Abstract
I propose the gender gap in delinquency is linked to adolescents’ orientation to gender-normative behavior and empathic development. I use longitudinal data on 1,525 youth from the Denver Youth Survey to analyze relationships among gender, empathy, and delinquency. I find girls exhibit higher levels of empathy across adolescence than do boys, and these differences emerge in preadolescence. Empathy is inversely related to delinquency, and is predictive of fraud and theft, but not violent delinquency. Finally, empathy partially mediates the effect of being male on delinquency. I therefore argue the gender gap in delinquency can be explained—in part—by adherence to gender norms governing empathetic expression.
Introduction
The gender gap in offending, in which males commit more offenses than do females, is one of the most enduring findings in criminology. Consequently, criminologists have examined how and why being male accounts for the gender gap in offending (Broidy & Agnew, 1997; Daly & Chesney-Lind, 1988; Lauritsen et al., 2009). A pronounced focus on the negative components of masculinity, however, has resulted in conspicuous inattention to prosocial emotional states, traits, and abilities (Kruttschnitt, 2013). In particular, female-coded prosocial traits, such as empathy, have received little attention overall as they relate to offending, and although male-coded predictors of offending such as aggression and impulsivity are regularly tested for “gender neutrality” (see Hagan et al., 1985; Ishoy & Blackwell, 2019; Steffensmeier & Allan, 1996), predictors of female (non)offending are rarely tested on boys and men. Therefore, our collective understanding of how and why male-coded behaviors are associated with offending in general, and the gender gap in particular, far outpaces our knowledge regarding the deterrent effects of female-coded behaviors. As a result, the implications of gendering prosocial behaviors and the role of this process in the persistence of the gender gap in offending remain under-theorized.
This article addresses these limitations by focusing on empathic ability—that is, the ability to transport oneself into the thinking, feeling, and acting of another person (Dymond, 1950; Ellis, 1982; Eysenck et al., 1985). Empathy requires both emotional intelligence and future-orientation, in that individuals must recognize the emotional states of others and predict how and why behaviors affect others’ emotional states. As with offending, there are age and gender differences in empathic ability. Children generally possess less empathy than adolescents and adolescents less than adults (Ellis, 1982; Lennon & Eisenberg, 1987; Van Vugt et al., 2011). Empathic ability increases from childhood through adolescence and is stable throughout most of adulthood (Eysenck et al., 1985; Grühn et al., 2008). Furthermore, males generally receive lower empathy scores on indices and tests than do females (Baron-Cohen & Wheelwright, 2004; Davis, 1983). Empathy has been argued to inhibit offending in that an individual’s ability to predict the negative emotional reactions of others resulting from their own antisocial behaviors disinclines the individual toward offending (Feshbach, 1975). Furthermore, there is evidence that gender differences in empathy are associated with gender differences in offending and may explain gender gaps in offending (Broidy et al., 2003). Numerous studies have found evidence of both empathy/offending and gender/empathy links (Davis, 1983; Jolliffe & Farrington, 2004, 2007; Van Langen et al., 2014), but few have centered the role of gender socialization and development in their explanations, and none have directly tested whether empathic ability mediates the effect of gender on offending. Therefore, empathy’s role in the perpetuation of gender gaps in offending is not wholly understood and may be underestimated.
I apply an interactionist framework of gender-differentiated development across childhood and adolescence to address previous findings of girls’ higher levels of empathic ability and boys’ higher involvement in delinquency. I theorize the gender gap in delinquency is partly a consequence of empathy’s strong positive association with femininity and, therefore, empathy’s negative association with masculinity (Goffman, 1977; West & Fenstermaker, 1995; West & Zimmerman, 1987). This, in conjunction with increased pressure to conform to these gender norms in adolescence, results in pronounced gender gaps in both empathic ability and delinquency. In sum, more empathy means less delinquency, and girls’ socialization into empathic roles means that—on average—their empathic ability is greater than that of boys. Therefore, girls engage in less delinquency than boys.
The Denver Youth Survey (DYS) provides an opportunity to trace empathy and delinquency over time and to specify their temporal relationship in statistical models. Analyses support previous findings that empathic ability varies within individuals across the life course (Baron-Cohen & Wheelwright, 2004; Davis, 1983; Kohlberg & Gilligan, 1971). I trace changes in empathic ability across childhood and adolescence, disaggregated by gender, and extend these findings to explore links among gender, empathy, and delinquency. Significantly, this research is the first of its kind to support a causal sequence in which empathic ability predicts delinquency and mediates the effect of gender on delinquency.
Theoretical Framework
There are many explanations for why boys offend more than girls. One of the most popular emphasizes offending as a masculinity-making resource incompatible with femininity (Messerschmidt, 2013; Steffensmeier & Allan, 1996). This explanation suggests boys and men engage in crime to demonstrate masculinity (Rios, 2009), whereas girls and women are socialized to value an “ethic of care” which decreases taste for risk but explains offending in defense of valued relationships, or “survival” crimes like running away from home (Chesney-Lind & Pasko, 2013; Joe-Laidler & Hunt, 2001; Steffensmeier & Allan, 1996). These arguments rely on interactional explanations of human behavior: assuming adherence to a gender binary is part of taking the role or perspective of an imagined other in pursuit of social acceptance and “unity of self” (Mead, 1930, 1934). For example, when Messerschmidt (2013) argues males use aggression and offending to advance “masculine performance,” he assumes (a) social actors are taking the perspective of an imagined “masculine man”; (b) the actor’s audience will provide interactional feedback confirming these behaviors are masculine; and (c) masculinity is desirable. In this way, boys learn to behave as masculine men and eventually believe those behaviors reflect natural inclinations (Goffman, 1977; Mead, 1930, 1934; West & Zimmerman, 1987). Where offending can reinforce masculine standing, empathic expression can undermine it due to associations with femininity (Connell, 1987; Dean, 2013; Pascoe, 2011). Empathic expression supports claims to femininity, as girls are socialized into care-oriented roles from a young age (Conlon et al., 2014; Giordano et al., 2007; Karniol et al., 1998). Thus, gender norms governing emotional expression incentivize empathic behaviors for girls and women, but discourage them among boys and men. These gendered processes and pressures have been argued as contributing to gender gaps in both empathy and delinquency (see Baron-Cohen & Wheelwright, 2004; Schrock & Schwalbe, 2009), and one may very well explain the other.
Application of this framework points to the processes through which gender differences in delinquency and empathy emerge, but offers limited guidance as to when or in what order. Multiple criminological studies, however, weave life course concepts such as developmental trajectories, turning points, and transitions throughout interactionist frameworks to explain moral and cognitive development (see Massoglia & Uggen, 2010; Matsueda & Heimer, 2004; Sampson & Laub, 1995). In doing so, they further contextualize time-dependent relationships such as those between empathy and delinquency. Furthermore, application of life course methods—such as longitudinal analysis of panel data—aids in theorizing the causal order of the empathy/delinquency relationship.
Currently, it is easy to miss when in the life course the gendering of certain behaviors and interactions is amplified or suppressed and in what order. For example, in early adolescence, risk of delinquency increases dramatically before plateauing and decreasing (D’Unger et al., 2002; Steffensmeier et al., 1989). This corresponds with increased pressure to adhere to gender norms regarding emotional and/or empathic expression and risk-taking (Hill & Lynch, 1983; Kohlberg & Gilligan, 1971). Criminologists have found empathy and offending are inversely associated (Jolliffe & Farrington, 2004; Van Langen et al., 2014; Van Vugt et al., 2011). Existing research, however, stops short of establishing whether empathy decreases delinquency, delinquency decreases empathy, or there are reciprocal effects. For example, Broidy et al.’s (2003) work on empathy, delinquency, and gender finds significant sex differences in empathy in their sample of delinquents versus nondelinquents and a moderating effect of sex between delinquency and empathy. They argue sex differences in empathy and delinquency are “partly a function of gendered socialization patterns that instill in women, more than men, an acute sensitivity to the needs of others” and theorize low empathy causes delinquency (2003, p. 506). Their cross-sectional data, however, prevent them from testing this hypothesis directly. Therefore, their work cannot tell us why differences in empathy and delinquency emerge, when, or in what order. Interactionist theories on gender, role taking, and identity suggest learned adherence to gender norms as the “why.” The integration of life course concepts into this theory leads to research questions and hypotheses in response to timing and order.
Although there is literature on the process of learning to do gender, the time-ordered effects of gendered experiences remain understudied. Thus, my hypotheses leverage criminological findings and pick up where interactionist theory leaves off—supporting a temporal sequence that begins in childhood and continues through adolescence in which gender influences empathy, empathy influences delinquency, and empathy is implicated in the gender gap in delinquency.
Hypotheses
An Empathic Transition: Childhood to Adolescence
As individuals transition from childhood to adolescence, social interaction becomes increasingly demanding and requires taking on a variety of roles and perspectives, or “an expanding capacity for empathy” (Kohlberg & Hersh, 1977, p. 56). In childhood, social interaction is largely based on egocentric concerns and nonabstract others’ evaluations of behavior. Interactions involve few new social actors, rarely involve consideration of the social order, and consist mostly of simple, cause, and effect calculations (Piaget et al., 1977). Limited empathy is required. In adolescence, interactions are increasingly based on both egocentric concerns and loyalty to the social order. The circle of people with influence over adolescents’ behavior widens, maintenance of the social order sometimes takes precedence over personal preference, and behavioral calculations become more complex (and more fraught).
In addition to these social changes, biological changes associated with puberty play direct and indirect roles in empathic development. There is evidence, for example, that higher levels of testosterone are associated with lower levels of empathy, and that males have higher levels of testosterone than do females (Chapman et al., 2006; Knickmeyer et al., 2006), leading to a gender gap in empathy. Specific to adolescence, scholars have argued that increases in boys’ testosterone levels in puberty causes “an increase in competitive behavior,” reducing empathy (Van der Graaff et al., 2014, p. 885). It is worth noting, however, that testosterone levels increase for both males and females during puberty and into the late teens and early 20s (Ankarberg & Norjavaara, 1999; Lee et al., 1974), albeit much more so for males. If empathic development were strictly biological, we would expect to find that: (a) empathic development is suppressed starting at the onset of puberty and (b) empathic ability is lowest in the late teens, when testosterone levels are at their peak. We might also expect these patterns to be more pronounced for males than for females, because males produce substantially more testosterone than females. This pattern, however, is not widely supported in existing research on empathic development or in the sample used in this study. Girls’ empathic trajectories are linear and positive throughout adolescence. By contrast, boys’ empathic development has been shown to decelerate (or even reverse) in early adolescence, though their empathic ability is generally higher at age 18 than at age 10 (Lennon & Eisenberg, 1987; Van der Graaff et al., 2014). Testosterone is not the only hormone associated with empathic development (Mehta & Josephs, 2010; Zilioli et al., 2015), but it is the most prominently discussed in literature on sex differences in empathy. However, as stated, there is reason to doubt testosterone development alone explains these developmental patterns. More likely, gender socialization gives meaning to the biological sex differences associated with puberty. For example, the emergence of secondary sex characteristics may be indirectly associated with empathic development in that appearing more “manly” or “womanly” changes patterns of social interaction. Specifically, more developed bodies amplify gender awareness among adolescents, abruptly orienting them to traditional gender norms with much emphasis on conformity to “natural” behavior (Hill & Lynch, 1983; Kohlberg & Gilligan, 1971). In turn, puberty-linked social pressures lead to a variety of cognitive and behavioral divergences between boys and girls, including gender differences in empathy. Given the sociological thrust of this research, I focus on social pressures. However, I do not dispute that many of the adolescent gender dynamics I argue as contributing to the gender gap in empathy are related to biological changes associated with puberty.
Importantly, there is reason to doubt the existence of a gender gap in the ability to feel empathy, particularly in studies that use self-report data (Eisenberg, 2000; Heimer et al., 2006). Furthermore, it is possible that the primary difference between genders is in expressed empathy, rather than felt empathy. Eisenberg and Strayer (1990) elaborate, observed sex differences on self-report measures may be due, at least in part, to differences in sex-role norms . . . Whether there is a true sex difference in emotional reactions to the distress of others, as opposed to reports of these reactions, is not yet as clear. (p. 195)
Nonetheless, I deliberately use the term “empathic ability” in this text. I do so based on the assumptions that repeated suppression of empathic expression degrades empathic ability over time, expression indicates ability, and expression cannot be entirely disentangled from ability. I define empathic ability as the ability to transport oneself into the thinking, feeling, and acting of another person. In keeping with psychological research, I argue it is an essential component of human development that changes with age and practice (Chase-Lansdale et al., 1995; Dymond, 1950; Ellis, 1982; Eysenck et al., 1985).
My first hypothesis tests for gender differences in empathic ability:
If supported, this hypothesis affirms prior research showing girls are generally more empathic than boys (Baron-Cohen & Wheelwright, 2004; Davis, 1983; Lawrence et al., 2004). In addition, my data allow me to trace empathic trajectories from ages 7 to 18, capturing empathic ability before, during, and after survey respondents transition from childhood to adolescence. If this hypothesis is supported, so too is a key assumption upon which this research is predicated: Empathy is coded as feminine. Interactionist theory suggests this turns boys off from engaging in empathic interactions as they weaken claims to masculine identities (Connell & Messerschmidt, 2005). As a result, boys’ empathic development slows during peak offending years. Assuming empathy is inversely associated with delinquency, the consequences of these divergent pathways likely include a widening gender gap in delinquency from the beginning of adolescence onward.
Establishing the Relationship Between Empathy and Offending
Although there is some evidence that “empathic overarousal” increases antisocial behaviors, overall empathy is associated with prosocial behavior and is considered a prosocial ability or trait (D. Cohen & Strayer, 1996; Eisenberg, 2000; Smith, 2010). Scholars have argued empathy and offending are associated, though few analyses include girls and women, and none have used longitudinal data (Jolliffe & Farrington, 2004; Van Langen et al., 2014; Van Vugt et al., 2011). Furthermore, empathic effects may vary by offense type. To start, empathic ability may be especially important for more deliberate types of delinquency. Longer decision-making processes could make it more likely potential offenders will encounter empathic triggers and less likely they will prioritize other traits relevant to offending decisions. For instance, retaliatory offenses—such as punching someone for coming on to a significant other—may be more closely linked to impulsiveness or masculinity than empathic ability, decreasing empathy’s impact. Conversely, stealing a car generally requires planning and deliberation, making future-oriented traits such as empathy more salient in the decision-making process. In addition, some delinquent behaviors may be more likely to elicit negative appraisals from others, and empathic youth may be more likely to notice and/or care about these appraisals than less empathic youth. For instance, although shoplifting may provide “sneaky thrills” (Katz, 1988), stealing money from a friend or family member may be more likely to elicit negative reactions from important others if revealed. Therefore, I conduct hypothesis tests by offense type—(a) fraud, (b) theft, and (c) violent delinquency—and present disaggregated results.
Given previous findings that empathy is negatively associated with antisocial behavior (Johnson & Ferraro, 2000; Staats et al., 2006; Walker, 2009), I expect that as empathic ability decreases, delinquency will increase. I assume this will hold true for both boys and girls, but test for an interaction between gender and empathy to determine whether gender moderates the empathy/delinquency relationship as found by Broidy et al. (2003). Thus, my second hypothesis predominantly functions to confirm or disconfirm previous findings regarding the empathy/delinquency relationship.
My third hypothesis tests for the ordering of the empathy/delinquency relationship. Most, but not all, theorizing regarding empathy and offending implies low empathy predicts offending. If an individual understands their actions as harmful and is responsive to the needs of others—in short exercises empathy—it follows their propensity to offend will decrease (Marlow et al., 2012; Marshall et al., 1995). Conversely, the experience of offending may also explain differences in empathy between offenders and nonoffenders (Broidy et al., 2003; Goldstein & Higgins-D’Alessandro, 2001; Marlow et al., 2012). Longitudinal data and analyses allow me to adjudicate these competing hypotheses.
Whether empathy predicts delinquency, delinquency predicts empathy, or empathy and delinquency are unrelated is a question of importance to scholars of cognitive development, crime, and deterrence. Null findings contradict conventional wisdom regarding empathy and delinquency, whereas significant findings in either direction back up long-made claims with empirical evidence. If there are gender gaps in empathy (H1), empathy is associated with delinquency (H2), and empathy predicts delinquency (H3), empathy is positioned as an explanation of the gender gap in delinquency rather than a simple correlate.
Empathy and the Gender Gap in Delinquency
While largely untested, the argument that gender differences in empathy explain the gender gap in offending has been around for some time. A. K. Cohen (1955) suggested girls reject violence because the “characters” they aspire to encourage affiliation and nurturance—behaviors which require empathic ability (p. 143). Likewise, Steffensmeier and Allan (1996) argue women are socialized to be more responsive to others’ needs than men, inhibiting their desire to offend. Gender scholars have also made the connection between empathy and offending. Connell and Messerschmidt (2005), for example, argue that pursuit of masculine power can result in both criminal activity and a “withering of empathy” (p. 852). These arguments position gendered interactions as key to empathic expression and offending behaviors and situate empathy as a possible cause of the gender gap in delinquency. Thus, my final hypothesis is as follows:
I anticipate empathic ability will decrease the total effect of being male on delinquency, partially mediating the effect of gender on delinquency. My theory suggests gendered pressures to behave according to traditional gender roles in adolescence will deter boys, but not girls, from practicing empathy. This positions empathy as a protective component of femininity and indicates testing empathic ability apart from gender will reveal a mediating effect. If the effect of being male is smaller in models where empathy is included, it implies male deficits in empathic ability help explain the gender gap in delinquency. Although I am limited by my nonexperimental survey data in regard to the establishment of a causal relationship between empathy and the gender gap in delinquency, affirmative H4 findings would suggest that empathy is one of several important correlates of the gender gap in delinquency.
Data
The Denver Youth Survey (DYS)
To test my hypotheses, I use the DYS. The DYS is a 10-wave longitudinal study initiated in 1988 on delinquency in “high-risk” Denver, Colorado neighborhoods, with questions on empathy and delinquency in the first five waves (for more on sampling methodologies, see Huizinga, 2017). Interviewer-conducted surveys were administered annually and the survey attrition rate is less than 10% for Waves 1–5. In addition, 98 respondents (6%) left and re-entered the survey during this period. The DYS includes five cohorts, aged 7–8, 9–10, 11–12, 13–14, and 15–16 in Wave 1. Respondents were given a child (ages 7–10), youth (ages 11–18), or adult (ages 19+) survey based on age at time of interview. Overall, the DYS allows me to include 1,525 individuals ranging in age from 7 to 18 (6,877 person years). This age range includes individuals leaving early childhood and beginning to experience an increase in gender-specific socialization and individuals at the peak of the age-delinquency curve.
Measures of Key Concepts
Empathy
The DYS includes a series of “yes or no” questions drawn predominantly from the empathy section of the Eysenck Personality Questionnaire (Eysenck et al., 1985; 1980). It is important to note that measuring an individuals’ ability to transport oneself into the thinking, feeling, and acting of another person is no easy task and any effort to do so comes with qualifiers. Empathy is generally understood as a broad, multidimensional concept comprising both cognitive and visceral reactions to the needs and dispositions of others (Davis, 1983; Eisenberg & Fabes, 1990; Miller & Eisenberg, 1988). Although the DYS contains questions that touch on both these dimensions, most of the questions measure visceral feelings of sympathy, concern, and personal distress. Thus, we must be careful in interpreting results as they relate to empathy’s cognitive dimensions such as the adoption of others’ viewpoints or competent interpretation of others’ emotional states. Finally, some scholars argue personal distress may indicate lower levels of social functioning and should not be used as a standalone measure of empathic ability or concern (Eisenberg, 2000; Fiske, 2018; Lawrence et al., 2004). Nonetheless, Eysenck and Eysenck’s Scales have been used with success by social scientists who wish to study personality characteristics such as thrill-seeking, impulsiveness, and empathy (Aluja & Blanch, 2007; Burt et al., 2014; Mak, 1991; Richendoller & Weaver, 1994).
Ten questions on empathy were posed to youth respondents in Waves 1 through 5, and four of these questions were posed to child respondents in Waves 1–3. As a result, the DYS contains two sets of empathy questions: a small, four-question set for children and youth, and a larger 10-question set for youth-only. In practice, this means I am limited to the child and youth question set (Q1–Q4) for analyses including respondents aged 10 and under, but I am able to use the more comprehensive youth-only question set (Q1–Q10) in analyses restricted to youths aged 11–18. Therefore, I tested and developed two empathy measures. To construct these measures, questions were evaluated per previous literature and the extent to which they capture empathy as a latent construct using confirmatory factor analysis (CFA):
Q1. Does it bother you very much when one of your friends seems upset?
Q2. Do you get very upset when you see someone cry?
Q3. Do you find it hard to understand why some things upset people so much?
Q4. Would you find it hard to break bad news to someone?
Q5. Do you tend to be nervous when others around you seem nervous?
Q6. Would you feel sorry for a lonely stranger in a group?
Q7. Do you ever get deeply involved with the feelings of a character in a film, play, or book?
Q8. Do you worry about how other people feel?
Q9. Can you be in a good mood even if those around you are depressed?
Q10. Do you become more irritated than sympathetic when you see someone cry?
Results from tetrachoric correlations, alpha tests, and principal components analyses (PCAs) paralleled previous efforts to evaluate Eysenck and Eysenck’s empathy measurement. Correlation coefficients are lower in early waves where respondents are younger and increase as the sample ages. PCAs indicate Q3, Q9, and Q10 are weak indicators of empathy relative to other questions in the set. As such, these questions were removed from consideration. CFAs were run wave-by-wave, and results did not vary significantly by wave. Ultimately, two sets of questions provided satisfactory standardized factor loadings ranging from 0.55 to 0.80, and root mean square error of approximation (RMSEA) values of less than 0.10 (Bollen & Long, 1993). I therefore generated two indices. For the Child and Youth Empathy Index, each respondent was assigned an empathy score ranging from 0 (exceptionally limited empathy) to 3 (high empathy). This score was determined by counting affirmative responses to three questions. The Youth-Only Empathy Index ranges from 0 (exceptionally limited empathy) to 7 (exceptionally high empathy) and was determined by counting affirmative responses to seven questions. The scales correlate at .84, indicating they capture the same or similar latent constructs. Indices were lagged as necessary in hypothesis tests and descriptive statistics are displayed in Table 1.
Descriptive Statistics.
Note. GPA = grade point average.
Significant gender differences in the mean (i.e., z-test statistics greater than 1.645 or less than -1.645). bLagged at t − 1 in some analyses, annotated accordingly. cAlso lagged at t − 1 to control for prior offending. dAlso squared to account for nonlinear offending and empathy patterns. eCoefficients not displayed in tables.
Independent sample t tests run to determine whether or not there were significant empathy differences between respondents who were and were not eligible for inclusion in analyses, which indicate that, overall, there are no differences in mean child and youth empathy scores, with one minor exception: Boys included in the analyses have slightly higher mean empathy scores (µ = 1.94) than do boys who were excluded from the analyses (µ = 1.83). Therefore, the effect of being male on empathic ability may be understated as some boys with lower scores are missing in H1 analyses.
Delinquency
Questions were asked of all youth respondents regarding how many times they had engaged in delinquent behaviors in the past year. Following previous research, these variables were summed to create three delinquency count indices, measuring fraud (using stolen or forged checks or credit cards, selling items under false pretenses, and selling stolen items), theft (theft of items worth US$5 or more, shoplifting, pickpocketing, and theft from or of an automobile), and violent delinquency (simple and/or aggravated assault, assault by throwing and object with intent to harm, and robbery).
Alternative Explanations
Previous research on offending, in general, and the gender gap in offending, in particular, suggest alternative explanations for these phenomena outside of differences in empathic ability. Therefore, I test several variables representing alternative explanations alongside empathic ability. To start, prior delinquency may be a strong predictor of future delinquency. Furthermore, delinquency and/or criminal justice involvement may influence empathic development (Broidy et al., 2003). Therefore, prior delinquency was included in models predicting delinquency and empathy to test the direction of this relationship. Prior delinquency for the focal delinquent offense is included in models. For example, if empathy is regressed on violence, the previous delinquency variable included is the previous year’s violent delinquency count.
Gender differences in parental control are likely contributors to the gender gap in juvenile delinquency. Cultural norms regarding the protection of girls’ virtue and boys’ risk-taking lead to gender differences in types and levels of parental supervision, and empirical research supports the hypothesis that differential levels of parental control are associated with the gender gap in delinquency (Hagan et al., 1985; Heimer & DeCoster, 1999). The DYS includes measures of parental control derived from Hagan et al.’s (1985) work on power-control theory. I use two measures recommended in this literature: (a) how likely parents are to “know who [respondent] is with when [they] are away from home” and (b) how likely parents are to “know if [respondents] are home on time.” Responses range from 1 (parents almost never know these things) to 3 (parents always know these things). I take the mean value of the sum of these responses as my time-varying measure of parental control experienced by respondents. There is some nonrandom missingness on this variable: Younger respondents were not always asked both or either question. Therefore, I control for missing parental control values using a dummy variable (0 = non-missing value; 1 = missing value). Coefficients for missingness controls are not displayed in tables.
Gender differences in impulsivity have also been suggested as an explanation for gender gaps in delinquency, and research on correlates of antisocial behavior suggests empathy and impulsivity may be linked, albeit contrasting concepts (Jolliffe & Farrington, 2006). As with empathy, it is likely that both biological and social factors generate observed sex differences in impulsivity and contribute to gender gaps in antisocial behavior (Burt et al., 2014; Lenroot & Giedd, 2010). Control theorists argue impulsivity is a fixed characteristic by adolescence, but that differences in social control (i.e., levels of parental control or attachment to school) suppress impulsive behaviors (Hirschi, 1977; Hirschi & Gottfredson, 1995). Therefore I include a measure of impulsivity. I take the mean value of the sum of responses to questions on impulsivity: (a) “I get restless if I have nothing to do,” (b) “I act without stopping to think,” (c) “I am impatient,” and (d) “I get upset when I have to wait for something.” Responses range from 1 (disagree) to 3 (agree). I then average within-respondent scores to form a time-invariant measure of impulsivity for each respondent.
Finally, peer influences on delinquent behavior have been suggested as causes of delinquency, in general, and the gender gap in delinquency, in particular. Boys are more likely to have deviant friends and acquaintances than are girls, because boys’ friendship networks consist mostly of other boys, and boys are more likely to be deviant than are girls (Warr, 2002). To that end, I include two measures of peers’ delinquency in final models: peers’ violent crime and peers’ property crime. These variables range from 1 (no offending peers in friend group) to 5 (all offending peers in friend group) and are an aggregation of several reports of peers’ violent (standardized α = .83) and property crime (standardized α = .90).
Background and Contextual Variables
Previous research has found both socioeconomic status and school performance reduce the effect of gender on empathic ability (Jolliffe & Farrington, 2007) and are inversely associated with delinquency (Elliott & Ageton, 1980; Hirschi, 1977). Therefore, I use income and single-mother households as proxies for socioeconomic status and grade point average (GPA) as a direct measure of school performance. I lag GPA to ensure models run were consistent with the temporal order suggested by my theoretical framework. As a result, lagged GPA could not be included in ordinary-least squares (OLS) regressions for 7-year-olds. As displayed in Table 1, the majority of DYS respondents are non-White. Dummy variables were included for race and ethnicity, with White used as the reference category. In addition, I control for missing GPA values via dummies (0 = nonmissing value; 1 = missing value). Coefficients for missingness controls are not displayed in tables.
Model Selection
OLS regression analyses were performed to evaluate H1: Boys exhibit lower levels of empathy than girls in adolescence, but not in childhood. H1 was tested using both age-stratified and pooled analyses (not shown). Respondents in H1 tests range in age from 7 to 18 and are roughly equal parts male and female. Where the Youth-Only Empathy Index was used, child respondents were dropped from the sample.
I use negative binomial regression in tests of H2, H3, and H4: Empathy is inversely associated with delinquency; Empathy predicts future delinquency; and Empathy partially mediates the effect of gender on delinquency. Because delinquency counts likely follow a Poisson distribution, OLS yields biased and inconsistent estimates of regression parameters (Long, 1997). Poisson regression analyses reveal overdispersion—the variance of the dependent variables exceeds the mean—in each model. Therefore, I use negative binomial models, which include an additional parameter to account for overdispersion. 1 Given most respondents are represented at multiple time points in these models, I follow King and Roberts (2015) and compared robust standard errors (clustered on respondents) to classical standard errors to determine the extent to which observation dependence biases nonlinear model estimates. I found—as have others who have used the DYS to investigate offending outcomes (e.g., Leverso & Matsueda, 2019)—that robust and classical standard errors were similar across these models, suggesting negative binomial estimates are not biased due to clustering on respondents. Therefore, I report negative binomial results with classical standard errors. Respondents in H2, H3, and H4 analyses range in age from 11 to 18 and are roughly equal parts male and female. Therefore, I use the lagged (t − 1) Youth-Only Empathy Index. Given associations between age and delinquency are nonlinear, age polynomials were included. As with models predicting empathy, I ran a series of t tests to determine whether or not there were significant delinquency differences between respondents who were and were not eligible for inclusion in these models. In total, I ran six gender-split t tests (two for each delinquency type) and three nonsplit t tests (one for each delinquency type). None of these t tests indicate significant differences in delinquency counts for respondents who were and were not included in final models.
Results
H1: An Empathic Turning Point
As shown in Figure 1, empathic ability differs by gender across childhood and adolescence. Age-stratified OLS regressions support a nonlinear empathic development process with statistically significant differences emerging at or immediately prior to the onset of puberty. Girls’ empathic development is positive and linear. Up to age 10 there are no significant gender differences in empathic ability. After age 10, however, boys’ empathic ability dips, increasing only slightly through age 18. The magnitude of the gender effect more than doubles between the ages of 11 and 13, when secondary sex characteristics are increasingly conspicuous for both boys and girls (Euling et al., 2008). Boys’ empathic ability increases only slightly through age 18 and they are poised to enter young adulthood lagging behind their female peers. Results support H1, as gender is not associated with empathic development in early childhood, but becomes significant as respondents transition from childhood to adolescence. Results were similar for both empathy indices, with results for the Child and Youth Empathy Index displayed in Table 2.
Mean youth and child empathy scores by age.
Age-Stratified Effects of Male on Empathy Using the Child and Youth Empathy Index (OLS Analyses).
Note. Standard errors appear in parentheses. OLS = ordinary-least squares; GPA = grade point average.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Pooled analyses provide additional support for H1: male, age, age2, and age3 are all statistically significant. Differences in empathic ability are proportionately similar across indices—boys score about 16% lower than girls. On the whole, control variables have a limited effect on the relationship between empathy and gender in childhood and adolescence. Notably, the effect of prior delinquency (t − 1) was so small and insignificant that I dropped this variable in final analyses.
The most significant predictor of empathic ability outside of gender is racial identification. Specifically, identifying as Black is inversely associated with empathic ability in pooled and (some) age-stratified analyses. Asian, Native American, and Other Race/Ethnicity identifying respondents also see their empathy scores differ significantly from White respondents at a handful of ages, though they are not significantly associated with empathic ability in pooled analyses. To be clear, I do not interpret these results to mean racial identification alone influences empathic ability. First, there are a number of structural disadvantages and experiences with bias specific to people of color in general, and Black Americans in particular, which are not and cannot be captured in these analyses. For example, Black girls are more likely than White girls to experience “exclusionary discipline” in schools—an experience that may discourage empathy-building behaviors (Morris, 2016). Furthermore, there is evidence that emotional suppression is practiced and encouraged by Black parents wishing to spare their children (potentially life-threatening) confrontations with stereotypes regarding Black anger and discontent (Dow, 2016). Thus, in addition to having less access to empathy-building interactions, Black respondents may be particularly susceptible to social desirability bias in their responses—responding with what is expected rather than how they truly feel. Although not the focus of this article, results make clear that race, ethnicity, and gender influence empathic development, and future researchers are encouraged to explore these relationships.
H2 and H3: The Empathy/Delinquency Relationship
Tests of H2 and H3—Empathy is inversely associated with delinquency; and Empathy predicts future delinquency tests—were performed on 11- to 18-year-olds using the lagged Youth-Only Empathy Index.
Table 3 presents the results of negative binomial regressions. Empathy is inversely associated with fraud and theft, but not violent delinquency. In addition, empathic ability is only statistically significant for fraud and theft. Empathic ability reduces fraud and theft by 17% and 14%, respectively. 2 Therefore, Model I results support H2 and H3 for fraud and theft, but not for violent delinquency. In Model II, the addition of prior delinquency, parental control, impulsivity, and peers’ violent/property crime reduces but does not eliminate the inverse effect of empathic ability on delinquency. A 1-unit increase in empathic ability is significantly associated with a 13% decrease in fraud, and a 12% decrease in theft. Again, empathy is not associated with or predictive of violent delinquency. Taken together, results indicate low empathic ability predicts a large proportion of delinquent behavior. 3 Although the addition of “alternative explanations” for the gender gap in delinquency does not substantially alter the relationship between empathy and delinquency, several of these variables are statistically significant. Both impulsivity and peers’ violent/property crime are positively associated with all three types of delinquency, prior delinquency is associated with a small increase in theft, and parental control is inversely associated with violent delinquency. Furthermore, male is no longer a statistically significant predictor of theft, though it is otherwise positive and significant across analyses and delinquency types.
The Effect of Empathy on Delinquency for Youth Age 11–18 (Negative Binomial Analyses).
Note. Standard errors appear in parentheses. Missing GPA and parental control dummy controls included but not shown. BIC = Bayesian information criterion; GPA = grade point average.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
There were some statistically significant results for background and contextual variables. Reported income has a small effect on both fraud and violent delinquency in Model II analyses, and being in a single-mother household is positively associated with fraud in both Model I and II analyses. GPA is inversely associated with all three delinquency types in Model I, but not in Model II. Individuals who identified as Asian or Hispanic/Latinx were significantly less likely to violently offend than individuals who identify as White. However, it is important to note there were only 57 observations and 16 respondents who identified as Asian. Likewise, Other Race and Multiracial also had small sample sizes.
Finally, Model II analyses were run with a male×empathy interaction term to determine whether gender moderates the relationship between empathy and delinquency, as found by Broidy et al. (2003). No significant interaction effects between empathic ability and gender on fraud or violent delinquency were found. However, the male×empathy term was significant in models predicting theft (−0.23**). These results indicate empathy deters fraud and violence in equal measure for boys and girls, but that empathic deficits predict boys’ but not girls’ theft. Results directly contradict Broidy et al.’s finding that empathic deficits are more criminogenic for girls than for boys. In addition, z-statistics were generated as a secondary check of equality of boys’ and girls’ empathy regression coefficients (see Paternoster et al., 1998). Z-scores for fraud (0.06) and violence (0.37) indicate no significant gender differences in empathy regression coefficients. The z-score for theft (−2.13), however, indicates that empathy coefficients for males and females are unequal. In gender-split models, empathy is not a significant predictor of girls’ theft, though they are inversely associated. 4 It is, however, significantly associated with a 19% decrease in boys’ theft counts.
H4: The Gender Gap in Delinquency
To test for a mediating effect of empathy on gender and delinquency, I first test the unmediated effect of male on delinquency (using Model II variables). I then add empathic ability to these models and see whether the magnitude of effect for male has decreased. 5 Analyses indicate the total effect of male on delinquency is positive across all delinquency types and significant in models predicting fraud and violent delinquency. The magnitude of effect reduces when empathy is added to fraud and theft models. Being male is associated with a 177% increase in the number of fraud offenses per year, but the inclusion of empathy reduces this effect to 136%. For theft, the effect of being male on number of theft offenses is reduced from 23% to 1%. Surprisingly, the effect of being male on violent delinquency actually increases in models where empathic ability is included, though the coefficient is not statistically significant. Results therefore support the hypothesis that empathy partially mediates the effect of male on delinquency for fraud and theft, but not for violent delinquency.
Discussion and Conclusion
Results support a relationship between gender, empathy, and delinquency that varies across childhood and adolescence. The DYS allows me to test empathic ability before, during, and after the onset of adolescence, and to determine empathy is implicated in the gender gap in delinquency. The first hypothesis, H1: Boys exhibit lower levels of empathy than girls in adolescence, but not in childhood, is wholly supported. In support of an interactionist, life course framework, emergence, and growth of significant differences in empathic ability in late childhood and early adolescence suggest empathic differences are a result of gender-differentiated socialization patterns and processes. If empathy were gender-neutral and these findings a result of lagged emotional maturation in boys, we would expect boys’ developmental trajectory to remain directionally consistent, albeit with a decreasing slope. However, not only do boys fall behind girls in empathic ability, they experience a period of anti-empathic development between the ages of 11 and 14, which I interpret as a period of increased awareness of gender-normative behavior (Hill & Lynch, 1983). Furthermore, if, as hypothesized in some developmental literature, sex differences in empathy were mostly or wholly biological, we might expect empathic development to be suppressed throughout adolescence and well into the late teens—especially for boys (Knickmeyer et al., 2006). However, as illustrated by these results, this is not the case. The increasingly negative effect of male on empathic ability from ages 7 to 14, combined with the switch to significance as individuals transition from children to adolescents, and the sizable change in effect at or around the emergence of secondary sex characteristics illustrate that empathy and gender socialization are deeply entangled.
Young adolescents are increasingly aware of threats to their gender identities, and exert and experience social pressure to “do gender” in attendance to perceived “natural” differences between the sexes (Kohlberg & Gilligan, 1971; West & Zimmerman, 1987). Thus, adolescents may understand empathy not just as a thing girls are better at than boys, but as a natural and inherent component of femininity—hopelessly inaccessible to “real” men (Dean, 2013; Goffman, 1977). Boys’ empathic ability does appear to bounce back: Around age 14 their empathy slope regains positivity. I take this to mean that as social interactions become more demanding, empathic ability necessarily increases, as does coherence, organization, and complexity of the self (Matsueda, 1992; Mead, 1934). Regardless, at age 18 men in my sample lag behind women on measures of empathic ability.
There are many consequences of empathic expression’s association with femininity. Gender differences in conflict resolution and friendship formation have been traced to empathic ability, and adolescent boys have been found to lag behind girls when it comes to problem-solving and intimacy as a result of lesser empathic ability (Chow et al., 2013; Wied et al., 2007). There are negative consequences for girls and women as well. Women may be more likely to experience exceptional personal distress when seeing others suffer—a tendency linked to higher incidence of depression (Broidy & Agnew, 1997; Helgeson & Fritz, 1998). Furthermore, women perform a disproportionate amount of “emotion work” in family units, a burden that increases psychological distress, decreases marital satisfaction, and generally takes up time and energy men are not expected to spend (Hochschild, 1979; Strazdins & Broom, 2004). Both personal distress and emotion work require and are a result of empathic ability, and empathy’s association with femininity means girls and women are expected to experience and perform personal distress and emotion work more so than boys and men. The consequences of gendered empathic development deserve interrogation, and I begin by elaborating on the relationship between empathy and delinquency.
The hypotheses that Empathy is inversely associated with delinquency (H2) and Empathy predicts future delinquency (H3) are not supported for violent delinquency, but are supported for fraud and theft. The competing hypothesis that delinquency predicts empathy is not supported for any delinquency type. Notably, interaction terms, z-scores, and gender-split models indicate empathy predicts boys’ theft, but not girls’ theft. The theoretical framework of this article suggests several reasons as to why such differences may exist. To start, boys are more likely to be penalized than girls for exhibiting empathic behavior (Pascoe, 2011), and if this penalty takes the form of exclusion from antisocial group activities, it may actually decrease boys’ opportunity to engage in theft. In addition, given empathy is a female-coded behavior, boys may have to work harder, or more actively to develop this trait, which may in turn increase the likelihood that empathic concerns will be prioritized over impulsive concerns. This could explain why impulsivity is a significant predictor of girls’ theft, but not of boys’ theft.
The hypothesis that Empathy partially mediates the effect of gender on delinquency (H4) is supported for fraud and theft. However, I find empathic ability is not a correlate, predictor, or mediator of gender for violent delinquency. This finding contradicts most, but not all, previous studies on empathy, violence, and aggression. Specifically, studies of incarcerated populations comparing violent offenders to nonviolent offenders have produced mixed results (Fernandez & Marshall, 2003; Goldstein & Higgins-D’Alessandro, 2001; Hoppe & Singer, 1976). In addition, it is possible empathy’s effect on violent delinquency has less to do with how much or how often one offends, but whether one transitions from nonoffender to violent offender in the first place. To that end, I ran logistic regressions comparing violent offenders to respondents who had not reported any violent delinquency and found no significant empathy effects. These results provide additional support for the argument that empathy—as measured in this article—is not associated with violent offending. It is possible, however, that the specific type of empathy measured in these analyses is not associated with violent delinquency, but other measures would be. In particular, cognitive empathy may be more important in preventing or deterring violent delinquency than visceral empathy. Further research is needed to validate this finding.
Although the protective effect of empathy is negligible for violent delinquency, empathic ability is a correlate, predictor, and mediator of gender for fraud and theft. I propose three explanations for the empathy/delinquency relationship that apply specifically to fraud and theft, but not to violent delinquency. The first explanation is deliberation. Fraud and theft may require more deliberation than the violent offenses measured here, which means future-oriented traits such as empathy are more salient in the decision-making process. Notably, impulsivity is positively and significantly associated with all three delinquent outcomes, implying it is important across all offenses. However, only for violent delinquency is low parental control also a significant predictor of delinquency. Therefore, I hypothesize impulsivity and parental control take priority in the sometimes-very-brief periods of deliberation prior to engaging in violent delinquency, thus reducing—or perhaps moderating (see Jolliffe & Farrington, 2006)—empathy’s influence on violent offending decisions. Second, fraud and theft offenders may be more likely to be confronted with negative appraisals from peers if and when their delinquent behaviors are revealed to others. Although negative appraisals are made of violent offenders, I speculate these appraisals are less damaging to one’s conception of self because violent delinquency is tied to status on the street, particularly for inner-city boys and men (Anderson, 2000). In addition, peers may be more likely to confront fraud and theft offenders than violent offenders. An empathic individual should be more capable of recognizing negative appraisals and less likely to engage in behaviors that cause them. Third, male’s positive effect on fraud and theft indicates some masculine payoff is embedded in these behaviors but unlike violent delinquency, empathy significantly reduces the effect of male on these offenses. This indicates two things: (a) refusal to engage in these behaviors is less of a threat to adolescents’ masculinity than refusal to engage in violent delinquency and (b) gender gaps in empathic ability help explain gender gaps in fraud and theft.
In addition to supporting the central tenants of tested hypotheses, findings raise questions for future researchers. First, the Eysenck and Eysenck Empathy Questionnaire is one of many possible empathy measures. Subsequent scholars have credibly argued empathy can be separated into visceral and cognitive components, and that these components should be tested separately (Davis, 1983; Fernandez & Marshall, 2003; Jolliffe & Farrington, 2007; Marshall et al., 1995; Van Langen et al., 2014). This raises the question of whether or not the gender, empathy, and delinquency relationships found in these analyses can be replicated across empathy measures. Finally, although one of several goals in this article is to emphasize the importance of investigating prosocial, feminine-coded traits, and emotions as they relate to the gender gap in delinquency, it is worth noting that a growing body of literature that suggests negative feminine-coded traits and emotions such as guilt and shame are also worth interrogation (Hay, 2003; Rebellon et al., 2016). Given empathy may trigger some of these feelings, future researchers would do well to include empathic ability in models predicting deterrent, affective states.
The contributions of this study are clear. There has been no research effort wherein the causal order of the empathy/delinquency relationship is strongly supported, or in which empathy’s role in maintenance of the gender gap in delinquency is directly tested. Low empathy is not simply associated with delinquency; it is a predictor of future fraud and theft in lagged models. Moreover, results support application of an empathy-emphasizing interactionist framework and demonstrate the utility of life course methods in understanding emotional and cognitive development as it relates to crime and gender. Finally, these analyses directly test empathy’s role in the gender gap in delinquency and finds it help explain a persistent criminological finding. The gender gap in delinquency may be explained—in part—by significant differences in empathic ability between boys and girls across the life course. In the future, scholars of the gender gap in offending should attempt to identify not just criminogenic components of traditional masculinity, but protective components of traditional femininity. This will require gender- and emotion-inclusive longitudinal data collection. For better or worse, criminology’s legacy is built on theories developed by men to explain male offending. This legacy extends into contemporary data collection in which measures of traditionally masculine, antisocial behaviors are staple survey items but the inverse is untrue (Block et al., 2010; Cook, 2016; Kruttschnitt, 2013). Collecting and analyzing such data will yield additional insight into offending and gender and can inform programs and policies aimed at deterring delinquency.
In addition, given strong evidence that empathic development is linked to both gender- and sex-specific social and biological factors, researchers who wish to better understand gender differences in empathy should develop and use data that account for both social and biological change in adolescence. Furthermore, while outside the scope of this article, there is evidence that gender gaps in delinquency may also be associated with biological changes in adolescence (see Pratt et al., 2016). Therefore, although I argue it is improbable that biological factors are driving what I argue to be social, further research is needed in this area. Finally, the hypotheses and analyses resulting from an additional emphasis on gender support the utility of an integrated interactionist theory. In calling for integrating gender literatures and life course concepts into an interactional framework, I believe future scholars will be able to further interrogate the causes and consequences of the gender gap in offending.
Footnotes
Acknowledgements
The author thanks Ross Matsueda, Julie Brines, Callie Burt, and John Leverso for their feedback throughout the writing process; Karina Rider for comments on early drafts; and Alexes Harris and the UW Gender Group for feedback in later stages. She also thanks the anonymous reviewers for their careful and considerate feedback. Finally, she would like to thank her husband Brendan for his love, support, and copyediting.
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
