The empathic personality profile: Using personality characteristics to reveal genetic, environmental, and developmental patterns of adolescents' empathy

Objective: How do genetic and environmental processes affect empathy during early adolescence? This study illuminated this question by examining the aetiology of empathy with the aetiology of other personality characteristics. Method: Israeli twin adolescents rated their empathy and personality at ages 11 ( N = 1176) and 13 ( N = 821) (733 families, 51.4% females). Parents rated adolescents' emotional empathy. Adolescents performed an emotion recognition task, indicating cognitive empathy. Results: Using a cross-validated statistical learning algorithm, this study found emotional and cognitive “empathic personality profiles,” which describe and predict self-reported empathy from nuanced Big-Five personality characteristics, or “nuances” (i.e., individual items). These profiles predicted empathy moderately ( R 2 = 0.17– 0.24) and were stable and robust, within each age and between ages. They also predicted empathy in a new sample of older nontwin adolescents ( N = 96) and were validated against non- self-report empathy measures. Both emotional and cognitive empathy were predicted by nuances representing positive attitudes toward others, trust, forgiveness, and openness to experiences. Emotional empathy was also predicted by nuances representing anxiousness and negative reactivity. Twin analyses revealed overlapping genetic and environmental influences on empathy and the empathic personality profiles and overlapping environmental influences on empathy– personality change. Conclusions: This study demonstrates how addressing the complexity of individuals' personalities can inform adolescents' empathy development.


| INTRODUCTION
Empathic tendencies-the tendencies to understand and feel others' emotions (Uzefovsky & Knafo-Noam, 2016)-are vital for social competencies and relationships (Allemand et al., 2015;Wolgast et al., 2020). Adolescence, a period when meaningful social and neurophysiological changes take place (Blakemore & Mills, 2014;Forbes & Dahl, 2010;Smetana et al., 2006), is also characterized by a substantial change in empathy (Allemand et al., 2015;Van der Graaff et al., 2014). What contributes to empathy development during adolescence? To inform this question, this study investigated the development of empathic tendencies as part of adolescents' developing personality, using a genetically informative longitudinal sample.
Broadly speaking, empathy is divided into two separate, yet related, tendencies or abilities. The first is emotional empathy-the tendency to experience an emotion similar to that of another person while maintaining selfother distinction and other-oriented focus. The second is cognitive empathy-the ability to recognize, understand, and mentalize others' emotions (Decety, 2011;Uzefovsky & Knafo-Noam, 2016;Zaki & Ochsner, 2012). Emotional and cognitive empathy are at least partially distinct from one another and therefore require separate investigations, based on evidence from behavioral (Jordan et al., 2016), neuroscientific (Zaki & Ochsner, 2012), developmental ( Van der Graaff et al., 2014) and behavioral genetic (Abramson, Uzefovsky, Toccaceli, et al., 2020) research. Another, closely related construct to empathy is empathic concern (also called sympathy or compassion)-warm feelings of concern for the well-being of the other, partially overlapping with empathy but particularly motivated by caring for others (Eisenberg, 2010;Zaki & Ochsner, 2012). Empathic concern is different from emotional empathy because it does not necessarily involve resonating with others' emotions (Eisenberg, 2010) and is more influenced by motivational considerations such as values and prosocial motivations (Decety & Yoder, 2016;Eisenberg, 2010;Jordan et al., 2016). To keep this study's focus limited, we concentrate only on the core components of the empathy system, namely emotional and cognitive empathy (Uzefovsky & Knafo-Noam, 2016).

| Illuminating the genetic and environmental architecture of empathy using other personality traits
Like most psychological constructs (Plomin et al., 2016), both emotional and cognitive empathy were shown to be influenced by genetic and environmental processes, although to different extents (Abramson, Uzefovsky, Toccaceli, et al., 2020). The goal of this study was to further illuminate these processes during early adolescence, by connecting them to the possibly parallel development of other fundamental individual differences. Theoretically, individual differences in empathy result, in part, from individual differences in biologically based processes underlying reactivity, regulation, and approach tendencies. Specifically, reactivity and regulation were proposed to be important mainly for emotional empathy, because emotional empathy requires emotionally reacting to others' feelings and regulating this reactivity to maintain the necessary other-oriented focus Decety, 2011;Eisenberg, 2010;Weisz & Cikara, 2021) (of note, relations between traits reflecting regulation like conscientiousness and cognitive empathy were also found, for example, Airagnes et al., 2021;Melchers et al., 2016). Approach tendencies underlie the motivation to interact and engage with others' feelings and are thus considered to contribute to both emotional and cognitive empathy Decety, 2011;Zaki, 2014).
Reactivity, regulation, and approach processes stand at the basis of other personality characteristics as well (DeYoung, 2015a;Yarkoni, 2015), and therefore may be influenced by the same genetic and environmental factors. Indeed, phenotypic covariance between behavioral constructs is often caused by genetic covariance, indicating that different phenotypic differences have a common genetic basis (Plomin et al., 2016). For example, genetically based dopaminergic and oxytocinergic activations affect stress and reward processing, thus affecting reactivity and approach-avoidance behavior (DeYoung, 2015a; Rodrigues et al., 2009;Yarkoni, 2015). Accordingly, they should stand at the basis of traits like neuroticism, extraversion, and openness to experience (Caspi & Shiner, 2007;Chong et al., 2019;DeYoung, 2015a), as well as empathy (Gong et al., 2017;Pearce et al., 2017;Uzefovsky et al., 2014Uzefovsky et al., , 2015. Indeed, a Genome-Wide Association Study (GWAS) found a genetic overlap between cognitive empathy and openness to experiences, as well as a marginal genetic overlap with brain volume in the dorsal striatum, a region enriched with dopaminergic input (Warrier et al., 2018). Nevertheless, direct twin-model evidence for a personality-empathy genetic overlap, to our knowledge, is currently lacking.
Although trait associations are often attributed to genetic influences (Plomin et al., 2016), environmental factors may also cause traits to associate. This study focused on early adolescence, a period of major environmental changes such as the transition to junior high, which may result in changes in peer relationships, social status, and affiliation to social groups (Blakemore & Mills, 2014;Smetana et al., 2006). Additionally, adolescence is characterized by heightened reactivity of individuals to their environment (Blakemore & Mills, 2014). Such considerations suggest that during early adolescence, environmental factors may have a meaningful impact on individuals, which may affect both their empathy and other tendencies. Accordingly, we hypothesized that associations between | 755 adolescents' empathy and other personality traits would be accounted for by overlapping genetic, as well as environmental effects.
To capture the full extent to which genetic and environmental processes are shared between empathy and other personality traits there is a need to first form a construct which (a) maximizes the prediction of empathy from various personality traits, and (b) describes the refined relations between personality and empathy. In line with that premise, we used a data-driven technique to build an overall "personality profile of empathy."

| The empathic personality profile
Emotional and cognitive empathy are integrated within a broader set of personality traits (Knafo & Israel, 2012;Mooradian et al., 2011). Studies that focused on late adolescence and adulthood, which operationalized individual tendencies by the "Big-Five" model of personality, and studies that focused on childhood, which investigated temperament traits as early emerging dispositions, have shown consistent phenotypical relations between empathy and various tendencies. Specifically, emotional and cognitive empathy have been shown to associate with traits reflecting approach, such as agreeableness, openness to experiences, and extraversion (Airagnes et al., 2021;Jolliffe & Farrington, 2006;Melchers et al., 2016;Song & Shi, 2017;Vernon et al., 2008;Villadangos et al., 2016), interest (Liew et al., 2011), and low shyness (MacGowan & Schmidt, 2021). In addition, cognitive empathy has been shown to positively associate with conscientiousness, which reflects control and regulation abilities (Airagnes et al., 2021;Melchers et al., 2016;Song & Shi, 2017;Vernon et al., 2008). Emotional empathy has been shown to associate with traits reflecting emotional reactivity, like negative emotionality Edwards et al., 2015;Eisenberg, 2010) and neuroticism (Airagnes et al., 2021;Jolliffe & Farrington, 2006), although the direction of these associations may depend on the specific emotional reactivity trait that is measured (e.g., fear, anger, or sadness) Edwards et al., 2015).
Focusing on the early adolescence period, this study looked at many personality traits as captured by the "Big-Five" model research, the most pervasive taxonomy of individual differences in the personality literature (John et al., 2008). This taxonomy is often described as including a hierarchical organization of personality (DeYoung, 2015b;DeYoung et al., 2007;Mõttus et al., 2020), although the exact structure of the lowerorder level is subject to debate (DeYoung, 2015b). While some researchers focus on the two meta-traits of stability and plasticity, the most researched level is the domains level, which measures five domains reflecting openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism (John et al., 2008). Each domain is further composed of two aspects, reflecting unique and shared variance of the higher-order domain (e.g., volatility and withdrawal are two separate yet related aspects of neuroticism). Below this level is the facets level, which includes more fine-grained personality traits (DeYoung, 2015b;DeYoung et al., 2007).
Regardless of the investigated level, personality research traditionally separates between these psychometrically distinct traits. In real life, however, individuals' personalities are composed of various characteristics that are not necessarily differentiated (Mõttus et al., 2017). Specifically, the division of domains to aspects and facets can be done in many different, partially overlapping ways, which also differ in their relations with other behavioral traits and neural processes (DeYoung, 2015b;DeYoung et al., 2007;Hou et al., 2017;Soto & John, 2017). Accordingly, it has been claimed that there is no strong theoretical reason to a priori prefer one trait differentiation over others (Mõttus et al., 2017). Moreover, it is becoming acknowledged that the genetic and neurobiological architecture of individual differences does not map into the psychometrically defined phenotypical models of personality (Mõttus et al., 2017;Yarkoni, 2015). This is not because of theoretical or empirical weaknesses but because there is real lack of isomorphism between physiological systems (i.e., genetic, neurological) and conventional divisions of personality measures (Yarkoni, 2015). Accordingly, an approach that uses an accumulative measure of phenotypically relevant traits, regardless of the psychometric differentiation between them, might better reflect biologically based personality-empathy relations.
An accumulative personality-trait analytical strategy fits a growing approach in personality research (Hall & Matz, 2020;Mõttus et al., 2017Mõttus et al., , 2020Rozgonjuk et al., 2021;Sindermann et al., 2021;Stewart et al., 2022). This approach proposes that better description and prediction may be obtained with data-driven exploratory techniques that focus on nuanced personality characteristics, or "nuances," typically operationalized as personality questionnaire items. Indeed, single-item personality characteristics were shown to have adequate trait-like properties of stability and raters' agreement reliability, and most of them were shown to have unique developmental patterns, heritability, and outcome predictability (McCrae, 2015;Mõttus et al., 2017Mõttus et al., , 2020Sindermann et al., 2021;Stewart et al., 2022). These findings support the examination of single items not only as part of the aggregated Big-Five domains, as traditionally done, but also as stand-alone nuanced traits.
A data-driven, item-level approach may also complement the Big-Five domains approach by providing new insights that could not be detected with a priori aggregated scales (Mõttus et al., 2020). For example, whereas the adult literature suggests mainly negative relations between neuroticism and empathy (Airagnes et al., 2021;Melchers et al., 2016;Song & Shi, 2017;Vernon et al., 2008), research with children shows that some, more fine-grained negative affect traits have positive relations with empathy Edwards et al., 2015). Examining nuances of neuroticism, without prior assumptions on how these nuances converge, may help understand the continuity between children and adults' findings and form new hypotheses.
In line with that notion, this study aimed to identify an "empathic personality profile," that is, an assembly of individual characteristics which together describe the person who is likely to be empathic. Instead of describing the relations between empathy and each trait separately, we studied the joint contribution of many personality nuances, that is, of a personality profile, and estimated the extent to which indexes based on these nuances not only describe, but also predict, emotional and cognitive empathy (namely, the extent to which they predict empathy in an independent sample). Then, we investigated the extent to which common genetic and environmental factors contribute to these constructs' joint development.
During early adolescence, personality stability is significantly lower than in older ages (Borghuis et al., 2017) and empathic tendencies continue to develop (Allemand et al., 2015;Van der Graaff et al., 2014). Such change in empathy may be driven by various causes. For example, partially due to puberty processes (Forbes & Dahl, 2010), early adolescence is characterized by substantial hormonal changes, brain maturation, and heightened sensitivity (Blakemore & Mills, 2014;Forbes & Dahl, 2010). In addition, as mentioned, it is characterized by major social changes (Blakemore & Mills, 2014;Smetana et al., 2006). Understanding if the same genetic and environmental sources change both personality and empathy, or if different sources change them independently, may, therefore, direct to specific sources affecting adolescents' empathy. To understand how personality and empathy co-develop during early adolescence, we examined personality-empathy relations in a two time-points longitudinal design, at age 11 and age 13. This enabled us to examine the phenotypical stability of empathy prediction from personality nuances during early adolescence. Importantly, using the twin design in a longitudinal fashion enabled us to estimate genetic and environmental effects on possible changes in personality-empathy relations.

| The present study
To illuminate the genetic and environmental processes behind adolescents' empathy development, this study investigated how, and to what extent, these processes stand also at the basis of other personality traits. We used a datadriven approach to find the personality profiles which best predict and describe individuals' emotional and cognitive empathy, namely the "empathic personality profiles." To ensure robustness of our findings, we used ridge regression-a statistical learning algorithm that reduces estimation uncertainty, and performed validation analyses to examine our model's predictive accuracy when applied to new samples. Importantly, we also tested the associations of the models' predicted scores with non-self-report measures of empathy. After establishing the empathic personality profiles, we estimated the contributions of shared genetic and environmental factors to the associations between these profiles and empathy, as well as to patterns of stability and change in personality-empathy associations.

| Participants
Hebrew speaking twin adolescents of Jewish descent, who were born during 2004-2005, participated in an online study as part of the Longitudinal Israeli Study of Twins (Vertsberger et al., 2019). Data collection was conducted during the years 2015-2019, in two separate time-points, when twins were approximately 11 and 13 years old. Overall, 1402 adolescents from 737 families participated at age 11 (M = 11.1 years, SD = 0.22, range-10.8-12.6, 51.4% females) and 827 adolescents from 456 families participated at age 13 (M = 13.3, SD = 0.22, range: 12.8-14.1, 53.4% females) (705 adolescents participated at both time-points). The sample size was determined based on the longitudinal study's existing pool of participants and resource constraints. This sample size is large enough for phenotypical ridge regression (Finch & Hernández Finch, 2016). To verify its appropriateness for genetic analyses, we performed a simulated power analysis of a cross-sectional bivariate ACE model (Verhulst, 2017) (see Appendix 1 for details). We based our simulated effect sizes on findings from previous meta-analyses of empathy and other personality traits (Abramson, Uzefovsky, Toccaceli, et al., 2020;Vukasović & Bratko, 2015), noting that this study's age range was less examined and so its actual effect sizes may differ. The analysis showed the sample size had reasonable power to detect most effects of moderate size (e.g., additive genetic effects and genetic correlation at age 11, based on the effects previously found for emotional empathy), and lower power to detect effects of lower size (e.g., genetic correlation at age 13, based on the effects previously found for cognitive empathy). Accordingly, we conducted the genetic analyses while noting that given the sample size, some conclusions should be taken more cautiously.
Participants' household income, assessed by describing the average income in Israel and asking the mothers to rate their total household income relative to it (1 = significantly below average, 5 = significantly above average) was around the midpoint of the scale (M = 3.60, SD = 1.17). Maternal years of education (0-12 years: 19%, 13-15 years: 27.7%, 16 years or more: 53.3%), reported in previous waves when participants were 5 or 3 years old, were similar to those in the Jewish Israeli population of women in this age range (0-12 years: 25.2%, 13-15 years: 27.5%, 16 years or more: 46.6%) (Israel Central Bureau of Statistics, 2020).
After eliminating participants due to technical reasons (see Appendix 1), the final sample included 1380 adolescents (1176 at age 11; 821 at age 13; 51.4% females) from 733 families (617 participants from 344 families had data at both time-points). At age 11, this sample included 249 MZ twins (117 complete pairs), 498 DZ-same sex twins (DZS) (235 completed pairs), and 411 DZ-opposite sex twins (DZO) (189 complete pairs). At age 13, it included 180 MZ twins (78 complete pairs), 359 DZS twins (159 pairs), and 279 DZO twins (123 complete pairs). Zygosity was uncertain for 18 participants at age 11 and 3 participants at age 13. These participants were included in the phenotypic analyses but excluded from the genetic analyses. For same-sex twins, zygosity was measured using DNA data (available for 59% of the same-sex twins in the sample) or by a parent questionnaire of physical similarity (Goldsmith, 1991).

| Procedure
When the twins were approximately 11 and 13 years old, their parents were contacted by phone and invited to partake in an online study. Due to limited internet access or other reasons, some families chose to receive and return printed questionnaires via postal service (10.0% and 11.1% at ages 11 and 13, respectively). Except for cognitive empathy and the empathic personality profile at age 13, empathy and personality ratings did not significantly differ between adolescents who filled the questionnaires in an online or a "pencil and paper" format, nor between adolescents who participated in one or two time-points (see the Results section).
As part of a broader battery (Vertsberger et al., 2019), adolescents filled an empathy questionnaire, followed immediately by a personality questionnaire. Before filling the questionnaires, adolescents who completed the study online performed an additional computerized task. At age 13, they performed an emotion recognition task, described in the Measures section (66 adolescents received an abbreviated battery without this task to increase participation rate). At age 11, they conducted a task examining perception of nonverbal social behavior (Costanzo & Archer, 1989). Because post-hoc analyses showed this task had extremely poor reliability in our sample (Cronbach's alpha: 0.05), possibly because it examines not only emotional understanding but multiple features of social behaviors, we decided not to include this task in the study's analyses. Parents also filled a battery of questionnaires that included items regarding their children's emotional empathy. As compensation for participation, families were offered 3-4 tickets to the movie theater or a museum, or coupons worth 100 shekels (around $28). The study received ethics approval from the ethics committee of the Hebrew University of Jerusalem. Adolescents provided informed assent to participate, and parents provided informed consent for their children's participation and their own.

| Measures
A brief description of the study's measures is provided here, whereas a more detailed description of all the measures can be found in Appendix 1.

Self-reported empathy
Self-reported emotional and cognitive empathy measures were obtained with the Basic Empathy Scale Questionnaire (BES) (Jolliffe & Farrington, 2006), in which participants rated their perceived levels of empathy on a 5-point scale ranging from 1 (strongly disagree) to 5 (strongly agree). The study's larger battery of questionnaires also included items expressing empathic concern. However, as noted in the introduction, in this study we focused only on the core components of emotional and cognitive empathy to limit the study's scope. The BES was originally validated on adolescents and was shown to have good reliability and validity (Jolliffe & Farrington, 2006). In this sample, which included younger adolescents, the scales showed good internal consistency as well (Cronbach's alpha, age 11: 0.71-0.72, age 13: 0.77).

Parent-reported emotional empathy
Parents rated their children's emotional empathy by rating five items from the children's Empathy Quotient-Systemizing Quotient (EQ-SQ) (Auyeung et al., 2009), on a 3-point scale ranging from 0 (not true or rarely) to 2 (very true or often). Mothers' reports were available for 1096 adolescents at age 11 and 645 adolescents at age 13. Fathers' reports were available for 687 adolescents at age 11 and 397 adolescents at age 13. The full EQ-SQ showed good reliability (Auyeung et al., 2009). In the present sample, the chosen items showed acceptable internal consistency as well (Cronbach's alpha, age 11: 0.66-0.67, age 13: 0.65-0.68).

Emotion recognition abilities
Adolescents' emotion recognition, considered an indicator of cognitive empathy (Warrier et al., 2018), was examined at age 13 with the Jerusalem Facial Expressions of Emotion test (JeFEE) (Yitzhak et al., 2017) (final available N = 626-627). This task was designed to examine the recognition of dynamic and subtle emotion displays, conveyed in an unconstrained, nonprototypical manner. Participants were presented with 28 short videoclips (approximately 12-14 s, including 2 initial seconds of white screen presentation) of actors dynamically displaying one of seven emotional states: anger, fear, disgust, happiness, sadness, surprise, and neutral. Each emotional state was displayed four times in a random order, by two male and two female actors. Participants indicated which label best described the displayed expression, with all seven emotional states serving as response options and an unlimited time to respond.
Proportion of accurately recognized displays per emotion category were calculated for each participant. An exploratory factor analysis (Appendix 1; Table S1) showed that recognition of the seven emotional states reflected two correlated factors (r = .28), explaining overall 43.16% of the variance: (1) negative emotion recognition, on which recognition of anger, fear, disgust, and sadness loaded positively and (2) positive/neutral emotion recognition, on which recognition of happiness, surprise, and neutral emotional states loaded positively. The respective emotional states for each factor were averaged to yield two scores: negative and positive emotion recognition. In addition, a global emotion recognition score averaging all emotional state categories was computed.

| Personality
Self-reported personality nuanced characteristics were obtained with the 44 items from the Big-Five Inventory (BFI; John & Srivastava, 1999), answered on a 5-point scale ranging from 1 (strongly disagree) to 5 (strongly agree). To minimize prediction differences between items that stem only from the difference in the items' variances, ratings were standardized so that each item's mean would be zero and SD would be 1 (Friedman et al., 2010;Lee, 2015).

| Statistical analyses
The data and analyses scripts are openly available at https://osf.io/5pwfe/ ?view_only=33dbe f46d7 5447b a99b2 4e759 0259a10. For brevity, we report only the main analyses in the main text. We refer the reader to Appendix 2 for a detailed description, explanation, and justification of all the statistical analyses.

| Handling missing values
Up to 2% of BFI ratings in each item were missing and had to be imputed for the regression analysis. We imputed these missing values using multiple imputations with a predictive mean matching procedure (Little, 1988) (see Appendix 2 and Table S2 for more details).

| Predicting empathy from nuanced personality characteristics
To predict emotional and cognitive empathy from all the 44 personality items, we used linear ridge regression (Friedman et al., 2010)-a supervised statistical learning algorithm that reduces the estimation uncertainty resulting from predictors' multicollinearity. It does so by adding a bias parameter that penalizes regression coefficients proportionally to their absolute value, which has the effect of shrinking the coefficients of correlated predictors (Friedman et al., 2010). The penalty parameter was chosen in a nested cross-validation procedure (see Appendix 2). Analyses were done separately for emotional and cognitive empathy in the R program, using the "glmnet" package (Friedman et al., 2010). Because we were interested in examining stability and change of personality-empathy relations, we conducted two separate analyses for age 11 and age 13. To examine the extent of similarity between the ages, we also conducted a third analysis, in which the model fitted with age 11 was tested on data from age 13.
To avoid over-fitting the regression model to the specific data and increase the results' generalizability, we used a sixfold cross-validation procedure (Yarkoni & Westfall, 2017). In each iteration, the model's performance was assessed by quantifying the correspondence between the model-predicted score and the empathy outcome in the test set (i.e., mean squared error [MSE] and correlation coefficient). Finally, the model performance was averaged over the six iterations, such that the end result reflected the testing of the entire data set. Because the model predicts empathy solely based on the individual's personality questionnaire ratings, we refer to this model-predicted empathy as an empathic personality profile score.
Our data included two types of observation interdependence. First, it included pairs of twins. Second, in the joint analysis that included both ages, the same participants were measured twice, once at age 11 and once at age 13. To ensure the independence of the testing and training data sets, measures of related twins for both time-points were always entered together into the same testing set. This way, we could make sure the relation between testing and training sets was not confounded by the factors of individuals' identity and family relatedness. Statistical significance of the models' overall predictive accuracy and the significance of each personality item coefficient was determined with nonparametric permutation tests (see Appendix 2 for more details). While building the reference distributions for the permutation tests, we made sure that scores of twins from the same pair were assigned together to another, randomly chosen twin pair. Thus, the statistical tests simulated the interdependence of observations in our study and considered it in the analyses. Since many items were found to significantly predict empathy, we opted to create a conservative, robust, and narrow list that includes only the most predictive empathy items. Thus, we considered items as significant only if their p value was lower than 0.01. Yet, to provide the reader with full information, we also marked items significant at p < .05 (Figure 1).
Finally, to produce empathy predictive personality scores that are consistent across all participants and can be used in further analyses, we generated scores that represent each child's empathic personality profiles, using a single model across all participants (i.e., no subset was left aside for testing purposes). Training the entire dataset produced nearly identical models to the models produced in the cross-validation procedure (Appendix 2). This process was done separately for age 11 and age 13, to allow for age differences in personality-empathy relations.

| Genetic and environmental relations between empathy and empathic personality profiles
Using the Mx structural equation modeling software (Neale et al., 2004), genetic and environmental influences on empathy, empathic personality profiles, and the associations between these constructs within and across ages, were investigated with a biometric cross-lagged model (Saudino et al., 2018)-one for each empathy construct (emotional and cognitive). This model assumes that the genetic and environmental associations between the phenotypes are proportional to the phenotypic association and is therefore more parsimonious than other models (Turkheimer, 2016;Turkheimer et al., 2014). Importantly, unlike other models (e.g., Cholesky decomposition models), it does not assume directionality of causal effects between variables measured at the same timepoint (e.g., empathy and personality) and is, thus, more suitable to this study's design and assumptions. The cross-age associations function as phenotypic partial regression coefficients. These coefficients indicate the phenotypes' phenotypic stability (e.g., the stability of empathy from age 11 to age 13) and the effects of each phenotype at the earlier age on the other phenotype at the later age (e.g., the effect of empathic personality profile at age 11 on empathy at age 13), independent of their early-age association and the stability effects. The variances of the phenotypes at each age are decomposed into their genetic and environmental components, and the correlations between these components represent the genetic and environmental overlap of the phenotypes.
The variance components affecting each phenotype can be examined by two models: in an ACE model, the variance of each observed variable is decomposed into additive genetics (A), shared environment (C), and nonshared environment + error. In contrast, in an ADE model, the shared environment component is not estimated and instead, nonadditive genetic effects are examined (D) (see Appendix 2 for details). In this study, we decided which model to use based on the raw twin intraclass correlations (ICCs). An ADE model was chosen if the correlation between MZ twins was higher than twice the correlation between DZ twins.
The overall fit of the model was assessed by calculating twice the difference between the negative log-likelihood (−2LL) of the model and that of a saturated model (i.e., a model in which all variances and covariances for MZ and DZ twins are estimated). The difference in −2LL is asymptotically distributed as χ 2 , with degrees of freedom equal to the difference in the number of parameters in the full and the saturated model (Saudino et al., 2018). A reduced model, in which paths that were estimated at almost zero were dropped, was compared to the full model, in which all parameters were estimated. We did not drop other nonsignificant paths (i.e., paths whose estimation, although not significant, showed some effect on the phenotype), since dropping such paths can inflate the estimation of the other paths and bias the model's results.

| Empathy prediction accuracy
Predictive accuracy measures for each testing iteration are presented in Table S3. At age 11, the mean correlation F I G U R E 1 Personality items prediction of emotional and cognitive empathy. Item coefficients were averaged across the six folds of the ridge regression. Personality items were standardized, whereas the empathy outcomes were kept in their original scale. Numbers, therefore, represent non-standardized coefficients. (a) Emotional empathy-age 11; (b) Cognitive empathy-age 11, (c) Emotional empathy-age 13; (d) Cognitive empathy-age 13. Colors represent the original Big-Five domain to which the item belongs; A = Agreeableness; C = Conscientiousness; E = Extraversion; N = Neuroticism; O = Openness to experience. *p < .05; **p < .01.
To assess the generalizability of the empathic personality profiles, we examined the prediction accuracy of the model built in the twins' sample on an independent test set. This set included adolescents who completed the self-reported questionnaires via an online survey company (N = 96, Mean age = 15.67, SD = 1.58, range: 12.7-18.9, 57.3% females; see Appendix 3 for more details). Table S4 provides the models' prediction measures when performed on that test set. For both cognitive empathy (R 2 = 0.14, MSE = 0.24-0.25) and emotional empathy (R 2 = 0.34-0.39, MSE = 0.26-0.28), the prediction accuracy was similar to, or higher than, the prediction accuracy in the original twin dataset. This suggests that the results can be generalized beyond samples of twins and to later stages of adolescence. Figure 1 presents the weight coefficients of all personality items, averaged across folds, at both ages. For better integration with the previous literature, we describe the empathic personality profiles' content both at the level of nuanced individual items and at the level of the Big-Five domains to which the items belong. Both emotional and cognitive empathy were predicted, to some degree, by items from all five personality domains. Nevertheless, some domains had more items with high predictive power than others. For emotional empathy, the items with the highest (positive) coefficients were mostly items from the neuroticism and agreeableness domains, as well as some items from the openness to experience and extraversion domains. Interestingly, not all neuroticism items were important for emotional empathy. The high and significant items were items reflecting anxiety, emotional volatility, and negative reactivity to events in the environment (e.g., "is emotionally stable, not easily upset" [reversed], "can be tense," "remains calm in tense situations" [reversed]). In contrast, coefficients of items that reflect general moodiness (e.g., "can be moody," "is depressed, blue") were low and insignificant.

| Description of the empathic personality profiles' content
The agreeableness items that were most predictive of emotional empathy were items reflecting a positive attitude toward others (e.g., "is generally trusting," "has a forgiving nature"), and kindness (e.g., "is considerate and kind to almost everyone"). In contrast, items reflecting cooperation and respectful behaviors e.g., "likes to cooperate with others," "starts quarrels with others [reversed]") were less predictive of emotional empathy. Two items from the extraversion domain that may express social vitality and focus on positive interpersonal interactions ("is talkative," "generates a lot of enthusiasm") were also predictive of emotional empathy (at age 11, the item "is talkative" was significant at p < .05).
For cognitive empathy, the items with the highest positive coefficients were mostly items from the agreeableness and openness to experience domains, as well as some items reflecting extraversion (mainly at age 13). Similarly to emotional empathy, the agreeableness items with the highest coefficients were mostly items reflecting kindness and a positive view of others. At age 13, compared to age 11, items from the extraversion domain reflecting social vitality, assertiveness, and dominance ("is talkative," "tends to be quiet [reversed]," "has an assertive personality") had higher predictive power. Finally, one item belonging to the conscientiousness domain which reflects social responsibility ("is a reliable worker") predicted cognitive empathy at both ages.
The correlation across the 44 items' prediction coefficients at age 11 and age 13 was 0.76 [%95 CI: 0.60-0.87], p < .0001, for emotional empathy and 0.60 [%95 CI: 0.37-0.76], p < .0001, for cognitive empathy. This indicates relatively high stability along age, not only in terms of the overall ability to predict empathy from personality nuances, but also in terms of specific nuances' ability to predict empathy.

| Validating the empathic personality profiles with non-self-report empathy measures
To test the validity of both the empathy and empathic personality profile scores, we examined their correlations with parent-reported emotional empathy (at both ages) and with an emotion recognition test, indicating cognitive empathy (at age 13). Correlations between all variables are presented in Table S5. Self-reported emotional and cognitive empathy measures showed positive correlations of small to moderate size with the other empathy measures (r = 0.12-0.29). Importantly, correlations of the non-self-report empathy measures with the empathic personality profile scores (r = 0.10-0.29) were similar to their correlations with the self-reported empathy outcomes. This was the case also for the emotion recognition test, and especially for the measure of negative emotions recognition. This pattern suggests that the variance captured by the empathic personality profiles represents parts of participants' empathic tendencies, rather than irrelevant information specific to how participants answered the questionnaires (e.g., self-report response tendencies).

| Preliminary analyses
Before conducting the genetic analyses, we examined possible attrition effects by examining if participation status was associated with the main variables. Except for cognitive empathy and empathic personality profile at age 13 (Cohen's D = 0.16-0.22), there was no difference between adolescents who participated in one or two time-points (Table S6). There was also no difference between adolescents who had parental reports from at least one parent and adolescents whose parents did not complete the questionnaires at the same time-point (all p's > .10). Similarly, there was no difference at age 13 between adolescents who filled the abbreviated or full online survey version (all p's > .65).
We also examined if sex (Table S7), questionnaire format (online or pencil and paper) (Table S8), and participants' age within each time-points, were associated with the empathy and empathic personality profile scores. Except for cognitive empathy and cognitive empathic personality profile at age 13 (Cohen's D = 0.25-0.32), there was no difference in the variables' means between adolescents who filled the questionnaire online and adolescents who filled the questionnaire in a "paper and pencil" format. Similarly, age variation within study wave did not correlate with variables at the same time-point (r range: −0.06 to 0.02). Because these variables did not systematically differ between the study outcomes, we did not control for them in the main analyses. For robustness, we report in the supplementary materials on an additional analysis for cognitive empathy, which uses residual scores cleaned from the effects of sex (see the next paragraph), participation status (one/two time-points) and questionnaire format (Appendix 4; Figure S1). This analysis was done because cognitive empathy and empathic personality at age 13 showed small but significant differences in those variables. We note that the results stayed highly similar.
Sex was related to all variables. Specifically, females reported higher emotional and cognitive empathy and empathic personality profiles in both ages. As sex differences were not of main interest in this study, and because the study's sample size was relatively small for performing sexlimitation models, we took two steps. First, we combined the DZS and DZO twins into one group (Bates, 2020a). Before doing so, we verified that they could be treated similarly in terms of phenotypical similarity, by comparing the ICC (of type ICC1) of DZS and DZO twins, using the Fisher-Z test (Table S9). Except for emotional empathy at age 13 (Z = 4.11, p < .001), none of the variables showed significant differences in the ICC of DZS and DZO twins, thus supporting the decision to combine the two groups. Second, we residualized the variables by regressing them on participants' sex and saving the regression residuals as new variables. These residualized variables, controlling for sex, were used in subsequent analyses.

| Genetic analyses
ICC (of type ICC1) within MZ and DZ twin pairs for the self-reported empathy and empathic personality profile variables are presented in Table S10. MZ twins were substantially more strongly correlated than DZ twins, suggesting that twin similarity reflected genetic relatedness. Specifically, except for cognitive empathy and cognitive empathic personality profile at age 11, the correlations between MZ twins were higher than twice the correlations between DZ twins, indicating the presence of non-additive genetic effects and the absence of shared environmental effects. Accordingly, we examined ADE models in both the emotional and cognitive empathy analyses.

Model comparisons
The biometric cross-lagged models' fit statistics are presented in Table S11. In the full model of cognitive empathy, the variances at age 11 stemming from non-additive genetic effects were estimated at 0.00 for both empathy and the empathic personality profile. These paths and the correlation between them were, therefore, dropped from the final model, without reducing the model's fit (Table S11).
In the full model of emotional empathy, the variance of the empathic personality profile stemming from additive genetic effects at age 11 was estimated as 0.00 [0.00-0.24].
Nevertheless, since additive genetics represent the genes' main effect whereas non-additive genetics could represent interaction effects between genes (Purcell & Sham, 2004), and since dropping the main effect from a statistical model may cause artificial inflation of the interaction product, we decided to retain this component (Bates, 2020b). The correlations between the additive genetic paths in both ages had a 95% confidence interval of [−1 to 1]. This result, which may have stemmed from the very low and nonsignificant estimations of the A components, indicates that the additive genetic empathy-personality correlations are practically estimated at zero and are not meaningful. These paths were, therefore, dropped without reducing the model's fit (Table S11).

Phenotypical stability and cross-lagged effects
In the final models of both emotional and cognitive empathy (Figure 2), empathy and the empathic personality profiles showed moderate stability. The empathic personality profiles at age 11 showed moderate cross-lagged effects on empathy at age 13. The cross-lagged effects of empathy at age 11 on the empathic personality profiles at age 13 were somewhat lower (and nonsignificant for emotional empathy).

Genetic and environmental influences at age 11
At age 11, emotional and cognitive empathy and empathic personality profiles were moderately affected by genetic effects. The variables in the cognitive empathy model were influenced by additive genetics, whereas the variables in the emotional empathy model were affected mostly by nonadditive genetics. In both models, the genetic effects of personality and empathy (additive for cognitive, nonadditive for emotional) were highly or fully correlated. This indicates that most of the genetic factors affecting empathy also affect the personality characteristics disposing adolescents to be empathic. The nonshared environmental effects on empathy and personality correlated moderately. This indicates that in part, the same environmental processes affect both empathy and the empathic personality profiles.

Genetic and environmental influences at age 13
Figure 2 presents only novel genetic and environmental effects which are unique to age 13. Table S12, in contrast, presents the estimations of the genetic and environmental effects to the general age 13 variances. Overall, genetic effects at age 13 (mostly nonadditive) accounted for 24%-26% of the variance in emotional empathy and 40%-44% of the variance in cognitive empathy (Table S12). The genetic variance unique to age 13 explained 19%-20% and 32%-33% of the variance in the emotional and cognitive empathy analyses, respectively. Although this novel variance was not significant for most variables, perhaps due to the division of the genetic influences into A and D effects, these results suggest that new genetic effects emerge at age 13 and account for a joint change in adolescents' empathy and the respective empathic personality profiles. The new environmental effects at age 13 were significant and substantial for all the variables. Importantly, in both the emotional and cognitive analyses, the new environmental effects on personality and empathy were moderately correlated. This suggests that in part, the same new environmental processes change both empathy and empathic personality profiles during early adolescence.

| DISCUSSION
This study aimed to illuminate empathy development during early adolescence by investigating how genetic and environmental processes connect empathy to other relevant personality traits. By using a nuanced, single-item investigation approach, a cross-validated statistical learning algorithm, and a longitudinal twin design, it presents several advances. First, at the phenotypical level, it demonstrates a theoretically meaningful, fine-grained, and robust way to predict and describe individuals' emotional and cognitive empathy by many personality characteristics together, thus capturing the overall "empathic personality profiles." Second, at the aetiological level, it estimates to what extent empathy and the empathic personality profiles are influenced by the same genetic and environmental processes. Importantly, it shows that from age 11 to age 13, new environmental factors emerge that change both adolescents' empathy and empathy relevant personality traits. We start by discussing the phenotypical findings regarding the empathic personality profiles. Then, we discuss what new insights these phenotypes may reveal about empathy's genetic and environmental architecture.

| The empathic personality profiles
This study used a cross-validated ridge regression to predict emotional and cognitive empathy from nuanced personality characteristics. Because these indicators reflect characteristics from all the Big-Five personality domains, thus covering the most common and widely supported personality traits, the models' results reflect the combination of personality characteristics disposing individuals to be empathic. This analytic approach enabled us to gain moderate predictive accuracy and high generalizability across samples. The predictions were stable and robust, within each age (11 and 13) and between the ages, and in an older, nontwin sample. The relations between empathy | 765 ABRAMSON et al.

F I G U R E 2
Parameter estimates from the final biometric cross-lagged models of emotional and cognitive empathy and empathic personality profiles. Path estimates at age 11 represent the genetic and environmental contributions to the total variance. Path estimates at age 13 represent residual, novel effects unique to age 13 (thus reflecting the genetic and environmental contributions to the change in the phenotypes). The estimations represent standardized partial regression coefficients, and the square of these paths represents the genetic and environmental variances. For ease of interpretability, the ADE partial regression coefficients are presented as the squared root of the ADE variances. Numbers in brackets represent the 95% confidence intervals. Black lines represent significant effects at p < .05. and specific personality items also showed high age consistency, indicating phenotypical stability of nuanced personality-empathy relations during early adolescence. Importantly, non-self-reported measures of empathy were related to the empathic personality profiles, and these relations were similar in magnitude to the relations with the self-reported empathy measures. Thus, although the correlations with the non-self-reported measures were relatively low (as often is found in the empathy literature, e.g., Smith et al., 2019;Sunahara et al., 2022), they suggest that the empathic personality profiles reflect real variance in individuals' empathy.
Our analytic approach joins a new wave of studies attempting to represent an individual's predicted disposition to a certain outcome, based on many personality predictors together (Hall & Matz, 2020;Rozgonjuk et al., 2021;Sindermann et al., 2021;Stewart et al., 2022). Using these predicted scores in further scientific questions could help empirically investigate how a general personality profile (as opposed to a specific trait) contributes to empathy development. For example, in the present study, we utilized these scores to investigate how genetic and environmental factors broadly affect both the personality profile of empathy and adolescents' actual empathy. Other studies could use this approach to examine how an overall set of empathy-relevant traits interacts with specific environmental factors (e.g., parenting, school climate, socioeconomic status) in contributing to one's empathy.

| Understanding the relations between empathy and other personality characteristics
This study investigated the relations between empathy and personality during early adolescence, a developmental period that was relatively less addressed in previous studies on empathy-personality relations. Importantly, its item-based approach enabled not only to robustly predict empathy but also to describe the fine-grained nuances of personality that are important for empathy and were less researched. In line with previous literature (e.g., Jolliffe & Farrington, 2006;Melchers et al., 2016), emotional and cognitive empathy were predicted by items representing some features of agreeableness. Although there are similarities between some features of agreeableness and some empathy definitions (mainly when referring to empathic concern-a construct not investigated here) (Hou et al., 2017;Mooradian et al., 2011), we note that the way empathy was defined and operationalized in this study was very different from the examined agreeableness characteristics. Specifically, cognitive empathy was composed of items about understanding others' emotions, and emotional empathy was composed of items about being carried away by others' emotions (see Appendix 1 for the full item list). Thus, the relations found in this study probably reflect real associations between core features of empathy and positive attitude toward others, rather than trivial connections between empathy and items of shared content.
Interestingly, the agreeableness items that predicted empathy more strongly were items reflecting a positive attitude toward others and a trusting, prosocial, or forgiving nature. In contrast, items reflecting cooperative behaviors, which do not clearly state an emotional attitude toward others, had lower predictive power (one such item was significant for cognitive empathy prediction at age 11). This finding suggests that not all agreeable behaviors are necessarily important for empathy. Instead, what may be most crucial is the internal motivation to approach, trust, and accept other people.
Cognitive empathy was predicted mainly by items reflecting openness to experience, in line with previous studies (e.g., Jolliffe & Farrington, 2006;Melchers et al., 2016). This finding suggests that a motivation to explore new ideas helps to understand other people's emotions (DeYoung, 2015a). Since at least some features of cognitive empathy entail complex cognitive abilities, another possibility is that openness to experience is related to cognitive empathy due to the relations of both traits with intelligence (DeYoung, 2015a; Jolliffe & Farrington, 2006). Future studies incorporating intelligence tests into the learning algorithm could help differentiate between the two explanations.
Except for items of agreeableness, emotional empathy was predicted mainly by items of neuroticism, indicating that to be sensitive to others' feelings, one needs not only affiliation to other people but also a basic propensity to experience negative emotions (Edwards et al., 2015;Eisenberg, 2010;Wolf et al., 2015). Interestingly, items with higher predictive power were those reflecting anxiousness and situational reactivity. In contrast, items representing general moodiness were less related to emotional empathy. This finding shows that the relation between emotional empathy and neuroticism is more nuanced than previously found and that neuroticism may be positively related to empathy specifically through the tendency to emotionally react to the environment. Further support for this idea comes from multiple findings showing that the tendency to be emotionally reactive is a marker for environmental sensitivity (Belsky & Pluess, 2009;Slagt et al., 2016). That is, the socio-emotional development of individuals who are more emotionally reactive is more influenced by the environment they are exposed to, for better or for worse. Accordingly, the association found between reactivity and emotional empathy might represent an  (Abramson, Uzefovsky, Toccaceli, et al., 2020;Briley & Tucker-Drob, 2014), individual differences in emotional and cognitive empathy and their personality profiles were influenced both by genetic factors and by the nonshared environment. The genetic factors affecting empathic tendencies and empathic personality profiles were substantially correlated, suggesting that most of the genetic factors contributing to empathy also contribute to empathy-relevant personality traits.
Emotional empathy was affected mainly by nonadditive genetic factors. Cognitive empathy was affected by additive genetic factors at age 11, whereas new, nonadditive genetic factors contributed to the unique variance at age 13 (albeit significant only for the empathic personality profile). Nevertheless, the lack of additive genetic effects should be considered carefully due to this study's relatively small sample size for genetic analyses. Accordingly, we interpret the results in terms of the overall genetic contribution to empathy and personality, rather than giving specific interpretations to the division between additive and nonadditive genetics.
The common genetic influences on empathy and empathic personality profiles suggest that genetically based hormonal and neural systems affect both tendencies (Plomin et al., 2016). The behavioral content of the empathic personality profiles directs attention to systems affecting stress reactivity (for emotional empathy), affiliation, and approach tendencies, such as the oxytocinergic and dopaminergic systems. Indeed, genetic variations which encode for differences in the expression of oxytocin and dopamine were associated with empathy (Ben-Israel et al., 2015;Chong et al., 2019;Gong et al., 2017;Pearce et al., 2017;Rodrigues et al., 2009;Uzefovsky et al., 2014Uzefovsky et al., , 2015 as well as with affiliation, extraversion, and stressreactivity tendencies (Anacker et al., 2013;Fischer et al., 2018;Pearce et al., 2017;Rodrigues et al., 2009;Zheng et al., 2020). Although the specific effects of candidate genes do not consistently replicate, possibly due to low statistical power, small effect sizes, and numerous gene-environment interaction processes (Halldorsdottir & Binder, 2017;Montag et al., 2020), this body of literature and the present study's findings highlight the potential of investigating oxytocinergic and dopaminergic genetic contributions to both empathy and personality. Particularly, incorporating physiological phenotypes within a twin design (e.g., saliva oxytocin levels, eye blink, which is indicative of dopamine levels, Barkley-Levenson & Galván, 2017;Tummeltshammer et al., 2019) to the examination of personality and empathy heritability may elucidate the pathways connecting genes, personality, and empathic behaviors.
The nonshared environmental factors affecting empathy and the empathic personality profiles also correlated at age 11, albeit to a lower extent. Interestingly, at age 13, new environmental factors emerged that explain the common variance between them, thus reflecting a personalityempathy joint change. This finding suggests that some experiences which change adolescents' empathic tendencies may do so by changing their general tendencies to approach and emotionally react to their surroundings. For example, while most nuanced personality-empathy predictions remained stable from age 11 to age 13, one noticeable change was that at age 13, items representing sociability, social dominance, and assertiveness, became more predictive of cognitive empathy (this change might have accounted for the relatively lower age stability of items' predictability for cognitive empathy compared to emotional empathy). Environmental changes in adolescents' social structures during those years (e.g., the transition to junior high, Smetana et al., 2006), other school climate considerations (Barr & Higgins-D'Alessandro, 2007), or increase in use of online social media (Booker et al., 2018) may create a relation between being empathic and being sociable and socially valued, as was indeed found in adolescents around the age of 13 (Huang & Su, 2014;Wölfer et al., 2012).
Development of interpersonal relationships and their effects on adolescents' cognitive attachment system may also change both empathy and approach tendencies (Silke et al., 2018;Stern & Cassidy, 2018). Indeed, differences in adolescents' attachment security were found to moderate the extent of empathy change during late adolescence, such that those with lower security at age 14 showed an increase in empathic abilities from 16 to 18 years, whereas those with higher security showed higher, more stable empathic skills, perhaps due to an earlier empathic maturation (Stern et al., 2021). Interestingly, attachment security in that study also predicted the extent to which teens' friends asked for their support. This finding is in line with the possibility raised here, that environmental processes affect teens' empathy as well as how socially valued they are. Again, future twin studies that incorporate measures of adolescents' social environments into empathy-personality genetic analyses could investigate such mechanisms.

| Limitations and future directions
This study has several strengths, including cross-validated replications across different samples and ages, and validation of the personality profiles with non-self-report measures. However, it also has a few limitations that should be addressed in future studies. First, although being large enough for phenotypical investigations, the study's sample size was relatively small for multivariate genetic analyses. Second, as this study was administered online, our emotion recognition task scores may have included irrelevant variance stemming from differences in participants' viewing conditions (e.g., internet speed, screen size). Third, the emotional empathy items on the BES pertained to either general or negative emotions, but not to positive emotions. This item composition may have accounted for the high prediction of emotional empathy by items that reflect negative reactivity. Future studies should investigate whether this prediction also generalizes to positive empathy, or whether it is unique to empathy to negative emotions.
A fourth limitation is that while the learning algorithm estimated the additive contribution of all the personality items, it did not assess the predictive role of interactions between personality nuances. We chose not to examine such interactions to allow for simple theoretical interpretability, and because the immense number of possible inter-item interactions would have resulted in a disproportional number of predictors compared to the study's sample size. Nevertheless, interactions between personality traits could have theoretical meaning for empathy. For example, the interaction between emotional reactivity and effortful control or regulation abilities was found important for empathy expression Eisenberg, 2010;Eisenberg et al., 1994). Thus, the examination only of main effects may explain why items reflecting conscientiousness and control were not meaningful for empathy prediction in this study. Future studies with larger sample sizes should strive to examine interaction products between empathy predictors. Lastly, although this study's sample had demographic characteristics similar to the Jewish population in Israel, future studies should attempt to replicate its findings with other ethnic groups and in different cultures.

| CONCLUSIONS
This study illuminated the genetic and environmental processes affecting adolescents' emotional and cognitive empathy by connecting them to the aetiological origins of other personality traits. Using a data-driven, nuanced item-based approach, it provided robust, detailed, and theoretically meaningful constructs that reflect the overall personality profiles disposing adolescents to be empathic. The use of these constructs in further analyses showed that empathy and personality share overlapping genetic and environmental influences, and that during early adolescence, new overlapping environmental factors emerge, changing both empathy and personality. Addressing the complexity of adolescents' personalities could inform empathy development and open new avenues for the research of emerging individual differences.