Who Is Interacting With Whom? Assessing the Relationship Between Personality Traits and Preferences for Interaction Partners in Real Life

Human social situations are complex. For instance, various characteristics of interaction partners can affect social outcomes (Schwerdtfeger et al., 2020; Vanderhasselt et al., 2018). In social relationships, individuals are initially attracted to and satisfied by persons who are very similar to them in a variety of personality traits (Montoya et al., 2008). According to the reinforcement-affect explanation, similarity in values reinforces individuals’ opinions and feelings, thereby triggering an implicit affective response that increases attraction (Clore & Byrne, 1974; Izard, 1960). At the same time, complementary personality traits can also be adaptive for lasting bonds (Markey & Markey, 2007). In general, findings on whether similar or complementary personality profiles promote positive outcomes are mixed. Individuals might prefer personality profiles in their interaction partners that promote relationship success (Sacco & Brown, 2018). Participants rate social interaction quality as higher when partners are similarly extraverted but also report lower interaction quality with similarly disagreeable partners (Cuperman & Ickes, 2009). Personality similarity can also adversely affect marital satisfaction. Negative marital satisfaction outcomes increase when partners are similarly extraverted and conscientious (Shiota & Levenson, 2007). However, men and women generally prefer partners who seem to be more conscientious, extraverted, and agreeable but less neurotic than themselves (Figueredo et al., 2006).

Most studies on mutual personality influences are conducted in strictly controlled laboratory experiments and/or refer to similarity in close bonds such as friendships or romantic relationships (e.g., Barelds & Barelds-Dijkstra, 2007; Decuyper et al., 2012; Leikas et al., 2018; Selfhout et al., 2009). It is therefore unclear whether these similarity effects also translate to preferences for (dis)similar interaction partners in versatile everyday-life settings with familiar and less familiar interaction partners. To overcome previous limitations, the present research aimed to expand the social network to all natural interactions that occur in daily life (e.g., friends, romantic partners, work colleagues, strangers) in different social contexts (e.g., in private, at work, in public locations such as restaurants or bars). As an everyday-life example, contextual effects like (un)familiar background music bias the personality perception of other persons. With increasing levels of neuroticism, familiar music led to an increase in perceived dissimilarity in the extraversion and openness of another person in a laboratory study (Moss et al., 2007). Thus, environmental influences may inhibit or even cancel out actual (dis)similarity effects. However, less is known about how the interrelation between a person’s traits and the traits they attribute to their interaction partner affect social preferences for interaction partners and the evaluation of the interaction. The mutual influence of a person’s personality traits and perceived traits of a partner were recently tested using the trust game (Weiß et al., 2021). The study showed that participants scoring high on anxiety (a facet of neuroticism) entrusted more money to trustees which were represented by faces rated higher on conscientiousness. Extending previous approaches to multiple personality traits and more diverse social interaction settings, the current study focusses on participants’ own Big Five personality traits and the perceived personality traits of the interaction partners in their everyday lives and how they predict the (positive) evaluation of a social interaction.

To gain a comprehensive impression of (perceived) personality trait interactions in everyday life, we used ecological momentary assessment (EMA). EMA studies repeatedly assess individuals’ behavior and experiences in naturalistic settings (Moskowitz et al., 2009). Key benefits of this approach are increased ecological validity and a reduced retrospective bias (Shiffman et al., 2008), which in turn facilitate more accurate representations of subjectively perceived processes (Trull & Ebner-Priemer, 2013). Ecological validity is especially improved as the (social) contexts captured via EMA are those that matter to the participant as compared to laboratory experiments (Shiffman et al., 2008). Thus, we expected to collect data in a multitude of different social contexts so that we could deduce average preferences for (dis)similar interaction partners.

To narrow down the number of hypotheses based on the literature (e.g., Cuperman & Ickes, 2009; Figueredo et al., 2006), we focused on participants’ (self) and their interaction partners’ (partner) extraversion, agreeableness, and neuroticism.

We expected that participants may prefer interaction partners whom they rate as highly extraverted (H1a) and interaction partners whom they rate as highly agreeable (H1b), reflected by a significant positive main effect of partner agreeableness and extraversion on preference to interact. In contrast, we assumed that partner neuroticism may be associated with a decrease in preference to interact with that person (H1c), reflected by a significant negative main effect of partner neuroticism. Moreover, we assumed that the participants’ personality may interact with the personality of the interaction partner (H2), reflected by significant two-way interactions between each self-trait (i.e., extraversion, agreeableness, and neuroticism) and each partner trait (i.e., extraversion, agreeableness, and neuroticism). However, due to the mixed results on personality (dis)similarity and its impact on social interactions (Cuperman & Ickes, 2009; Figueredo et al., 2006; Shiota & Levenson, 2007), we did not specify a direction for H2.

As there was no previous study investigating subjective preferences to interact with someone based on their personality, we approached the estimation of sample size as follows. Using prototypical personality faces as targets of preferences, Sacco and Brown (2018) reported correlations of r = .14 for matching preferences of trait open male and female participants for openness faces and of r = .20 for trait neurotic female participants preferring agreeable male, but less agreeable female faces (r = -.16). However, inferring personality only from a neutral facial expression might be more complicated than integrating all information from a real-life social interaction (e.g., mimic, gestures, facial expression, verbal communication). First, we conducted a safeguard power analysis (Perugini et al., 2014) on the lower bounds of a moderate effect size (r = .30) using an 80% confidence interval. This resulted in a sample size of N = 118. To account for the repeated measures design, we also performed a follow-up power analysis using the ema.powercurve function from the R package EMAtools (Kleiman, 2017). This analysis indicated that a sample of N = 130 with 6 prompts per day for 7 days and an expected mean completion rate of 85% would be required to detect a small effect size (i.e., a Cohen’s d of 0.2) with 80% power and an intraclass correlation coefficient (ICC) of .12 (based on a previous study on person-situation interactions; McCormick et al., 2019). To motivate participants for high compliance, they were rewarded with a €10 show-up fee and an additional €0.50 per completed prompt (i.e., max. €21 additional reward). The R Script for the power analysis can be found at https://osf.io/h4y3s. We aimed for a gender-balanced sample and continued recruiting until 130 (i.e., approx. 65 women and 65 men) valid datasets were available (see below for exclusion criteria). We advertised the study via an online recruitment portal of the University of Würzburg, Germany. The protocol was approved by the local ethics committee. All participants provided informed consent.

Inclusion criteria were fluency in German and exclusion criteria were current pregnancy or lactation period, cardiovascular illnesses, chronic neurological disorders, acute psychiatric disorders, other severe medical illnesses, psychotropic medication, and visual and motoric impairments. We included infrequency-items and instructed-items in trait questionnaires for attention checks following recommendations for online studies (i.e., not in a controlled laboratory) on inter-individual differences (Zorowitz et al., 2021). We excluded prompts with incorrect answers for more than one of the instructed attention check catch items (e.g., “If you are paying attention to these questions, please select ‘A little’ as your answer”; Gillan et al., 2016) or infrequency items (“Over the last two weeks, how much time did you spend worrying about the 1977 Olympics?”, “Have there been times of a couple days or more when you were able to stop breathing entirely (without the aid of medical equipment)?”, “I would feel bad if a loved one unexpectedly died.”, “I would be able to lift a 1 lb (0.5 kg) weight.”; Zorowitz et al., 2021). To increase the probability of achieving a mean completion rate of 85% across all participants, we included only those participants into the analyses who had provided data in at least 60% of prompts.

Eligible participants were invited to a pre-session and first filled in sociodemographic data and trait questionnaires (see below). Afterwards, the participants practiced the correct use of the EMA questionnaire application (app). Supervised by the experimenter, they answered the EMA survey in two imaginary situations (social, non-social) before receiving the study smartphones. The participants’ planned schedule for the seven measurement days was enquired to individually adjust the 12-hour measurement time windows, starting one hour after the usual wake-up time.

The trait questionnaires and sociodemographic data were collected using SoSci Survey (https://www.soscisurvey.de). Subjective and contextual data in everyday life were assessed on a study smartphone within the Android-based app movisensXS (movisens GmbH). To ensure equitable access to the study, we provided all participants with smartphones running a compatible version of the Android operating system. This was necessary as the movisensXS app is only available on Android devices with certain versions of the operating system. By providing all participants with the necessary equipment, we aimed to ensure that everyone had the ability to fully participate in the study. In the individually adjusted time window, acoustic alarms reminded the participants to answer the prompts.

During the pre-session, participants answered the IPIP-NEO-120 personality questionnaire (Johnson, 2014) as we were interested in their broad Big-Five personality factors. For exploratory purposes, participants answered the social interaction anxiety scale (SIAS; Heimberg et al., 1992; Mattick & Clarke, 1998) and the 8-item Patient Health Questionnaire as depression inventory (PHQ-8; Kroenke et al., 2009). The results of exploratory analyses with these assessments provide relevant information for future clinical studies.

Participants underwent a seven-day-long EMA (six prompts per day, max. 42 prompts per person). If the social situation was a group setting, participants were instructed to focus on the main interaction partner for all ratings. All numeric ratings were provided on Likert scales from 1 to 7 with item-specific anchors. Of note, the target questions we focus on in the current study were placed at the first positions to avoid a bias caused by the other (exploratory) questions (for the full list of items and scales, see Supplemental Tables S1 and S2).

In each survey, participants were asked to indicate the time of their most recent social interaction (“now”, “within the last 30 min”, “more than 30 min ago”). For social interactions within the last 30 min, they were presented with the social interaction questionnaire. Otherwise, they answered an alternative activity questionnaire (see Supplement). We first assessed the approximate start and duration of the interaction (slider scales with three anchors: “<1 min”, “15 min”, “30 min”), the type of interaction (“via online channels”; “face to face”; Schwerdtfeger et al., 2020), the number of interaction partners (“1” to “5 or more”), the main interaction partner’s gender (“female”, “male”, “diverse”), and the relationship with the main interaction partner (“partner”, “family”, “friend”, “colleague”, “acquaintance”, “stranger”; Venaglia & Lemay, 2017). Participant were then asked about the main person with whom they had interacted. The items are presented in Table 1. Next to a rating on the preference to interact with this person (i.e., one dependent variable), participants were asked how they evaluate the perceived personality of their interaction partner using ten unipolar ratings (i.e., two per Big Five factor) according to the Ten-Item Personality Inventory (TIPI; Gosling et al., 2003). Although we focused on extraversion, agreeableness, and neuroticism, we also asked about openness and conscientiousness for exploratory purposes. We included ratings on the perceived physical attractiveness and familiarity of the (main) interaction partner to control for potential physical attractiveness and familiarity biases (Venaglia & Lemay, 2017; Voit et al., 2021). In addition, as there is evidence that weather influences affect (Denissen et al., 2008), we included an adapted item from the Daily Stress Inventory (Brantley & Jeffries, 2000) and combined it with a modified instruction for social interactions from the weather item by Zenk and colleagues (2017): “Did bad weather (e.g., rain, too hot, too cold) make it more difficult to engage in social interactions preceding the last prompt?”. Here, we used a binary answering format (yes vs. no).

We analyzed data using linear mixed-effect models (LMMs) with the lme4 package (Bates et al., 2015) in R Studio. P-values were obtained via the Satterthwaite approximation of degrees of freedom (Luke, 2017). We reported descriptive statistics regarding average self-rated trait ratings of participants (“self-trait”) and perceived traits of the interaction partners (“partner trait”), as well as correlations between participant self-traits, partner traits, and the preference to interact with an interaction partner. Based on the outlined hypotheses, we were particularly interested in the trait characteristics of an interaction partner and self x partner personality trait interactions of extraversion (E_partner/E_self), agreeableness (A_partner/A_self), and neuroticism (N_partner/N_self). Based on these three personality dimensions, we calculated four different models for the preference to interact. Each model had a random intercept per participant and included the perceived trait ratings of the interaction partner (E_partner, A_partner, and N_partner) as predictors. To account for the complexity of the data, we specified both a random intercept model and a random slopes model for each analysis. The random slopes model (i.e., day|participant as random effect) included measurement day as a random slope to allow for the estimation of within-subject changes over time and to account for non-constant within-subject variance and correlated errors in the repeated measures design. We compared the fit of the random intercept and the random slopes models by conducting a Likelihood Ratio Test using the anova function. We then used the model that provided the better fit. In case of a non-significant model comparison, we selected the model with the lower Akaike information criterion (AIC). In model 1 addressing hypothesis 1a-1c, we did not include participants’ personality traits. In models 2-4 addressing hypothesis 2, participants’ trait extraversion (model 2-E), agreeableness (model 3-A), and neuroticism (model 4-N), respectively, were included as predictor to test for participant trait x perceived target trait interactions. In the present research, we did not explicitly focus on gender effects. However, as gender is a crucial part of social interactions (see e.g., Glynn et al., 1999; Qi et al., 2021), we included a binary gender variable that encoded same-gender vs. other-gender interactions in all four models. In addition, we included physical attractiveness, familiarity of the (main) interaction partner as well as weather as covariates in all four models. In the model for hypotheses 1a-1c, we allowed for two-way interactions between gender x one partner trait (E_partner or A_partner or N_partner). In the models for hypothesis 2, we allowed for three-way interactions between gender x one partner trait (E_partner or A_partner or N_partner) and the model-specific participant self-trait (e.g., E_self or A_self or N_self). For instance, in the extraversion model, this resulted in the interaction terms gender x E_partner x E_self, gender x A_partner x E_self, and gender x N_partner x E_self. All between-person predictor variables were grand-mean centered and all within-person predictor variables were person-mean centered to facilitate the interpretation of model parameters (Hofmann & Gavin, 1998; Schwartz & Stone, 1998). The general equation for the random intercept mixed model testing hypothesis 1a-1c was:

The general equation for the random intercept mixed models testing hypothesis 2 was:

The winning models were updated by removing any outliers identified as observations with a standardized residual > 2.5 SD from 0 using the romr.fnc function from the LMERconveniencefunctions package. To account for model-wise multiple comparisons, we used a type I error rate of 0.05/4 (number of models for hypothesis testing) = α of .013, which means that there is a 1.3% chance of rejecting the null hypothesis when it is actually true.

Data collection took place between February and October 2023. 133 participants were recruited. Two of them were excluded after having informed the experimenter post data collection that they had misunderstood several of the EMA items. Another participant failed the inclusion criterion of completing more than 60% of the EMA prompts. From the 5362 EMA prompts of the remaining 130 participants, 725 prompts were excluded due to more than one failed attention check item. The completion rate of the remaining prompts averaged 35.7/42 prompts (i.e., 85%) which conformed exactly with the lower bound of the desired completion rate. The final data reduction consisted of removing the missed EMA prompts and those capturing social interactions longer than 30 minutes ago (1529 in total) which resulted in a final data set of 3108 valid EMA prompts comprising social interactions from 130 participants (mean age = 25.43, SD = 7.7; 107 women). Please note that we did not achieve the desired gender balance. The data reduction process is shown in Figure 1.

Descriptive statistics of the analyzed variables are summarized in Table 2 . Multilevel correlations between numeric variables computed with the misty package are illustrated in Figure 2 .

Since the results of the partner traits and the covariates in the model answering hypothesis H1 are comparable to those of the models answering hypothesis H2, these results are presented in detail only in the section on H1. The 95% confidence intervals (CI) are reported for all significant results.

The model comparison for H1 showed a better fit of the random slopes model. However, as the model did not converge after removing the outliers based on standardized residuals, we report the random intercept model in the following. We found that higher scores for both A_partner and E_partner related to a higher preference to interact (A: B = .164, s.e. = .014, CI = [.136, .191], p < .001; E: B = .103, s.e. = .012, CI = [.079, .127], p < .001). In contrast, higher N_partner was associated with a decrease in the preference to interact (B = -.060, s.e. = .012, CI = [-.084, -.036], p < .001). Same-gender interactions also increased the preference to interact compared to other-gender interactions (B = .158, s.e. = .039, CI = [.082, .234], p < .001). The covariates familiarity and physical attractiveness of the interaction partner were both positively related to the preference to interact (familiarity: B = .226, s.e. = .010, CI = [.206, .246], p < .001; attractiveness: B = .097, s.e. = .010, CI = [.077, .117], p < .001). Finally, weather as well as the two-way interactions between gender and partner personality showed no significant results (ps ≥ .035).

For the model including trait extraversion (H2), the random intercept model was the winning model. The interaction between E_self and E_partner (B = .002, s.e.< .001, CI = [.001, .004], p = .013) was significant (Figure 3, Panel A), indicating that extraverts prefer to interact with others they perceive as being extraverted, too. A simple slopes analysis confirmed a significant slope of E_self when E_partner was 1 SD above the mean (p < .001). The main effect of E_self (p = .024) as well as the three-way interaction with gender (p = .036) were significant only at the unadjusted level of α = .05. No other effect was significant on the pre-registered level of α < .013.

As with model l, the random slopes model including trait agreeableness (H2) did not converge after removal of the outliers based on the standardized residuals, which is why we report the random intercept model. The main effect of A_self indicated that more agreeable individuals showed a higher preference to interact with others (B = .019, s.e. = .007, CI = [.006, .032], p = .005). There was also a significant two-way interaction between A_self and N_partner (B = .004, s.e. = .001, CI = [.002, .006], p < .001). As illustrated in Figure 3, Panel B, the interaction indicates that more agreeable persons prefer to interact with others they perceive as neurotic. A follow-up simple slopes analysis confirmed that the slope of A_self was significant when N_partner was at the mean or 1 SD above the mean (ps < .001).

With respect to the model including trait neuroticism (H2), the random slopes model was the winning model. Only the main effect of N_self was significant, indicating that more neurotic individuals showed a lower preference to interact with others (B = -.001, s.e. = .005, CI = [-.021, -.003], p = .01).

In an exploratory analysis, we wanted to gain an overall impression of the relationship between trait and perceived differences in the factors E, A, and N and the preference to interact with another person. To this end, we calculated the Euclidean distances between standardized trait personality (i.e., E_self, A_self, and N_self) and the corresponding standardized perceived personality of the interaction partners, resulting in one distance value per interaction. We estimated a model including the main effects of gender and trait distance, their two-way interaction, and the same covariates as in the confirmatory models. We detected a main effect of trait distance (B = -.162, s.e. = .028, CI = [-.217, -.110], p < .001), indicating that greater distances between self-reported traits and the perceived traits of one’s interaction partners correspond to decreasing preferences to interact with these persons (Figure 4 ).

In this study, we investigated whether and how one’s own personality traits and the perceived personality traits of one’s interaction partners influence an individual’s preference to interact with other persons in daily life. Our results showed that the partner personality traits influence this preference to interact, with higher partner agreeableness and extraversion being beneficial and higher partner neuroticism being detrimental. Moreover, personality trait interaction effects revealed that extraverted individuals preferred other extraverts, while agreeable individuals preferred neurotic interaction partners.

Extraversion can be a protective personality trait as reflected in positive relations to mental health but also a risk factor when extraverts are limited in their ability to live out their personality (Weiß, Baumeister, et al., 2022; Weiß, Rodrigues, et al., 2022). Our data show that in everyday life, extraverts particularly value social interactions with like-minded people – that is, with those who appear to be as resilient and sociable as themselves. This is in line with the similarity effect of individuals reporting higher interaction quality when partners are similarly extraverted (Cuperman & Ickes, 2009) and with several theories suggesting that individuals generally prefer to interact with similar others (e.g., similarity-attraction theory, Byrne, 1971; social identity theory, Tajfel, 2010; self-categorisation theory, Turner, 2010). A potential mechanism behind this effect could be a personality-congruent selection effect (see, e.g., Hannuschke et al., 2020), meaning that individuals prefer interaction partners whom they perceive as similar to themselves (Selfhout et al., 2010). As our models are based on the participants’ perception rather than the self-rated personality traits of the respective interaction partners, we cannot rule out that interaction partners may act as more extraverted than they factually are in the presence of an extraverted person (i.e., our participants) because they strive for similarity. Thus, we cannot be sure that the detected extraversion interrelation is based on the true personality of both interaction partners. This alternative explanation stems from research showing that not only “true” extraversion, but also experimentally induced extraversion (i.e., acting as being extraverted) increases well-being (Margolis & Lyubomirsky, 2020).

For agreeableness, we found support for a complementary or dissimilarity perspective on personality trait interrelations. The data showed that the interaction preference rating increased with higher levels of self-agreeableness and higher partner neuroticism. One explanation for this effect might be that agreeableness is generally associated with helping behavior (Habashi et al., 2016), while individuals scoring high on neuroticism have been shown to emotionally benefit from perceived positive interactions with family members and friends (e.g., Mueller et al., 2021; Shackman et al., 2018). In consequence, more neurotic individuals might be more likely to approach agreeable persons who are in turn more likely to positively react to their (neurotic) interaction partners given their inherent helping motivation.

Importantly, similar to previous work (e.g., Hampson, 2012), individuals scoring high on neuroticism showed an overall reduction in the preferences to interact with others. In other research, higher scores on neuroticism predicted that a person perceives their interaction partner’s behavior as more negative and threatening than those scoring lower on neuroticism (Finn et al., 2013). In consequence, neurotic individuals might be prone to a negative bias in their interpretation of social cues, which might cause lower scores in the preference to interact (e.g., Bunghez et al., 2023). Such a negative bias is associated with negative attention, memory, and interpretation biases (e.g., Chen et al., 2023). There is further evidence linking neuroticism to other deficits such as low self-esteem that might negatively affect the preference to interact (Watson et al., 2002) and increase the sensitivity to potential interpersonal threat (Denissen & Penke, 2008).

Our exploratory analysis implies an overall effect of personality trait dissimilarity. We conducted this aggregated analysis as we wanted to gain an overall impression of how the trait distance between two individuals affects the preference to interact. However, we refrain from using this analysis to draw conclusions about whether trait similarity or trait dissimilarity is more beneficial for social interactions as our confirmatory results indicate that both directions – similarity in extraversion and dissimilarity in agreeableness-neuroticism – can lead to subjectively higher preferences for social interactions, depending on which personality trait is in the spotlight.

All significant covariates – gender, attractiveness, and familiarity – showed effects in the expected direction. Individuals particularly value same-gender interactions or benefit from them. This effect was previously reported on the autonomous level (i.e., heart rate and heart rate variability; Gründahl et al., 2023) as well as in subjective reports (e.g., a higher preference to be friends with same-gender students; Friebel et al., 2021). Our finding that higher familiarity is related to an increasing preference to interact fits well with previous EMA studies in which familiarity showed a beneficial (i.e., reducing) effect on subjectively reported state anxiety (Gründahl et al., 2023; Hannah Lee, 2021). Finally, the interaction preference-increasing effect of physical attractiveness reported in the current study is also in line with previous studies showing an attractiveness bias both in laboratory studies (e.g., Voit et al., 2021) and in daily life (Lemay et al., 2010). Note that perceived familiarity and attractiveness were significantly and positively correlated, thus showing a potential perception bias (i.e., familiar persons are also perceived as more attractive) that should be considered when studying social interactions.

Despite broad advertising of the study, we did not succeed in recruiting a gender-balanced sample. Therefore, the results are not generalizable to both genders. The larger proportion of female participants matches recent evidence showing that there is a selection bias towards younger women in EMA research (Stone et al., 2023). Furthermore, the frequent absence of social interactions prior to EMA prompts reduced the amount of data for analyses. Future studies should increase the number of prompts so that more interactions can be captured to prevent this data loss.

Moreover, it would be interesting to collect EMA data from all interaction partners. Such partner studies would allow for testing universal vs. situation-specific influences of personality traits on social interactions and for conducting interaction analyses based on true interaction partner personality traits. Our EMA approach results in data reflecting very natural behavior. That said, given that individuals were free to choose their interaction partners, some trait interactions occurred more frequently than others. To investigate more heterogenous personality combinations, future research should use combined EMA-laboratory designs, i.e., designs that are applied to collect data both in real life and under more standardized conditions in the laboratory (e.g., using Standardized State Assessments, see Freudenstein et al., 2023) to introduce more diverse interaction partner personalities.

The present EMA study tested the effect of participants’ and partners’ personality traits on the preference to interact with others in real life by analyzing more than 3000 social interactions. Replicating results from the laboratory, we found that preferences to interact with others augment with increasing scores on extraversion and agreeableness, decrease with increasing scores on neuroticism, and are modulated by gender, attractiveness, and familiarity. Extending previous results, our EMA data revealed that individuals scoring high on agreeableness prefer interactions with individuals that they perceive as neurotic. Our findings regarding the effect of personality trait interactions on social preferences are important for social and clinical settings in which multiple individuals need to live or work together and are thus likely to frequently influence each other.

The authors thank Kilian Stenzel for his assistance in collecting the data.

Contributed to conception and design: MW

Data collection: AJ

Drafted and/or revised the article: MW, MG, AJ, GH

Approved the submitted version for publication: MW, MG, AJ, GH

M.W. was supported by funds of the Bavarian State Ministry of Science and the Arts and the University of Würzburg to the Graduate School of Life Sciences (GSLS), University of Würzburg. This publication was supported by the Open Access Publication Fund of the University of Würzburg.

The authors declare that there is no conflict of interest.

The anonymized data, corresponding R scripts, a codebook and Supplementary Material are openly available at the Open Science Framework (https://osf.io/h4y3s). We used the renv package to manage R packages and package dependencies in a reproducible way to ensure that the R scripts are reproducible by other users or on different machines.