Abstract
Traditional bullying and cyberbullying are common problems faced by today’s youth. Research seeking to explain bullying perpetration has often invoked Agnew’s general strain theory (GST). However, research to date has often explored within a given study only a single emotion at a time that can result from strain. Further, prior research has tended to take the causal ordering arguments of GST at face value. The current study seeks to focus on the correlation of strains related to socioeconomics with traditional and cyberbullying perpetration and negative emotions. Utilizing the Add Health data and path modeling in Mplus, results suggest that socioeconomic strain positively correlates with bullying perpetration and recent negative emotions. However, results suggest a potential causal chain that is the opposite of an expectation of GST, with bullying perpetration potentially affecting negative emotions, and not the other way around. Implications of the results for theory and policy are discussed.
Bullying has been a concern going back several decades, with numerous scholars seeking to explain this behavior (Besag, 1989; Ericson, 2001; Gaffney et al., 2019; Olweus, 1978; Olweus et al., 2019; Patchin & Hinduja, 2011; Tattum, 1989). Early studies sought to explain the prevalence of bullying (Boulton & Underwood, 1992; Finkelhor et al., 2005; Haynie et al., 2001; Seals & Young, 2003; Stephenson & Smith, 1989), while more recent research has attempted to explain the etiology of bullying behaviors (Borg, 1998; Connell et al., 2016; Hawker & Boulton, 2000; Hinduja & Patchin, 2008; Mulvey et al., 2018; Patchin & Hinduja, 2011; Rigby, 2003; Roland, 2002; Smith et al., 2019). One theory that has frequently been cited as a potential explanation for bullying behavior is general strain theory (GST). GST argues that when individuals experience stressors, or “strains,” they may experience negative emotions and cope through deviant behavior, that is, strain → negative emotions → deviance (Agnew, 1992, 2006a).
While the causes of bullying have been frequently studied, and GST has been cited as an explanation for bullying behavior, gaps remain in this literature. For instance, the literature on bullying behavior that uses GST as a theoretical framework has not often enough explored the possible negative emotions that can simultaneously result from strain, and how they may correlate with different types of bullying behavior. More often, research has focused on a single negative emotion in isolation within a given study, usually either anger or depression (see Hinduja & Patchin, 2008; Patchin & Hinduja, 2011). Additionally, important theoretical controls, such as parenting, school factors, and peer deviance, have occasionally been absent from past studies (see Jang et al., 2014).
Lastly, prior studies in the bullying literature have taken the causal ordering of GST at face value, and not explored alternative models of the relationship between strain, negative emotions, and bullying (Hinduja & Patchin, 2008; Patchin & Hinduja, 2011). Newer statistical software packages allow researchers to examine differing causal models than those that strictly adhere to a particular theory, allowing researchers the ability to gage whether a theory’s proposed ordering of variables fits better than an alternative modeling strategy.
The focus of the current study is to partially fill these gaps in the literature on bullying. To do so, we examine the correlation between experiencing socioeconomic strain and three negative emotions, in a global measure, and both traditional and cyberbullying in a nationally representative sample of adolescents in grades 5 to 10. Prior research suggests that low socioeconomic status increases negative emotions (Gallo & Matthews, 2003) and bullying behavior (Tippett & Wolke, 2014).
We propose and test hypotheses based on a re-specification of GST utilizing path modeling in Mplus. In the following sections, we will review the literature on bullying and GST, present our hypotheses, describe our sample and methodology, present results, and ultimately discuss the findings in light of their potential importance for theory and policy.
Traditional Bullying and Cyberbullying
Bullying behavior during adolescence, a period traditionally defined as going from the onset of puberty to legal adulthood, has been a major issue of concern in educational institutions for decades (Boulton & Underwood, 1992; Olweus, 1978; Patchin & Hinduja, 2010). The U.S. Centers for Disease Control and Prevention has defined bullying as “any unwanted aggressive behavior(s) by another youth or group of youths who are not siblings or current dating partners that involves an observed or perceived power imbalance and is repeated multiple times or is highly likely to be repeated” (Gladden et al., 2014). More recent advances in technology have contributed to the development of cyberbullying, which has been defined as “willful and repeated harm inflicted through the use of computers, cell phones, and other electronic devices” (Hinduja & Patchin, 2011).
Due to differences in the measurement and definition of bullying across studies, various estimates of the prevalence of bullying among youth have been suggested (Gladden et al., 2014). Modecki et al. (2014) reported a prevalence rate of 35% for traditional bullying perpetration among adolescents. Suggested lifetime rates of cyberbullying have varied, with the average offending rate being about 16% (Patchin & Hinduja, 2016). Cyberbullying is a unique form of bullying, as it can occur anywhere and anytime, and can also occur in conjunction with traditional bullying (Mehari & Farrell, 2018; Patchin & Hinduja, 2006). Currently, there are no national estimates of the number of adolescents who engage in or have engaged in both types of bullying. Research suggests that bullying victims tend to be passive, anxious, insecure, submissive, and rejected by their peers, while those that bully tend to be older, male, less empathetic, aggressive, and have greater histories of trouble at home, such as parental rejection and divorce (Bernstein & Watson, 1997). Children who engage in bullying behavior are also less likely to come from high socioeconomic backgrounds (Tippett & Wolke, 2014).
General Strain Theory and Bullying
Prior research has emphasized the consequences and the prevalence of bullying. More recently, research has sought to identify why adolescents engage in bullying behavior. While several theoretical explanations for bullying behavior have been offered, general strain theory (GST) has arguably been cited most often (Hinduja & Patchin, 2008; Jang et al., 2014; Moon et al., 2011, 2012; Patchin & Hinduja, 2011). GST proposes that the stressors or “strains” that individuals experience can lead to deviant behavior as a means of coping with these strains (Agnew, 1992). Agnew (1992) argues that there are three types of strain that can lead to deviance: the inability to achieve positively valued goals, the removal of positively valued stimuli, and the presentation of noxious stimuli. In the present study, we focus on a type of strain that may reflect all three categories, resource deprivation in the form of socioeconomic strain.
Agnew (1992) argued that strained individuals experience negative emotions, such as anger and/or depression, and that deviance is one possible way to cope with these emotions. Low socioeconomic status in particular has been found to correlate with negative emotions, such as depression, hopelessness, anxiety, and hostility (Bøe et al., 2012; Gallo & Matthews, 2003; Huurre et al., 2007). Furthermore, Baron (2004) found that youths who experience deprivation, specifically regarding their financial situation, are angrier, and that anger is correlated with higher rates of criminal activity. Youths may experience strain but not engage in deviant behavior if the strain is not paired with experiencing negative emotions.
In the time since GST was proposed, a solid body of research has been established supporting the theory (Agnew, 2006a; Agnew et al., 2002; Agnew & White, 1992; Aseltine Rh et al., 2000; Brezina, 1996; Broidy, 2001; Kaufman, 2009), including studies that have shown how important measures coming from other theories should be accounted for when examining a GST model (see Agnew, 1999; Piquero & Sealock, 2004; Watts, 2015; Watts & McNulty, 2013).
While a fair number of studies have looked at bullying victimization as a type of strain (see Cullen et al., 2008; Hay & Meldrum, 2010; Hay et al., 2010; Hinduja & Patchin, 2007; Wallace et al., 2005), fewer studies have looked at bullying perpetration as a potential outcome of strain (Patchin & Hinduja, 2011). Bullying perpetration among adolescents makes sense as a potential response to strain, in particular when considering strains related to socioeconomics. Youths experiencing strain regarding their perception of their family’s socioeconomic standing and/or home life, whether it is abuse, poverty, or generally poor relations with parents, may feel as though they can regain a sense of power and improve how they feel about themselves by harassing others, whether in person or online (Baron, 2004; Olweus, 1978, 1993; Patchin & Hinduja, 2011; Rigby & Slee, 1993). The bad feelings caused by strain, such as feeling angry, depressed, anxious, and frustrated, could potentially be alleviated by asserting dominance and superiority over one’s peers, thus making the individual engaging in bullying behavior feel better about their relative social position in comparison to others (Patchin & Hinduja, 2011). Studies on deprivation suggest that deviant behavior and criminal activity are adaptations to the strain produced by the deprivation (Agnew, 1992; Baron, 2004; Stiles et al., 2000), thus indicating that youth who perceive their home life as being deprived may engage in bullying as a means to alleviate their negative feelings.
Several previous studies have examined bullying perpetration as the outcome of strain. Patchin and Hinduja (2011) found that feelings of strain increased feelings of anger/frustration and both traditional bullying and cyberbullying. However, their study examined only anger/frustration, and their models did not include any variables from other theories of deviance as controls. This latter point is particularly important, given that GST is clear that the strain-negative emotions-deviance relationship can be conditioned by variables coming from other criminological theories, for which these variables should at least be controlled (Agnew, 2006a). In particular, past studies that have tested GST have focused on variables representative of social learning theory and social bonding theory (Watts, 2017; Watts & McNulty, 2013). Moon and Jang (2014) looked at the effects of strain on negative emotions and bullying while considering variables from other theories of deviance, but it was within a small, non-representative sample, and they did not consider cyberbullying. In a study looking at cyberbullying only, Hinduja and Patchin (2008) found that measures that constitute strain, including problems in school and bullying victimization, increased cyberbullying perpetration, but negative emotions were not considered at all. Moon et al. (2011) investigated the effects of both strain and negative emotions on bullying alongside other theories. However, they only considered traditional bullying, and their study was not a strict test of GST, as they did not examine the effects of strain on negative emotions or whether negative emotions mediated the effect of strain on traditional bullying. In a similar study, Moon et al. (2012) looked at how variables from other prominent theories condition the relationship between strain and traditional bullying perpetration, but again they did not test the effects of strain on negative emotions, in this case, anger only, or whether anger mediated the effects of strain on traditional bullying. Jang et al. (2014) found in a sample of Korean youth that traditional bullying victimization increased cyberbullying perpetration, but their study did not consider negative emotions or variables from other important theories of deviance. To address this issue, Yang et al. (2018) found that negative emotion had a mediating effect on peer victimization and its contribution to bullying perpetration. Students who were victimized were more likely to experience negative emotions and use perpetration as a means to alleviate those negative emotions.
While the traditional model of GST examines how strain influences negative emotions that then lead to deviant behavior, there is little discussion on whether this model can be examined through another path: strain to deviant behavior to negative emotions. There is a wide body of literature that examines how strain contributes to negative emotions (e.g., Agnew, 2006b; Ganem, 2010; Jang & Johnson, 2003; Moon et al., 2009), but there is a dearth of literature exploring the relationship concerning how strain might lead to deviant behavior and how that behavior might then influence one’s negative emotions. However, several articles have examined how bullying and harm are mediated or involved with negative emotions. Menesini and Camodeca (2008) found that adolescents were more likely to have feelings of guilt if harm was caused intentionally versus unintentionally, and feelings of both shame and guilt, specifically in situations where the behavior might constitute a moral failing. Similarly, Ahmed and Braithwaite (2004) found that shame management, alone and in conjunction with other variables such as personality, family, and school, helped reduce bullying behaviors. Additionally, the shame associated with bullying showed displacement of anger toward others. Roberts et al. (2014) found that bullying behavior in adolescents is associated with guilt and anger, and that guilt and anger is influenced by the level of empathy an adolescent possesses, with more empathetic adolescents experiencing more guilt and less anger.
While there is significant support for the original model of GST, and support that there are negative emotions associated with bullying, there is a need to examine how committing deviant acts like bullying, as a result of strain, could potentially influence one’s emotions. Newer statistical software allows researchers to check the validity of proposed causal relationships against proposed alternatives.
The Present Study
We seek to build on the literature on bullying in three ways. First, while several previous studies have utilized GST as a framework to explain bullying perpetration, to date most studies have only looked at one negative emotion at a time and have not generally included measures of resource deprivation in their models. It has been argued in the GST literature that numerous negative emotions may result from strain and shape offending (Agnew, 1992, 2006a; Brezina, 1996; Iratzoqui, 2018). Second, when studies have examined the effect of strain on bullying perpetration, there have been several instances where important variables from other prominent criminological theories have been absent. It has been argued in the GST literature that important variables from other criminological theories may condition or confound the effect of strain on offending, and they should therefore be controlled for when possible (Agnew, 2006a). Finally, most previous research in the bullying literature has taken the causal model of GST at face value, and despite a few studies that show bullying behavior can result in negative emotions (i.e., Ahmed & Braithwaite, 2004; Menesini & Camodeca, 2008; Roberts et al., 2014), have continued to rigidly test the model as being strain-negative emotions-bullying. Due to the above research that suggests that negative emotions can result due to bullying perpetration, and because of the limited research surrounding the strain-negative emotions-bullying pathway, in our analyzes, we will examine an alternative model: strain-bullying-negative emotions.
We draw on our re-specification of GST to derive three hypotheses about the relationship between strain, negative emotions, and bullying perpetration. Hypothesis 1 predicts that perceived socioeconomic strain will positively correlate with both traditional bullying and cyberbullying behavior. Hypothesis 2 predicts that perceived socioeconomic strain will positively correlate with recent negative emotions. Hypothesis 3 predicts that traditional bullying and cyberbullying behavior will positively correlate with recent negative emotions, and there will be a significant indirect effect of perceived socioeconomic strain on recent negative emotions through traditional bullying and cyberbullying behavior, which would be suggestive of mediation. These hypotheses make sense because research suggests that low socioeconomic status increases both negative emotions (Gallo & Matthews, 2003) and bullying behavior (Tippett & Wolke, 2014).
Data and Method
Sample
The current study utilizes data from the Health Behavior in School-Aged Children (HBSC) study, 2009 to 2010. HBSC is a study that is a collaborative, ongoing international project conducted by the World Health Organization (WHO) with a focus on examining a variety of behaviors that contribute to the health and wellbeing of individuals between the ages of 10 to 17 years of age. Specifically, this survey monitors a variety of behaviors that have been linked to health-risks among youth including behaviors that lead to violence and unintentional injuries; high-risk sexual behaviors; the use of tobacco products; the use of alcohol and drugs; body image, and risky dieting behaviors; and physical inactivity (Iannotti, 2013). The HBSC study has been conducted every 4 years since 1985 to 1986. Data for this study came from the 2009 to 2010 HBSC survey conducted in the US, which is the most current publicly available US HBSC data at the time of analysis.
The 2009 to 2010 HBSC contains a nationally representative sample of students in grades 5 to 10 across the United States that was obtained by using a three-stage, stratified design (Iannotti, 2013). All students in grades 5 to 10 in both public and private schools were included in the target population. Data were collected from all 50 states and the District of Columbia (Iannotti, 2013). Following the three-stage, stratified sample design, census divisions, and grades were used as strata, and school districts were sorted into primary sampling units (PSUs) by the size of the county in which they are located. African American and Hispanic students were oversampled to obtain nationally representative samples of these groups. Approximately 475 schools were eligible for participation in the 2009 to 2010 data collection, from which 314 schools participated. The day the survey was administered, 675 student respondents who were eligible to complete the survey were absent. However, within a few days of the original administration date, 301 of the absent students completed the survey. Ultimately, the 2009 to 2010 HBSC had a completion rate of just over 90% (Iannotti, 2013). The data are for each student, students are not considered nested within schools, and there are no cluster effects (Iannotti, 2013).
For the national survey, the questionnaire was sent to each participating school for school representatives to administer to students (Iannotti, 2013). Some researchers have argued that allowing school officials to administer surveys to students could limit the data collection by allowing for bias, missing those respondents who dropped out of school, and influencing students to give more socially acceptable answers to please the official (Brownfield & Sorenson, 1993). However, this is a very common method of data collection, and numerous national studies, such as Monitoring the Future and the Youth Risk Behavior Surveillance Study, are conducted this way. School representatives (i.e., teachers, counselors, etc.) read a script, which briefly introduced and explained the survey to the participating students. Additionally, school representatives recorded information, such as the grade level and the number of students enrolled in the sample classes. By recording this information, researchers are later able to verify sample selection and weight data. The questionnaire was conducted in a regular classroom setting and took approximately 45 minutes to complete.
The HBSC suits the current study well, as it has a specific section devoted to adolescents’ experiences with traditional and cyberbullying perpetration. Additionally, it has a specific set of survey questions that deal with adolescents’ perception of their home life, and numerous questions concerning emotions. The analytic sample in the current study consists of 11,579 individuals. 1 Descriptive statistics for all of the study variables outlined below can be found in Table 1. 2
Descriptive Statistics.
Note. Because these statistics are weighted and adjusted for survey design, standard errors are produced rather than standard deviations.
Measures
Dependent Variables
The dependent variables under study consist of one global construct, comprised of three variables, measuring recent negative emotions and three separate measures of bullying perpetration. The three variables within the global construct measuring recent negative emotions all have similar wording, all cover the same time, and all have the same answer categories. The three questions all asked “in the last 6 months how often have you had the following,” with the three questions ending with “feeling low,” “irritability or bad temper,” and “feeling nervous.” While certainly not reflective of the entirety of the concepts, these variables can be considered as partial, flawed representations of depression, anger, and anxiety, respectively. For all three questions, the answer categories consisted of about every day (1), more than once a week (2), about every week (3), about every month (4), and rarely or never (5). These ordinal measures were reverse coded so that higher scores denoted greater negative emotions, and they were combined
The three bullying measures consist of cyberbullying, traditional bullying, and a combined bullying measure. Cyberbullying consists of four combined measures that asked about bullying in the past couple of months. The questionsincluded (1) “I bullied another student(s) using a computer or e-mail messages or pictures”; (2) “I bullied another student(s) using a cell phone”; (3) I bullied others outside of school using a computer or e-mail messages or pictures;” and (4) “I bullied others outside of school using a cell phone.” The response options for these questions included: I have not bullied another student in the past couple of months (1), it has only happened once or twice (2), two or three times a month (3), about once a week (4), and several times a week (5). These four measures were combined (summed) into a global measure of cyberbullying (4–20, alpha = .94).
Traditional bullying consists of seven combined measures that also cover the past couple of months and have the same answer categories as the cyberbullying measures. The questions included (1) “I called another student(s) mean names, and made fun of, or teased him or her in a hurtful way”; (2) “I kept another student(s) out of things on purpose, excluded him or her from my group of friends, or completely ignored him or her”; (3) “I hit, kicked, pushed, shoved around, or locked another student(s) indoors”; (4) “I spread false rumors about another student(s) and tried to make others dislike him or her”; (5) “I bullied another student(s) with mean names and comments about his or her race or color”; (6) “I bullied another student(s) with mean names and comments about his or her religion”; (7) “I made sexual jokes, comments, or gestures to another student.” These seven measures were combined (summed) into a global measure of traditional bullying (7–35, alpha = .90). The combined bullying measure simply consists of combining (summing) all eleven bullying measures into a single, universal measure of bullying (11–55, alpha = .94). Prior research has supported the use of these bullying measures in the HBSC as valid (Roberson & Renshaw, 2018).
Independent variable
The independent variable perceived socioeconomic strain consists of seven items that largely reflect strain concerning socioeconomic status. Three questions asked, (1) “Do you have your own bedroom for yourself?” (2) “Does your family own a car, van, or truck?” and (3) “How many computers does your family own?” with response options of none, one, two, or more than two. Another question asked, (4) “How often do you have an evening meal together with your mother or father?” with response options of never, less than once a week, 1 to 2 days a week, 3 to 4 days a week, 5 to 6 days a week, and every day. Another item asked, (5) “Some young people go to school or bed hungry because there is not enough food at home. How often does this happen to you?” with response options of always, often, sometimes, and never. An item asked (6) if a respondent’s father had a job (yes/no), while a final item asked (7) “How well off do you think your family is?” with response options of very well off, quite well off, average, not very well off, and not at all well off. Items were reverse coded where necessary such that higher scores meant more stressful socioeconomic situations. Since the items were measured on different metrics, they were standardized and then combined into the global scale of perceived socioeconomic strain. The use of these measures as indicators of SES in the HBSC has been validated in prior research (Currie et al., 1997; Inchley & Currie, 2016).
Controls
Numerous controls are included in the path analyzes. Several basic demographic controls include biological sex, age, race/ethnicity, and grade in school. Biological sex is represented by male (1 = respondent identifies as male, female is referent). As indicted in Table 1, the sample is approximately 51% male, 49% female. The age question included the categories 10 years or younger, 11, 12, 13, 14, 15, 16, and 17 or older. White (1 = yes, non-white is referent) and Hispanic (1 = yes, non-Hispanic is referent) are the race/ethnicity variables. As indicated in Table 1, the sample is approximately 54% white, 46% non-white, and 28% Hispanic, 72% non-Hispanic. Grade includes the categories of fifth, sixth, seventh, eighth, ninth, and tenth. We additionally control for several variables of potential theoretical import that may relate to either strain or bullying perpetration and are thus important for avoiding model misspecification and omitted variable bias. Controlling for theoretically important variables is also important because GST has argued that important variables from other theories of crime may confound the relationship between strain, negative emotions, and deviance, and their inclusion is important to isolate the unique effects of strain on negative emotions and deviance (Agnew, 2006a).
These variables most closely represent concepts from social bonding theory (Hirschi, 1969) and social learning theory (Akers, 1985). Parental warmth consists of four items. The four questions asked the respondent if their parent/guardian helps them as much as they need, is loving, understand their problems and worries, and makes them feel better when they are upset. Answers included, (1) always, (2) sometimes, (3) almost never, and (4) don’t have or don’t see parent/guardian. These items were reverse coded and combined into a single measure where higher scores represent greater parental warmth (alpha = .82). Adolescents who feel more connected to and supported by their parents engage in less school misconduct (James et al., 2015).
School attachment consists of four items, as well. One question asked respondents how they felt about school at present, with answers that included (1) I like it a lot, (2) I like it a bit, (3) I don’t like it very much, and (4) I don’t like it at all. Three other questions asked the respondent if other students in their classes enjoy being together, are kind and helpful, and accept the respondent as they are, with answers ranging from (1) strongly agree to (5) strongly disagree. All four items were reverse coded and combined into a single measure where higher scores represent greater school attachment (alpha = .69). Students who have high levels of commitment to school engage in less misconduct (James et al., 2015; Stewart, 2003; Watts et al., 2019). Substance-using peers includes three combined items that measure a respondent’s peer’s substance use. The questions asked how many of a respondent’s friends smoke cigarettes, drink alcohol, and smoke/use marijuana, with answers including (1) none, (2) a few, (3) some, (4) most, and (5) all. The three items were combined into a single measure where higher scores denote more peer substance use/delinquency (alpha = .89). Peer delinquency is a major correlate of an individual adolescent’s own delinquency (Agnew, 1991; Haynie, 2002).
Analytic Strategy
To examine the potential mediating relationship between socioeconomic strain, bullying perpetration, and negative emotions, the current study utilizes path analysis, a form of stuctural equation modeling, within Mplus. Path analysis is particularly suited to the present study because it allows for the estimation of direct, indirect, and total effects simultaneously, which aids in claims of mediation. Traditional regression methods typically require multiple models to be estimated to give evidence for mediating or moderating effects, with potential mediators being effectively treated as covariates (Baron & Kenny, 1986). And with all relationships not modeled simultaneously, regression assumptions are violated (namely, independence of error terms and omitted variables), resulting in parameter estimates that are not efficient or valid (Freedman, 1987). By observing all of the relationships within a single model, we can better understand underlying causal processes (Iratzoqui & Watts, 2019).
Additionally, the Mplus software produces modification indices that can suggest pathways to improve model fit that were not originally specified. This ability of Mplus is of particular interest because in the present study we have proposed a direction (deviance to emotions, rather than emotions to deviance) that runs counter to the expectations of a highly validated theory in GST. The modification indices help in producing confidence in the causal claims being made, despite the data being cross-sectional. However, it should be noted that given the cross-sectional nature of the data, combined with a methodology that differs from a traditional experiment with a two-group, pre- and post-test design and random assignment, any findings pointing toward causality or mediation should be considered suggestive and preliminary.
Figure 1 presents the set of relationships modeled in the path analyzes, as well as the controls accounted for within each path. 4 Separate models were run for each of the three bullying variables. The models were estimated using maximum likelihood estimation with robust standard errors (MLR). This method is the most appropriate because the dependent variables are continuous but have non-normal distributions.
Empirical model.
The appropriate weighting variables were utilized in the path analyzes to account for the sampling design of the HBSC and to unbias coefficients. 5
Results
Table 2 presents results for the three path models run in Mplus. Each model is different only in that they model the effect on and of the three different bullying measures: cyberbullying, traditional bullying, and combined bullying. Goodness of fit for structural equation models is generally determined from several different indicators. Three commonly referenced indicators are the root mean square error of approximation (RMSEA), the Comparative Fit Index (CFI), and the Tucker Lewis Index (TFI) (Tankebe, 2013). 6 In general, good model fit gives a measure of how well the specified model reproduces the relationships as they exist in the true population of interest, and good model fit is thought to be indicated by an RMSEA < 0.06, a CFI > 0.95, and a TLI > 0.95 (Hu & Bentler, 1999). The models presented in Table 2 represent the best fitting models for the data.
The direct and indirect effects of socioeconomic strain and bullying on negative emotions.
Note. Standardized coefficients reported.
N = 11,579.
p < .01, ***p < .001.
Of note, the models support the proposed deviance to negative emotions pathway, rather than the negative emotions to deviance pathway traditionally proposed by GST. If the paths are flipped, with negative emotions predicting bullying, model fit is very poor. 7 Further, if reciprocal paths are specified between bullying and negative emotions, model fit is poor, and the net of the reciprocal effects between them suggests the effect is actually unidirectional, from bullying to negative emotions (results not presented).
Table 2 presents standardized coefficients, r-sq. values, and the model fit statistics for the three models. Results in Table 2 support the study hypotheses. Across the three models, socioeconomic strain consistently has a significant, positive correlation with bullying (supporting Hypothesis 1 and the traditional GST model); socioeconomic strain consistently has a significant, positive correlation with negative emotions (supporting Hypothesis 2 and the traditional GST model); bullying consistently has a significant, positive correlation with negative emotions (supporting Hypothesis 3, in opposition to the traditional GST model); and, as shown by significant indirect effects, evidence is given that bullying potentially partially mediates the relationship between socioeconomic strain and negative emotions (also supporting Hypothesis 3, in opposition to the traditional GST model).
Among the independent variables, all of the proposed relationships are statistically significant across all three models. For the socioeconomic strain-bullying relationship, the coefficients suggest socioeconomic strain has a stronger correlation with traditional bullying versus cyberbullying. The correlation between socioeconomic strain and negative emotions appears to be of similar strength across all three models. The relationship between bullying and negative emotions appears to be strongest in the case of traditional bullying. Where the control variables have statistically significant relationships, they are consistently in the expected directions.
The best-fitting model appears to be the cyberbullying model, while the model for traditional bullying has the most explained variance (highest r-sq. values) for both bullying and negative emotions, with there being more explained variance in negative emotions than bullying across all three models. Lastly, the indirect effect of socioeconomic strain on negative emotions through bullying appears to be greater for traditional bullying versus cyberbullying.
Discussion and Conclusion
This paper examines the relationship between perceived socioeconomic strain/resource deprivation, bullying perpetration, and negative emotions in a nationally representative sample drawn from the HBSC. Similar to past research in the bullying literature, GST is utilized as a theoretical framework for understanding how these variables relate. However, we propose hypotheses based on a re-specified version of the model proposed by GST, wherein bullying perpetration potentially shapes negative emotions, and not the other way around. This model and the associated hypotheses are supported in analyzes run in the Mplus software. Utilization of path modeling and Mplus adds confidence to the mechanisms identified, despite them going in part in the opposite direction proposed by a prominent criminological theory. The Mplus software identified the proposed model as the best fitting model for the data, and the traditional GST model of the independent and dependent variables and their relationship fit very poorly if specified in the way traditionally proposed by GST (strain-negative emotions-deviance, see footnote seven). The current study aligns with prior research that examines how socioeconomic strain is associated with negative emotions (Bøe et al., 2012; Gallo & Matthews, 2003; Huurre et al., 2007) and supports previous studies that have examined how negative emotions can occur as a result of bullying perpetration (Ahmed & Braithwaite, 2004; Menesini & Camodeca, 2008; Roberts et al., 2014), but expands these relationships further to fill the gap in the literature.
Results from the present study suggest several potential policy recommendations that could affect change among youths. First, youths need to be given the tools to reconcile stressful life circumstances at home or elsewhere without resorting to externalizing behaviors that cause them to take out their frustrations on their peers. The findings of this study and others suggest that schools should provide self-management skills through health education programing (De Wolfe & Saunders, 1995; Hampel et al., 2008; McCraty et al., 1999; Patchin & Hinduja, 2011). In-class teaching modules could potentially cover stress management while reminding students that mistreating others is an ineffective way to handle their feelings (Matheny et al., 1993; Miller et al., 1996; Patchin & Hinduja, 2011).
Second, since strain produces the need for corrective action (Thaxton & Agnew, 2004), it is better if this corrective action is positive rather than negative. Therefore, those who serve youth need to make positive outlets available at school and elsewhere as a means for adolescents to unburden themselves of their stress. These positive outlets can include extracurricular activities that tax youths physically and/or mentally, allowing them a way to find self-worth and satisfaction while pursuing personal hobbies and goals (Frydenberg & Lewis, 1993; Miller & McCormick, 1991; Patchin & Hinduja, 2011).
The results of this study expand on the broader literature on the causes and correlates of bullying behavior, particularly work by Patchin and Hinduja (2011). We observed the strain-negative emotions-bullying behavior relationship within a nationally representative sample, with some important theoretical controls accounted for, modestly filling in a few small gaps in the literature in the process. A significant difference of note in the findings between this study and Patchin and Hinduja (2011), among others, is that the present study did not rigidly adhere to the traditional GST model of strain-negative emotions-deviance. In proposing an alternative ordering that went somewhat against expectations of a prominent and validated theory of crime and delinquency, we uncovered a model that our statistical software supported as a superior fit for the data. This suggests that even time-tested and valid theories of crime and delinquency are worth revisiting and refreshing, as our “theoretical blinders” can sometimes cause us to miss how variables most accurately relate.
While this study contributes to the literature, limitations should be noted. The main limitations of this study revolve around the data set that was used and the measures it contains. The 2009 to 2010 version of the US HBSC dataset was the most recent version that was publicly available for data analysis. Cyberbullying has certainly advanced since this time period, therefore findings utilizing data more recently collected could better illustrate the relationship between strain, negative emotions, and cyberbullying in particular. Also, while the bullying measures in the HBSC have been validated (Roberson & Renshaw, 2018), it is worth noting as a limitation that they do differ from more widely accepted bullying measures (see Solberg & Olweus, 2003). Similarly, the measures used to reflect negative emotions are very narrow in how much of the concepts of depression, anger, and anxiety they measure. These concepts have many facets, and the current study utilizes measures that are low in content validity.
An additional limitation of this data set is the self-report nature of the survey. Individuals may have under- or over-reported their experiences with socioeconomic strain, bullying perpetration, and negative emotions. The bullying perpetration measures in particular may have led individuals to provide answers that are socially desirable (Brownfield & Sorenson, 1993), but there’s no way to know how much of this occurred and no independent way to verify the accuracy of the self-reports. In particular, the bullying perpetration measures make use of the term “bully,” which may automatically bias respondents from giving truthful answers if they do not define their behavior as bullying or understand that to be a highly loaded term. Future surveys and research would do well to avoid references to “bully,” “bullying,” and “bullies,” and instead focus on the behaviors that make up the definition of bullying behavior. This could result in more accurate reporting of both bullying behavior and bullying victimization estimates.
Finally, the cross-sectional design of this study should be noted as the biggest limitation. Given the overlap of time in the collection of the proposed independent, dependent, and mediating variables, strong claims to causality cannot be made with confidence based on these results, and they should be considered preliminary and suggestive. We have attempted to stay with references to correlation rather than causation throughout the sections describing the analysis and findings. Though the utilization of path modeling methods in Mplus could add to claims of causality, for all the reasons previously noted, these results should be considered initial and exploratory rather than as direct evidence for causality in the relationships under study.
In conclusion, this study provides evidence that socioeconomic strain correlates with bullying and negative emotions, and the relationships between these variables are best captured by a re-specified GST model within this data. Future research would do well to further tease out the various pathways uncovered in the present study with longitudinal data that will better serve claims of causality, and with measures of bullying and negative emotions that have higher content validity.
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) received no financial support for the research, authorship, and/or publication of this article.
