Method

Participants

Forty‐five healthy infants (M = 6.47 months, SD = 0.89, min = 5.07, max = 7.97, 17 females, 28 males) and their caregivers (M = 34.2 years, SD = 3.31, 39, min = 27.72, max = 42.74, 39 females, six males) participated in the current study. We selected 5–7‐month‐old infants to investigate early behavioural and neural responses to ostracism before more advanced social‐cognitive abilities emerge. This age range is based on evidence that by 7–9 months infants can already respond to ostracism in others (Prendergast 2019). Moreover, infants by 6 months show flexible attentional allocation between social and non‐social stimuli (Bremner and Wachs 2010; Fogel 2015), suggesting the presence of relevant precursors in early infancy. They were recruited via a local participant data base and advertisements. Two dyads were excluded due to discomfort with the EEG cap and disinterest in the ball game and due to the experimenter deviating from neutral facial expressions during the ball game; therefore 41 participants were included in the affective and behavioural analyses and the correlation analysis between infants’ and caregivers’ affective and behavioural responses. Additional dyads to the four excluded in behavioural analysis were excluded depending on the type of neural analysis. For theta and alpha power in infants, four infants were excluded due to technical problems (e.g., technical errors in the EEG system and unsaved markers in EEG data), and three infants due to less than 50% of trials remaining for each event, resulting in 34 infants. For theta and alpha power in caregivers, four caregivers were excluded after EEG preprocessing due to the technical problems (e.g., technical errors in the EEG system and unsaved markers in EEG data), resulting in 37 caregiver datasets. For inter‐brain synchrony analysis and correlation between parental bonding and inter‐brain synchrony values, seven dyads were excluded because one or both pairs had less than 50% of valid trials after preprocessing, as well as technical errors in the EEG system and unsaved markers in EEG data, resulting in 34 infant‐caregiver dyads. Fourteen additional dyads were tested to reach the sample size of 34 participants in line with the power analysis for theta power (Supplementary Materials 1). The study was approved by the local ethics committee of the Bremen University (reference number 2024–2011) and conducted in line with the Declaration of Helsinki. Methods, hypotheses, and analyses were preregistered with the Open Science Framework (https://osf.io/c3n6g).

Procedure

Upon arrival, the caregivers read and signed the data protection and informed consent form. After applying EEG for both caregivers and infants, they were invited to an adjacent room to play a ball‐tossing game on the table with two other experimenters. Infants sat on the caregiver's laps. The caregiver supported the infant to toss a ball back and forth with the experimenter. In the inclusion block, all the players had equal opportunity to toss the ball toward each other (14 tosses each). Subsequently, the two experimenters ignored the infant‐caregiver‐dyad and tossed the ball toward each other with equal opportunity (eight tosses each) in the exclusion block. Finally, the experimenters included the infant‐parent‐dyad in the game again in the reinclusion block by tossing the ball towards them (five tosses each). The experimenters maintained a neutral, friendly facial expression and eye contact during the entire game (Figure 1). The game was recorded via the webcam for the behavioural analysis. In order to synchronize the recorded videos with the EEG data, one of the experimenters pressed a button on a small keypad connected via Python/Psychopy, which displayed a green circle on the monitor behind the dyads (outside their visual field) at the start of the ball game and sent markers to the EEG data. In addition, a white circle was displayed every second on the same screen while a simultaneous marker was sent. This procedure enabled us to synchronize the video and EEG recordings and provided accurate markers for subsequent EEG analysis.

FIGURE 1: Procedure of the main condition in which the caregiver (wearing the EEG net) plays with the infant (a), infant‐control condition in which an experimenter plays with the infant while the caregiver watches (b), cargiver‐control condition in which the caregiver plays while the infant watches (c) and video‐control condition in which the caregiver and infant watch a video together (d).

Dyads participated in three counterbalanced control conditions if they were willing to continue. In the caregiver condition, the caregiver played the ball game (inclusion and exclusion) with two experimenters while the infant, held by another experimenter within 60 cm, observing the game. In the infant condition, the infant played the game on the experimenter's lap while the caregiver observed from 60 cm away. In the video condition, dyads watched a pre‐recorded ball game (recorded from their own perspective) (Figure 1). After completing the control conditions, caregivers filled out the Postpartum Bonding Questionnaire (PBQ; Reck et al. 2006), demographic information, and were fully debriefed. Finally, infants received a certificate and a small gift for their participation.

Measurement

Affective and Behavioural Responses

During the ball game, behavioural responses were recorded with a Logitech C920 PRO HD camera. The recording started when the experimenter pressed a button on the small keypad, which simultaneously showed the circle on the screen to send markers via a parallel port for EEG synchronization. The video recording and circles on the monitor were controlled via Python/Psychopy.

EEG Recording

EEG was recorded at a sampling rate of 500 Hz from 32 actiCAP slim active electrodes, mounted in the electrode cap (actiCAP snap; Brain Products GmbH, Gilching, Germany). FCz was used as a reference during the online recording, which was filled with gel and kept below 40 kΩ to ensure adequate signal quality. The EEG was recorded on two separate computers (one for the infant and one for the caregiver) using BrainVision Recorder software (v1.21.0402, Brain Products 15 GmbH, Gilching, Germany) for infants and caregivers separately. Electrodes F7, F3, Fz, F4, F8, FC1, FC2, Cz (approximately corresponding to electrode clusters used in a study by Van Noordt et al. (2015a) with different electrode montages) and Oz were filled with electrode gel to maximize efficiency of the set up, and if the infant was happy, additional occipital and parietal electrodes (O1, O2, P3, Pz, P4) were filled with gel for both the infant and caregiver. Electrode impedances were kept below 40 kΩ for all electrodes. For infant participants, electrodes were filled with an infant‐friendly electrode cream.

Postpartum Bonding Questionnaire‐16 (PBQ‐16)

The PBQ was designed as a self‐report scale to identify bonding difficulties (Brockington et al. 2001). In the current study, we used the short version of the PBQ (Reck et al. 2006). Caregivers rated each statement such as, “I feel close to my baby” or “My baby irritates me” on a 6‐point Likert scale ranging from 0 (“never”) to 5 (“always”). Higher PBQ‐16 scores reflect a poor level of bonding. Internal consistency was acceptable (Cronbach's α = 0.77), with sum scores ranging from 3 to 27 (M = 10.17, SD = 5.52).

Analysis

Affective and Behavioural Coding

Affective and behavioural responses were coded on five scales for infants and caregivers as follows: (1) positive emotionality, (2) negative emotionality, (3) visual attention, (4) attention‐seeking behaviour, and (5) social referencing (See Supplementary Materials 2). The videos were coded by two experimenters and one blind coder and the inter‐rater reliability (Fleiss' Kappa) was computed for around 10% of the videos. Each subscale was coded frame by frame using BORIS software (Friard and Gamba 2016) (See Supplementary Materials 2 for checking the reliability). The scores of the different subscales were normalized by dividing by the total frame subscales of the relevant block and then multiplied by 100 to get a percentage of the sum score. Finally, normalized subscales were summed up to calculate a score for each scale.

EEG Processing

Prior to preprocessing, the video recordings of the ball‐tossing game were coded to generate three types of markers: (1) “inclusion” markers identified at the time when the experimenters tossed the ball toward the infant‐caregiver dyad, (2) “not my turn markers” identified when the experimenters tossed the ball to each other during the inclusion block and (3) “exclusion” markers identified when the experimenters tossed the ball to each other during the exclusion block. The main focus of this study was to compare “not my turn” to “exclusion” conditions as the perceptual characteristics of these events were identical and there was no movement required from the participants in both of these events. All markers were calculated manually based on the green circle indicating the start of the ball game, and the updated markers were implemented on the continuous data of each participant using MATLAB. Moreover, the quality of the reference channel (Oz) was checked before including data in the analyses, based on both the impedance (meaninfant = 8.23, SDinfant = 6.08, maxinfant = 28, meancaregiver = 13.09, SDcaregiver = 9.49, max caregiver = 39) and visual inspection. EEG data were processed using identical analysis protocols for adult and infant data using the Maryland analysis of developmental EEG (MADE) pipeline (Debnath et al. 2020) in MATLAB using the EEGLAB toolbox (Delorme and Makeig 2004). First, FIR filters with a high‐pass boundary of 2 and a low‐pass boundary of 15 Hz were applied by using FIRfilt plug‐in of EEGLAB (Widmann 2015). This filter was chosen to include the theta and alpha bands of interest in both infants and caregivers, while minimizing slow drifts, movement‐related artifacts, and high‐frequency noise. The wider lower and upper bounds were used to provide a buffer, thereby minimizing edge artifacts. Bad channels were detected and removed by the EEGLAB plug‐in FASTER (Nolan et al. 2010). To ensure the data quality, channels flagged by the FASTER algorithm were also visually inspected to confirm poor signal quality, and the remaining channels were visually checked to ensure they had adequate signal quality. Finally, a copy/ICA procedure was performed on a range of 9 to 14 channels (see Debnath et al. (2020) for details). For this purpose, a copy of the original data was created, a high‐pass filter (1 Hz) was applied to the copy, and channels and epochs with excessive artifacts were removed from the copy. Independent component analysis (ICA) was performed on the copy of the remaining channels of interest, as the rest of the channels were not filled with gel, and then, ICA weights were transferred back to the original data. Artifactual Independent Components (ICs) caused by eye movement were removed from the original data manually based on the visual inspection (component topographies, time courses, and power spectra). The data was epoched around the triggers for inclusion, not my turn and exclusion markers, including one second of data recordings for each condition. A baseline correction was applied using a 200 ms time interval before the trigger. A voltage threshold rejection (±150 µV for infants and ±100 µV for caregivers) was applied to the epochs, and channels were removed if they exceeded the voltage threshold. In the next step, the channels identified by the FASTER function were interpolated. Since it was not possible to fill all electrodes with gel for infants due to time constraints, the data was not re‐referenced to the average. As the aim was to measure theta and alpha power in frontal/midfrontal regions, the data was re‐referenced to the Oz channel. After preprocessing, dyads with fewer than 50% of trials in each condition (Inclusion ≤ 7, Not My Turn ≤ 7, Exclusion ≤ 8) were excluded from further analysis. The overall mean and SD of the number of trials included in all neural analyses are as follows: infants: Minclusion = 12.5, SDinclusion = 1.78, Mexclusion = 13.5, SDexclusion = 2.50 and Mnot my turn = 12.5, SDnot my turn = 1.67, caregivers: Minclusion = 12.8, SDinclusion = 1.62, Mexclusion = 14.0, SDexclusion = 2.30 and Mnot my turn = 12.8, SDnot my turn = 1.57). (Detailed means and SD for each individual analysis are provided in Supplementary Materials 3 and 4).

Alpha and Theta Power

Alpha and theta power were calculated using the Matlab function pwelch. Each epoch was divided into 50% overlapping 1‐sec segments. Each segment was windowed with a 1‐s Hanning window, and a Fast Fourier Transformation was applied to obtain spectral density. The frequency bands were adjusted to age‐related shifts in frequency band (Cuevas and Bell 2022; Turk, Endevelt‐Shapira et al. 2022): We used frequency bands ranging from 3 to 6 Hz for theta and 6 to 9 Hz for alpha in infants (Orekhova et al. 2006). For adults, we used frequency bands ranging from 4 to 8 Hz for theta power and 8 to 13 Hz for alpha power (Kulke et al. 2023; Strijkstra et al. 2003). The power extractions for each frequency band were then averaged across participants. The mean and SD of the remaining trials after preprocessing for theta and alpha power analyses are provided in Supplementary Materials 3 and detailed alpha power analyses are reported in Supplementary Material 6.

Inter‐brain Synchrony

To calculate caregiver–infant inter‐brain synchrony, Phase‐locking value (PLV) was computed: the EEG data were first filtered to the desired frequency band using a finite impulse response filter (infant: 2.9–6.1 Hz and caregiver: 3.9–8.1 Hz) (Lachaux et al. 1999). After that, the instantaneous phase was extracted using the Hilbert transform. For each trial, the PLV was computed by calculating the complex phase difference between infants’ and caregivers’ signals for each trial (1 s window) (the equation can be found in Lachaux et al. 1999; Marriott Haresign et al. 2023). The PLV was calculated for each of the matched trials between infant‐caregiver dyad and subsequently separated according to each condition (inclusion, exclusion, and no my turn). To assess whether the observed PLV exceeded chance levels, a permuted analysis was performed by pairing infants and caregivers randomly 1000 times and calculating the PLV for permuted dyads in the same vein as for the real dyad to create a null distribution of synchrony values expected by chance. In the next step, we compared the real PLV with the permuted PLV distribution for each condition. Differences between the real and permuted data were assessed using a non‐parametric permutation‐based approach (Cohen 2014; Maris and Oostenveld 2007). The p‐value was calculated as the proportion of permuted values that were equal to or greater than the real PLV. Real PLV values exceeding 95% of the surrogate distribution (p < .05) were considered statistically significant. Subsequently, multiple comparisons across channel pairs were controlled using False Discovery Rate (FDR) correction. Moreover, the observed PLV was compared to the PLV of dyads in the control conditions (infant, caregiver, and video condition) using paired‐t tests. Additional measures of inter‐brain synchrony (power correlation and partial directed coherence (PDC)) are described in Supplementary Materials 5.