022; Pan et al. 2023), and investigated how effort expenditure influences the subjective value of feedback involving material reward (vs. no reward). In daily life, reward can be not only produced through monetary compensation but also can be produced in the context of social interactions, such as social connection or agreement (Davis et al. 2023; Li et al. 2023). Prior studies have found that the feedback showing agreeance with others’ opinions can evoke a sense of reward comparable to that of monetary reward (Klucharev et al. 2009; Zaki et al. 2011; Bhanji and Delgado 2014; Ruff and Fehr 2014). Little known is about how effort expenditure influences subjective value of feedback involving social interaction rewards. Exploring this can offer insights into the ubiquitous scenario mentioned above, namely, how effort expenditure influences subjective value evaluations in feedback involving self–other (dis)agreement.

In addition to subjective value associated with reward, the second-order subjective value (i.e. confidence) associated with the reward is also worthy of attention (Hagura et al. 2023). Recent studies have considered “confidence” as the second-order subjective value, which is a metacognitive belief about the subjective value-based judgments (Lebreton et al. 2015; Lopez-Persem et al. 2020; Mazor et al. 2020; Shapiro and Grafton 2020). The “confidence” is defined as subjective evaluation of one’s judgment in a task, resulting in a degree of certainty that the judgment is optimal (though it is not actually optimal) (Pouget et al. 2016; Rausch et al. 2020). Previous studies also found the “effort paradox” emerged in the relation between effort expenditure and confidence: some studies have proposed that individuals feel more confident in having responded correctly (i.e. receiving the reward) when they invested high effort (Arkes and Blumer 1985; Turner et al. 2021; Hagura et al. 2023). Conversely, other studies have demonstrated that individuals maintain a balance between expected accuracy (ACC) and cost (Levy and Glimcher 2012). Increasing the cost should reduce the expected ACC (i.e. confidence in judgments) (Hagura et al. 2023). In particular, the cost includes effort expenditure, and it is suggested that participants would have lower confidence after exerting high effort (Shapiro and Grafton 2020). However, similar to the question above, how effort expenditure influences second-order subjective value (confidence) evaluations in feedback involving self–other (dis)agreement remains uncharacterized.

The processing of feedback evaluations is an implicit processing rather than observable behavior. The development of neuroscience techniques such as event-related potentials (ERPs) offers insights into the implicit neural dynamics of feedback evaluation (Butterfield and Metcalfe 2006; Frömer et al. 2021). In particular, numerous ERP studies have observed that 2 ERP components reflect the variance of subjective value and confidence during feedback evaluations, namely, the reward positivity (RewP) and P3a. First, the RewP component is a neural index of reward processing, which includes reward valuation (Proudfit 2015; Glazer et al. 2018). The RewP, which is measured at fronto-medial electrodes, is a positive waveform that peaks at approximately 200–350 ms following the presentation of a reward (Holroyd et al. 2011; Proudfit 2015; Muir et al. 2021). The RewP is related to the subjective value of rewards (Umemoto et al. 2019; Pan et al. 2023). Emerging ERP evidence concerning the influence of effort expenditure on subjective value associated with rewards has shown that more positive RewP amplitudes were evoked after participants exerted high efforts, likely because investing in effort can increase an individual’s subjective value of a reward (Harmon-Jones, Clarke, et al. 2020; Harmon-Jones, Willoughby, et al. 2020; Bogdanov et al. 2022; Pan et al. 2023). However, other studies have observed attenuated RewP responses to reward stimuli after high-effort (vs. low-effort) tasks, likely because the subjective value associated with rewards was discounted by high effort expenditure (Bowyer et al. 2021; Wu and Zheng 2023). Following the RewP, P3a, a sub-component of P3 that peaks between 300 and 600 ms after feedback onset, is distributed in the frontocentral region (Delplanque et al. 2006; Polich 2007; Valt et al. 2020). The P3a (also labeled as “novelty P3”) represents the attention mechanisms engaged in evaluating incoming stimuli in the frontal brain region (Polich 2007). The P3a magnitude is associated with attentional capture by salient or distracting feedback stimuli, and enhanced attentional capture is reflected by enhanced P3a in the frontal lobe (Metcalfe et al. 2015). Previous studies have reported that the P3a is related to how individuals’ confidence is influenced, such that if individuals with high confidence in their judgments receive feedback suggesting that their judgments were incorrect (or fail to get a reward), the P3a increases because the violation of the high confidence leads to metacognitive mismatch (Butterfield and Mangels 2003; Butterfield and Metcalfe 2006). This metacognitive mismatch attracts more attentional resources (Butterfield and Mangels 2003; Butterfield and Metcalfe 2006; Ungan et al. 2019; Frömer et al. 2021).

In addition to time domain ERP analysis, the time–frequency analysis measures energy changes in the synchrony of the underlying neuronal populations at different frequencies during the feedback evaluations (Cohen et al. 2007; Wang et al. 2017), which can provide important insights into the effect of effort expenditure on feedback evaluations involving rewards (Umemoto et al. 2023). Previous time–frequency studies concerning the effects of effort expenditure on feedback evaluations have focused on frontal midline theta (FMT) and frontal midline delta (FMD) activity; in particular, FMT, consisting of 4–8-Hz neural oscillations distributed over frontal–central areas of the scalp, is examined during cognitive processing involving effort expenditure (Holroyd and Umemoto 2016; Umemoto et al. 2023). Increased FMT power is related to increases in effort expenditure (Arnau et al. 2023). In addition, FMD, consisting of 1–3-Hz neural oscillations distributed over frontal–central areas of the scalp, can be observed in the processing of metacognitive monitoring (Putman 2011; Harmony 2013). Increased FMD power reflects the monitoring of metacognitive mismatches (e.g. the violation of high confidence) because metacognitive mismatch captures increased attention, and increased FMD is related to allocating more attention to stimuli (Fazio and Marsh 2009; Metcalfe et al. 2012; Williams et al. 2018).

Based on these studies, we employed electroencephalography (EEG) methods (including ERP and time–frequency analyses) to examine how effort expenditure influences subjective value and confidence in feedback evaluations involving self–other (dis)agreement. Regarding subjective value evaluations (indexed by the RewP), we proposed 2 opposing hypotheses according to the “effort paradox”—H_a: after exerting high effort, participants might attach greater subjective value to the feedback showing self–other agreement (i.e. reward in the context of social interaction) and exhibit greater subjective value sensitivity (increased RewP difference) to whether they receive self–other agreement feedback; and H_b: after exerting low effort, participants might exhibit a greater sensitivity (increased RewP difference) to whether they receive self–other agreement feedback. Regarding confidence (indexed by the P3a), also based on the “effort paradox,” we proposed 2 opposing hypotheses—H_c: after exerting high effort, participants might have high confidence in their judgments, and the self–other disagreement feedback would violates this high confidence (larger P3a); and H_d: after exerting low effort, participants might have a high confidence in their judgments, and self–other disagreement feedback would violate this high confidence (larger P3a). Furthermore, no direct studies have examined the influence of effort expenditure on FMT during feedback evaluations involving reward; therefore, we explored how effort expenditure influences FMT during feedback evaluations involving social reward without concrete hypotheses. Additionally, since studies have shown that the FMD has similar indicator functions (Watts et al. 2017), it was predicted that the FMD might have the similar pattern as the P3a (H_e).

Materials and methods

Participants

The sample size was determined by a priori power analysis for a 2 (effort expenditure level: low- vs. high-effort) × 2 (self–other [dis]agreement) within-participants design, using G*Power 3.1.9 (F tests, analysis of variance [ANOVA]: repeated-measures, within factor, and power = 0.90; effect size f = 0.25; α = 0.05) (Faul et al. 2007). According to this power analysis, at least 30 participants would ensure 90% statistical power in the case of small to medium effect sizes. In other words, the power analysis revealed 90% statistical power with an alpha of 0.05 to obtain an effect size of 0.25 for a sample of n = 30. To ensure a sufficient sample size, 46 participants (20 women, M_age = 22.34 years, SD = 3.21) were recruited from the undergraduate students at Hunan Normal University. All participants reported normal or corrected-to-normal vision and were right-handed without traumatic brain injury. No one was colorblind. This study was conducted following the Declaration of Helsinki, and written informed consent was obtained from all participants. After the experiment, each participant was paid ¥60 as basic compensation, plus a preset bonus ranging from ¥10 to ¥30 (see details in Procedure section). The study protocol was approved by the Internal Research Ethics Committee of the Department of Psychology, Hunan Normal University, China (ethics approval number: 2020-068). Seven participants were excluded because of an excessive number of EEG artifacts (i.e. more than 50% of the trials were rejected) (n = 1), failure to complete the entire experiment (n = 4), and voicing suspicion about the cover story (n = 2). As a result, 39 participants were included in the final analysis (18 females, M_age = 21.28 years, SD = 2.49).

Task

For the manipulation of effort expenditure (low-effort vs. high-effort), we used a modified version of the number judgments task (Harmon-Jones, Clarke, et al. 2020; Harmon-Jones, Willoughby, et al. 2020). This task has been reported to effectively distinguish between effort expenditure levels. The specific manipulation methods are as follows:

Low-effort tasks

Numbers (1–9, excluding 5) were shown in yellow or blue font: when the number was presented in yellow, participants were required to judge them greater (press “F” key on the keyboard) or less (press “J” key) than 5. This was a magnitude judgment task. When the number was presented in blue, participants needed to judge them as odd (press “F” key) or even (press “J” key). This was a parity judgment task. Participants were required to make either the magnitude judgment or parity judgment in one block.

High-effort tasks

Numbers (11–19, excluding 14) were also shown in yellow or blue font: when the number was presented in yellow, participants were required to divide it by 2 and then judge whether this number was greater (press “F” key) or less (press “J” key) than 7, which was a magnitude judgment task. Here, we considered potentially biased responses when participants were performing the magnitude judgment task; the number of the “greater than 7” and “less than 7” trials was carefully balanced (Supplementary Material). When the number was presented in blue, participants needed to add 1 into this number and then judge if it was odd (press “F” key) or even (press “J” key), which was a parity judgment task. Participants were required to make both magnitude and parity judgments in each trial depending on the color (blue or yellow) of a given number in one block.

Procedure

Upon arrival, each participant was informed that a pair of people (i.e. the participant and a same-sex confederate unknown to each participant [actually pretended by a same-sex undergraduate from another university]) would complete the number judgment task together. All the participants were instructed that their judgments would be transferred to the local network and compared with their corresponding confederate’s judgments in each trial.

All participants were required to complete 4 blocks (2 low-effort and 2 high-effort task blocks) (The order of blocks was randomly presented across participants.). Each block was preceded by a task cue (lasting for 1000 ms). For low-effort task blocks, one low-effort task block was preceded by the yellow circle (yellow [RGB = 240, 228, 66]), which indicated that participants should make a magnitude judgment in the following trials; the other low-effort task block was preceded by a blue circle (blue [20, 154, 232] (The RGB of each colors were chosen to maximize ease of discrimination and suitability for people (Wong 2011))), which indicated that the participants should make parity judgment in subsequent trials. For high-effort task blocks, both high-effort task blocks were preceded by a half-blue and half-yellow circle, which indicated that participants should switch between magnitude and parity judgment in the following trials) (see Fig. 1A). There was a self-paced break between neighboring blocks.

Each block comprised 65 trials. As shown in Fig. 1B, each trial began with a white fixation (500 ms), and then 2 successive numbers were presented with an interval (500 ms): each number in one trial was presented until participants made a judgment (maximum of 800 ms), and if participants did not respond within 800 ms in either “number” surface, the trial was terminated and the sentence saying “too slow!” was presented on the screen (1000 ms). Only if the 2 successive numbers judgments were given, participants were asked to rate their subjective confidence degree to their judgments on a 9-point scale (“How confident are you that you made the previous number judgments optimally?”) (Explicit confidence ratings are commonly used within metacognitive research to assess the effectiveness of metacognitive monitoring, and it is generally elicited in an “online” fashion, after each item on a task (Double and Birney 2019).) (i.e. the judgments were correct; and the average speed of twice judgments was within the mean speed among normal undergraduates (Each participant was informed that among average undergraduates, the average speed of judgments for low-effort tasks was 600 ms, and that for high-effort tasks was 750 ms in the instruction.))?” from “1 = not confident at all” to “9 = extremely confident,” maximum of 1500 ms for rating. This rating was unknown to the confederate. Here, according to the definition of confidence (i.e. evaluating the level of certainty that the decision is optimal; Rausch et al. 2020), if participants clearly realized that they had pressed a wrong key on the keyboard, they should respond with “0” on the keyboard. This trial was terminated. Only when subjective ratings (1–9) were given, a sentence indicating “Comparing your responses…” lasted for 1000 ms, and then followed by a blank screen (800–1000 ms). Afterwards, feedback reflecting whether participants’ judgments agreed with the confederate was presented for 1000 ms. Specifically, the red arrow that appeared on the left side of the fixation point represented the participants’ judgments, and the blue arrow that appeared on the right side of the fixation point represented the judgments of the confederate. If participants’ judgments agreed with the confederate’s regardless of ACC (i.e. both of 2 successive numbers judgments in one trial were the same, or both of their speed of judgments were within or outside the average speed among 30 undergraduates), the red and blue arrows all pointed to the fixation point. If participants’ judgments disagreed with the confederate’s (i.e. either of 2 successive numbers judgments was different or the judgment speed was within the average speed), the arrows pointed away from the fixation point, and the head of the red arrow pointed to the left, whereas the head of the blue arrow pointed to the right. Finally, an interval of 1000 ms was displayed simultaneously. Before the experiment, each participant was instructed that ACC and speed were equally critical for judgments in this task. Their performance (veridical ACC of judgments and speed of judgments) across all trials would be accumulated after having completed the task, and they would gain the amount of money based on their performance in addition to participation compensation.

Unbeknownst to the participants, all self–other disagreement and agreement feedback was preprogrammed and not provided by a real person. Instead, feedback was randomly set up such that the probabilities of receiving feedback indicating that self–other “agreement” and “disagreement” were both 50%. Moreover, to enhance the credibility of the task, by preprogramming, some feedback following additional trials showing “the confederate is too slow” (i.e. self–other judgments could not be compared) was set to be presented randomly in each block (4–5 trials in low-effort task blocks and 8–9 trials in high-effort task blocks), which was based on the pilot experiment (see Supplementary Materials). The entire formal experiment lasted approximately 50 min for each participant. After the experiment, each participant was asked “Did you believe that you just completed this task with a real person simultaneously?” and required to answer “yes” or “no.” If the participants answered “no” for this question (expressed their doubt about the credibility of this cover story), the data form them would be excluded for further analysis.

Behavioral and EEG data recordings

During the experiment, each participant sat approximately 75 cm away from a 19-inch LED computer monitor (refresh rate: 60 Hz; resolution: 1,440 × 900 pixels). All stimulus presentations and behavioral data acquisition were conducted using E-Prime software (version 3.0; PST, Inc., Pittsburgh, PA, USA). Brain electrical activity was recorded at 32 scalp sites using tin electrodes mounted on an elastic cap (Brain Products, Munich, Germany), with an online reference to the left mastoid and an offline algebraic re-reference to the average of the left and right mastoids. Vertical electrooculograms (EOGs) were recorded using electrodes placed above the left eye. All interelectrode impedance was maintained below 5 kΩ. We amplified the EEG and EOG using a 0.01–100 Hz bandpass and continuously sampled at 500 Hz/channel for offline analysis.

Data analysis

The EEGLAB toolbox for MATLAB (MathWorks, Natick, MA, USA) (Delorme and Makeig 2004) was used to for offline EEG data analysis. Major artifacts such as eye movements, eye blinks, and muscle-related potentials were corrected using independent component analysis (Jung et al. 2000). Then the data were digitally filtered (low-pass 30-Hz; 24 dB/octave) and segmented from 200 ms before to 800 ms after the onset of feedback stimulus presentation. After baseline correction (−200 to 0 ms), the trials with amplitudes exceeding ±70 μV were excluded to eliminate contamination from larger artifacts. The number of artifact-free trials for the ERP analysis after preprocessing is shown in Table 1. No significant differences were observed across experimental conditions (all Fs < 3.701, all P > 0.05). The epochs were then averaged separately for each condition and participant.

In particular, based on previous studies (Luck 2014; Zhang D et al. 2022), we analyzed the mean amplitude because it is considered a relatively unbiased measure owing to its high tolerance for high-frequency noise. Accordingly, the mean amplitudes of the RewP (250–300 ms) and P3a (350–480 ms) in the frontal region (i.e. average across the F3, Fz, F4, FC1, and FC2 electrodes) were extracted based on visual detection of grand-averaged data and prior studies (Hu et al. 2017; Chandrakumar et al. 2018; Zhang et al. 2021).

Time–frequency analysis of the EEG signals was performed using EEGLAB (Delorme and Makeig 2004). For the time–frequency analysis, epochs were segmented into 2000 ms epochs beginning 1000 ms before the feedback stimuli. Time–frequency characteristics were extracted from the EEG time series using a windowed Fourier transform (WFT) with a fixed 500-ms Hanning window (implemented in the EEGLAB tool) for FMT and FMD acquisition (Deng et al. 2022). Specifically, the WFT is a common method for time–frequency decomposition, which could effectively isolate stimulus-induced neural responses from background EEG activity and noise-related artifacts (Hu 2019). Importantly, in this study, we focused on low-frequency components such as delta and theta power. Previous studies have highlighted the effectiveness of the WFT in balancing time and frequency resolution when exploring delta and theta power (Hu et al. 2014; Deng et al. 2022). The present analysis compared brain oscillations at frequencies of 1–30 Hz. To avoid boundary effects, the period between −300 and −200 ms pre-stimulus was used as the baseline, and we used decibel conversion to normalize the time–frequency power data. The conversion formula was dB = 10 × lg (energy after baseline/average energy baseline). According to the activation pattern and topography distribution, to quantify the experimental condition-related changes, FMT was extracted from the band of 4–8 Hz (Wascher et al. 2014; Holroyd and Umemoto 2016; Umemoto et al. 2023) during the 300–400 ms after feedback onset, and FMD was extracted from the band of 1–3 Hz (Harmony et al. 1996; Li et al. 2016; Wang et al. 2016) during the 300–500 ms after feedback onset at the frontocentral clusters (i.e. F3, Fz, F4, FC1, and FC2) for time–frequency analysis.

Statistical analysis

Statistical analyses were performed using IBM SPSS Statistics (version 26.0; IBM Corp., Armonk, NY). Descriptive data are presented as the arithmetic mean (M) ± standard error. The paired t-test was conducted to compare the mean ACC, response time (RTs), and confidence rating between low-effort and high-effort tasks. The amplitudes of the RewP, P3a, FMT, and FMD were analyzed separately using a 2-way repeated measures ANOVA of 2 (effort expenditure level) × 2 (self–other [dis]agreement). These were within-participant factors. Moreover, based on previous studies regarding the RewP (Angus et al. 2015; Wang et al. 2018; Zhang et al. 2022b), to rule out potential effects of the P3a on RewP, we created a difference RewP by subtracting the RewP following “self–other disagreement feedback” trials from that following “self–other agreement feedback” trials and used this “difference RewP” as a dependent variable. Greenhouse–Geisser correction was used whenever appropriate, unless otherwise stated. Statistical significance was set at P < 0.05. The partial eta-squared (ƞ_p²) was reported as a measure of the effect size, where 0.05 represents a small effect, 0.1 represents a medium effect, and 0.2 represents a large effect (Cohen 1973; Kirk 1996).

Results

Behavioral results

The mean ACC, RTs, and confidence rating were calculated by averaging the ACC, RTs, and confidence ratings for “low-effort” and “high-effort” tasks, respectively. See Table 2 for the statistical analysis results of the ACC, RTs, and confidence ratings. The differences in the ACC, RTs, and confidence ratings between the “low-effort” and “high-effort” tasks were significant (all P < 0.001). In particular, we compared the mean ACC with the chance-level rate (i.e. 50%) and found that significant differences between in “low-effort” tasks, t(38) = 28.86, P < 0.001, and in “high-effort” tasks, t(38) = 17.31, P < 0.001, respectively. This showed that participants did not complete the task by chance.

ERP results

Figure 2a displays the scalp topographies of voltage differences between self–other agreement and self–other disagreement feedback in RewP and P3a time windows for separate trials in low-effort and high-effort tasks. Figure 2b displays the grand-average ERP waveforms at the Fz electrode site. Figure 2c displays the bar graphs showing the mean values of the amplitudes of RewP and P3a for each condition.

Table 3 shows the statistical analysis of ERPs results. First, the RewP showed a significant main effect of effort expenditure, F(1, 38) = 7.38, P = 0.010, ƞ_p² = 0.16. This suggested that the RewP evoked in low-effort tasks (6.15 ± 0.45 μV) was greater than that evoked in high-effort tasks (4.69 ± 0.52 μV). The main effect of self–other (dis)agreement was significant, F(1, 38) = 20.70, P < 0.001, ƞ_p² = 0.35, suggesting that the RewP evoked by self–other “agreement” feedback (5.85 ± 0.43 μV) was larger than that evoked by self–other “disagreement” feedback (4.99 ± 0.41 μV). A significant interaction of effort expenditure × self–other (dis)agreement was observed, F(1, 38) = 4.62, P = 0.038, ƞ_p² = 0.11. The simple effect analysis revealed that, in the low-effort tasks, the RewP difference between self–other agreement and disagreement feedback was significant (P = 0.040); self–other agreement feedback evoked a larger RewP (6.39 ± 0.51 μV) than self–other disagreement feedback (5.91 ± 0.43 μV). In the high-effort tasks, the RewP difference between self–other agreement and disagreement feedback was also significant (P < 0.001); self–other agreement feedback evoked a larger RewP (5.30 ± 0.50 μV) than self–other disagreement feedback (4.07 ± 0.58 μV).

Additionally, two 1-sample t-tests revealed RewP amplitudes between self–other agreement and self–other disagreement feedback to be significantly greater than 0, in both the low-effort tasks, 0.48 ± 0.22 μV; t(38) = 2.13, P = 0.040, Cohen’s d = 0.34, and the high-effort tasks, 1.89 ± 0.32 μV; t(38) = 5.84, P < 0.001, Cohen’s d = 0.94. A paired-sample t-test with the difference RewP showed that the difference RewP in high-effort tasks was larger than that in low-effort tasks, t(38) = 3.82, P < 0.001, Cohen’s d = 0.61.

Second, the results of P3a revealed that a significant main effect of effort expenditure level, F(1, 38) = 9.76, P = 0.003, ƞ_p² = 0.20. This suggested that the P3a evoked by low-effort tasks (10.26 ± 0.71 μV) was larger than that evoked by high-effort tasks (8.13 ± 0.56 μV). The main effect of self–other (dis)agreement was also significant, F(1, 38) = 9.04, P = 0.005, ƞ_p² = 0.19, showing that the P3a evoked by self–other disagreement feedback (10.25 ± 0.70 μV) was larger than that evoked by self–other agreement feedback (8.14 ± 0.58 μV). A significant interaction of effort expenditure level × self–other (dis)agreement was observed, F(1, 38) = 7.38, P = 0.010, ƞ_p² = 0.16. The simple effect analysis showed that, in low-effort tasks, a significant difference was observed in the P3a between self–other agreement and disagreement feedback was observed (P < 0.001), indicating that, in low-effort tasks, the self–other disagreement feedback evoked a larger P3a (12.28 ± 0.93 μV) compared with the self–other agreement feedback (8.24 ± 0.81 μV). However, this P3a difference was not observed for high-effort tasks (P = 0.858).

Time–frequency results

Table 4 shows the results of the time–frequency analysis. First, concerning FMT, only a significant main effect of effort expenditure level emerged, F(1, 38) = 5.31, P = 0.027, ƞ_p² = 0.12, showing greater FMT power elicited by high-effort tasks than by low-effort tasks. Neither the main effect of self–other (dis)agreement, F(1, 38) = 0.03, P = 0.862, ƞ_p² < 0.01, or the interaction of effort expenditure level × self–other (dis)agreement was significant, F(1, 38) = 2.28, P = 0.140, ƞ_p² = 0.06.

Second, concerning the FMD, there was no significant main effect of effort expenditure level, F(1, 38) = 1.89, P = 0.178, ƞ_p² = 0.05. A significant main effect of self–other agreement was observed, F(1, 38) = 16.94, P < 0.001, ƞ_p² = 0.31, indicating that participants exhibited a stronger FMD power after receiving self–other disagreement feedback compared with receiving self–other agreement feedback. A significant interaction between effort expenditure level × self–other (dis)agreement was also observed, F(1, 38) = 6.04, P = 0.019, ƞ_p² = 0.14, and the simple effects analysis revealed that, in low-effort tasks, participants exhibited higher delta power when they received self–other disagreement feedback (P = 0.001). However, the participants did not significantly differ in delta power in high-effort tasks (P = 0.812). Figure 3 shows the time–frequency spectrograms showing the spectral power for each condition over time (i.e. time window of 300–400 ms in the theta band [4–8 Hz, marked with black squares], and the time window of 300–500 ms in the delta band [1–3 Hz, marked with black squares]) at the Fz electrode site.

Discussion

Using EEG methods (including ERP and time–frequency analyses), this study investigated how effort expenditure influenced subjective value and confidence in feedback evaluations involving self–other (dis)agreement. The ERP results showed that, after exerting high (vs. low) effort, participants exhibited an increased RewP difference in response to whether they received self–other agreement feedback. After exerting low effort, participants reported having high confidence in their judgments, and the self–other disagreement feedback evoked a larger P3a. The time–frequency results showed that high-effort tasks evoked higher FMT power. Furthermore, compared with high-effort tasks, the FMD power was increased for self–other disagreement feedback in low-effort tasks.

Regarding the ERP results, at the early, and automatic processing stage, we found that after completing high-effort tasks, participants exhibited an enhanced RewP difference in response to whether they received self–other agreement feedback. This finding is consistent with our H_a. The RewP can index subjective value levels of reward feedback (Umemoto et al. 2019; Pan et al. 2023). Therefore, this result suggests that, after exerting high effort in the task, individuals exhibited a heightened sensitivity to self–other agreement feedback in subjective value evaluations. Previous studies have shown that exerting more effort can increase the subjective value associated with a reward (Kelley et al. 2019; Harmon-Jones, Willoughby, et al. 2020; Bogdanov et al. 2022; Pan et al. 2023). According to the cognitive dissonance theory (Festinger 1962; Harmon-Jones and Harmon-Jones 2012), individuals often assume that they have chosen to engage in the effort and their effort should be associated with a reward (Harmon-Jones, Willoughby, et al. 2020). People will attach greater subjective value to the deserved reward after exerting high effort (Norton et al. 2012). Emerging evidence has shown that people have a common response to social and monetary rewards (Oumeziane et al. 2017; Wake and Izuma 2017). Accordingly, individuals’ subjective value evaluation exhibits a heightened sensitivity to self–other agreement feedback, and then elicits an enhanced RewP difference.

Following the RewP, at the later elaborate and controlled processing stage, we found that after the low-effort tasks, participants reported having high confidence in their judgment, and the self–other disagreement feedback evoked a larger P3a. This is consistent with our H_d. The P3a consistently reflects the recruitment of attentional resources through salient or distracting feedback (Nieuwenhuis et al. 2004; Polich 2007; Zhu et al. 2022). The P3a increased with metacognitive mismatch because metacognitive mismatch attracts more attentional resources (Butterfield and Mangels 2003; Butterfield and Metcalfe 2006; Ungan et al. 2019; Frömer et al. 2021). Previous studies have regarded the P3a as an index reflecting changes in confidence (second-order subjective value) such that if individuals with high confidence in their responses receive feedback indicating failure to receive a reward, their P3a increases because this violation of their confidence leads to metacognitive mismatch (Butterfield and Mangels 2003; Butterfield and Metcalfe 2006). This finding also suggests that self–other disagreement feedback (feedback in the context of social interaction) has a similar effect to feedback showing that participants’ judgments fails to get a monetary reward (feedback in the context of non-social interaction), which can produce a metacognitive mismatch for participants who experience high confidence after completing the low-effort tasks.

In addition, prior studies have demonstrated that the processing of subjective value temporally emerges before second-order subjective value (confidence) (Shapiro and Grafton 2020). Using ERPs, our findings further confirmed that the interaction effect of effort expenditure and self–other (dis)agreement on subjective value temporally occurred before confidence.

Regarding the time–frequency results, first, we found that FMT power was sensitive to effort expenditure, such that FMT power was greater for high-effort tasks than for low-effort tasks. This finding is consistent with prior findings showing that FMT power increases with increasing cognitive effort, since it is a reliable marker of distinct changes in cognitive processing with increasing fatigue (Wascher et al. 2014; McFerren et al. 2021; Umemoto et al. 2023). Second, similar to the interaction effects seen for the P3a, we found that participants exhibited higher FMD power in low-effort tasks when they received self–other disagreement feedback. However, this effect did not emerge for high-effort tasks. This finding is consistent with H_e. Following interpretations of the meaning of the P3a, prior studies have suggested that increased FMD power is related to attentional resource allocation during mental tasks (Harmony et al. 1996; Harmony 2013). Therefore, the violation of high confidence (i.e. metacognitive mismatch) was also reflected in FMD.

In summary, using electrophysiology, the present study examined how effort expenditure influenced the subjective value evaluations in feedback involving self–other (dis)agreement. These findings suggest that, in the earlier feedback processing stage, after exerting high effort, individuals are more sensitive to subjective value evaluation in self–other agreement feedback (indexed by RewP). Moreover, individuals reported having high confidence in judgments, and self–other disagreement feedback violated this high confidence (metacognitive mismatch) after low-effort tasks (indexed by P3a amplitudes and FMD).

The current study has several theoretical or practical implications, as follows. First, this study expands on previous studies on the impact of effort on subjective value associated with material rewards and demonstrated its impact on subjective value associated with social rewards. Second, we further extended the focus on subjective value to second-order subjective value (i.e. confidence) in feedback evaluations. Third, previous studies have shown that people with social impairment-related disorders, such as those with schizophrenia, overestimate the effort associated with social interaction (Catalano and Green 2023). Therefore, it is reasonable that individuals with social impairment-related disorders may have specific neurological responses in the task used in the current study. From a clinical perspective, the current study can serve as a reference for future research on diagnosing potential social impairment-related disorders.

Funding

This study was funded by the National Natural Science Foundation of China (No. 32000769), youth project of the Natural Science Fund of Hunan Province (No. 2021JJ40337), the MOE (Ministry of Education in China) Project of Humanities and Social Sciences (23YJC190013), the Social Science Foundation of Guandong Province (GD23XXL13).

Conflict of interest statement: None declared.

Data availability

All data generated for this study are available in the Open Science Framework (OSF) at https://osf.io/cdk3a/.

References

Author notes