The Basic Empathy Scale: Evidence of Internal Structure in the Brazilian Context

This study aimed to investigate evidence for the internal structure of the 20 item Basic Empathy Scale (BES-20), and verify the parameters of each of its items. The participants totaled 300 subjects from the community sample who answered both the BES-20 and demographic questions. The results indicated a two-factor measure involving: affective empathy ( α = .84 and ω =.84) and cognitive empathy ( α =.84 and ω =.84). Analyses using Item Response Theory demonstrated that the items of the BES-20 properly discriminated between participants, requiring low levels of latent trait for agreement/checking-off responses in each category. In addition, we developed a 10 item (short-form) version (BES-10) which when compared to the full version yielded equivalent psychometric parameters and information scores. Overall, both the BES-20 and BES-10 were confirmed as qualified for psychometric use in future studies to measure empathy in Brazil.


Introduction
In a global sense, empathy involves the reaction experienced when observing the experience of another (Mitsopoulou & Giovazolias, 2015) and constitutes a personal competence which contributes to initiating and maintaining desirable interpersonal relationships (Zych et al., 2020).Empathy allows an individual to perceive the emotions of others, whether emotionally or cognitively, along with considering their perspective.It is distinguishing between one's own emotions and those of others (Eisenberg, 2000;Riess, 2017).Empathy enables a relationship of congruence between the observer and the target subject (Eisenberg, Shea, Carlo, & Knight, 1991).The contemporary conceptualization of this construct as a psychological phenomenon brings a multidimensional perspective that encompasses both its affective and cognitive dimensions (Ang & Goh, 2010).
Affective empathy refers to the ability to experience the same emotions as others (i.e., emotional congruence; Bryant, 1982).A direct consequence of this ability is seen in the motivation of people to express altruistic behavior towards others, to maximize their positive emotions (e.g., happiness) or reduce their negative emotions (e.g., fear; Davis, 1996).Cognitive empathy, in turn, refers to the ability to recognize and understand the emotions of others (Hogan, 1969), the ability to adopt someone else's perspective and, by extension, infer their thoughts and feelings (Preston & De Waal, 2002).In practical terms, cognitive empathy allows a given individual to be able to predict the behavior of another person, facilitating dialogue, and social understanding (Smith, 2006).
Empathetic affective and cognitive processes can be understood as mechanisms that sometimes promote or sometimes inhibit behavior (Decety, Bartal, Uzefovsky, & Knafo-Noam, 2016).In practical terms, studies have shown that the empathy construct has implications for understanding situations in daily life, such as antisocial and pro-social behaviors.The literature has documented that low levels of empathy (in non-virtual contexts) are specifically associated with perpetration of bullying in the school environment (Zych et al., 2019a), with reductions in both ethnic and religious confrontation, and with racial prejudice (Cikara, 2015).On the internet, antisocial acts such as cyber-bullying (Zych et al., 2019b) and "internet trolling" behavior (i.e., cybertrolls) generate conflict on the internet by provoke others (Sest & March, 2017).However, both affective and cognitive empathy present demonstrated pro-social behavior motivating functions (See Eisenberg, Eggum, & Di Giunta, 2010), for example, volunteering or donating to charities, or inhibiting aggressive behavior (Rodriguez, Mesurado, & Moreno, 2019).
In addition to predicting behavior, empathy is related to the more stable individual characteristics, or personality.With regard to studies modeled on the five great personality factors (i.e., the Big Five), although still inconclusive, studies show that the traits of agreeableness and conscientiousness are the most consistent predictors of affective and cognitive empathy (Melchers et al., 2016), presenting medium-sized effects.Yet all of the socially aversive personality traits (i.e., narcissism, psychopathy, and Machiavellianism) are negatively associated with affective empathy (r = -.21;-.52, and -.40, respectively), indicating deficits in the ability to feel the same emotions as others.On the other hand, the data shows that such traits do not present significant losses in cognitive empathy, with only narcissism being positively and significantly related to this factor, and presenting a low magnitude correlation (Wai & Tiliopoulos, 2012).
Similarly, various studies have raised sociodemographic correlates for empathy.As for sex, the commonly accepted stereotypes suggest that women have a greater capacity to understand the thoughts and feelings of others than men (Klein & Hodges, 2001).In fact, empirical studies corroborate that women have higher levels of empathy than men (e.g., Schieman & Van Gundy, 2000), this is a consistent finding in research (Jolliffe & Farrington, 2006).Certain studies have sought to prove factorial invariance in relation to the participants' sex with instruments ensuring the interpretability of the data (e.g., the Basic Empathy Scale -BES; Anastácio et al., 2016;Pechorro et al., 2018).In relation to other variables, it is evident that as age advances, empathy increases (Oh et al., 2020), and that being married contributes to higher levels of empathy, as well as having children (Park et al., 2016;Wang et al., 2017).Finally, having siblings is also associated with higher levels of empathy (Park et al., 2016).
Although in recent research and psychological practice the empathy construct has received more attention (Romera et al., 2019).Instruments to measure the variable were already available in the late 1960s; specifically, the Hogan Empathy Scale (HES) (Hogan, 1969), the Questionnaire Measure of Emotional Empathy (QMEE; Mehrabian & Epstein, 1972) and the Interpersonal Reactivity Index (IRI: Davis, 1983).However, these measures present certain limitations.Initially, concerning the QMEE and IRI measures, Jolliffe and Farrington (2006) pointed out that both instruments treat empathy and sympathy constructs as synonymous variables, yet empathy implies an emotional reaction quite similar to the target person's emotion, while sympathy may not necessarily generate the same emotion (Eisenberg & Strayer, 1987).Additionally, none of the three scales, according to Jolliffe and Farrington (2006), measure cognitive empathy.For example, though IRI possesses a "perspective-making" factor which is similar to the cognitive empathy variable, it does not imply a specific ability to understand the other's emotions, but only the ability to assume their perspective.
As a result of the above scenario, Jolliffe and Farrington (2006) proposed the Basic Empathy Scale (BES), designed to measure empathy as a process involving both understanding and sharing someone else's emotional state or context (Cohen & Strayer, 1996).It allows consideration of affective congruence (i.e., affective empathy) and understanding the other's emotions (i.e., cognitive empathy).Thus, BES, with its structure involving two latent factors, appears to overcome the limitations of the other instruments, since HES measures only cognitive empathy, QMEE measures only affective empathy, and IRI, although it measures both dimensions, is, as already mentioned, imprecise in its operationalization of the cognitive facet.Jolliffe and Farrington (2006) constructed the initial version of the BES instrument (consisting of 40 items).In their first study, the authors used a sample of 363 English adolescents with a mean age of 14.8 years.Using principal component analysis (PCA), the authors reported a structure with two factors, affective and cognitive empathy, which respectively explained 19.5% and 7.6% of the total variance.Subsequently, the authors eliminated items with individual factorial loads of less than .40(i.e., empirical criteria), and in the final version, the BES-20 was developed using 11 items for the affective dimension (Cronbach's alpha = .85)and 9 items for the cognitive (Cronbach's alpha =.79).From a new sample of 357 adolescents, the authors then performed confirmatory factor analysis on the total database (N = 720), and in summary, they observed adequate fit indices for the two-factor solution (e.g., goodness of fit index GFI > .85,and an adjusted goodness of fit index AGFI >.80), which was superior to the alternative, single-factor model.
In summary, given that the BES-20 is one of the most used measures of empathy in the world (Zych et al., 2020), as well as the fact that it presents satisfactory psychometric properties in differing contexts, the present study had the general objective of adapting the BES-20 to the Brazilian context.As a specific objective, we sought to provide a brief version of the instrument, with operational advantages such as shorter administration time, greater participant engagement, and greater response accuracy (Kruyen et al., 2013), yet that also presents the ability to execute large-scale studies that measure various psychological constructs (Ziegler et al., 2014).For this compact version, based on the internal structure and precision of the measure: two evidence hubs for validity were considered: (1) evaluation of the factorial structure and internal consistency of the measure, and (2) investigation of its discrimination and difficulty parameters, and its information curve.This, to select the psychometrically most appropriate measure items (i.e., the least redundant), to maintain the measuring power for the construct (Rammstedt & Beierlein, 2014).

Translation and adaptation of the BES to the Brazilian context
The procedures for translation and adaptation of BES were conducted according to the guidelines of the International Test Commission (ITC, 2010), and standardized for Portuguese by Borsa, Damásio, and Bandeira (2012).Thus, considering the original language of the BES, two independent translators proceeded to translate the measurement items from English into Portuguese.In addition to the translation itself, experts were asked to recode all inverse items (i.e., those which measure in the opposite direction of the construct), this, because of their likely negative impacts on the psychometric properties of the measure (Suárez-Alvarez et al., 2018).This has already been observed in adaptation processes in other cultural contexts (e.g., Poland; Zych et al., 2020) in which some of the inverted items were not properly understood by the participants (Heynen et al., 2016).Back translation (Portuguese to English) was then performed by a new independent translator.After these procedures, the original and back-translated versions of the instrument were compared by psychologists -evaluators, which concluded that the scales were semantically analogous.In addition, semantic validation of the measure was performed in the target population, with eight adults participating.As a result of this last phase, the final version of this scale was reached (Table 1).

Instruments
Participants answered an online questionnaire that, in addition to sociodemographic questions (age, sex, sexual orientation, marital status, and religion), contained the Basic Empathy Scale (BES-20).Originally developed by Jolliffe and Farrington (2006).This scale consists of a 20-item self-report measure designed to assess empathy from a two-factor structure: affective empathy (11 items; e.g., "I am easily influenced by the feelings of my friends."),and cognitive empathy (9 items; e.g., "I often notice when people are happy.").The set of items was answered on a five-point ordinal response scale (ranging from 1 = strongly disagree to 5 = strongly agree).

Procedures
Data collection was performed online.People were contacted through social networks (e.g., Facebook, Instagram), and asked to respond voluntarily to self-administered instruments.In addition, the prerogatives provided for in resolutions 510/16 of the National Health Council regarding the regulation of research with human beings were respected, with each participant declaring his consent by signing a Free and Informed Consent Form.This research was approved by the Ethics Committee for Research with Human Beings at the Federal University of Paraíba (Opinion no.3,939,602), and on average, it took the participants 10 minutes to complete their responses to the study.

Data analysis
Evidence of validity based on the internal structure and accuracy of the BES-20 was investigated via Exploratory Factor Analysis (EFA), using the Unweighted Least Squares (ULS) estimator, and the polychoric correlation matrix as input, given the measure's nature in being composed of ordered categories (Holgado-Tello et al., 2010).To determine the number of factors to be extracted, the Hull method was applied in order to find an ideal estimate that best represented the relationship between the fit of the model and the number of parameters.This procedure was verified using the Comparative Fit Index indicator (CFI ≥ .90;Lorenzo-Seva et al., 2011).For internal consistency, Cronbach's alpha, and McDonald's omega indexes were used.The analyses were performed using the Factor program version 10.8.01 (2018) (Lorenzo-Seva & Ferrando, 2013).
For Item Response Theory (IRT) analysis, R language was used with software version 3.3.2(R Development Core Team, 2015), considering a Graduated Response Model (Samejima, 1969), since the response scale was polytomous.The statistical package mirt (Chalmers, 2012) was used to estimate the discrimination and difficulty parameters, and the item information score as well.

Evidence of Validity: internal structure and BES accuracy
The data were analyzed using the Factor program, with exploratory factor analysis (EFA).Initially, the data matrix was favorable (Kaiser-Meyer-Olkin = .83;Bartlett's Sphericity Test = 2803.4 [190]; p < .001).As for EFA, using the Hull method (i.e., factor retention method) the results indicated a structure with two factors.It is noteworthy that interpretation of the data matrix used the oblique rotation method, (weighted oblimin; Lorenzo-Seva, 2000), since it was assumed that the two factors are correlated.In short, for the BES a factor structure was observed formed by two factors: affective empathy and cognitive empathy, which respectively explained 30% (eigenvalue = 8.34) and 15% (eigenvalue = 3.02) of the total variance (See Table 1).
As noted in Table 1, the Affective Empathy factor brought together 10 items, all of which presented adequate saturation levels (>.30), ranging from .39 (Item 4) to .71(Item 17).This factor also presented satisfactory reliability indexes (α = .84and ω = .84).Similarly, the second factor, Cognitive empathy, was formed by 10 items with adequate saturation levels (>.30), ranging from .31 (Item 9) to .82(Item 20).As in the previous case, the reliability indexes were adequate (α = .84and ω = .84).It is noteworthy that unlike the original study, it was observed that Item 1, "I am sensitive to the feelings of my friends" (in theory originating from the affective empathy factor), saturated the cognitive dimension factor.It was therefore disregarded in subsequent analyses for presenting cross factorial loads (i.e., empirical exclusion criterion; Hair, Jr. et al., 2014).

BES Item Parameters: discrimination, difficulty, and information curves
As for analytical procedures via IRT (See Table 2), it was observed that the items of the Affective Empathy factor, when taken together, were highly discriminative (Mean = 1.73,SD = .53),the least discriminating items were item 4 (a = .78),and item 13 (a = .95)whose discrimination was moderate.The most discriminating item was item 2 (a = 2.39), with very high discrimination.The other items presented high discrimination (a > 1.35).As for the item response thresholds in relation to the difficulty parameter (b1 -b4), items were found that required low amounts of latent trait -on average (Mean = -.89;SD = .48),ranging from -1.74 (item 8) to -.20 (item 15), as shown in Table 2.When taken together, the Cognitive Empathy Factor items presented very high discrimination (Mean = 2.30, SD = .53),with item 9 (a = .93)being the least discriminative, being moderate together with item 13 (a = 1.31).The most discriminating item was item 20 (a = 4.42), with very high discrimination.The other items also presented very high discrimination (a > 1.70).As for the response thresholds of the items in relation to the difficulty parameter (b1 -b4), the existence of items was observed that required, on average, a low amount of latent trait (M = -1.79;SD = .56),ranging from -2.82 (item 9) to -1.66 (item 6).In summary, all presented low levels of difficulty, as did the affective empathy items (See Table 2).
Finally, for research purposes seeking to provide a more concise measure yet without losing psychometric quality, it was decided to test the information parameters using the 10 BES items distributed equally between factors (i.e., BES-10).Specifically, the decision to select the items was based on the item discrimination index (parameter a), and the information score provided by the two item sets.Thus, for the affective empathy factor, the items with the highest levels of discrimination and the greatest capacity for measuring theta information (θ) in the range of -3 to +3 were selected, that is, items: 7, 17, 2, 8 and 5, respectively (Figure 1A).Based on the mentioned criteria, for the cognitive empathy factor, items 20, 16, 10, 12 and 6 were respectively selected, (Figure 1C).The information scores for both BES factors are shown in Figure 1.
As can be seen in Figure 1, the information tests reveal the accuracy of the measurements.Specifically, both versions of BES, the original structure (Figures 1A and 1C) and the reduced version (Figures 1B and  1D) are capable of capturing the largest amount of information, approximately, in the range of -4.01 to 3.20, being greatly informative measures.In other words, BES is best suited to measure empathy in people presenting a theta (θ) in the mentioned range.Further, in addition to the original version of the BES, it was demonstrated as possible to have a short version of the instrument, the BES-10, with appropriate individual parameters.As shown in Table 3, the 10-item version maintains, in addition to the information score, "very high" discrimination scores for the affective (Mean = 2.09, SD = .27;α = .82,and ω = .83)and cognitive empathy facets (Mean = 2.90, SD = 1.03; α = .83,and ω = .85).

Discussion
The present study aimed to provide evidence for validity based on the internal structure of the Basic Empathy Scale (BES; Jolliffe & Farrington, 2006), as well as investigating the parameters of its items using a Brazilian adult sample.Despite the importance of the empathy construct in the performance of interpersonal and social functions, and providing the emotional bases for promoting pro-social behavior (Riess, 2017), empirical studies are still poorly systematized in the Brazilian context, such that certain gaps still need to be explored (e.g., studies in different stages of the human life cycle and contexts; Azevedo et al., 2018).In this sense, our efforts initially focused on adapting and validating an up to date empathy measure widely used in several countries around the world.Initially, in view of the complexity of the procedures for adapting psychological measures to other contexts (Cassepp-Borges et al., 2010), we considered the important steps in the process.Translating the instrument from the source language (i.e., English) into  the target language (i.e., Portuguese), synthesis of the initially translated versions, evaluation of the synthesized version by expert judges, and finally, evaluation of the measure using a target audience sample (Sireci et al., 2006) were all performed.Before the empirical testing (with data collection and analysis), we aimed to provide an improved measure by investigating the qualifications of the instrument through consideration of its basic characteristics (e.g., factorial structure, item semantics, instructions, etc.).
As for the dimensionality of the BES-20, the analytical procedures demonstrated its structure in two factors, formed by the affective and cognitive dimensions of empathy.The structure converged theoretically and empirically with studies conducted for this purpose, both in the original study (Jolliffe & Farrington, 2006) as well as in other countries around the world (e.g., Albiero et al., 2009;D'Ambrosio, et al., 2009;Geng et al., 2012, etc.).We note that the Hull method was applied to investigate the BES factorial solution (Lorenzo-Seva et al., 2011).
As to internal consistency, from the polychoric correlation matrix, both of the empathy factors presented adequate McDonald's omega levels.In addition to Cronbach's alpha internal consistency indicator, we opted for McDonald's omega because it presents greater reliability for instruments, and avoids underestimating internal consistency (Sijtsma, 2009).In Cronbach's alpha, this occurs due to the basic assumption of tauequivalence, which calculates reliability assuming that all items have equal factor loads (Raykov, 1997).McDonald's omega estimator suggests an alternative index for psychological instrument reliability (Dunn, Baguley, Brunsden, 2014), to allow greater precision in analysis of psychological measures.
Using the IRT procedures, adequate discrimination indexes were initially observed for the BES-20, such that most of the items (in accordance with the classification proposed by Baker, 2001), presented discrimination magnitudes ranging from moderate to very high.Altogether, the items of the affective and cognitive empathy factors presented very high discriminations (Baker, 2001).Such evidence assure quality in measuring and differentiating people with similar magnitudes in the latent empathy trait.
As for the difficulty parameter, in view of Pasquali's guidelines (2007), taken together, all items of the BES-20 presented low levels of difficulty, that is, the items required low latent trait levels to be agreed to and selected.Such results have theoretical support, since empathy is both a socially desirable attribute, and opposes individual characteristics considered undesirable (e.g., lack of remorse with people, insensitivity, etc.; Wai & Tiliopoulos, 2012).
Finally, as explained, a reduced version of the measurement composed of 10 items, the BES-10, was tested and also presented adequate psychometric indices.Specifically, the test information curves were analyzed for both versions of the instrument; the procedure graphically represents the items' contribution to the information total (Castro et al., 2010) and allows selecting items that share more information, and which are consequently more central to the construct.In other words, the short version of the measure maintained both adequate discrimination scores and the ability to capture information concerning the latent trait of empathy.Concise instruments bring certain advantages (e.g., reducing the impact of inattention), and allow (in research contexts) greater measure accuracy in terms of validity and precision indicators (Rammstedt & Beierlein, 2014), as well as for inferences from empirical data.

Final considerations and future directions
In summary, this study presents a Portuguese version of the BES, an instrument to measure affective and cognitive empathy (Jolliffe & Farrington, 2006).Preliminary psychometric evidence corroborated the plausibility of the scale (i.e., internal structure and reliability) and its individual items (i.e., discrimination, difficulty, and information scores).The present study also aims to bridge operational gaps involving use of the construct in Brazil, since empathy is considered a fundamental variable both for the regulation of social interactions, as well as for social cohesion in general (Durlak et al., 2011).Once the initial step of providing an instrument with preliminary psychometric properties has been completed (Borsa et al., 2012), the measure will allow advancement of future studies in Brazil.Specifically, future research will be able to examine relationships between empathy and theoretically relevant psychological constructs; monitoring changes in empathy that may result from intervention programs (Zych et al., 2020), as well as helping in the construction of instruments for diagnostic purposes, making its use in clinical settings (as well as in research) possible (Pechorro et al., 2015).
Despite the above findings, corroborating evidence commonly found in the literature, the present study is not without limitations.First, there is the non-probabilistic character of the sample, or being composed only of young adults, which makes it impossible to generalize any current results to the general Brazilian population.Second, the empathy measure is a self-reporting instrument; it is not possible to control the effects of social desirability on the participants' response.
For future directions, new studies are suggested that take into account: other sample strata (e.g., children, adolescents, and adult populations of 30 years old or more), social desirability (e.g., relationships to social desirability measures, and/or construction of implicit measures of the construct), as well as further evidence of convergent and predictive validity in Brazil (e.g., pro-social and antisocial behaviors, personality, etc.).Finally, in view of the recent theoretical discussions around this construct, confirmatory models (e.g., structural equation modeling) would be useful to test both the two-factor factorial model now found in the present study, as well as the triadic conception of empathy, (a perspective involving functional and dysfunctional components of empathic processes and responses in both adults and adolescents, i.e., emotional contagion, emotional disconnection, and cognitive empathy) (Carré et al., 2013).

Figure 1 .
Figure 1.Test information curves: affective empathy, versions with 10 and 5 items, and cognitive empathy, versions with 9 and 5 items.

Table 1 .
Factorial structure of the Basic Empathy Scale(BES)

Table 2 .
Parameters of BES items (Gradual Response Model)

Table 3 .
Parameters of the BES-10 items(Gradual Response Model)