Studies of empathy have typically used self-reported measures to assess empathy in individuals, the most popular being Mehrabian and Epstein's (1972) Questionnaire Measure of Emotional Empathy (QMEE), Bryant's Index of Empathy (Bryant, 1982), and the Interpersonal Reactivity Index (IRI; Davis, 1983). The QMEE was developed to measure emotional (affective) empathy in adults, with 33 questions falling under the subscales of “Susceptibility to Emotional Contagion”, “Appreciation of the Feelings of Unfamiliar and Distant Others”, “Extreme Emotional Responsiveness”, and “Tendency To Be Moved by Others' Positive Emotional Experiences”. Responses to each item ranges from +4, suggesting very strong agreement, to−4, indicating very strong disagreement. Bryant's Index of Empathy (Bryant, 1982) adapts the QMEE (Mehrabian and Epstein, 1972) to be suitable for use with school-aged children and adolescents. This new scale consists of 22 questions where respondents are asked to answer questions with a yes/no format. While the QMEE and Bryant's Index of Empathy primarily focus on the affective component of empathy, the IRI (Davis, 1983) encompasses both affective and cognitive components. The IRI consists of 28 items that fall under four different subscales, that is, “Perspective-Taking”, “Fantasy”, “Empathic Concern”, and “Personal Distress”. Respondents are required to rate the questions on a 5-point Likert scale ranging from “Does not describe me well” to “Describes me very well”.

Conceptual limitations have been evidenced in the QMEE and IRI, as questions measuring affective empathy are perceived reactions, rather than shared emotions (Jolliffe and Farrington, 2006). Most self-reported measures are also created for an adolescent and/or adult population and are not suitable for young children. Although the Index of Empathy (Bryant, 1982) provides a scale for a younger population (6+ years), a systematic review (Sesso et al., 2021) showed this measure to have one of the lowest internal consistencies out of the 16 measures in review. Further to the Index of Empathy (Bryant, 1982), Bensalah et al. (2016) presents a more recent self-reported measure suitable for middle-aged children (6–12 years old). This measure is adapted from the adult version of the Basic Empathy Scale (BES; Jolliffe and Farrington, 2006) and contains 20 items involving cognitive and affective empathy that are rated on a 5-point Likert scale. While similarities between the adult version (BES-A) and the child version (BES-C) are acknowledged, highlighting the scale's potential for studying empathy in middle childhood, the authors also indicate a limitation in the scale's low internal consistency.

Beyond the issues discussed, there are other validity concerns worth acknowledging in self-reported measures (Eisenberg and Miller, 1987). Specifically, reported answers can be affected by social desirability (Neumann et al., 2015), language ability (Wang and Wang, 2015), and even more fundamentally, whether young children are aware of what they are feeling and can/are willing to report their emotional states accurately (Eisenberg and Fabes, 1990). Thus, further investigations are needed.

Picture-story methods have also been used to measure empathy in children, with the most widely used being Feshbach and Roe's Affective Situations Test for Empathy (FASTE; Feshbach and Roe, 1968; Zhou et al., 2003). This measure involves the child being told a story while shown pictures of protagonists in emotionally charged situations (Zhou et al., 2003) and then being asked how the protagonist feels and how they, themselves, feel (Feshbach and Roe, 1968). Although the FASTE was an important early method (Eisenberg-Berg and Lennon, 1980; Iannotti, 1985), this measure captures only the affective component of empathy. Furthermore, concerns lie in the stories ability to evoke sufficient affective responses (Lennon et al., 1983; Eisenberg and Miller, 1987), alongside social demands (Eisenberg-Berg and Lennon, 1980) and gender-biases (Eisenberg and Lennon, 1983; Lennon et al., 1983).

A more recent performance-based measure called the Kids' Empathic Development Scale (KEDS; Reid et al., 2013) developed for 7–11-year-olds, also exposes children to emotional stories but is inclusive of a cognitive component. Children are asked to provide a justification that explains their emotional inferences. However, this measure has elicited its own criticisms. Particularly, picture scenarios may restrict the depth and capacity for which perspective-taking is measured, as cognitive empathy should relate to the understanding of emotional situations that are specific to people's emotional states (Bensalah et al., 2016). The affective component of the measure has also been criticized, as it requires children to draw on an emotional inference, rather than the explicit sharing of emotions (Bensalah et al., 2016). Overall, self-reported measures whether conducted through questionnaires or picture-story methods each have their existing limitations that need to be considered further.

Researchers have developed observational methods for measuring empathy, which assess children's facial, gestural, and behavioral responses (Eisenberg and Fabes, 1990). Generally, following Zahn-Waxler et al. (1992) observational coding scheme, infants' reactions to their mother's and/or an experimenter's distress is measured. Due to its ability to reduce self-presentational biases and its capacity to lessen language comprehension barriers (Zhou et al., 2003), studies have been able to measure empathy through observational approaches in children and in infants as young as 15 months old (see Miller et al., 1996; Zahn-Waxler et al., 1992). In more recent studies, rare cases of comforting, as well as empathic concern have also been evidenced in infants as young as 8 or 9 months old (see Roth-Hanania et al., 2011; Vreden et al., 2025).

While observational measures may provide a more objective measure of children's empathy, they still present limitations. Arguably, they do not fully capture the cognitive component of empathy. Although children's exploration or attempt to understand a distressed victim, known as ‘hypothesis testing', or ‘inquiry behavior', gives a signal that they are trying to understand the situation, this behavior does not indicate what the child actually understands (Davidov et al., 2021). Thus, hypothesis testing, as captured under the cognitive component of empathy in existing observational measures (Zahn-Waxler et al., 1992) should not be used as a pure assessment of cognitive empathy (Davidov et al., 2021).

In sum, the measurement of empathy remains a challenge. Administration, language barriers, validity of scales, and the lack of cognitive and affective aspects have each presented a limitation to the methodological approaches used to measure empathy. Self-reported measures, though frequently used in studies of adults and children, can present biases that affect the accuracy of participants' answers. While observational measures seem to reduce biases, the cognitive component of empathy cannot be inferred from behavior alone (Hoffman, 2000).

Two story stems, Hurt Knee (Woolgar and Murray, 2010) and Three's a Crowd (Bretherton and Oppenheim, 2003) were chosen, as each scenario consists of an emotionally charged story beginning that involves a child being put in a position of potential distress or needing care. These scenarios present the child participant with an opportunity to elicit a possible empathic response toward the protagonist in the story. Children are invited to choose dolls that represent themselves and their family members, as a means of accessing their representations of their relationships and their own and others' behaviors.

Hurt Knee Description. In these examples the participant child is called Sarah, and her friend is called Molly. The experimenter says, “it's a new day and the family are at the park. Mummy and daddy are sitting over here. Sarah is playing with Molly. Sarah runs, falls, and cuts her knee. Sarah says, “ow! I've really hurt my knee!” Show me and tell me what happens next”. While the child is manipulating the dolls to finish the story, the experimenter will also ask, “how does Sarah feel when she fell over? Why does she feel this way?” and “how does mummy feel when Sarah fell over? Why does she feel this way?”. These questions are only asked of the child participant doll who falls over, and of the primary caregiver doll who is witnessing the situation.

Three's a Crowd Description. In these examples, the participant child is called Sarah. Ava is her friend, while Sophie is another child that is brought into the story. The experimenter says, “it's a new day and the family are at another park. Mummy and daddy are sitting over here. Sarah is playing with Ava and her new ball. Sophie comes along and asks if she can play. Sarah says, “sure”, but Ava says, “no way, I don't want to play with Sophie”. Show me and tell me what happens next”. While the child is manipulating the dolls to finish the story, the experimenter will also ask, “how does Sarah feel when Sophie couldn't play? Why does she feel this way?” and “how does Sophie feel when she couldn't play? Why does she feel this way?”. These questions are only asked of the child participant doll who is witnessing the situation, and of the child doll who could not play.

Coding of cognitive empathy is captured by three categories in each of the two tasks, that is, whether the doll play shows (1) helping, (2) care/concern, and (3) perspective-taking. Within each of the three categories, there are two subscales of behaviors that can exist. See Tables 2, 3 for the subscales and examples of participant responses in each of the two tasks.

These subscales were created with the intention of capturing a variation of behaviors possible, as shown by the child participants. Under helping, the child could seek help from an adult or show direct helping through physical behavior in the doll play. Under care/concern, the child may verbalize comments that show care or concern (e.g., “are you okay”, “just sit down now and rest”). Under perspective-taking, the participant child must correctly answer the experimenter's questions of how particular characters are feeling and why, or they must state character's emotions during the doll play. These behaviors are coded as present (1) or absent (0). A final empathy score is given at the end of both tasks, by summing all the scores of the categories together. Ultimately, the maximum score a child can receive is 12, representing the highest level of cognitive empathy in this coding scheme. See Appendix A for an example of the coding, and the Supplementary materials for a detailed description.

The three categories of helping, care/concern and perspective-taking were generated through exploring existing measures combined with literature, and viewing the data repeatedly, to see what responses children were showing during each of the story stem tasks.

Observational studies of empathy in existing literature have commonly used Zahn-Waxler et al. (1992) version of the MacArthur Longitudinal Twin Study coding scheme (Emde et al., 1992). Many of the responses coded involved facial, behavioral, or gestural expressions for the victim (e.g., sustained expressions, brows furrowing, crying, etc.). However, the behaviors shown in the story stem tasks did not elicit such affective responses. Cognitive empathy is, therefore, the focus of this measure.

Helping. Helping is defined by actions that provide assistance and/or resources to benefit a needy other (Eisenberg and Mussen, 1989). This category draws influence from the MNCS, as prosocial behaviors like helping, are categorized under the Empathic Relations theme. In the Hurt Knee task, children often demonstrated helping behaviors, for example, when a plaster was given to the hurt child, or in the Three's a Crowd task, one might be helping the excluded child be included in the play.

Care/concern. Caring is defined as having an emotional commitment to an individuals' well-being that can encourage prosocial efforts to support them (Weiner and Auster, 2008). Concern for others is the tendency to express interest in others' emotional distress (Rhee et al., 2013). These behaviors were coded together, as in the Hurt Knee and the Three's a Crowd tasks, children would often express care or concern with verbal comments such as “are you okay?” or “it'll be okay, let's not run again”. Drawing influence from the MNCS, behaviors demonstrating reassurance (e.g., “don't worry, I'm here”) were also categorized under the Empathic Relations theme.

According to literature, prosocial behaviors should be considered as a part of the ‘empathic repertoire', as children may demonstrate an emotional reaction (i.e., an attempt to alleviate the distress) that conveys some sort of understanding or concern for a victim in distress (Moreno et al., 2008).

Perspective-taking. Perspective-taking is arguably, the essential factor of cognitive empathy, as it represents one's ability to recognize and understand the feelings of others. One of the most crucial features of a story stem task is that the experimenter explicitly asks the child, “how does X feel?” and “why does X feel this way?”. These questions serve to elicit cognitive understanding and allow researchers to determine whether a child recognizes how particular characters in the stories are feeling and why.

Strengths and Difficulties Questionnaire. The Strengths and Difficulties Questionnaire (SDQ, school-aged version for 4–17 years; Goodman, 1997) has been widely used in clinical and educational research (Goodman and Scott, 1999) to understand children's behavior. This study uses only the prosocial behavior subscale of the SDQ, which contains five questions. Respondents are asked to rate each item on a three-point scale (0 = not true, 1 = somewhat true, or 2 = certainly true) according to the child's behavior within the last 6 months. Where parental consent for contact with the child's school was given, the child's class teacher was also invited to complete the SDQ. Parent and teacher reports are widely used in studies of children's mental health and wellbeing (Murray et al., 2021), as each provides a different perspective and teachers and parents are reporting on what they see across different settings (i.e., home vs. school). Studies have reported satisfactory psychometric properties related to reliability and validity when using the SDQ (Mieloo et al., 2012).

Story Stem Narratives. Children's prosocial behavior was also coded using the story stem method. This was a novel scale developed from the initial study trial, using ideas from the MNCS but rated on a 0–3 scale to assess the intensity of behaviors shown. A different coding scheme was used but involved the same two tasks of Hurt Knee and Three's a Crowd. An additional story stem task called Spilt Juice (see The MacArthur Story Stem Battery; Bretherton et al., 1990; Bretherton and Oppenheim, 2003), was also used in this coding scheme to explore prosocial behavior. However, there was little variance in children's expression of empathy during piloting of coding using this story stem and thus, it was excluded. Prosocial behaviors under this measure were coded on a scale consisting of 0 = no prosocial behaviors, 1 = minimal, 2 = moderate, or 3 = extreme. Minimal prosocial behaviors were evidenced if characters were friendly toward each other but not going out of their way to include or help (e.g., “I clean up the juice). Moderate behaviors were evidenced if characters were explicitly helpful, considerate, comforting or friendly (e.g., “mum helps me clean up the juice”). Extreme behaviors were evidenced if characters go above and beyond to help, or if characters modify their behavior to help (e.g., “since dad drank all the juice, I go to the shop to get him some more”).

Callous-unemotional traits (CU traits) were assessed using a seven-item scale, as reported by the primary caregiver. Four items were taken from the SDQ prosocial scale [i.e., ‘considerate of other people's feelings' (reverse-scored)], while three items were taken from the Inventory of Callous-Unemotional Traits (ICU; Frick, 2004) [i.e., ‘does not show feelings or emotions']. This combined scale has been used by previous groups and satisfactory psychometric properties have been reported (see Dadds et al., 2014; Takahashi et al., 2021).

Cohen's Kappa Cohen (1960) was used to determine the inter-rater agreement between two postgraduate researchers trained on this coding scale (double-coding 7.5% of the videos). The coders reached 0.91 agreement for the total cognitive empathy score—the main outcome of interest. Scores between 0.81 and 1.00 represent an excellent level of agreement (McHugh, 2012).

Cronbach's alpha Cronbach's (1951) was used to test for internal consistency. The total cognitive empathy score had an alpha of 0.68, which reached ‘acceptable' reliability, that is, within 0.6–0.7 (using the rule-of-thumb guidelines; Nunnally, 1967, 1978; Nunnally and Bernstein, 1994).

Females scored higher than males for the overall, total cognitive empathy score (mean difference = 1.15, 95% CI [0.46, 1.84]).

There was no evidence of an association in cognitive empathy by age (r = 0.07, p = 0.32), although the range of age was not very large (6.7–9.6 years).

A moderation analysis was performed by adding an interaction term (age by gender) with cognitive empathy. This additional test was conducted to determine whether the relationship between cognitive empathy and age differs by gender. The interaction between age and gender was not statistically significant (b = −0.69, p = 0.20). This suggests that gender does not change how age relates to empathy.

There was a small positive association found between parent-reported prosocial behavior (SDQ) and cognitive empathy (r = 0.16, p = 0.02). However, no association was found between teacher-reported prosocial behavior and cognitive empathy (r = 0.09, p = 0.33), though fewer teachers (n = 97) reported. A strong positive association was found between child-reported prosocial behavior and cognitive empathy using story stems (r = 0.61, p = 0.00).

There was an association between callous-unemotional traits (CU traits) and cognitive empathy with a small negative correlation (r = −0.15, p = 0.02).