https://orcid.org/0000-0003-2630-3790
Figures
Abstract
Understanding how individuals evaluate social content in affective scenes is crucial for research in emotion, social cognition, and decision-making. However, standardized image databases often lack fine-grained ratings of social dimensions beyond basic emotional content. We present a normative dataset of 296 affective scenes rated by 424 adult participants across six social-affective dimensions: social relevance, emotion sharing, action sharing, interpersonal equality, scene pleasantness, and participatory arousal. Each image was independently coded for objective features such as number of people, face visibility, and gaze direction. All rating dimensions showed excellent interrater reliability. Pleasantness ratings demonstrated strong convergent validity with established valence norms from the International Affective Picture System (IAPS), while arousal ratings differentiated between experienced and participatory arousal, suggesting that the latter captures a distinct affective component. Correlation analyses revealed that social relevance was strongly associated with scene pleasantness and a composite “engagement” factor combining emotion sharing, action sharing, and equality. Notably, the association between pleasantness and social relevance was especially pronounced in scenes depicting single individuals. Linear mixed models further indicated that extraversion showed a tentative positive association with the engagement–social relevance relationship, which did not survive correction for multiple comparisons and requires replication. Other personality traits and empathy did not significantly moderate these relationships. This open-access dataset provides a reliable, multidimensional tool for selecting affective scenes based on both emotional and social-interpersonal features. It supports improved stimulus control and design in a wide range of behavioral and neurocognitive research. By systematically mapping social meaning in static affective scenes, our study advances existing methodological resources and offers empirical insight into the structure of social-affective appraisals. All image codes, mean ratings, and analysis scripts are publicly available via the Open Science Framework.
Citation: Mitschke V, Grassi F, Doğaner A, Schacht A (2026) Mapping social relevance in affective scenes: A large-scale multidimensional rating study. PLoS One 21(10): e0331156. https://doi.org/10.1371/journal.pone.0331156
Editor: Jinhwan Kwon, Kyoto University of Education: Kyoto Kyoiku Daigaku, JAPAN
Received: August 12, 2025; Accepted: September 15, 2026; Published: October 8, 2026
Copyright: © 2026 Mitschke et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data underlying the results are available from the Open Science Framework (OSF) at: https://osf.io/9m3ef. This includes the full dataset of image ratings, objective image annotations, and analysis scripts used in the study.
Funding: This work was supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation; 454648639 – SFB 1528). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Detecting and interpreting social information from visual cues is crucial for navigating complex environments. Humans are adept at extracting affective and social meaning from visual scenes, often within milliseconds of exposure. Experiments involving the perception and evaluation of social information have been integral to our understanding of complex interactions between emotional, motivational, and cognitive processes. While research has extensively characterized emotional responses to visual scenes using databases such as the International Affective Picture System (IAPS; [1]), far less is known about how people perceive and evaluate the social qualities of these complex scenes, particularly when they depict interactions among multiple individuals.
Despite growing interest, there is no consensus on how to define or operationalize social information in complex scenes. Some studies rely on binary categories (e.g., presence of humans; [2]), while others focus on interpreting social cues [3] or assessing levels of social engagement [4]. In this study, we define social relevance as the perceived importance of a stimulus for understanding, predicting, or responding to others’ behavior. This includes cues related to interpersonal dynamics, intentions, and interactions [5,6]. Like emotional valence, social relevance must be inferred from contextual and relational information [7,8], making it a psychologically meaningful but methodologically challenging construct to quantify. Importantly, social relevance is related to but distinct from several neighboring constructs. First, it differs from social content, which refers to whether social agents are present (e.g., one or more people) but does not specify whether their behavior is interpretable or consequential. Second, it differs from social salience, which captures the extent to which social cues attract attention (e.g., faces, gaze) without necessarily being informative for understanding intentions or interpersonal dynamics. Third, it is not equivalent to affordances, which describe action possibilities in a scene more generally (i.e., “what one could do”), whereas social relevance concerns what the scene implies about others and the interactional meaning of their behavior. Accordingly, we treat social relevance as an appraisal-like dimension that depends on contextual and relational information rather than on any single perceptual cue.
Research on social perception has largely concentrated on either isolated features such as faces, their emotional expressions, or gaze-driven attentional cues, often neglecting the complexity of real-world social interactions. Such interactions are difficult to capture with isolated face stimuli; instead, they require scenes that depict full-body postures, multiple agents, and contextual (non)social cues. Recent studies suggest that social interaction influences attentional capture [9], and perceived pleasantness [10], beyond the simple presence of one or two humans.
This omission of interactional complexity is nontrivial. Social cues are not only a key driver of attention [6] but also modulate physiological and behavioral responses in affective tasks [11]. A growing body of research highlights the automatic attentional prioritization of social features such as faces, gaze direction, or interaction cues—even when emotional content is held constant [5,12–14]. These findings suggest that social relevance constitutes a distinct and powerful dimension of stimulus processing – interacting with, but not reducible to, emotional valence.
This perspective is increasingly supported by recent empirical findings from behavioral and neuroscientific paradigms. For example Kosonogov et al. [15] demonstrated that increasing the number of people in affective scenes systematically altered valence and arousal ratings, highlighting the influence of social content on affective evaluations. However, their approach treated social presence as a categorical feature and did not capture the nuances of interaction which are conveyed in social scenes. Our study extends this line of work by employing fine-grained, continuous ratings of key social-affective dimensions — including perceived social relevance, action and emotion sharing, and interpersonal equality — to better capture how social meaning is evaluated.
Additional evidence underscores the robustness and early impact of social relevance in complex scene processing. Social information can shift emotional priorities, alter visual salience, and guide attention—even under cognitive load [16]. For instance, emotional salience can override low-level visual features [17], and social relevance shapes both attentional dynamics and memory representations [3,18]. Neuroimaging findings further indicate that high-level social-affective features—rather than purely perceptual characteristics—primarily drive the neural encoding of observed actions [19]. Together, these findings reinforce the need for normative databases that include structured social-affective dimensions to better reflect real-world scene interpretation.
Structured neural representations further support the idea that social relevance is cognitively organized along interpretable dimensions. For example, Zhang et al. [20] showed that social decisions are encoded in a two-dimensional space of affiliation and power within hippocampal and parietal regions. Similarly, Wittmann et al. [21] found that the dorsomedial prefrontal cortex abstracts social interactions into basis functions that generalize across multi-agent contexts. These findings demonstrate that social meaning is systematically encoded and can be captured using standardized rating protocols.
A recent data-driven taxonomy by Santavirta et al. [22] provides further support. Using dynamic film clips, they identified both categorical and continuous dimensions of perceived social meaning, such as emotion sharing, power dynamics, and action coordination. Their results show that ecologically valid social features are consistently appraised across observers. This aligns closely with our approach, which aims to empirically map and validate key dimensions of social relevance in static visual scenes. Our study diverges from theirs by using affective static images—common in visual neuroscience—and by linking social dimensions to valence and arousal ratings. We further conduct exploratory factor analyses to uncover the latent structure of social-affective appraisals, offering an empirically characterized, open-access dataset and visualization tool for broader use.
Prior work on affective scene perception has primarily relied on normative affective dimensions such as valence and arousal (e.g., IAPS) and, when social information is considered, often operationalizes it coarsely (e.g., presence/number of people). In contrast, we treat social relevance as an appraisal-like construct that depends on interactional meaning and relational context, and we operationalize it via separable dimensions (emotion sharing, action sharing, equality, and participatory arousal). This enables stimulus selection and control along social-interaction dimensions that are not captured by valence/arousal norms alone. In doing so, our dataset complements existing affective norm sets by providing structured social-appraisal annotations for complex scenes.
The present study
Whereas previous research has often relied on binary classifications of social content or focused narrowly on facial features, our study offers a comprehensive, dimensional approach to mapping social meaning in affective scenes. By combining continuous ratings of key social-affective dimensions with objective image features and established valence/arousal norms, we provide a structured resource for selecting and analyzing social stimuli in experimental research. This multidimensional dataset enables researchers to move beyond coarse contrasts—such as social versus non-social images—and instead target specific interpersonal properties, such as emotion sharing, action coordination, and equality. In doing so, our work supports more precise stimulus control and contributes to a richer understanding of how individuals appraise social relevance in complex visual environments.
We conducted a large-scale rating study of 296 affective images, focusing on evaluations of social and interpersonal content. In addition to standard ratings of scene pleasantness and participatory arousal, participants provided judgments of emotion sharing, action sharing, and interpersonal equality. Objective features such as number of visible individuals, face visibility, and gaze direction were coded independently.
We adopted an exploratory approach to investigate social-affective processing in complex visual scenes. Rather than testing specific hypotheses, we pursued three primary objectives. First, we aimed to establish the reliability and construct validity of the rating dimensions, and to examine the relationship between perceived social features and valence. We assessed interrater agreement and tested convergent validity by comparing our pleasantness and participation ratings with IAPS norms [1]. Second, we explored whether individual differences in personality traits and empathy influenced social relevance ratings. Third, we examined how the number of individuals present in a scene affected ratings across dimensions—allowing us to differentiate between scenes with single agents and those depicting multi-agent social interactions.
Methods
Participants
Participants were recruited online using a convenience sample provided by MakeOpinion (https://www.makeopinion.com/) from 12/12/2023–11/01/2024. Of the 497 recruited participants, eight did not give their consent, and four did not complete the study. Additionally, 47 participants were excluded because they failed the attention checks, and 14 were further excluded because their responses indicated lack of attention to the questions (see the paragraph on data processing).
All analyses were run on the final sample of 424 participants, between 18 and 64 years of age (M = 41.47, SD = 12.42). 37.03% of the participants identified as female, 62.50% as male, 0.24% as transgender male, and 0.24% preferred not to declare. Table 1 shows further details of the demographics of the final sample.
Ethics
This study was approved by the Ethics Committee of the Georg-Elias-Müller Institute for Psychology at the University of Göttingen (Approval No. 293, follow-up to Approval No. 199). All participants provided written informed consent electronically before participation. Consent was obtained via an online form integrated with the MakeOpinion platform, where participants actively clicked to confirm consent after reading the information sheet and GDPR/data protection statement. Only participants aged 18 years or older were eligible. No minors were included in the study.
Image selection
For the purpose of this study, we selected images from the IAPS database [1], focusing on scenes that depict individuals either alone, in dyadic interactions, or as part of a group. The selected images primarily portray everyday scenarios such as shopping, walking down the street, waiting at a bus stop, or engaging in casual conversation. Additionally, the set includes some images representing more intense but commonly encountered situations as portrayed in media, such as scenes of war, plane crashes, traffic accidents, and interpersonal conflict. To ensure ecological validity and ethical applicability in future research, we excluded images depicting extreme positive (e.g., erotic content) and extreme negative (e.g., body mutilation) affective content. This approach was intended to maintain a focus on naturalistic social contexts. Based on these criteria, an initial pool of 424 images was compiled. Duplicates and overly repetitive themes (e.g., multiple images of dental visits or children) were removed to improve the distinctiveness of the stimuli. The final image set was refined to include a balanced sample of 100 images per interaction category (single individuals, dyads, and groups), distributed as equally as possible across negative, neutral, and positive affective contexts as defined by the normed database ratings of valence [1]. During compilation of the three interaction-category lists, three images were found to have been miscoded, resulting in duplicate assignment across lists (two images duplicated across two lists, and one image duplicated across all three lists). These four erroneously duplicated entries were excluded from all analyses, yielding a final set of 296 unique images (103 single-person, 94 dyad, and 99 group images). A complete list of IAPS codes and associated data for the final stimulus set is available on the Open Science Framework (https://osf.io/9m3ef/).
Procedure
After providing consent, giving image examples (exemplary images were not part of the rating), participants viewed 50 images randomly selected from one of three possible subsets of 100 images (order randomized across participants) and rated each one on perceptual and affective dimensions using 9-point Likert scales (for an overview of the rating procedure see Fig 1). First, participants were asked to assess the overall social relevance of the scene by responding to the question: “Please look at the scene as a whole. How social is this scene?” (1 = “not at all social,” 5 = “neither,” 9 = “very social”). Participants were then asked to imagine themselves as part of the depicted scene and evaluate its affective qualities. They rated scene pleasantness (“How pleasant do you think it is to be in the situation depicted?”; 1 = “very pleasant,” 5 = “neither,” 9 = “very unpleasant”) and scene participation (“Does the scene involve a lot of movement and active participation, or a slower and calmer participation?”; 1 = “very active,” 5 = “neither,” 9 = “very calm”). To avoid confusion with traditional arousal norms (e.g., SAM arousal), we interpret this rating as participatory arousal, i.e., the perceived degree of involvement/action-readiness elicited by the scene. Accordingly, this judgment may be informed by both observable movement intensity and the rater’s subjective sense of being actively involved in the depicted situation. Next, they indicated the number of individuals in the image by selecting whether they saw one person or more than one (1 = “one,” 2 = “more than one”). For scenes containing more than one individual, participants completed additional ratings on interaction-based dimensions. These assessed emotional sharing (“Are the individuals in the scene sharing the same emotional experience?”; 1 = “completely shared,” 5 = “neither,” 9 = “completely separate”), with instructions clarifying that, for example, two people laughing would indicate shared emotion, while one laughing and one crying would indicate separate emotional experiences. They also rated action sharing (“Are the actions depicted in the scene performed collectively, or is each individual doing something on their own?”; 1 = “joined action,” 5 = “neither,” 9 = “separate actions”), with examples such as two people carrying a sofa together (joint action) or performing unrelated tasks like reading and fishing (separate actions). To further distinguish the two dimensions, we note that emotion sharing and action sharing can vary independently. For example, a scene can show high emotion sharing but low action sharing when two individuals passively observe the same event and display the same affect (e.g., both laughing or both appearing sad) without coordinated action. Conversely, a scene can show high action sharing but low emotion sharing when individuals coordinate a joint task (e.g., carrying an object together) while displaying different emotional expressions or levels of affect. In many naturalistic interactions, both dimensions co-occur (e.g., celebrating together), but conceptually they capture distinct aspects of interpersonal coordination – shared affective experience versus shared goal-directed action.
The images shown are illustrative examples from the OASIS database [23] and were not included since pictures used in the present study are copyrighted.
Lastly, participants rated equality in the scene (“Are the individuals in the scene equals, or is one person more in control than the other?”; 1 = “fully equal,” 5 = “neither,” 9 = “one person in control”), with examples including differences due to social roles or context, such as a police officer making an arrest or a teacher instructing a student. The equality rating captures perceived asymmetry in control or power within the depicted interaction and may therefore reflect both enduring role-based hierarchies (e.g., police officer) and momentary situational control. Its association with the sharing scales likely reflects that coordinated or mutually engaged interactions often coincide with perceived equality; nevertheless, we retain equality as a distinct, perception-based index of asymmetry.
All ratings were obtained without time constrictions. All items were originally given in German (see Item List German in the online repository materials for the original wording). Image ratings were accompanied by two attention checks (after 30 and 50 images) featuring non-used images with a 9-point Likert scale asking participants to select option 9 on the scale.
After the image ratings, all participants answered two questionnaires assessing personality and empathy traits. Personality traits were assessed using the Big Five Inventory (BFI-K; [24]) and empathy was assessed using the German version of the Interpersonal Reactivity Index (SPF; [25])
The objective image features were assessed by three independent coders and included the number of visible people (N People) and faces (N Face), grouping of people (People Group), visibility of faces (Face Visibility), gaze sharing between people in the image (Gaze Direction Within), and eye contact with the viewer (Gaze Direction Viewer). The responses from the three coders were compared, and discrepancies were resolved by taking the majority response for each image and feature. Objective image features were then combined with participant responses.
Data processing
Processing of the data was conducted in R (version 4.5.0; [26]), using tidyverse (v2.0.0; [27]) and readxl (v1.4.3; [28]) packages. The first step of data processing involved excluding non-consenting participants, those who did not complete the study, and participants who failed to correctly respond to both attention checks. Participant responses regarding the social relevance of the image, scene pleasantness, scene participation, presence of single or multiple persons in the image, as well as emotion sharing, action sharing, and equality within the image were then extracted for further processing.
Images were subsequently categorized as single-person images if the objective feature People Group indicated only a single individual in the image, or as multiple-person images if People Group indicated a dyad or group of individuals. Since the study included additional questions when participants indicated that multiple people were depicted in an image, it was possible that some participants might have intentionally indicated that multiple-person images contained a single person to complete the study faster. To address this, participants who disagreed with the independent coders regarding the number of people in the image in more than 20% of the trials were excluded (N = 14). Moreover, as the number of people in the image was treated as an objective feature based on the rating of the independent coders, all trials where participants disagreed with the coders’ assessments were removed. This resulted in a loss of 4.41% of the trials.
The ratings for social relevance, scene pleasantness, scene participation, emotion sharing, action sharing, and equality were adjusted so that a value of 1 represents the lowest rating and a value of 9 represents the highest rating for each variable.
Participant scores on the SPF (empathy) questionnaire were calculated by first reversing the scoring on the negatively worded items and then averaging the scores of the items belonging to each of the questionnaire sub-scales (“-“ indicate reverse-scored items): perspective taking (items: 3-, 8, 11, 15-), fantasy (1, 5, 7-, 12-, 16), empathic concern (2, 4-, 9, 14-) and personal distress (6, 10, 13-). The same approach was used for the participant scoring on the BFI-K questionnaire, aggregating the scoring belonging to the sub-scales: extraversion (1-, 6, 11-, 16), agreeableness (2-, 7, 12-, 17-), conscientiousness (3, 8-, 13, 18-), neuroticism (4, 9-, 14, 19) and openness (5, 10, 15, 20, 21-).
Interrater reliability of image ratings
Interrater reliability of the subjective ratings was assessed using intraclass correlation coefficients (ICCs). Given that our design was partially crossed—each image was rated by a subset of participants (with participants rating only 50 of 100 images within one of three distinct image sets) – the appropriate approach was to calculate ICC(1, k), i.e., based on a mean-rating, absolute-agreement, one-way random effects model [29]:
Where is the variance attributed to differences between images,
is the residual variance, and
is the number of raters per image. This was done via the function lmer of the R package lme4 (v1.1-37, [30]). Confidence intervals for the ICC estimates were obtained through parametric bootstrapping using the bootMer function (N = 10000 bootstraps).
Convergent validity with IAPS scales
To assess potential convergent validity between our rating scales and the norms of the IAPS database, Spearman’s correlation coefficients were calculated for each rating variable (social relevance, scene pleasantness, scene participation, emotion sharing, action sharing, equality) with both the valence and arousal norms of the IAPS database.
Exploratory correlation analysis
Correlation analysis was conducted to explore relationships between image ratings and objective image features. Since these variables are of different scales (i.e., interval, ordinal, nominal), we used an approach based on so-called difference semimatrices proposed by Podani et al. [31]. Briefly, given a dataset with variables (rows) and
observations (columns), for each variable
a
matrix
is calculated where each
entry
quantifies the difference between observations
and
on variable
. The way these differences are calculated depends on the scale of the variable
(see Podani et al. [31] for details). In a second step, the correlation between two variables
and
(referred to as d-correlation to distinguish it from the regular Pearson correlation) is calculated from the two semimatrices
and
as:
The analysis was conducted on the mean image ratings (averaged across participants) separately for single- and multiple-person images, since ratings of emotion sharing, action sharing and equality were available only for the latter.
We further visualized the average image ratings of scene pleasantness against scene participation to explore whether the two variables showed the typical “U-shaped” relationship of the IAPS norms of valence and arousal (Fig 3).
Calculation of composite engagement rating
The correlation analysis showed an almost perfect association between ratings of emotion sharing, action sharing and equality (see Results section), and an additional factor analysis revealed very high loadings of the three variables on one latent factor (Table S4 in S1 File; note, however, that this three-variable single-factor model is just-identified with zero degrees of freedom and therefore does not permit assessment of global model fit). To mitigate multicollinearity concerns when incorporating these variables into subsequent analyses, we computed a composite measure. This was achieved by averaging each participant's ratings of emotion sharing, action sharing, and equality for each image. The resulting composite measure, referred to here as engagement rating, was utilized in the following analysis steps.
Analysis of the effect of scene pleasantness on social relevance
A Linear Mixed Model (LMM; [32]) was used to investigate the effect of scene pleasantness on ratings of social relevance. The model also controlled for potential effects of scene participation and gaze direction in respect to the viewer (i.e., direct or averted). These two terms were included as main fixed effects, as the exploratory analysis did not reveal any particular linear or non-linear relationship between these variables and scene pleasantness. Moreover, since the ratings of social relevance, scene pleasantness and scene participation showed different distributions for single- and multiple-person images (Fig 2A-2C), an additional factor person number (with levels: single, multiple) was added in interaction with both scene pleasantness and scene participation.
Each dot represents an image’s mean rating (averaged across participants), the central box indicates the interquartile range with the median bar, and the density portion illustrates the overall shape of the distribution.
Participant ID and image ID were included as random intercepts. To maintain a 5% Type I error rate, all identifiable random slopes and their correlation with random intercepts were included [33,34], specifically for scene pleasantness and scene participation within participant ID. The full model therefore was defined as:
The overall effect of scene pleasantness was tested using a full-null model comparison [35], where the null model lacked the main effect of scene pleasantness and its interaction with person number, but was otherwise identical to the full model.
Analysis of the effect of engagement on social relevance
Another LMM was used to investigate the effect of the composite rating of engagement on the social relevance of the images. Even in this case the model controlled for the additional effects of scene participation and gaze direction. However, since the rating of engagement was only available for multiple-person images, no further effect of person number was investigated. Participant ID and image ID were included as a random intercept, and engagement and scene participation within participant ID as random slopes.
The full model was therefore defined as:
The corresponding null model lacked the fixed effect of engagement but was otherwise identical to the full model.
Analysis of the moderation effects of personality traits
We further investigated whether participants’ personality traits could moderate the effects of scene pleasantness and engagement on social relevance. To this aim, two additional LMMs were defined, including the interaction effects of each sub-scale of the SPF and BFI-K with scene pleasantness or engagement, respectively. The moderation effect of personality traits was tested by comparing each model with a reduced one lacking the interaction term while being otherwise identical to the respective full model.
Even in this case, main effects of scene participation and gaze direction were included as control predictors, participant ID and image ID as random intercept, and scene pleasantness/engagement and scene participation within participant ID as random slopes.
All LMMs described above were fitted in R with the function lmer from the lme4 package. Before fitting the models, all numerical predictors were z-transformed (mean-centered and scaled) to provide a valid reference level. The significance of the fixed effects was tested using the Satterthwaite approximation [36] via the function lmer from the lmerTest package (version 3.1−3; [37]). The assumptions of normally distributed and homogeneous residuals were checked by visually inspecting QQ-plots of the residuals [38] and by plotting residuals against fitted values [39]. Model stability at the level of estimated coefficients and standard deviation was assessed by excluding one level of the random effect (i.e., participant ID and image ID) at a time and refitting the model [40], using a function kindly provided by Roger Mundry (University of Göttingen). Confidence intervals for the model estimates were obtained by parametric bootstrapping using the bootMer function from the lme4 package, with 1000 bootstraps. Due to the different computation methods, the interpretation of Satterthwaite approximated p-values and bootstrapped CIs might differ. The absence of collinearity issues was verified by calculating the Variance Inflation Factors (VIF) for a model without interaction terms, using the vif function from the car package (version 3.1−3; [41]).
Results
Descriptive statistics of all the image rating variables and image objective features are presented in Table 2, while Fig 2 shows the distribution of the image rating variables.
Interrater reliability of image ratings
ICC estimates and their 95% confidence intervals revealed excellent [29] interrater reliability for all image rating variables (Table 3). Per-image standard deviations across rating dimensions are reported in Table S1 in S1 File, confirming meaningful variability across stimuli.
Convergent validity with IAPS normative data (valence and arousal)
Table 4 shows the results of the Spearman’s correlation of each of the rating variables and the IAPS norms of valence and arousal. All rating variables, except scene participation in multiple-person images, presented a statistically significant correlation with valence norms, with a very strong positive association for social relevance and scene pleasantness, a strong positive association for emotion sharing, action sharing and equality and a weak association for scene participation in single-person images. Conversely, all rating variables showed a statistically significant correlation with arousal scores, moderately positive for scene participation, moderately negative for social relevance and pleasantness, and weakly negative otherwise.
Exploratory correlation analysis
The full results of the exploratory correlation analysis are presented in Tables S2-S3 in S1 File).
Three patterns in particular emerged from the analysis. First, social relevance showed a very strong positive correlation with scene pleasantness (single-person images: d = .95, multiple-person images: d = .92) and a strong association with emotion sharing, action sharing and equality (d = .78,.76,.70, respectively).
Second, pleasantness and participation showed only a weak association for single-person images (d = .31) and essentially no association for multiple-person images (d = −.04). The absence of a linear association does not necessarily exclude some other forms of quadratic or non-linear relationship between these two variables, as in the case of the typical “U-shape” of the IAPS valence vs arousal. However, as can be seen in Fig 3A, no clear association, linear or otherwise, can be observed between the two variables.
(B) IAPS norm of valence vs arousal for the same images used in the current study.
Third, emotion sharing, action sharing and equality showed strong or very strong positive associations among each other (emotion sharing vs action sharing: d = .95; emotion sharing vs equality: d = .90; action sharing vs equality: d = .81), suggesting that these three variables might tap one latent factor which can be interpreted as a measure of engagement in the image. An additional exploratory factor analysis was conducted to support this interpretation (see Table S4 in S1 File). Because this engagement composite was derived from only three highly correlated indicators, the corresponding factor model is just-identified and does not permit assessment of global model fit. Accordingly, the engagement score is treated here as a descriptive composite index introduced for data-reduction and to mitigate multicollinearity, rather than as a formally validated latent construct. Independent validation with additional indicators or in an independent sample will be an important direction for future work.
All LMMs used in the analyses showed no multicollinearity issue and good stability. Visual inspection of the model diagnostics indicated that the assumptions of normality and homogeneity were reasonably met. Summary of the LMMs is presented in Tables S5-S7 in S1 File.
Effect of scene pleasantness on social relevance
The effect of scene pleasantness on the ratings of social relevance was tested while controlling for scene participation, gaze direction and number of persons in the scene. The analysis showed a significant association between scene pleasantness and social relevance (full-null model comparison: χ² = 880.04, df = 2, p < .001). More specifically, the model showed a significant interaction of scene pleasantness and person number (β = −0.10, SE = 0.04, p = .006, 95% CI [−0.16, −0.03], Table S5 in S1 File). Increased ratings of scene pleasantness were associated with increased social relevance, and this association was slightly stronger for single-person images compared to multiple-person ones (Fig 4A).
Shaded areas represent 95% confidence intervals.
Effect of engagement on social relevance
Our second model tested the effect of engagement on social relevance, controlling for scene participation and gaze direction. The test revealed a significant association between engagement and social relevance (full-null model comparison: χ² = 340.94, df = 1, p < .001). As can be seen in Fig 4B, increased engagement ratings were associated with increased social relevance ratings (β = 0.63, SE = 0.03, p < .001, 95% CI [0.58, 0.69], Table S6 in S1 File).
Moderation of scene pleasantness effects by personality traits
Participants’ personality traits were tested as potential moderators of the effect of scene pleasantness on social relevance. This was done by fitting a full model including the interaction effects of each SPF and BFI-K sub-scale and scene pleasantness and comparing it to a reduced model lacking such interaction.
The comparison showed no significant difference between the full and the reduced model (χ² = 11.26, df = 9, p ≥ .05), suggesting that personality traits play no role in moderating the effect of scene pleasantness on ratings of social relevance.
The absence of a moderation effect does not imply that personality traits alone might not have a direct effect on social relevance. To test this, the reduced model was compared to a null model lacking the SPF and BFI-K sub-scales predictors entirely. This comparison was also not significant (χ² = 16.07, df = 9, p ≥ .05), indicating that personality traits did not contribute incremental explanatory power beyond scene pleasantness (with scene participation and gaze direction as controls), either as main effects or as moderators of the pleasantness effect.
Moderation of engagement effects by personality traits
Personality traits were also tested as potential moderators of the effects of engagement on social relevance. In this case, the full-reduced model comparison was significant (χ² = 20.95, df = 9, p = .013), indicating that, as a whole, personality traits moderated the relationship between engagement and social relevance. More precisely, the analysis of the individual fixed effects showed a significant interaction of engagement and extraversion (β = 0.06, SE = 0.02, p < .012, 95% CI [0.01, 0.1], Table S7 in S1 File). Consistent with the results reported above, increased engagement was associated with increased ratings of social relevance, and this effect was stronger the higher the participant’s level of extraversion (Fig 4C). Given the multiple moderation tests, the p-values for the nine interaction terms were further evaluated using the Holm correction. The interaction effect between engagement and extraversion was no longer significant after this correction (p = .104) and is thus considered tentative. Independent replication is needed to confirm its robustness.
Discussion
This study set out to systematically map how individuals appraise social relevance in affective scenes by collecting continuous ratings along key social-affective dimensions and linking them to both objective image features and affective experience. We found that all rating dimensions showed excellent interrater reliability—indicating that participants applied the scales consistently and interpreted the constructs in a shared, reliable manner—and strong convergent validity with existing affective norms, supporting the robustness of the social-affective appraisal framework. Critically, social relevance was most strongly associated with scene pleasantness and a composite measure of interpersonal engagement—reflecting shared emotion, joint action, and perceived equality—highlighting that social meaning in complex scenes is both reliably perceived and structured along interpretable dimensions.
All rating dimensions showed excellent interrater reliability, confirming that participants applied the scales consistently and interpreted the constructs similarly. The very high interrater reliabilities observed across rating dimensions (ICCs > .98) warrant careful interpretation. While high ICCs could in principle arise from trivial or coarse judgments, this explanation is unlikely in the present study. All stimuli depicted individuals, eliminating a simple people–versus–no-people distinction, and substantial variability was intentionally introduced within this constrained stimulus space, spanning positive, neutral, and negative scenes as well as diverse social contexts. Importantly, ICC reflects the ratio of between-stimulus variance to residual (rater-specific) variance; consequently, homogeneous or weakly differentiated stimulus sets would be expected to reduce ICC estimates rather than inflate them. The exceptionally high ICCs observed here therefore indicate that images were reliably differentiated along the assessed dimensions, and that participants converged on shared subjective judgments rather than responding uniformly. In this sense, high reliability reflects both a well-structured stimulus set and the meaningfulness of the rated constructs. Pleasantness ratings showed strong convergent validity with the original IAPS valence norms, indicating a close correspondence between experienced valence during passive viewing and evaluations of pleasantness when imagining oneself within the scene. In contrast, participation ratings diverged from IAPS norms, likely reflecting the distinct nature of our participatory arousal measure, which captures imagined engagement (e.g., active vs. calm participation) rather than stimulus-driven physiological arousal. This distinction suggests that participatory arousal may tap into a different facet of affective processing—one that is more closely tied to imagined involvement and action readiness than to passive emotional intensity. Nevertheless, the moderate correlations between participatory arousal and IAPS arousal (rho = .41 for single-person images and .58 for multiple-person images) indicate substantial empirical overlap; future work should clarify what unique variance participatory arousal captures beyond stimulus-driven physiological arousal. To situate the present norms within established affective picture research, we examined convergent validity with published IAPS valence and arousal ratings for the subset of overlapping images. The expected correspondence supports the validity of our affective ratings while underscoring that social relevance and the interaction-structural dimensions assessed here provide complementary information that is not captured by valence/arousal alone. Taken together, this dataset extends existing normative resources by moving beyond coarse indicators of “social content” and providing structured annotations of social meaning in complex scenes – capturing shared emotion, shared action, interpersonal equality, and participatory arousal. This enables researchers to select and control stimuli along interpretable social dimensions and to test hypotheses about how social relevance shapes attention, affective evaluation, and memory under ecologically richer conditions.
Social relevance was most strongly associated with scene pleasantness and the degree of interpersonal engagement in the scene, as reflected in shared emotions, actions, and perceived equality. This suggests that social scenes are often perceived as more meaningful when they are also emotionally positive and interactive—capturing reciprocal exchanges between individuals. Interestingly, the association between pleasantness and social relevance was strongest in scenes with limited social context (i.e., images depicting a single individual), indicating that even subtle interpersonal cues may activate social appraisal mechanisms when interpreted as affiliative or inviting. The relationship between social relevance and pleasantness was not significantly moderated by most personality traits or empathy, pointing to a partially universal structure of social-affective evaluation across perceivers. The only exception was extraversion, which was associated with a stronger link between engagement and social relevance. This may reflect a greater attunement among extraverted individuals to opportunities for social connection, consistent with trait-level differences in sociability and reward sensitivity in social contexts [42,43]. However, the effect was modest and did not survive correction for multiple interaction tests. This finding should therefore be interpreted cautiously, warranting replication in independent samples.
Our findings offer a methodologically grounded contribution to the emerging effort to formalize the construct of social relevance in visual scene perception. In this context, our study complements and extends the recent work of Santavirta et al. [22], who proposed a data-driven taxonomy of social features based on observer ratings of dynamic film clips. Their findings revealed that dimensions such as emotion sharing, power asymmetries, and action coordination are reliably perceived and cluster into meaningful patterns – underscoring the value of systematic feature annotation in understanding social perception. Building on this rationale, our study applies a similar dimensional framework to static affective images, which remain the standard format in many experimental paradigms across cognitive and affective neuroscience. This methodological focus not only enhances compatibility with existing databases like the IAPS but also supports wider adoption of structured social-affective ratings in stimulus selection and experimental design.
The strong correlation we observed between social relevance and pleasantness in single-person images suggests that even minimal interpersonal cues can influence evaluative appraisals. This pattern is consistent with appraisal theories, particularly the Component Process Model [44,45], which posit that early relevance checks—such as novelty, intrinsic pleasantness, and goal conduciveness – are sensitive to social contextual features. Our findings provide a concrete operationalization of these early appraisal components, showing how relational cues embedded in affective scenes systematically influence perceived social meaning. In this sense, our dimensional ratings capture not just descriptive scene features but also the psychological mechanisms through which individuals interpret and evaluate complex social contexts.
The strong association between perceived pleasantness and social relevance likely reflects the fundamental human need for belonging [46], which not only motivates social behavior but also shapes perceptual and evaluative processes. When individuals identify potential opportunities for interaction—particularly affiliative ones—these scenes are appraised as both socially meaningful and emotionally positive. This interpretation aligns with prior findings demonstrating that social cues can shift emotional processing priorities in complex scenes [3], and continue to guide attention even under cognitive load [18]. Such robustness is echoed in our own finding of high interrater agreement across social-affective dimensions, despite the complexity and variability of the scenes.
At the neural level, recent work has shown that social meaning is represented in structured, multidimensional formats. Zhang et al. [20] identified a network—including hippocampus and precuneus—that encodes individuals’ positions within a social space defined by power and affiliation. Similarly, Wittmann et al. [21] demonstrated that dorsomedial prefrontal cortex abstracts social interactions into compressed basis functions, enabling efficient representation of multi-agent contexts. Our study complements these findings by showing that static images—rather than interactive or dynamic tasks—can elicit reliably structured appraisals of social relevance. Moreover, we provide an empirically characterized tool to quantify these appraisals, facilitating systematic investigation of social relevance across diverse visual paradigms.
The fit between social relevance and affective pleasantness was not meaningfully influenced by individual differences in personality traits or empathy. Our results suggest that, as long as affective properties such as pleasantness and participation are accounted for, dispositional factors play a limited role in shaping social appraisals of visual scenes. This finding points toward a relatively universal and robust evaluation process, wherein observers consistently detect and respond to social cues—such as shared emotion or interpersonal equality—regardless of their baseline level of sociability, emotional sensitivity, or perspective-taking. Such consistency supports the view that certain social-affective appraisals may operate as early, automatic processes, sensitive to contextual features of the scene rather than stable internal traits. From a theoretical standpoint, this aligns with appraisal models proposing that relevance detection (e.g., goal conduciveness or intrinsic pleasantness) constitutes a fast, low-threshold mechanism that guides attention and meaning-making across individuals [44,45]. While more nuanced individual differences may still influence downstream emotional responses or decision-making, our findings suggest that the initial appraisal of social relevance in affective scenes is largely stable across perceivers.
A central implication of the present norms is that perceived social relevance in complex scenes is tightly coupled to interaction structure, particularly emotion sharing and action sharing. Conceptually, emotion sharing captures the extent to which observers perceive a shared affective state across agents, aligning with accounts that treat interpersonal affect as a key cue for social meaning, coordination, and affiliation. Action sharing reflects coordinated goal-directed behavior and thus relates to theories emphasizing joint action, intentionality, and agency as organizing principles of social perception. Importantly, these dimensions show systematic associations with affective evaluations, suggesting that social meaning is not reducible to valence/arousal but interacts with them—e.g., scenes that depict shared affect or coordinated action may be evaluated as more socially relevant and, depending on context, more pleasant and engaging. This supports an appraisal-like view in which observers integrate relational cues (shared emotion, shared action, asymmetry) to derive social meaning and affective significance from scenes.
Our findings extend prior work by directly linking social features in visual scenes to affective valence and arousal, demonstrating that social and emotional appraisals are not independent but co-organized into structured, interpretable dimensions. Emotion sharing, action coordination, and perceived equality consistently covaried with pleasantness ratings and were robust across observers, reinforcing the idea that social meaning constitutes a core, rather than peripheral, dimension of affective scene perception. This structured co-variation offers a more nuanced account of how individuals evaluate complex scenes—one that incorporates interpersonal context as an integral component of emotional appraisal. By providing an empirically characterized and openly accessible dataset annotated along multiple social-affective dimensions, our study also addresses a practical need for standardized tools that enable targeted stimulus selection in experimental research. Together, these contributions support both theoretical refinement and methodological precision in the study of social relevance.
These findings have direct implications for the design of future experiments in affective neuroscience, social cognition, and related fields. Rather than relying on coarse categorical contrasts—such as the mere presence or absence of human figures—researchers should consider selecting stimuli based on multidimensional social-affective properties. For example, studies on empathy, moral judgment, or interpersonal decision-making may benefit from using stimuli that systematically vary in engagement-related aspects, such as action sharing or perceived equality. Additionally, incorporating objective image features – such as number of individuals depicted, face visibility, or gaze direction – can facilitate hybrid models that integrate low-level perceptual input with high-level social meaning. Our dataset provides a structured foundation for such approaches, supporting more ecologically valid and theoretically precise investigations of socio-affective information processing.
Several limitations should be acknowledged. First, although our stimuli spanned a range of emotional and social content, all ratings were based on static images presented without contextual or temporal information. While this format is widely used in affective neuroscience, it may limit ecological validity and underrepresent the dynamic nature of social interactions. Nonetheless, by working within the constraints of a standardized and widely adopted image set, we provide an empirically grounded resource for improving stimulus selection. Second, the final stimulus set was not fully balanced across all dimensions of interest, such as interaction type, valence, and arousal. As a result, some combinations of features, e.g., negative group scenes with high equality, were underrepresented due to natural constraints in the IAPS database, potentially limiting the generalizability of fine-grained comparisons. Additionally, the images were rated by an exclusively German sample. Although this sample was not limited to students or younger age groups, it nevertheless restricts the cultural generalizability of the findings. In particular, cultural norms regarding hierarchy, collectivism/individualism, and emotional expression may systematically shape appraisals of interpersonal equality and emotion sharing in depicted interactions. Cultural differences may influence attentional patterns as well as emotional and social appraisals of visual scenes; therefore, further research is needed to provide a more nuanced understanding of social relevance across cultural contexts. Cross-cultural norming efforts (e.g., collecting ratings in additional countries and testing measurement invariance across samples) would help extend the present findings beyond a German sample and provide a more comprehensive understanding of social relevance across cultures. Methodologically, future extensions could strengthen inference by combining static-scene ratings with (i) cross-cultural norming, (ii) more fine-grained individual-difference measures (e.g., empathy, social anxiety), and (iii) behavioral/physiological indices (e.g., eye tracking, pupil dilation) to link subjective social dimensions to attentional dynamics and arousal. Such work would help clarify boundary conditions and improve generalizability while preserving the practical utility of normative ratings for stimulus selection. Finally, while our rating dimensions were both theory-driven and empirically supported, they do not exhaust the full range of social meaning. Future extensions might incorporate additional constructs such as perceived affiliation, threat, or intentionality to further refine the structure of social-affective appraisal.
Taken together, our study provides a theoretically informed, methodologically rigorous framework for quantifying social relevance in visual scene perception. By bridging the gap between standardized affective image sets and the complex interpersonal appraisals that guide human social behavior, we offer a novel resource for investigating how emotional and social meanings co-construct perceptual experience. As research increasingly emphasizes the relational structure of social cognition—including spatial, hierarchical, and affective dimensions—our work contributes an accessible and empirically characterized toolset for operationalizing these features in experimental designs. Future research may use our dataset to test mechanistic models of social relevance computation, explore developmental or cultural differences in social appraisal, or integrate behavioral and neural measures to advance comprehensive accounts of how humans perceive, interpret, and respond to social meaning in complex environments.
In conclusion, the present dataset complements established affective norm sets by providing structured ratings of social relevance and interaction dimensions in complex scenes. By decomposing social meaning into emotion sharing, action sharing, interpersonal equality, and participatory arousal, it enables researchers to select and control stimuli along interpretable social dimensions beyond valence and arousal. At the same time, the norms should be extended via cross-cultural validation and targeted individual-difference measures to test robustness and boundary conditions. We provide the full dataset and accompanying tools to facilitate these future efforts.
Supporting information
S1 File. Results of the exploratory correlation analysis; factor analysis on engagement ratings; and LMM summaries (Tables S1–S7; Figures S1–S2).
https://doi.org/10.1371/journal.pone.0331156.s001
(DOCX)
S1 Fig. Social relevance as a function of scene participation, for single- and multiple-person images.
Shaded areas represent 95% confidence intervals.
https://doi.org/10.1371/journal.pone.0331156.s002
(TIF)
S2 Fig. Social relevance as a function of scene participation, for direct and averted gaze direction.
Shaded areas represent 95% confidence intervals.
https://doi.org/10.1371/journal.pone.0331156.s003
(TIF)
Acknowledgments
We thank Roger Mundry for making functions for the analyses available and Weicheng Lian for assistance in data collection.
References
- 1.
Lang PJ, Bradley MM, Cuthbert BN. International affective picture system (IAPS): Affective ratings of pictures and instruction manual. Gainesville, FL: University of Florida. 2008.
- 2. Colden A, Bruder M, Manstead ASR. Human content in affect-inducing stimuli: A secondary analysis of the international affective picture system. Motiv Emot. 2008;32:260–9.
- 3. Dudarev V, Wardell V, Enns JT, Kerns CM, Palombo DJ. Social cues tip the scales in emotional processing of complex pictures. Psychol Res. 2024;88(8):2221–33. pmid:39167127
- 4. Teh EJ, Yap MJ, Liow SJR. PiSCES: Pictures with social context and emotional scenes with norms for emotional valence, intensity, and social engagement. Behav Res Methods. 2018;50(5):1793–805. pmid:28842854
- 5. Fischer A, Postin D, Meiners L, Kulke L, Vrticka P, Schacht A. Social content mitigates the negativity bias in affective scene processing – a co-registered EEG and eye movements study. OSF. 2024.
- 6. Schacht A, Vrtička P. Spatiotemporal pattern of appraising social and emotional relevance: Evidence from event-related brain potentials. Cogn Affect Behav Neurosci. 2018;18(6):1172–87. pmid:30132268
- 7. Mumenthaler C, Sander D. Social appraisal influences recognition of emotions. J Pers Soc Psychol. 2012;102(6):1118–35. pmid:22288528
- 8. Teufel C, von dem Hagen E, Plaisted-Grant KC, Edmonds JJ, Ayorinde JO, Fletcher PC. What is social about social perception research?. Front Integr Neurosci. 2013.
- 9. Skripkauskaite S, Mihai I, Koldewyn K. Attentional bias towards social interactions during viewing of naturalistic scenes. Q J Exp Psychol (Hove). 2023;76(10):2303–11. pmid:36377819
- 10. Silva HD, Campagnoli RR, Mota BEF, Araújo CRV, Álvares RSR, Mocaiber I, et al. Bonding Pictures: Affective Ratings Are Specifically Associated to Loneliness But Not to Empathy. Front Psychol. 2017;8:1136. pmid:28740473
- 11. Cacioppo JT, Berntson GG, Decety J. Social neuroscience and its relationship to social psychology. Soc Cogn. 2010;28(6):675–85. pmid:24409007
- 12. End A, Gamer M. Preferential processing of social features and their interplay with physical saliency in complex naturalistic scenes. Front Psychol. 2017;8:418. pmid:28424635
- 13. Flechsenhar A, Rösler L, Gamer M. Attentional selection of social features persists despite restricted bottom-up information and affects temporal viewing dynamics. Sci Rep. 2018;8(1):12555. pmid:30135443
- 14. Rösler L, End A, Gamer M. Orienting towards social features in naturalistic scenes is reflexive. PLoS One. 2017;12(7):e0182037. pmid:28742873
- 15. Kosonogov V, Martínez-Selva JM, Torrente G, Carrillo-Verdejo E, Sánchez-Navarro J. Does social content influence the subjective evaluation of affective pictures? Span J Psychol. 2020;23:e6. pmid:32460951
- 16. Nummenmaa L, Calder AJ. Neural mechanisms of social attention. Trends Cogn Sci. 2009;13(3):135–43. pmid:19223221
- 17. Niu Y, Todd RM, Anderson AK. Affective salience can reverse the effects of stimulus-driven salience on eye movements in complex scenes. Front Psychol. 2012;3:336. pmid:23055990
- 18. Martinez-Cedillo AP, Dent K, Foulsham T. Social prioritisation in scene viewing and the effects of a spatial memory load. Atten Percept Psychophys. 2024;86(4):1237–47. pmid:37563513
- 19. Dima DC, Tomita TM, Honey CJ, Isik L. Social-affective features drive human representations of observed actions. Elife. 2022;11:e75027. pmid:35608254
- 20. Zhang L, Chen P, Schafer M, Zheng S, Chen L, Wang S, et al. A specific brain network for a social map in the human brain. Sci Rep. 2022;12(1):1773. pmid:35110581
- 21. Wittmann MK, Lin Y, Pan D, Braun MN, Dickson C, Spiering L, et al. Basis functions for complex social decisions in dorsomedial frontal cortex. Nature. 2025;641(8063):707–17. pmid:40074892
- 22. Santavirta S, Malén T, Erdemli A, Nummenmaa L. A taxonomy for human social perception: Data-driven modeling with cinematic stimuli. J Pers Soc Psychol. 2024;127(6):1146–71. pmid:39374128
- 23. Kurdi B, Lozano S, Banaji MR. Introducing the Open Affective Standardized Image Set (OASIS). Behav Res Methods. 2017;49(2):457–70. pmid:26907748
- 24. Rammstedt B, John OP. Kurzversion des Big Five Inventory (BFI-K): Entwicklung und Validierung eines ökonomischen Inventars zur Erfassung der fünf Faktoren der Persönlichkeit. Diagnostica. 2005;51:195–206.
- 25.
Paulus C. Der Saarbrücker Persönlichkeitsfragebogen SPF(IRI) zur Messung von Empathie: Psychometrische Evaluation der deutschen Version des Interpersonal Reactivity Index. 2009. https://psycharchives.org/en/item/2b090feb-4f9e-4ea2-9524-13151172f615
- 26.
R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. 2025.
- 27. Wickham H, RStudio. Tidyverse: easily install and load the “tidyverse.”. 2023. https://cran.r-project.org/web/packages/tidyverse/index.html
- 28.
Wickham H, Bryan J. readxl: Read Excel Files. Posit, PBC. 2023. https://cran.r-project.org/web/packages/readxl/index.html
- 29. Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155–63. pmid:27330520
- 30. Bates D, Mächler M, Bolker BM, Walker SC. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67:1–48.
- 31. Podani J, Schmera D, Bagella S. Correlating variables with different scale types: A new framework based on matrix comparisons. Methods Ecol Evol. 2023;14(4):1049–60.
- 32. Baayen RH, Davidson DJ, Bates DM. Mixed-effects modeling with crossed random effects for subjects and items. J Mem Lang. 2008;59:390–412.
- 33. Barr DJ, Levy R, Scheepers C, Tily HJ. Random effects structure for confirmatory hypothesis testing: Keep it maximal. J Mem Lang. 2013;68(3):10.1016/j.jml.2012.11.001. pmid:24403724
- 34. Schielzeth H, Forstmeier W. Conclusions beyond support: overconfident estimates in mixed models. Behav Ecol. 2009;20(2):416–20. pmid:19461866
- 35. Forstmeier W, Schielzeth H. Cryptic multiple hypotheses testing in linear models: overestimated effect sizes and the winner’s curse. Behav Ecol Sociobiol. 2011;65(1):47–55. pmid:21297852
- 36. Luke SG. Evaluating significance in linear mixed-effects models in R. Behav Res Methods. 2017;49(4):1494–502. pmid:27620283
- 37. Kuznetsova A, Brockhoff PB, Christensen RHB. lmerTest Package: Tests in Linear Mixed Effects Models. J Stat Softw. 2017;82:1–26.
- 38.
Field A. Discovering statistics using SPSS. London: Sage Publications Ltd. 2005.
- 39. Quinn GP, Keough MJ. Experimental Design and Data Analysis for Biologists. Cambridge: Cambridge University Press. 2002.
- 40. Nieuwenhuis R, te Grotenhuis M, Pelzer B. Influence.ME: Tools for detecting influential data in mixed effects models. R J. 2012;4:38–47.
- 41.
Fox J, Weisberg S. An R Companion to Applied Regression. SAGE Publications, Inc. 2018.
- 42. Smillie LD, Cooper AJ, Wilt J, Revelle W. Do extraverts get more bang for the buck? Refining the affective-reactivity hypothesis of extraversion. J Pers Soc Psychol. 2012;103(2):306–26. pmid:22582899
- 43. Wilkowski BM, Ferguson EL. Just loving these people: Extraverts implicitly associate people with reward. J Res Personal. 2014;53:93–102.
- 44. Scherer KR. Appraisal Considered as a Process of Multilevel Sequential Checking. Appraisal Processes in Emotion. Oxford University PressNew York, NY. 2001. p. 92–120.
- 45. Sander D, Grandjean D, Scherer KR. A systems approach to appraisal mechanisms in emotion. Neural Netw. 2005;18(4):317–52. pmid:15936172
- 46. Baumeister RF, Leary MR. The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychol Bull. 1995;117(3):497–529. pmid:7777651