Participants were 41 healthy undergraduate students of Leiden University and adolescents of local schools. Four adults and two adolescents were excluded from the analyses because they did not believe the cover story, showed no variation in their behavioral responses, or did not have enough ERP segments in one of the critical conditions. This resulted in a final sample of 18 adults (age: range = 19–24, M = 20.61, SD = 1.15; 8 female; 16 right-handed) and 16 adolescents (age: range = 14–17, M = 15.19, SD = 0.91; 11 female; 15 right-handed; handedness scale [28]) who were included in the analyses. All participants gave written informed consent for participation in the study; written informed consent was also obtained from the parents of the adolescents. None of the participants reported any psychiatric or neurological conditions or impairments and all participants had normal or corrected-to-normal vision. Participants received financial compensation for their participation. The study was approved by the Leiden University ethics committee.

Participants played a modified version of the Ultimatum Game (UG). They were instructed that they would play a game with another person of same age and gender. In reality, they played against the computer. We informed the participants that they would be the proposer during the entire game and that the other player would be the responder. To increase credibility of playing against another person, during preparation of the EEG equipment, a confederate walked into the room to ask how long preparations would take before the experiment could start, upon which participants were told that this was the experimenter responsible for instructing the other player. Additionally, fake calls were made between the two labs before starting the experiment and after each break.

In the current version of the Ultimatum Game two sets of dichotomous distributions for sharing the stake between the proposer and responder were presented. For a visual display of these distributions see Fig 1. In both sets the unfair distribution of seven coins for the proposer and three coins for the responder (hereafter 7/3 offer) was pitted against an alternative offer: (i) the fair-alternative (5/5 offer), or (ii) the hyperfair-alternative (3/7 offer). The participant was informed that the other player saw the same screen as the proposer and, thus, was aware of which two distributions the participant could choose from. Participants were instructed that the coins won during the game would be converted into real money and paid out at the end of the study.

The acceptance and rejection rate of the offers was manipulated such that in the fair-alternative condition, approximately 60% of the fair offers and 40% of the unfair offers was accepted. In the hyperfair-alternative condition, approximately 60% of the hyperfair offers and 40% of the unfair offers was accepted. Acceptance or rejection of the responder followed an algorithm allowing the responses to be random, but falling within a 60–40% division.

Each trial started with a jittered 500–700 ms fixation cross. After the fixation, two alternative distributions were presented (symmetrically positioned compared to fixation). The visual angel between fixation and the end of a distribution was 5.5°. The visual angel of the total height of the distribution was 4.0°. Participants had a maximum of 2500 ms in which they could choose which of the two distributions to offer. If the participant did not make an offer within 2500 ms a feedback screen with “Too late” was presented for 1200 ms. The position of the unfair offer was counterbalanced in each condition. Participants could make an offer with a button press with their left middle or index finger. Upon response the chosen distribution was outlined in red for 800 ms plus the remaining time of the offer presentation (2500 ms—reaction time). Next, a waiting screen with a fixation cross was presented for 1400–5000 ms. During this waiting screen the fictive responder made a choice to accept or reject the offer. After this waiting screen, the feedback of the responder (i.e., acceptance or rejection of the offer) was presented for 1400 ms. This was done by the words”YES” or “NO” in the center of the screen to indicate acceptance or rejection of the offer, respectively. Subsequently, a new trial started.

Participants played eight practice trials. The experimental task itself consisted of four blocks. Each block consisted of 52 trials in which the fair alternative (5/5 vs 7/3) and hyperfair alternative (3/7 vs 7/3) conditions were presented 26 times each, in a random order, resulting in 208 trials in total. The total duration of the experiment, including self-paced breaks between blocks during which the participant remained seated, was approximately 35 minutes. Participants completed a funnel interview at the end of the study to test whether they believed the cover story and were subsequently debriefed.

The EEG signals were recorded through an Active-Two amplifier system (Biosemi, Amsterdam, The Netherlands) from 21 scalp electrodes according to the 10–20 system (Fz, F1/2, FCz, F3/4, FC1/2, FC3/4, Cz, C1/2, C3/4, CPz, CP1/2, Pz, P1/3). The 21 Ag/AgCl electrodes were mounted in an elastic cap. Six additional electrodes were attached to the left and right mastoids, two outer canthi of the eyes (HEOG), and infraoribital, and supraorbital regions of the left eye (VEOG). Two additional scalp electrodes were used to serve as reference and ground electrodes. Online signals were recorded from DC to 134 Hz. All signals were digitized with a sample rate of 512 Hz and 24-bit A/D conversion.

Offline, a mathematically linked average mastoid reference was applied and EEG and EOG activity was filtered with a bandpass of 0.10–30 Hz (phase shift-free Butterworth filters; 24dB/octave slope). Ocular correction was applied according to the Gratton, Coles and Donchin [29] algorithm. Artifact rejection criteria were minimum to maximum baseline-to-peak allowed voltage -100 to +100 μV, and a maximum allowed voltage skip (gradient) of 75 μV per sample point. The data were segmented in epochs of 900 ms, from 100 ms pre-target onset, serving as baseline, to 800 ms post-target onset. The mean 100 ms pre-stimulus period was used for baseline correction. The average number of trials included in the ERP analyses for the adults for Fair Accept was 44 (range = 30–56), for Fair Reject 29 (range = 11–43), for Unfair Reject 45 (range = 26–62), and for Unfair Accept 30 (range = 17–38). The average number of trials included in the ERP analyses for adolescents for Fair Accept was 44 (range = 26–63), for Fair Reject 29 (range = 12–40), for Unfair Reject 38 (range = 25–64), and for Unfair Accept 26 (range = 14–49).

The average number of trials for the alternatives was not high enough for a reliable signal, however we do report the average number and range of those trials for clarity reasons. In the fair-alternative condition (7/3 versus 5/5) the average number of accepted 7/3 trials for adolescents was 10 (range = 4–28), rejected trials 16 (range = 4–34). For adults the average number of 7/3 trials was for accepted 17 (range = 2–39) and for rejected 11 (range = 0–27). In the hyper-fair alternative condition (7/3 versus 3/7) the average number of accepted 3/7 trials for adolescents was 22 (range = 4–33), rejected trials 14 (range = 4–28). For adults the average number of 3/7 trials was for accepted 17 (range = 0–35) and for rejected 9 (range = 0–25).

The percentage of unfair offers was examined in the fair-alternative condition in which an unfair distribution was pitted against a fair distribution (7/3 versus 5/5) and in the hyper-fair alternative condition where an unfair distribution was pitted against a hyperfair distribution (7/3 versus 3/7) (see Fig 2 for the percentages of the different offers made by the participant). The Condition (2) x Age group (2) repeated measures ANOVA for unfair choices resulted in a main effect of Condition (F (1,32) = 102.09, p < .001, ƞ²partial = .76), showing that participants made more unfair offers in the hyperfair-alternative (69%) than in the fair-alternative (27%) condition. There was neither a main effect of Age group (F (1,32) = 1.72, p = .20, ƞ²partial = .05), nor an interaction with Age group, (F (1,32) = 1.18, p = .29, ƞ²partial = .04) indicating that adolescents and adults made similar offers.

Grand averages were calculated for the fair offers (5/5) in the fair-alternative (5/5 vs 7/3) condition and unfair offers (7/3) in the hyperfair-alternative condition (3/7 vs 7/3). We focused on these two offers because the alternative offer types did not have enough trials for a reliable signal-to-noise ratio. Based on previous studies with the Ultimatum Game [16, 17, 20, 21, 23] and based on visual inspection, we defined the MFN component as the mean activity in the 240–320 ms time window after feedback presentation. To minimize potential effects of overlap of the MFN and other ERP components, such as the P300, we calculated difference waves (e.g. [16, 17, 20, 21, 30]). Both the amplitudes and the differences waves were analyzed and reported below.

We performed two repeated measures ANOVA to (1) test whether the MFN is modulated when a fair offer (5/5) is rejected compared to when it is accepted, reflecting a violation of fairness expectations and (2) to test whether the MFN responds to a violation of fairness-related expectations specifically or to unexpected events in general, by examining the MFN following unexpected acceptance.

The first hypothesis we tested was whether the MFN would be more pronounced when a fair offer (5/5) was rejected compared to accepted. Therefore, we performed a 2 (Electrode: FCz, Cz) x 2 (Feedback: Accept vs. Reject) x 2 (Age group: Adolescents vs. Adults) repeated measures ANOVA for the MFN amplitude of fair offers (5/5) in the fair-alternative condition (5/5 vs 7/3). This analysis resulted in a main effect of Feedback, F (1,32) = 6.12, p = .019 and Electrode, F (1,32) = 48.49, p < .001. As can be seen in Fig 3, the MFN was larger for reject than for accept trials in both adolescents and adults. The MFN was also negative at FCZ, but there was no Feedback x Electrode interaction.

Next, we computed a difference wave by subtracting Fair Reject from Fair Accept. To test whether this difference of Fair Accept—Fair Reject significantly differed from zero we performed a one-sampled t-test on the difference wave of Fair Accept—Fair Reject (see Fig 3) for both FCz and Cz. The one-sample t-tests showed that the MFN as measured by the difference between Fair Accept—Fair Reject significantly differed from zero at FCz, t (33) = 2.46, p = .020 and at Cz, t (33) = 2.53, p = .016. Although the repeated measures ANOVA showed no interaction with Age group we performed an independent sample t-test with Age group as a between subjects factor to confirm there were no differences between adults and adolescents. Indeed, we did not find significant differences between adults and adolescents neither for FCz, t (32) = .18, p = .855, nor for Cz, t (32) = .11, p = .914.

To test the second hypothesis examining whether the MFN responded to violation of fairness-related expectations specifically or to unexpected events in general, we focused on unexpected acceptance of unfair offers. That is, when the MFN would be larger for acceptance than for rejection of unfair offers in the hyperfair-alternative condition this would indicate a general expectancy violation. Therefore, we performed a similar 2 (Electrode: FCz, Cz) x 2 (Feedback: Accept vs. Reject) x 2 (Age group: Adolescents vs. Adults) repeated measures ANOVA for the MFN amplitude of unfair offers (7/3) in the hyperfair-alternative condition (3/7 vs 7/3). The results showed a main effect of Feedback, F (1,32) = 7.63, p = .009, and Electrode, F (1,32) = 61.24, p < .001. The MFN was larger for reject than for accept trials in both adolescents and adults. The MFN was also negative at FCz, but again, there was no Feedback x Electrode interaction.

Next, we performed one-sample t-tests for the difference waves of Unfair Accept—Unfair Reject and found that the difference wave significantly differed from zero in FCz, t (33) = 2.88, p = .007 and in Cz, t (33) = 2.58, p = .014. Again, when we performed an independent sample t-test with Age group as a between subjects factor we found no significant differences between adults and adolescents for neither FCz, t (32) = .02, p = .984, nor for Cz, t (32) = .01, p = .990.

When directly comparing the rejection trials after a fair offer (5/5) in the fair-alternative condition (5/5 vs. 7/3) and after an unfair offer (7/3) in the hyperfair-alternative condition (3/7 vs. 7/3), the 2 (Electrode: FCz, Cz) x 2 (Feedback: Reject after fair offer vs. Reject after unfair offer) x 2 (Age group: Adolescents vs. Adults) repeated measures ANOVA revealed a main effect of Feedback, F (1,32) = 18.29, p < .001 and Electrode, F (1,32) = 65.48, p < .001. The MFN was larger (more negative; see also Fig 3) when a fair offer was rejected compared to when an unfair offer was rejected.

Finally, we performed a one-sample t-tests for the difference between Fair Reject—Unfair Reject showing that the difference wave significantly differed from zero in FCz, t (33) = 3.58, p = .001 and Cz, t (33) = 4.45, p = .001. However, there was no difference between the two age groups for neither FCz, t (32) = 1.36, p = .183, nor Cz, t (32) = 0.81, p = .426.

Thus for all analyses, there was no interaction between Feedback (accept vs. reject) and Age group, nor were there significant differences in the difference waves, indicating that in the current study adolescents and adults did not differ in manifestation of the MFN during rejection of fair offers in the fair-alternative and rejection of unfair offers in the hyperfair-alternative conditions.

In the current study we chose to have participants play continuously with the same player during the entire experiment to buildup social expectations [17, 21]. However, such an experimental setup may alter expectancies as the session progresses. Previous research has shown that the FRN/MFN changes with expectancy when participants behaviorally learn (e.g. [18, 53, 54, 55, 56]). However, there also have been reports in which the FRN remained constant as response accuracy increased, or even that participants learn without any clear FRN [45, 57, 58], (see [31] for a review). Future studies should take into account that the buildup of expectancies may alter the FRN during the UG when playing with the same player. Furthermore, it may be possible that the color of the given feedback may have inflated the MFN, such that the negative reject feedback in red elicited an even larger MFN compared to when the negative reject feedback would have been given in a more neutral color. However, even if the color of the feedback inflated the MFN this would not have affected the critical comparison of negative reject feedback after a fair proposal and negative reject feedback after an unfair proposal, because in both conditions the feedback had a red color.

Taken together, the current study was the first to show that the MFN, a neural signal originating from the ACC, is sensitive to rejection and social prediction errors of fairness considerations in bargaining situations. Whereas prior studies focused on the modulation of the MFN in responders who receive unfair offers, the current study examined the neural response in the proposer and showed that the MFN is also present when fair offers of proposers are rejected. Prior studies revealed that the MFN response to norm violations is dependent on social context. For example, it was previously found that the error-related negativity (ERN), which is observed when people make errors, is larger for people with low social status [59]. In addition, the MFN is more negative going when receiving an unfair offer from a stranger (when status still needs to be determined) than when receiving an unfair offer from a friend ([21], but see [23]). An important direction for future research is to examine the determinants of rejection sensitivity of proposers, for example by manipulating the status of the proposer and responder. These results are expected to provide important information about how individuals make choices for fairness and reciprocity, which are important processes for developing social ties.