Cognitive/Affective Empathy,Pro-Bullying Beliefs,and Willingness to Intervene on Behalf of a Bullied Peer

Abstract

The purpose of this study was to determine whether pro-bullying attitudes mediate the relation between affective and cognitive empathy and a student’s willingness to intervene in support of a bullied peer. Participants were 764 early adolescents (372 boys, 392 girls) from the Illinois Study of Bullying and Sexual Violence (ISBSV). As predicted, pro-bullying attitudes successfully mediated the prospective relation between affective empathy and bystander intervention but failed to mediate the relation between cognitive empathy and bystander intervention. These results indicate that affective empathy may promote bystander willingness to intervene on behalf of a bullied peer by inhibiting pro-bullying attitudes. Intervention strategies designed to enhance affective empathy and challenge pro-bullying attitudes in bystanders may be of assistance in reducing bullying behavior.

Keywords

affective empathy cognitive empathy pro-bullying attitudes willingness to intervene on behalf of a bullied peer

School bullying involves not only bullies and their victims, but bystanders as well. Research indicates that between 80% and 90% of bystanders do not intervene when faced with a bullying situation (Craig & Pepler, 1998; Trach, Hymel, Waterhouse, & Neale, 2010). Some will join in or encourage the bullying behavior, a few will intervene, but most will adopt a passive stance in which they either watch or ignore the bullying. This can be harmful, however, because it often provides the bully with the “supportive” audience he or she desires. Common reasons given for not intervening include the following: “I didn’t want to become a target of the bullying myself,” “the victim probably deserved it,” and “it wasn’t any of my business” (Padget & Notar, 2013). By not intervening, the passive bystander plays an important role in maintaining the bullying behavior. It is therefore imperative that we understand what makes some youth willing to go against the grain and defend a victim of bullying. To this end, the current study examines two forms of empathy, cognitive and affective, and their effect on pro-bullying attitudes as a means of encouraging a bystander to intervene on behalf of a bullied peer.

Two theories have been advanced in an effort to explain bystander behavior in reaction to bullying and/or sexual harassment. The first of these theories is referred to as the situational model of bystander behavior (SMB: Latané & Darley, 1969). The SMB focuses on intrapersonal processes that lead the individual to notice a situation, interpret it as problematic, view themselves as responsible for taking action, know how to intervene, and take action. A number of personal characteristics have been found to correlate with willingness to intervene on behalf of a bullied peer, thereby providing support for the SMB model. Several of the more commonly mentioned characteristics are a felt moral obligation to intervene (Bennett, Banyard, & Garnhart, 2014), perceived peer pressure to intervene (Pozzoli & Gini, 2010), strong social self-efficacy to intervene (Cappadocia, Pepler, Cummings, & Craig, 2012), prior bullying victimization (Batanova, Espelage, & Rao, 2014), perspective-taking and empathy (Cappadocia et al., 2012), and female gender (Pozzoli & Gini, 2013). Several of these features reflect the subjective norms and behavioral control emphasized by a second theory with relevance to bystander behavior, the theory of planned behavior (TPB; Ajzen, 1991).

Derived from Ajzen and Fishbein’s (1980) theory of reasoned action, the TPB highlights the role of behavioral intentions and perceived behavioral control in determining a person’s voluntary behavior. According to TPB, behaviors are predicated on attitudes, intentions, perceived behavioral control, and subjective and social norms. Achieving a behavioral outcome, it is argued, depends on motivational (intentions), ability (perceived behavioral control), and cognitive (attitudes) factors, considered with a specific time frame and situational context. Proponents of TPB assume that people act on the basis of their belief that their behavior will yield a desirable outcome. This estimate is accompanied by an assessment of the risks and benefits of pursuing and attaining the outcome. With respect to observing bullying behavior, the TPB model considers the intentions, attitudes, and skills that support (e.g., pro-victim attitudes: Freis & Gurung, 2013) and impede (e.g., pro-bullying attitudes: Cappadocia et al., 2012) bystander willingness to intervene. The individual who believes bullying is wrong, perceives the social norms as antagonistic to bullying, and possesses the requisite attitudinal and behavioral skill set to stand up to bullying is the one most likely to intervene in a case of bullying.

Arguing that the SMB and TPB models are complementary rather than contradictory, Casey, Lindhorst, and Storer (2017) sought to integrate them, a synthesis they refer to as the situational-cognitive model of adolescent bystander behavior. Using both qualitative and quantitative data, Casey et al. (2017) uncovered preliminary support for their model. The current study also sought to integrate the SMB and TPB models but did so using a causal chaining approach in which components of the two models represented different links in the chain. The SMB component, an intrapersonal process that interfaces with the situational requirements of a bullying encounter, preceded the TPB component, an attitudinal process that interfaces with a person’s intentions to engage in a particular behavior. The TPB component, in turn, preceded the bystander’s willingness to intervene. Using standard research nomenclature, one component of the SMB, empathy, served as the independent variable, one component of TPB, pro-bullying attitudes, served as the mediating variable, and willingness to intervene served as the dependent variable in a three-wave causal mediation analysis.

Empathy, the first link in the causal chain model evaluated in the current investigation, may be too broad to serve as a meaningful predictor of bystander behavior (Schultz, Heilman, & Hart, 2014). Fortunately, it can be broken down into two neuroanatomically distinct processes, referred to as cognitive and affective empathy (Shamay-Tsoory, Aharon-Peretz, & Perry, 2009). Whereas cognitive empathy or perspective-taking involves a conscious attempt on the part of the individual to recognize and accurately label another person’s feelings, affective empathy is an automatic and largely unconscious emotional response to another person’s situation. A meta-analysis of the empathy-defending literature up through 2014 revealed effect sizes of comparable magnitude when defending behavior was correlated with cognitive (r = .14-.52) and affective (r = .12-.61) empathy (Nickerson, Aloe, & Werth, 2015). Two studies conducted after the Nickerson et al. (2015) meta-analysis, however, produced results showing that affective empathy may be a stronger correlate of willingness to intervene than cognitive empathy. Pozzoli, Gini, and Thornberg (2017), in fact, discovered that perspective-taking only correlated with bystander intervention when angry emotions served as a mediating variable. In the other study, Menolascino and Jenkins (2018) determined that affective but not cognitive empathy correlated with a greater likelihood of intervening in a bullying episode, although the effect was restricted to boys.

Studies conducted on the relation between pro-bullying attitudes and bystander willingness to intervene, like research on cognitive/affective empathy and defending behaviors, have produced mixed results. Using a social networking approach and longitudinal data, Espelage, Green, and Polanin (2012) noted that neither empathy nor pro-bullying attitudes were capable of predicting bystander willingness to intervene. Cappadocia et al. (2012), by contrast, discovered that lower levels of pro-bullying belief correlated with a greater willingness on the part of bystanders to intervene on behalf of a bullied peer. Pro-victim attitudes, on the other hand, failed to correlate with bystander intervention. As in the Menolascino and Jenkins (2018) study on affective versus cognitive empathy, the connection between pro-bullying attitudes and willingness to intervene in the Cappadocia et al. (2012) study was restricted to boys. Girls tend to report higher levels of empathy and defending behavior than boys (Pozzoli & Gini, 2010), and while there was no evidence of gender moderation in the Nickerson et al. (2015) meta-analysis, the significant affective empathy and pro-bullying attitudes effects from the Cappadocia et al. (2012) and Menolascino and Jenkins (2018) studies were restricted to males. Before probing for mediation, then, it is vital that gender moderation be evaluated.

Present Study

The purpose of this study was to explore the nature of the relation between cognitive and affective empathy, pro-bullying attitudes, and bystander intervention with three nonoverlapping waves of longitudinal data. Use of longitudinal or prospective data is one of the principal contributions the current investigation makes to the literature, given that the three previously reviewed studies in which affective empathy, low pro-bullying attitudes, and willingness to intervene were found to covary (i.e., Cappadocia et al., 2012; Menolascino & Jenkins, 2018; Pozzoli et al., 2017) were based exclusively on cross-sectional data. Variables like age and sex (Pozzoli & Gini, 2013), school connectedness (Ahmed, 2008), bullying victimization (Batanova et al., 2014), social support (Evans & Smokowski, 2015), and exposure to community and family violence (Rivers, 2013) have all been found to correlate with willingness to intervene and thus served as control variables in this study. Although other variables like caring, self-esteem, and parental knowledge have not been investigated in research on bystander intervention, they were included as control variables in the current study because they reflect personal (self-esteem), interpersonal (caring), and external (parental knowledge) forms of control that could clearly affect a person’s willingness to intervene in a bullying situation.

Four hypotheses were tested in this study:

Hypothesis 1: The five-item empathy scale used in this study should partition into two factors (cognitive empathy and affective empathy) and this two-factor model should demonstrate significantly better fit than a one-factor model in which all five items load onto a single factor.

Hypothesis 2: Sex moderation of relations involving empathy and bystander intervention should be evident such that the effects described in H3 and H4 should only be observed in male participants.

Hypothesis 3: Affective empathy should correlate significantly better with concurrently and prospectively measured bystander intervention than cognitive empathy.

Hypothesis 4: An affective-initiated pathway (affective empathy → low pro-bullying attitudes → willingness to intervene) should achieve significance but a cognition-initiated pathway (cognitive empathy → low pro-bullying attitudes → willingness to intervene) should not.

Method

Participants

Participants for this study were 764 early adolescents (age in years: M = 12.29, SD = 0.83, range = 10-15) from the Illinois Study of Bullying and Sexual Violence (ISBSV; Espelage, Low, Anderson, & De La Rue, 2014). These individuals were selected for the current study based on the fact that they had complete data on at least two of the first three waves of the ISBSV. There were 372 boys and 392 girls in a sample that was 50.5% African American, 31.5% White, 3.9% Hispanic, 1.2% Asian, and 12.8% Mixed/Other. The ISBSV was originally approved by the Institutional Review Board (IRB) at the University of Illinois at Urbana/Champaign and its use in the current secondary analysis was approved by the Kutztown University IRB.

Measures

The five-item empathy scale from the Teen Conflict Survey (Bosworth & Espelage, 1995) was partitioned into two subscales based on definitions of cognitive and affective empathy provided by Blair (2005): Cognitive empathy is defined as the ability to understand what a person is thinking or feeling without getting emotionally involved oneself, and affective empathy is defined as an emotional reaction by an observer to the affective state of another. Based on these definitions, the five empathy items were rationally sorted into a three-item Cognitive Empathy subscale (“I can listen to others”; “Kids I don’t like can have good ideas”; “I trust people who are not my friends”) and a two-item Affective Empathy subscale (“I get upset when my friends are sad”; “I am sensitive to other people’s feelings, even if they are not my friends”). Each item was rated by the participant on a 5-point scale (0 = never, 1 = seldom, 2 = sometimes, 3 = often, 4 = always), and a mean score per item calculated. The resulting cognitive empathy and affective empathy scores then served as independent variables in the present study. The mean inter-item correlation for the three cognitive empathy items was .36 (α = .63 in the 994 members of ISBSV with complete data at Wave 1) and the two affective empathy items correlated .44, indicating an acceptable level of internal consistency for both subscales.

The mediator variable in this study was pro-bullying attitudes or beliefs, as measured by the Illinois Positive Attitudes Toward Bullying Scale (POSATT: Espelage & Asidao, 2001). The POSATT consists of three items (“A little teasing doesn’t hurt anyone”; “I don’t care what mean things kids say, as long as it’s not about me”; “If other students are being teased too much, it’s not my problem”). Each item on the POSATT is rated on a 4-point Likert-type scale (1 = strongly disagree, 2 = disagree, 3 = agree, 4 = strongly agree). The three-item scores were then averaged to produce a mean score per item. This three-item scale achieved a Cronbach alpha coefficient of .68 at Wave 2 of the ISBSV (n = 848), but its mean inter-item correlation in the present sample of participants (r = .41) suggests good to excellent internal consistency (Clark & Watson, 1995).

Willingness to intervene on behalf of a bullied victim served as the dependent variable in this study. The University of Illinois Willingness to Intervene scale (Espelage et al., 2012) was used for this purpose. The Willingness to Intervene scale consists of five items (“If a kid is being teased, I will stick up for him or her”; “I will tell an adult if a kid is being teased a lot”; “I will tell another student to stop teasing someone”; “I will tell someone to stop teasing a friend”; “I will tell an adult that a good friend is being teased”), with each item rated on a 4-point Likert-type scale (1 = strongly disagree, 2 = disagree, 3 = agree, 4 = strongly agree). The individual item scores were then averaged to produce a mean score per item. The Willingness to Intervene scale displayed good internal consistency in the current sample of participants (α = .88).

There were three sets of control variables included in the current investigation: demographic measures, person variables, and situational indicators. The three demographic control measures were age (in years), sex (1 = male, 2 = female), and race (1 = White, 2 = non-White). The four person-control variables were a four-item school sense of belonging scale (“proud of belonging to my school”; “teachers respect me”; α = .65), a four-item bullying victimization scale (“other students picked on me”; “I got hit and punched”; α = .79), a four-item caring scale (“tell others I care about them”; “helped other kids”; α = .89), and a four-item self-esteem scale (“I am the kind of person I want to be”; “I can do things as well as others”; α = .84). There were also five situational control indicators: a nine-item measure of social support from family, friends, and adults (“number of adults, family, or friends I can talk to . . . who give good advice . . . who give practical help”; α = .85), an eight-item parental monitoring scale (“How often do your parents ask about homework . . . know if you come home on time . . . know about alcohol use”; α = .87), a three-item family violence scale (“How often is/are there yelling, quarreling, or arguing in the household . . . physical fights in the household”; α = .78), a two-item parental violence scale (“observe parents or guardians being hit by spouse”; “you have injuries from parents”; r_ϕ = .25), and a five-item neighborhood violence scale (“How often in your neighborhood do you hear guns being shot . . . see someone arrested . . . see someone beaten up”; α = .90).

The use of longitudinal data is important in establishing the temporal order of one’s data, but it is also important to establish the temporal direction of one’s variables. Temporal direction was established in the current study by accounting for preexisting differences in the two outcome measures (Cole & Maxwell, 2003). Therefore, prior or precursor measures of POSATT (the mediator variable) and Willingness to Intervene (the dependent variable) were incorporated into the design. The Wave 1 POSATT score, for instance, was included as a predictor in the regression equation predicting Wave 2 POSATT, and the Wave 1 Willingness to Intervene score was included as a predictor in the regression equation predicting Wave 3 Willingness to Intervene. Adding precursor measures of each predicted variable to the analysis transformed the research task from predicting static measures of Wave 2 POSATT and Wave 3 Willingness to Intervene, to predicting a change in POSATT from Wave 1 to Wave 2 and a change in Willingness to Intervene from Wave 1 to Wave 3.

Research Design

A three-wave longitudinal fixed panel design was employed in the present investigation. The first wave included two independent variables (cognitive empathy and affective empathy) and 14 control and precursor measures, the second wave contained one mediator variable (POSATT), and the third wave incorporated one dependent variable (Willingness to Intervene). Two pathways were tested as part of this study. One pathway ran from cognitive empathy to POSATT to Willingness to Intervene and the other ran from affective empathy to POSATT to Willingness to Intervene. Thus, each pathway had a different a path (from independent variable to mediator), but they shared the same b path (from mediator to dependent variable). There was a 6-month gap and no overlap between waves, thus qualifying this research design as prospective in nature.

Analytic Plan

A pair of confirmatory factor analyses (CFAs), each using a robust weighted least squares (WLSMV) estimator, were performed in an effort to determine whether the five empathy items conformed better to a one-factor model or to a two-factor (cognitive empathy, affective empathy) model. Absolute fit was assessed with the comparative fit index (CFI), Tucker–Lewis index (TLI), and root mean square error of approximation (RMSEA). The following rules of thumb (Hu & Bentler, 1999) were used to evaluate the three indices: CFI/TLI > .95 (good fit), .90 to .95 (borderline fit), and <.90 (poor fit); RMSEA <.06 (good fit), .06 to .08 (fair fit), .08 to .10 (borderline fit), and >.10 (poor fit). Relative fit between the one- and two-factor models was assessed using the DIFFTEST.

A path analysis was performed with a maximum likelihood (ML) estimator. The overall significance of the indirect effect was evaluated using confidence intervals from a bias-corrected bootstrap (5,000 repetitions), which research indicates does a better job of accounting for non-normality in the indirect effect and dependent variables than the Sobel (1982) normal theory test (Hayes, 2013; Preacher, 2015; Rucker, Preacher, Tormala, & Petty, 2011). Nearly half the Wave 1 data were missing and so a second analysis was performed using auxiliary variables and an ML with robust standard errors (MLR) estimator. Because bootstrapping cannot be applied in conjunction with MLR and auxiliary variables, the Monte Carlo method for assessing mediation (MCMAM) was used to test the significance of the two indirect effects (Preacher & Selig, 2012). MCMAM was performed with 20,000 repetitions, and significance was determined by a 95% confidence interval that did not include zero. All analyses were performed with MPlus 8.1 (Muthén & Muthén, 1998-2017) or SPSS Version 25 (IBM Corporation, 2017).

Sensitivity testing was performed to rule out omitted variable bias and endogenous selection bias as alternate explanations of the current results. The former was accomplished with the aid of Kenny’s (2013) “failsafe ef” procedure: (r_my.x) × (sd_m.x) × (sd_y.x) / (sd_m) × (sd_y). The coefficient produced by the “failsafe ef” procedure indicates how well an unobserved covariate confounder would need to correlate with the mediator and dependent variable, after controlling for the mediator and independent variable in the case of the latter, to eliminate the b path coefficient of the significant indirect effect. The second sensitivity test was designed to rule out endogenous selection bias, also known as a collider effect, as an explanation for the present findings. Because endogenous selection bias can sometimes occur when researchers condition on the precursor to an outcome (Elwert & Winship, 2014), it was tested by removing all precursor measures from the regression equations and re-performing the analyses.

Missing Data

A sizable minority of participants had complete data on all 18 variables (43.3%). Another 12.8% were missing data on one variable, 1.3% were missing data on two to four variables, and 43.7% were missing data on 13 or 14 variables. Nearly all participants missing data on 13 or 14 variables were missing data on all Wave 1 variables but had complete data on Waves 2 and 3 variables. This indicates that missing data were confined primarily to Wave 1 (≈42%, except for demographics), with less than 10% missing data for the Wave 2 mediator (4.7% missing data) and Wave 3 outcome (8.4% missing data) measures. Missing data in this study were handled with full information maximum likelihood (FIML). The FIML procedure estimates model parameters and standard errors for the entire sample from analyses performed on non-missing data. Research indicates that FIML produces significantly less biased results than traditional missing data procedures like simple imputation and listwise deletion (Allison, 2012; Peyre, Leplége, & Coste, 2011).

In addition to being less biased than traditional missing data approaches, FIML is also reasonably robust to violations of its basic assumptions (Collins, Schafer, & Kam, 2001; Young & Johnson, 2013). To further enhance the precision of FIML (Collins et al., 2001), 13 auxiliary variables (Wave 2 cognitive empathy, Wave 2 affective empathy, Wave 2 school belonging, Wave 2 bullying victimization, Wave 2 caring, Wave 2 self-esteem, Wave 2 social support, Wave 2 parental monitoring, Wave 2 family violence, Wave 2 parental violence, Wave 2 neighborhood violence, Wave 3 POSATT, and Wave 2 Willingness to Intervene) were added to the regression equations. Not only did the auxiliary variables from Wave 2 have less than 5% missing data each, they also correlated .49 to .63 with their Wave 1 analogues. It is worth noting that while auxiliary variables are used by FIML to estimate parameters and standard errors, they are not included in the analysis itself.

Results

Preliminary Analyses

Descriptive statistics and bivariate correlations for the 18 independent, mediator, dependent, control, and precursor measures included in this study are outlined in Table 1. Nearly half the correlations were statistically significant using a Bonferroni-corrected alpha level. There was no evidence of multicollinearity when predictors from the two regression equations of the path model were subjected to collinearity diagnostics: tolerance = .62 to .89, variance inflation factor (VIF) = 1.12 to 1.61.

Table 1.

Descriptive Statistics and Correlations for the 18 Independent, Dependent, Mediator, and Control Variables Used in This Study.

Variable	n	M	SD	Range	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
1. Age	592	12.29	0.83	10-15	−.08	.10	−.15	.04	−.08	.02	−.03	−.08	.09	−.03	.23^†	−.12	−.06	.21^†	.09	−.20^†	−.12
2. Sex	764	1.51	—	1-2		.01	.04	−.04	.23^†	.17	.09	.11	.10	.01	−.05	.07	.30^†	−.15	−.08	.25^†	.25^†
3. Race	764	1.68	—	1-2			−.17^†	−.07	−.17^†	.05	−.09	−.11	−.01	.02	.34^†	−.35^†	−.24^†	.22^†	.20^†	−.13	−.19^†
4. School belonging	438	2.91	0.63	1-4				−.10	.20^†	.22^†	.41^†	.29^†	−.14	−.03	−.30^†	.21^†	.26^†	−.20^†	−.09	.30^†	.28^†
5. Bullying victimization	435	0.59	0.82	0-4					.15	−.13	−.04	−.06	.26^†	.20^†	.06	.06	.12	−.06	−.09	.08	.21^†
6. Caring	436	2.07	1.20	0-4						.25^†	.34^†	.25^†	.05	.02	−.10	.27^†	.35^†	−.34^†	−.25^†	.43^†	.40^†
7. Self-esteem	435	2.71	1.10	0-4							.34^†	.41^†	−.09	−.10	−.04	.21^†	.19^†	−.02	−.08	.24^†	.18^†
8. Social support	435	1.37	0.42	0-2								.38^†	−.10	−.12	−.10	.31^†	.31^†	−.20^†	−.12	.33^†	.27^†
9. Parental monitoring	437	2.35	0.74	0-3									−.18^†	−.15	−.22^†	.25^†	.23^†	−.14	−.13	.38^†	.32^†
10. Family violence	432	1.32	1.26	0-5										.33^†	.31^†	.00	.05	.11	.05	.01	.01
11. Parental violence	431	0.17	0.30	0-1											.16	.01	.07	.02	−.02	.04	.13
12. Neighborhood violence	436	1.22	0.95	0-3												−.24^†	−.20^†	.21^†	.13	−.10	−.13
13. Cognitive empathy	436	1.73	0.83	0-4													.42^†	−.27^†	−.25^†	.24^†	.24^†
14. Affective empathy	436	1.70	1.18	0-4														−.30^†	−.29^†	.38^†	.43^†
15. POSATT-1	438	2.34	0.72	1-4															.48^†	−.36^†	−.32^†
16. POSATT-2	728	2.18	0.77	1-4																−.29^†	−.30^†
17. Willing to Intervene-1	438	2.90	0.64	1-4																	.56^†
18. Willing to Intervene-3	700	2.87	0.73	1-4

Note. Age = age in years; sex = male (1) versus female (2); race = White (1) versus non-White (2); school belonging = school sense of belonging measured at Wave 1; bullying victimization = bullying victimization measured at Wave 1; caring = caring and helping attitude toward other children measured at Wave 1; self-esteem = positive self-esteem measured at Wave 1; social support = social support from family, friends, and adults measured at Wave 1; parental monitoring = parental monitoring measured at Wave 1; family violence = family violence measured at Wave 1; parental violence = parental violence measured at Wave 1; neighborhood violence = neighborhood violence measured at Wave 1; cognitive empathy = cognitive empathy measured at Wave 1; affective empathy = affective empathy measured at Wave 1; POSATT-1 = pro-bullying attitudes measured at Wave 1; POSATT-2 = pro-bullying attitudes measured at Wave 2; willing to intervene-1 = willingness to intervene on behalf of bullying victim measured at Wave 1; willing to intervene-3 = willingness to intervene on behalf of bullying victim measured at Wave 3; n = number of non-missing cases; range = range of scores in current sample.

†

p < .00033 (Bonferroni-corrected alpha: .05/153 correlations).

First Hypothesis

The first hypothesis held that the five-item empathy scale used in this study would partition into two factors. The analysis was restricted to participants with complete data on all five Wave 1 empathy items (n = 432). The two-factor model achieved good absolute fit (CFI = .99, TLI = .97, RMSEA = .055), whereas the one-factor model showed borderline to good absolute fit (CFI = .96, TLI = .92, RMSEA = .088). Upon direct comparison, the two-factor model displayed a significantly better fit than the one-factor model, DIFFTEST χ²(1) = 9.80, p < .01. In addition, the two latent factors correlated significantly with one another (r = .76). These results are consistent with the first hypothesis and suggest that we were justified in portioning the empathy scale into two subscales, cognitive and affective empathy.

Second Hypothesis

The second hypothesis proposed that sex would moderate the empathy‒bystander intervention relationship, such that an effect would only register in boys. Interaction terms were computed between sex and cognitive empathy and between sex and affective empathy, using centered empathy scores. The interaction terms were then included as predictors in both regression equations (i.e., the one predicting Wave 2 POSATT and the one predicting Wave 3 Willingness to Intervene). None of the interaction terms proved significant (p > .10), thereby indicating that sex did not moderate the empathy‒bystander intervention nexus and obviating the need for separate analyses by sex.

Third Hypothesis

The third hypothesis maintained that affective empathy would correlate significantly better with bystander intervention than would cognitive empathy. This hypothesis was tested using correlations depicted in Table 1. Consistent with the third hypothesis, affective empathy eclipsed cognitive empathy as both a correlate, r = .38 versus .24, Steiger’s (1980) Z = 3.00, p < .01, and predictor, r = .43 versus .24, Steiger’s Z = 3.70, p < .001, of bystander intervention.

Fourth Hypothesis

The fourth hypothesis predicted that a pathway running from affective empathy to low pro-bullying attitudes to bystander intervention would prove significant but that a pathway running from cognitive empathy to low pro-bullying attitudes to bystander intervention would not. A path analysis revealed significant a and b-path coefficients for the affective empathy-initiated pathway, but only the b-path coefficient was significant for the cognitive empathy–initiated pathway (see Table 2 and Figure 1). The bias-corrected bootstrapped confidence intervals produced by the two indirect effects are summarized in Table 3 and indicate a significant affective empathy–initiated pathway and nonsignificant cognitive empathy–initiated pathway.

Table 2.

Results of a Maximum Likelihood Path Analysis Using Longitudinal Data (N = 764).

Predictor	b (95% CI)	β	z	p
POSATT-2 (outcome)
Cognitive empathy	−0.060 [–0.154, 0.032]	−0.065	−1.26	.209
Affective empathy	−0.078 [–0.145, –0.011]	−0.120	−2.27	.023
Age	−0.010 [–0.086, 0.065]	−0.010	−0.25	.800
Sex	0.035 [–0.092, 0.158]	0.023	0.55	.584
Race	0.096 [–0.037, 0.231]	0.058	1.40	.161
School belonging	0.086 [–0.036, 0.210]	0.071	1.37	.172
Bullying victimization	−0.034 [–0.119, 0.056]	−0.036	−0.76	.444
Caring	−0.029 [–0.102, 0.040]	−0.045	−0.82	.414
Self-esteem	−0.031 [–0.104, 0.039]	−0.045	−0.84	.401
Social support	0.069 [–0.118, 0.253]	0.037	0.73	.465
Parental monitoring	−0.024 [–0.143, 0.096]	−0.023	−0.39	.697
Family violence	0.019 [–0.048, 0.083]	0.032	0.58	.562
Parental violence	−0.059 [–0.301, 0.192]	−0.023	−0.47	.640
Neighborhood violence	−0.010 [–0.100, 0.73]	−0.012	−0.22	.822
POSATT-1	0.438 [0.328, 0.540]	0.409	8.13	<.001
Willing to Intervene-3 (outcome)
POSATT-2	−0.109 [–0.190, –0.028]	−0.116	−2.67	.008
Age	−0.012 [–0.077, 0.059]	−0.014	−0.35	.730
Sex	0.120 [0.011, 0.240]	0.082	2.08	.038
Race	−0.090 [–0.203, 0.017]	−0.058	−1.61	.107
School belonging	0.094 [–0.042, 0.231]	0.082	1.33	.182
Bullying victimization	0.126 [0.063, 0.194]	0.141	3.79	<.001
Caring	0.056 [–0.009, 0.120]	0.092	1.69	.091
Self-esteem	0.005 [–0.064, 0.073]	0.007	0.13	.893
Social support	0.022 [–0.176, 0.213]	0.012	0.22	.829
Parental monitoring	0.089 [–0.022, 0.203]	0.091	1.55	.121
Family violence	−0.033 [–0.093, 0.028]	−0.057	−1.08	.278
Parental violence	0.293 [0.092, 0.478]	0.120	2.96	.003
Neighborhood violence	0.000 [–0.080, 0.086]	0.000	0.01	.994
Cognitive empathy	−0.036 [–0.124, 0.060]	−0.041	−0.77	.443
Affective empathy	0.089 [0.028, 0.148]	0.145	2.89	.004
Willing to Intervene-1	0.366 [0.227, 0.495]	0.324	5.33	<.001
Empathy-C with Empathy-A	0.412 [0.309, 0.514]	0.419	7.87	<.001

Note. POSATT-2 (outcome) = regression equation with pro-bullying attitudes at Wave 2 serving as the outcome measure; willing to intervene-3 (outcome) = regression equation with willingness to intervene on behalf of a bullying victim at Wave 3 serving as the outcome measure; cognitive empathy = cognitive empathy measured at Wave 1; affective empathy = affective empathy measured at Wave 1; age = age in years; sex = male (1) versus female (2); race = White (1) versus non-White (2); school belonging = school sense of belonging measured at Wave 1; bullying victimization = bullying victimization measured at Wave 1; caring = caring and helping attitude toward other children measured at Wave 1; self-esteem = positive self-esteem measured at Wave 1; social support = social support from family, friends, and adults measured at Wave 1; parental monitoring = parental monitoring measured at Wave 1; family violence = family violence measured at Wave 1; parental violence = parental violence measured at Wave 1; neighborhood violence = neighborhood violence measured at Wave 1; POSATT-1 = pro-bullying attitudes measured at Wave 1; POSATT-2 = pro-bullying attitudes measured at Wave 2; willing to intervene-1 = willingness to intervene on behalf of bullying victim measured at Wave 1; Empathy-C with Empathy-A = covariance between cognitive empathy at Wave 1 and affective empathy at Wave 1; b (95% CI) = unstandardized coefficient and the lower and upper limits of the 95% confidence interval for the unstandardized coefficient; β = standardized coefficient; z = Wald Z test; p = significance level of the Wald Z test.

Figure 1.

Maximum likelihood path analysis of cognitive and affective empathy as predictors of bystander intervention with mediation by pro-bullying attitudes (N = 764).

Table 3.

Total, Direct, and Indirect Effects for Pathways Running From Cognitive and Affective Empathy to Bystander Intervention with Mediation by Pro-Bullying Attitudes (N = 764).

Pathways	BCBCI
Pathways	Estimate	Lower	Upper
Cognitive empathy → bystander intervention
Total effect	−0.029	−0.117	0.066
Direct effect	−0.036	−0.124	0.060
Total indirect effect	0.007	−0.002	0.023
Specific indirect effect
Cognitive empathy-1 → attitude-2 → intervene-3	0.007	−0.002	0.023
Affective empathy → bystander intervention
Total effect	0.098	0.036	0.158
Direct effect	0.089	0.028	0.148
Total indirect effect	0.009	0.001	0.022
Specific indirect effect
Affective empathy-1 → attitude-2 → intervene-3	0.009	0.001	0.022

Note. Cognitive empathy-1 = cognitive empathy at Wave 1; affective empathy-1 = affective empathy at Wave 1; attitude-2 = pro-bullying attitudes at Wave 2; intervene-3 = willingness to intervene on behalf of a bullying victim at Wave 3; BCBCI = bias-corrected bootstrapped 95% confidence interval (b = 5,000); estimate = unstandardized point estimate; lower = lower boundary of the 95% confidence interval; upper = upper boundary of the 95% confidence interval.

Sensitivity Analyses

According to the results of the “failsafe ef” procedure, an unobserved covariate confounder would need to correlate –.21 with POSATT-2 and –.21 with Willingness to Intervene-3, controlling for Affective Empathy-1 and POSATT-2 in the case of the latter, to completely eliminate the significant b path of the Affective Empathy-1 → POSATT-2 → Willingness to Intervene-3 pathway. These results indicate that the current results were moderately robust to the effects of omitted variable bias.

When the two precursor measures were removed from the regression equations and the path analysis recalculated, all path coefficients increased in size, to the point where even the a path of the cognitive empathy–initiated pathways was significant. These results should only be used to assess for endogenous selection bias, however, because the analysis suffers from model misspecification of the temporal direction type. In its capacity as a check on endogenous selection bias, this particular sensitivity test indicates that the current results probably cannot be attributed to a collider effect.

Supplemental Analysis

A supplemental path analysis was performed using an MLR estimator and 13 auxiliary variables designed to increase the precision of FIML. The a path of the affective empathy–initiated pathway was significant (p < .05), the a path of the cognitive empathy–initiated pathway approached significance (p = .052), and the b path that the two pathways shared also achieved significance (p < .01). Although the cognitive empathy–initiated path performed better in this analysis than it did in the main analysis, only the affective empathy–initiated pathway achieved a significant total indirect effect according to the MCMAM analysis (95% confidence intervals of 0.00073, 0.01906 vs. –0.00018, 0.02504).

Discussion

It has been argued that one of the best lines of defense against bullying is bystander intervention (Craig & Pepler, 1998; Padget & Notar, 2013). This underscores the value of appreciating the nature of bystander behavior in a bullying context and raises questions about the mechanisms responsible for this behavior. In the first hypothesis tested in this study, it was predicted that a five-item empathy scale would partition into two subscales (Cognitive Empathy and Affective Empathy). This hypothesis received support when one- and two-factor models of the five-item empathy scale were compared. The second hypothesis presumed that sex would moderate the affective empathy‒attitude–bystander intervention relationship. This hypothesis failed to gain support; an outcome inconsistent with some studies (Cappadocia et al., 2012; Menolascino & Jenkins, 2018) but consistent with the results of a meta-analysis by Dickerson et al. (2015). The third hypothesis held that affective empathy would correlate significantly better with willingness to defend a bullying victim than cognitive empathy. Findings consistent with this hypothesis, and with Menolascino and Jenkins (2018) and Pozzoli et al. (2017), were obtained when concurrent and prospective affective empathy‒willingness to intervene and cognitive empathy‒willingness to intervene correlations were compared. The fourth hypothesis, which predicted that the affective empathy–initiated pathway would be mediated by low pro-bullying beliefs but that the cognitive empathy–initiated pathway would not, also received support.

Limitations

There are several noteworthy limitations to this study. Foremost among these is the fact that the independent and mediator variables were assessed with scales composed of just two to three items each. Although the internal consistency of these three measures (cognitive empathy, affective empathy, pro-bullying attitudes) was good when assessed using the mean inter-item correlation or two-item correlation (Clark & Watson, 1995), short scales can present problems if participants misinterpret or skip an item. With longer scales, misreading or skipping an item will usually have little impact on the overall results. On a scale with just two or three items, skipping or misinterpreting an item can have major consequences. It could also be argued that two or three items, no matter how relevant, do not provide adequate coverage of as complex a concept as affective or cognitive empathy. In the current study, for instance, affective empathy was assessed with items that asked respondents if they were upset when friends were sad and if they were sensitive to other people’s feelings. Whereas these items clearly reflect affective empathy, they do not cover all of the emotions (fear, anger, and happiness) and contexts (other people’s situations or mental states: Jolliffe & Farrington, 2006) encompassed by affective empathy.

The fact that the affective empathy scale was composed of just two items also brings into question the results of the two-factor CFA. Our intention in performing the CFA was to illustrate how the three items identified as indicators of cognitive empathy and the two items identified as indicators of affective empathy would load on separate factors rather than on the same factor. Although the CFA results supported the notion that items on the five-item Teen Conflict Survey empathy scale loaded onto separate latent factors (i.e., cognitive empathy and affective empathy), a factor with fewer than three indicators is unstable and of questionable validity. Yet, if the latent factors in a two-factor CFA correlate well, as they did in the current study, and the error between indicators is uncorrelated, as it was in the current investigation, then a two-indicator factor may be interpretable, even though such a model is more susceptible to empirical under-identification than a model with three or more indicators (Brown, 2015).

A further limitation of this study is that all of the variables were assessed with measures completed by participants. Reliance on a single data source, in this case, participant self-report, can lead to a state of mono-operational bias and inflated path coefficients secondary to shared method variance (Shadish, Cook, & Campbell, 2002). This could be remedied in future research by using multiple data sources. There are two additional limitations that should be taken into account when interpreting the results of this study. First, participants came from a single jurisdiction and over two thirds had minority status. This raises questions about the representativeness of the sample and the generalizability of the results. Second, there were almost as many participants missing data on all Wave 1 variables (42%) as there were participants with complete data on all variables (43%). Thus, while FIML and auxiliary variables may have reduced the problems associated with missing data (i.e., representativeness, power, and bias), they did not eliminate them entirely.

Implications

A theoretical implication of this study is that emotional (affective empathy) and cognitive (low pro-bullying beliefs) factors may both need to be present for maximum student willingness to defend others against bullying. Cognitive empathy coupled with low pro-bullying beliefs, for instance, failed to increase a student’s willingness to intervene, whereas a combination of emotional and cognitive factors (affective empathy paired with low pro-bulling beliefs) distinguished between those who stated they would and those who stated they would not intervene. These longitudinal results are consistent with prior cross-sectional studies (Menolascino & Jenkins, 2018; Pozzoli et al., 2017) in showing that the affective features of empathy and cognitive intentions like low pro-bullying attitudes may, in amalgamation, motivate students to intervene. In an earlier study, Barhight, Hubbard, and Hyde (2013) discovered that children expressing an emotional reaction to a series of videos depicting bullying behavior were more likely to be rated by their peers as willing to intervene in an episode of bullying than children who displayed no emotional reaction to the videos. Another study determined that cognitive empathy had a short-term effect on bystander behavior in support of cyberbullying but that this effect tended to dissipate over time (Barlińska, Szuster, & Winiewski, 2015). This would seem to suggest that children are most likely to intervene when the relevant emotional (affective empathy) and cognitive (low pro-bullying beliefs) elements are present.

The fact that a reduction in pro-bullying attitudes may mediate the relationship between affective empathy and willingness on the part of bystanders to intervene in a bullying situation suggests that affective empathy neutralizes the pro-bullying beliefs that normally discourage a child from intervening in a case of bullying. In addition to the theoretical implications of these results, there are also practical implications in terms of bullying prevention. The value of identifying a causal chain that begins with affective empathy, continues with reduced pro-bullying attitudes, and ends with an increase in one’s willingness to intervene is that it provides tangible targets for intervention. Programs like the KiVa program in Finland, that target the development of prosocial behaviors in student bystanders, have produced impressive initial results (Juvonen, Schacter, Sainio, & Salmivalli, 2016; Polanin, Espelage, & Pigott, 2012). Perhaps these promising results can be reinforced, strengthened, and enhanced by efforts to increase affective empathy and decrease pro-bullying attitudes. There is still a great deal more research that needs to be done in this area before definitive conclusions can be reached, but findings from the current investigation indicate that combining cognitive and emotional influences, like low pro-bullying attitudes and high affective empathy, is one way of achieving relevance, predictability, and change in the field of youth bullying.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Research for the current study was supported by the Centers for Disease Control and Prevention (#1U01/CE001677) to D.L.E. (principal investigator).

ORCID iD

Glenn D. Walters

Author Biographies

Glenn D. Walters is a PhD and a professor in the Department of Criminal Justice at Kutztown University where he teaches classes in corrections, criminology, research methods, and substance misuse. His principal research interests include mediation analysis and the relationship between bullying and delinquency.

Dorothy L. Espelage, PhD, is a professor in the Department of Psychology at the University of Florida. She has authored over 140 peer reviewed journal articles, five edited books, and 30 chapters on bullying, homophobic teasing, sexual harassment, dating violence, and gang violence.

References

Ahmed

(2008). “Stop it, that’s enough”: Bystander intervention and its relationship to school connectedness and shame management. Vulnerable Children and Youth Studies, 3, 203-213. doi:10.1080/17450120802002548

Ajzen

(1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179-211. doi:10.1016/0749-5978(91)90020-T

Ajzen

Fishbein

(1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice Hall.

Allison

P. D.

(2012). Handling missing data by maximum likelihood (SAS Global Forum 2012). Cary, NC: SAS Institute.

Barhight

L. R.

Hubbard

J. A.

Hyde

C. T.

(2013). Children’s physiological and emotional reactions to witnessing bullying predict bystander intervention. Child Development, 84, 375-390. doi:10.1111/j.1467-8624.2012.01839.x

Barlińska

Szuster

Winiewski

(2015). The role of short- and long-term cognitive empathy activation in preventing cyberbystander reinforcing cyberbullying behavior. Cyberpsychology, Behavior, and Social Networking, 18, 241-244.

Batanova

Espelage

D. L.

Rao

M. A.

(2014). Early adolescents’ willingness to intervene: What roles do attributions, affect, coping, and self-reported victimization play? Journal of School Psychology, 52, 279-293. doi:10.1016/j.jsp.2014.02.001

Bennett

Banyard

V. L.

Garnhart

(2014). To act or not to act, that is the question? Barriers and facilitators of bystander intervention. Journal of Interpersonal Violence, 29, 476-496. doi:10.1177/0886260513505210

Blair

R. J. R.

(2005). Responding to the emotions of others: Dissociating forms of empathy through the study of typical and psychiatric populations. Consciousness and Cognition, 14, 98-718.

10.

Bosworth

Espelage

(1995). Teen Conflict Survey. Bloomington: Center for Adolescent Studies, Indiana University.

11.

Brown

T. A.

(2015). Confirmatory factor analysis for applied research (2nd ed.). New York, NY: Guilford.

12.

Cappadocia

M. C.

Pepler

Cummings

J. G.

Craig

(2012). Individual motivations and characteristics associated with bystander intervention during bullying episodes among children and youth. Canadian Journal of School Psychology, 27, 201-216. doi:10.1177/0829573512450567

13.

Casey

E. A.

Lindhorst

Storer

H. L.

(2017). The situational-cognitive model of adolescent bystander behavior: Modeling bystander decision-making in the context of bullying and teen dating violence. Psychology of Violence, 7, 33-44. doi:10.1037/vio0000033

14.

Clark

L. A.

Watson

(1995). Constructing validity: Basic issues in objective scale development. Psychological Assessment, 7, 309-319. doi:10.1037/1040-3590.7.3.309

15.

Cole

D. A.

Maxwell

S. E.

(2003). Testing mediational models with longitudinal data: Questions and tips in the use of structural equation modeling. Journal of Abnormal Psychology, 112, 558-577. doi:10.1037/0021-843X.112.4.558

16.

Collins

L. M.

Schafer

J. L.

Kam

C. M.

(2001). A comparison of inclusive and restrictive strategies in modern missing data procedures. Psychological Methods, 6, 330-351. doi:10.1037/1082-989X.6.4.330

17.

Craig

W. M.

Pepler

D. J.

(1998). Observations of bullying and victimization in the school yard. Canadian Journal of School Psychology, 13, 41-59. doi:10.1177/082957359801300205

18.

Elwert

Winship

(2014). Endogenous selection bias: The problem of conditioning on a collider variable. Annual Review of Sociology, 40, 31-53. doi:10.1146/annurev-soc-071913-043455

19.

Espelage

D. L.

Asidao

C. S.

(2001). Conversations with middle school students about bullying and victimization: Should we be concerned? Journal of Emotional Abuse, 2, 49-62. doi:10.1300/J135v02n02_04

20.

Espelage

D. L.

Green

Polanin

(2012). Willingness to intervene in bullying episodes among middle school students: Individual and peer-group influences. Journal of Early Adolescence, 32, 776-801. doi:10.1177/0272431611423017

21.

Espelage

D. L.

Low

S. K.

Anderson

De La Rue

(2014). Bullying, sexual, and dating violence trajectories from early to late adolescence (Report submitted to the National Institute of Justice Grant #2011-MU-FX-0022). Retrieved from https://www.ncjrs.gov/pdffiles1/nij/grants/246830.pdf

22.

Evans

C. B. R.

Smokowski

P. R.

(2015). Prosocial bystander behavior in bullying dynamics: Assessing the impact of social capital. Journal of Youth and Adolescence, 44, 2289-2307. doi:10.1007/s10964-015-0338-5

23.

Freis

S. D.

Gurung

R. A. R.

(2013). A Facebook analysis of helping behavior in online bullying. Psychology of Popular Media Culture, 2, 11-19. doi:10.1037/a0030239

24.

Hayes

A. F.

(2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York, NY: Guilford.

25.

L.-T.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1-55. doi:10.1080/10705519909540118

26.

IBM Corporation. (2017). IBM SPSS statistics for windows (Version 25.0). Armonk, NY: Author.

27.

Jolliffe

Farrington

D. P.

(2006). Development and validation of the Basic Empathy Scale. Journal of Adolescence, 29, 589-611. doi:10.1016/j.adolescence.2005.08.010

28.

Juvonen

Schacter

H. L.

Sainio

Salmivalli

(2016). Can a school-wide bullying prevention program improve the plight of victims? Evidence for risk × intervention effects. Journal of Consulting and Clinical Psychology, 84, 334-344. doi:10.1037/ccp0000078

29.

Kenny

D. A.

(2013). Mediation: Sensitivity analysis [Webinar]. Retrieved from http://davidakenny.net/webinars/Mediation/Sensitivity/Sensitivity.html

30.

Latané

Darley

J. M.

(1969). Bystanders “apathy.” American Scientist, 57, 244-268.

31.

Menolascino

Jenkins

L. N.

(2018). Predicting bystander intervention among middle school students. School Psychology Quarterly, 33, 305-313. doi:10.1037/spq0000262

32.

Muthén

(1998-2017). Mplus user’s guide (8th ed.). Los Angeles, CA: Muthén & Muthén.

33.

Nickerson

A. B.

Aloe

A. M.

Werth

J. M.

(2015). The relation of empathy and defending in bullying: A meta-analytic investigation. School Psychology Review, 44, 372-390. doi:10.17105/spr-15-0035.1

34.

Padgett

Notar

C. E

. (2013). Bystanders are the key to stopping bullying. International Journal of Educational Research, 1, 33-41.

35.

Peyre

Leplége

Coste

(2011). Missing data methods for dealing with missing items in quality of life questionnaires. A comparison by simulation of personal mean score, full information maximum likelihood, multiple imputation, and hot deck techniques applied to the SF-36 in the French 2003 decennial health survey. Quality of Life Research, 20, 287-300. doi:10.17105/spr-15-0035.1

36.

Polanin

J. R.

Espelage

D. L.

Pigott

T. D.

(2012). A meta-analysis of school-based bullying prevention programs’ effects on bystander intervention behavior. School Psychology Review, 41, 47-65. doi:10.1037/spq0000072

37.

Pozzoli

Gini

(2010). Active defending and passive bystanding behavior in bullying: The role of personal characteristics and perceived peer pressure. Journal of Abnormal Child Psychology, 38, 815-827. doi:10.1007/s10802-010-9399-9

38.

Pozzoli

Gini

(2013). Why do bystanders of bullying help or not? A multidimensional model. Journal of Early Adolescence, 33, 315-340. doi:10.1177/0272431612440172

39.

Pozzoli

Gini

Thornberg

(2017). Getting angry matters: Going beyond perspective taking and empathic concern to understand bystanders’ behavior in bullying. Journal of Adolescence, 61, 87-95. doi:10.1016/j.adolescence.2017.09.011

40.

Preacher

K. J.

(2015). Advances in mediation analysis: A survey and synthesis of new developments. Annual Review of Psychology, 66, 825-852. doi:10.1146/annurev-psych-010814-015258

41.

Preacher

K. J.

Selig

J. P.

(2012). Advantages of Monte Carlo confidence intervals for indirect effects. Communication Methods and Measures, 6, 77-98. doi:10.1080/19312458.2012.679848

42.

Rivers

(2013). Morbidity among bystanders of bullying behavior at school. In Srabstein

J. C.

(Ed.), Bullying: A public health concern (pp. 211-222). Hauppauge, NY: Nova Science.

43.

Rucker

D. D.

Preacher

K. J.

Tormala

Z. L.

Petty

R. E.

(2011). Mediation analysis in social psychology: Current practices and new recommendations. Social and Personality Psychology Compass, 5, 359-371. doi:10.1111/j.1751-9004.2011.00355.x

44.

Schultz

Heilman

Hart

K. J.

(2014). Cyber-bullying: An exploration of bystander behavior and motivation. Cyberpsychology: Journal of Psychosocial Research in Cyberspace, 8(4), Article 3. doi:10.5817/CP2014-4-3

45.

Shadish

W. R.

Cook

T. D.

Campbell

D. T.

(2002). Experimental and quasi-experimental designs for generalized causal inference. Boston, MA: Mifflin.

46.

Shamay-Tsoory

S. C.

Aharon-Peretz

Perry

(2009). Two systems for empathy: A double dissociation between emotional and cognitive empathy in inferior frontal gyrus versus ventromedial prefrontal lesions. Brain, 132, 617-627. doi:10.1093/brain/awn279

47.

Sobel

(1982). Asymptotic confidence intervals for indirect effects in structural equation models. Sociological Methods, 13, 290-312. doi:10.2307/270723

48.

Steiger

J. H.

(1980). Tests for comparing elements of a correlation matrix. Psychological Bulletin, 47, 245-251. doi:10.1037/0033-2909.87.2.245

49.

Trach

Hymel

Waterhouse

Neale

(2010). Bystander responses to school bullying: A cross-sectional investigation of grade and sex differences. Canadian Journal of School Psychology, 25, 114-130. doi:10.1177/0829573509357553

50.

Young

Johnson

D. R.

(2013). Methods for handling missing secondary respondent data. Journal of Marriage and Family, 75, 221-234. doi:10.1111/j.1741-3737.2012.01021.x