The Effect of Social, Verbal, Physical, and Cyberbullying Victimization on Academic Performance

ABSTRACT

Using data drawn from the 2015 School Crime Supplement to the National Crime Victimization Survey, we investigate whether specific types of bullying experienced by a youth influences his or her academic performance. The cross-sectional sample of adolescents is nationally representative and is composed of 4,610 middle and high school students ages twelve to eighteen (51% male, average age 14.7). Using General Strain Theory as a backdrop, we contribute to the extant literature by making an empirical distinction between social (also known as relational), verbal, physical, and cyberbullying victimization. Ordinal regression results show that while a composite measure of bullying victimization does attenuate a youth’s academic performance, most of this effect is due to social bullying victimization which remains robust notwithstanding a multitude of model specifications.

KEYWORDS:

• Adolescence
• bullying victimization
• social bullying
• academic performance
• ordinal regression
• relational bullying

Introduction

Research on bullying victimization in schools has developed into a robust body of literature since the early 1970s. Formally defined by Olweus (1994), “a student is being bullied or victimized when he or she is exposed, repeatedly and over time, to negative actions on the part of one or more other students and where a power imbalance exists” (p. 1173). Bullying can take the form of verbal, physical, social/relational, or cyberbullying and can engender numerous deleterious physical and emotional ancillary effects such as anxiety (Calvete, Fernandez-Gonzalez, & Gonzalez-Cabrera, 2018), depression (Moore et al., 2017), increased levels of aggression (Hawker & Boulton, 2000), and delinquency (Cullen, Unnever, Hartman, Turner, & Agnew, 2008) to name just a few.

While the occurrence of bullying in U.S. schools is not new, it has only recently become a public health issue due to its pervasive nature (Gladden, 2014). Recent findings from the Center for Disease Control and Prevention (2016) indicate that approximately twenty percent of high school students were bullied during school hours, and an equal amount were bullied in cyberspace. A reciprocated byproduct of widespread bullying victimization in schools is escapism and school avoidance behaviors among the student body (Astor, Benbenishty, Zeira, & Vinokur, 2002). For example, students who experience bullying often go out of their way to evade attending school by falsely claiming to be sick or by skipping their classes altogether to avoid further victimization (Rivers, 2000).

Due to the high frequency of bullying victimization, school districts potentially lose hundreds of thousands of dollars on an annual basis. Baams, Talmage, and Russel (2017) estimated that the state of California loses approximately 276 million dollars a year due to students missing school because they felt unsafe. Excessive absenteeism induced by bullying victimization also tends to decrease levels of academic performance among students (Gershenson, Jacknowitz, & Brannegan, 2017; McLeod, Uemura, & Rohrman, 2012). Academic achievement is an integral developmental variable in students’ lives as it is considered a critical measure of future success (Strenze, 2015) and plays a notable role in the career path they choose to pursue (Schnabel, Alfeld, Eccles, Koller, & Baumert, 2002; Strenze, 2007).

The current study specifically focuses on the effect of bullying victimization on academic achievement among adolescents (ages 11 to 18) and is guided by the theoretical mechanisms established by General Strain Theory (Agnew, 1992). Adolescence encompasses both the middle and high school years and serves as one of the most critical periods of psychosocial change in a student’s life. Not only does adolescence result in abrupt developmental and personality changes, but also changes in the perpetration of bullying. Direct forms of bullying victimization often decrease between the ages of eleven and nineteen because as children ascend from primary school, their bullying methods become subtler and more complex (Scheithaur, Hayer, Petermann, & Jugert, 2006). Indirect forms of social/relational bullying become much more pervasive than direct physical bullying during this period (Bureau of Justice Statistics, 2015). With the apparent emergence of different forms of bullying during the teenage years and the potential variability in the diverse types of bullying an individual can experience, it stands to reason that not all forms of victimization impact academic achievement in a similar manner.

Literature review

Defining verbal, social, physical, and cyberbullying victimization

Bullying occurs when someone takes an adverse action against another that inflicts intentional harm or discomfort (Olweus, 1994). The method of delivery, however, can substantially vary from slapping, name-calling, exclusion from groups, or even harassment/embarrassment on social media. Early research on the phenomenon of bullying spearheaded by Dan Olweus, has led to a more robust and systematic classification of the unique types of bullying students can experience (Olweus, 1994; Olweus & Limber, 2010a, 2010b). The Olweus Bullying Questionnaire (OBQ) is an extensively tested and effective survey instrument used to assess the prevalence and severity of bullying within schools and provides a foundation for formally operationalizing the unique forms of bullying victimization (Olweus & Limber, 2010a; Solberg & Olweus, 2003).

Since its most recent revision, the OBQ stratifies eight unique bullying types which include being verbally harassed or called names, being excluded from a group or being ignored, being bullied in a physical manner (pushing, kicking, punching, spitting), being the subject of rumors, having property destroyed or taken, being threatened or forced to do things, being the subject of racial discrimination, and being harrassed about sexual orientation (Solberg & Olweus, 2003). The continued development and classification of these bullying experiences has led other major survey instruments such as the National Crime Victimization Survey: School Crime Supplement (Bureau of Justice Statistics, 2015) and various forms of school climate surveys (Birkett, Espelage, & Koenig, 2009; Gage, Prykanowski, & Larson, 2014) to collect bullying victimization data in a similar, if not identical fashion.

Due to the extensive typology of bullying experiences recorded on the OBQ, contemporary researchers tend to cluster these items under verbal, physical, social, and even cyberbullying victimization to better formulate and test their hypotheses. It is important to note that while the types of bullying laid out in the OBQ are comprehensive, they are not always grouped and defined in the same manner in previous studies. For example, physical forms of bullying are often agreed upon in the current body of literature as being composed of any physical interaction between the bully(s) and the victim (Bradshaw, Waasdorp, & Johnson, 2015). There is, however, a lack of agreement on the operational definitions of verbal and social bullying (Underwood, 2003).

The bulk of studies define verbal bullying as all instances when gestural or oral aggression are administered as a means to threaten or inflict harm on a victim (Guerin & Hennessy, 2002; Ross & Horner, 2009). Social bullying on the other hand, often refers to any instance where bullies use indirect methods to exclude their victims from activities, spread rumors about them, or get other students to ignore them (Woods & Wolke, 2004). Lastly, when cyberbullying is defined as a type of bullying (as it is in the current study) rather than an environment (bullying occurred online rather than at school), it represents harassment that occurs across an electronic platform such as text messages, e-mails, or social media (Ybarra, Boyd, Korchmaros, & Oppenheim, 2012). The current study uses the above-listed operational definitions derived from the literature for parsing out the specific types of bullying, which allows for a more comprehensive analysis of the association between bullying victimization and a student’s academic achievement.

Prior research on bullying victimization and academic achievement

The nexus between bullying victimization and attenuated academic performance is well-documented. One often proffered position in the literature maintains that a youth’s academic performance suffers when a bully victimizes him or her. A meta-analysis (Nakamoto & Schwartz, 2010) along with several empirical studies produce support for the position that bullied youth have an enhanced proclivity for doing poorly in school (Dijkstra & Gest, 2015; Hammig & Jozkowski, 2013; Juvonen, Wang, & Espinoza, 2011; Lacey, Cornell, & Konold, 2017; Smith & Skrbis, 2016). While certainly informative, prior research has still tended to muddle the relationship between bullying victimization and academic achievement for several distinct reasons.

First, several studies focus on only one type of bullying victimization. To illustrate, Hammig and Jozkowski (2013) conducted a cross-sectional analysis to test whether violent victimization/bullying or the threat of violence exclusively affected academic achievement. Using a multinomial logistic regression model, they found that the academic performance of both male and female adolescents was influenced adversely by past bullying victimization. Students who earned mostly Ds and Fs experienced bullying victimization more often than those who earned As and Bs (Hammig & Jozkowski, 2013). While Hammig and Jozkowski’s study is a worthy contribution to the extant body of literature, their analytical prioritization on physical forms of bullying along with physical injury and intimate partner violence may have suppressed the stratified effects of each specific form of bullying victimization on academic achievement. Similarly, in an investigation of the effect of cyberbullying on academic achievement and other outcome variables, Li (2007) found that cyberbullying was weakly related to academic achievement. The potential effects of other forms of bullying victimization on academic achievement were not assessed.

The second way that prior research muddles the relationship between bullying victimization and academic achievement is that the different forms of bullying are often amalgamated into one composite measure (Arseneault et al., 2006; Eisenberg, Neumark-Sztainer, & Perry, 2003; Holt, Finkelhor, & Kantor, 2007; Iyer, Kochenderfer-Ladd, Eisenberg, & Thompson, 2010). For example, researchers often combined social bullying and verbal bullying (Arseneault et al., 2006), verbal bullying and physical bullying (Holt et al., 2007; Hoover, Oliver, & Hazler, 1992), or simply distinguish between cyberbullying and traditional forms of bullying (Beran & Li, 2007; Hinduja & Patchin, 2010). A potential problem with merging several different types of bullying into one comprehensive measure of victimization is that the specific type of bullying a youth experiences may have differential effects on the youth’s academic performance. This specific issue is dealt with in the current study because the effect of each specific form of bullying victimization on academic performance undergoes empirical investigation.

Third, while there is an abundance of support for the relationship between bullying victimization and academic achievement, some social scientists remain unconvinced that bullying victims have an enhanced tendency to perform poorly in academics because much of the negative relationship between bullying victimization and academic performance may be the result of omitted variable bias (Hanish & Guerra, 2002; Kowalski, Giumetti, Schoeder, & Lattanner, 2014). It is not bullying victimization that is salient in predicting a youth’s academic performance, but rather, demographic and school-related factors. For example, a study conducted by Woods and Wolke (2004) found no salient relationship between direct forms of bullying, such as verbal and physical bullying and academic achievement. Instead, they found that indirect/relational forms of bullying victimization did influence academic performance in a negative direction. Their results lead them to conclude that school size, class size, urbanization, the socioeconomic status of the school, and peer-rejection predicted academic achievement with more certainty than direct forms of bullying victimization (Woods & Wolke, 2004).

Ancillary effects of bullying victimization that lower academic achievement

A considerable amount of research has documented a variety of ancillary effects that are caused by bullying victimization and have a direct impact on a student’s scholastic abilities. While not an exhaustive list, these effects include psychosocial maladjustment, school engagement, and low levels of self-efficacy. Psychosocial maladjustment refers to a student’s inability to adjust to negative social stimuli in a positive/healthy manner (Nansel et al., 2001). Specific forms of psychosocial maladjustment include aggression, alcohol usage, smoking, fighting, and loneliness to list just a few (Hawker & Boulton, 2000; Nansel et al., 2001). These maladjusted behaviors have been shown in prior research to influence academic performance negatively (Glew, Fan, Katon, Rivara, & Kernic, 2005; Holt et al., 2007; Juvonen, Nishina, & Graham, 2000; Nansel et al., 2001; Nishina, Juvonen, & Witkow, 2005).

Another critical characteristic in explaining why bullying victimization affects academic performance resides in the level of school engagement. The school not only serves as a place for learning but also for youth to gather and to interact socially with other classmates on a daily basis. When a bully victimizes an adolescent, however, he or she may begin to develop a negative attitude toward school that ultimately causes the student’s level of classroom engagement to plummet (Iyer et al., 2010; Kochenderfer & Ladd, 1996; Ladd, Kochenderfer, & Coleman, 1997). Classroom engagement is an indicator of a student’s willingness to participate in class activities (which yield performance scores) and to abide by the social rules established for the classroom (Buhs & Ladd, 2001; Ford, 1985; Wentzel, 1991). Therefore, when students are bullied, classroom engagement naturally declines along with academic performance (Ladd, Birch, & Buhs, 1999).

Finally, bullying victimization is linked to decreased levels of self-efficacy, which has been shown to maintain a strong negative relationship with academic performance (Andreou & Metallidou, 2004; Kokkinos & Kipristsi, 2012; Natvig, Albrektsen, & Qvarnstrom, 2001). According to Lane, Lane, and Kyprianou (2004), self-efficacy can be defined as “the confidence individuals have in their ability to execute certain courses of action or achieve specific outcomes” (p. 247). Thus, when a bully victimizes a student, the experience teases out the functional and beneficial components found within higher levels of self-efficacy.

General strain theory

Despite the existence of several diverse explanations for the observed negative relationship between bullying victimization and poor school performance, one theoretical framework that seems particularly well suited to clarify this association, as well as the relationship between bullying victimization and nonacademic outcomes (Alavi et al., 2016; Juvonen & Graham, 2014), is Agnew’s (1992) General Strain Theory (GST). Like other stress-based theoretical perspectives, GST maintains that bullying victimization encourages a wide range of negative emotions among victimized youth that must be mollified in some manner. Specifically, bullying victimization introduces negatively valued/noxious stimuli into the victim’s life, which in turn creates intense levels of stress or strain. The youth who experience these stressors then require a means of coping. Hence, in a similar fashion to psychosocial maladjustment, these stressors serve as a conduit for social adaptations, which can lead to a decline in academic performance. These adapting behaviors, which are employed to assuage the negative emotions engendered by being the victim of a bully, can include escapism (e.g., school avoidance behavior), engagement in disruptive school behavior, or acting aggressively toward others (Kochenderfer & Ladd, 1997; Rigby, 1996; Smith, Talamelli, & Cowie, 2004).

While the above-listed studies on GST have established a recursive relationship between bullying and adverse outcomes, they have almost exclusively focused on direct and cyber forms of bullying victimization. This creates a dearth of knowledge on the unique influence of relational victimization (social bullying), which is more subtle than the other methods of bullying (Scheithaur et al., 2006). The parsing out of the specific types of bullying in the current analysis may serve to advance the theoretical knowledge on GST as the direct observation is made on whether a student experienced relational forms of bullying victimization. To clarify, some researchers claim that victimization experiences such as social exclusion and the spreading of rumors yield a unique category of strain known as relational strain.

Formally defined by Gallupe and Baron (2009), “relational strain involves any interpersonal situation resulting in negative affect. This includes such things as death, relationship severance, social unpopularity, breaches of trust, victimization by trusted others, and problems maintaining interpersonal relations” (p. 525). Research on relational strain originated with Agnew and Brezia (1997), and it remains an understudied niche of strain theory with prior studies focusing exclusively on gender differences in relational strain (Agnew & Brezia, 1997; Joon Jang, 2007; Watt & Sharp, 2001). It makes logical sense that levels of relational strain are higher among bullying victims as social bullying is more pervasive and difficult to control than verbal, physical, and cyberbullying (Owens, Shute, & Slee, 2000; Scheithaur et al., 2006). To the best of our knowledge, no published study has investigated the association between relational strain and academic achievement.

The current study

Using GST as a theoretical backdrop, the current study is designed to investigate the associations between the unique types of bullying victimization and academic achievement. Specifically, the goals of the current analysis are to (1) explore the unique relationship between verbal, social, physical, and cyberbullying victimization and self-reported grades; and (2) explore whether relational strain induced by forms of social bullying associate with academic performance. To accomplish these goals, we specifically use ordinal regression to probe the potentially unique effects of each type of bullying victimization on academic performance in a nationally representative sample of middle and high school students. The controls included in the current study account for a variety of factors identified in prior research as being salient for predicting a youth’s academic performance. These control variables help us to avoid basing our conclusions derived from the analysis on spurious or suppressed relationships.

Data & methodology

Data for the current study derive from the 2015 School Crime Supplement (SCS) of the National Crime Victimization Survey (NCVS). Using both computer-assisted telephone interviewing and computer-assisted personal interviewing, the SCS is administered after completion of the NCVS when a valid middle or high school student resides in the household. The SCS sample consisted of 4,768 students, and listwise deletion was used to deal with missing data. The final sample in the current analysis is composed of 4,610 students who were interviewed for various criminal victimizations and is nationally representative of the twelve to eighteen-year-old population in the United States (grades 6 through 12). The SCS has a weighted response rate of 58%. Demographically, 51% of the final sample consists of males and 49% females. The racial and ethnic makeup of the sample is 81% White, 12% Black, and 5% Asian. Hispanics represent 24% of the sample.

Dependent variable

Academic achievement is the dependent variable. Students were asked to identify the academic grades they mostly earned during the current school year (Bureau of Justice Statistics, 2015). Within the statistical model, the outcome variable of the students’ grades ordinally ranks as mostly As (coded 5), mostly Bs (coded 4), mostly Cs (coded 3), mostly Ds (coded 2), and mostly Fs (coded 1). Within each grade category, 42% reported earning mostly As, 42% reported earning mostly Bs, 14% reported earning mostly Cs, 1% reported earning mostly Ds, and 1% reported earning mostly Fs.

Bullying victimization variables

There are two bullying victimization variables of theoretical interest in this study. These variables include type of bullying victimization and frequency of bullying victimization. The first variable is dichotomous and measures the specific kind of bullying victimization experienced by an adolescent. These types include verbal, social, physical, and cyberbullying victimization and were composed of fourteen questions on the NCVS. Victims of verbal bullying answered yes to any survey question relating to an incident where someone made fun of them, called them names, insulted them in a hurtful way, or threatened them with harm. Victims of social bullying responded affirmatively to any survey question where someone spread rumors about them, tried to make people dislike them, or purposely excluded them from activities. Those youth who experienced physical bullying answered yes to any survey question intimating whether someone pushed them, shoved them, tripped them, spit on them, tried to make them do things that he or she did not want to do, or destroyed his or her property on purpose. Lastly, victims of cyberbullying answered yes to the question of whether any bullying they experienced occurred online or by text.

The second variable of interest is the frequency of the specific type of bullying experienced by a victim. The frequency of bullying is relevant because studies suggest that repeated bullying victimization has a more significant negative impact on a youth than more sporadic occurrences of bullying (Kochenderfer & Ladd, 1996; Kochenderfer-Ladd & Wardrop, 2001). The frequency of bullying victimization was measured as a direct observation of whether the youth was bullied once or twice a month, once or twice a week, or whether the bullying occurred almost daily. To quantify this data, several dummy coded variables were created to represent the different types of bullying victimization.

Control variables

A variety of factors are reported to influence a youth’s academic performance. These control variables are included in the analysis to help ensure that any observed relationship between the specific type of bullying victimization and academic performance is not spurious. Several control variables pertain to extracurricular activities, which take the form of participation in athletic clubs, spirit groups, performing arts, academic clubs, student government, and volunteering/community service clubs. Studies report that extracurricular activities are predictive of a youth’s academic performance due to increased levels of school engagement/involvement (Guest & Schneider, 2003; Silliker & Quirk, 1997). Economic disadvantage is another predictor of academic achievement in school (Considine & Zappala, 2002). The current study measures economic disadvantage with household income stratified by several dummy coded variables measuring income less than 19,000 dollars a year (reference category), income between 20,000 and 39,000 dollars a year, income between 40,000 and 74,000 dollars a year, and income 75,000 dollars or more a year.

Demographic factors such as a youth’s sex (Matthews, Pontiz, & Morrison, 2009), race and ethnicity (Kao, 1995; Ogbu, 1987), and age (Moss & St-Laurent, 2001) are also used as control variables because they have been shown in prior research to correlate with academic achievement. Race is composed of two dummy coded variables White and Asian with Black set to the reference category. Lastly, a measure of whether a student envisions a four-year degree is added to account for additional variation in the outcome variable. Research conducted on both middle school (Eccles, Vida, & Barber, 2004) and high school students (Boatwright, Chin, & Parr, 1992) found that the plan to enroll in college/university buttresses academic competence and performance. Table 1 reports the descriptive statistics for all variables used in the analysis.

Table 1. Descriptive statistics

Variable		Mean	Std. Dev.
Student academic achievement		4.236	.778
Student Experienced:
Verbal Bullying^a		.068	-
Social Bullying^a		.069	-
Physical Bullying^a		.035	-
Cyber Bullying^a		.018	-
Student is a member of:
Athletic Club^a		.379	-
Spirit Group^a		.082	-
Performing Arts^a		.281	-
Academic Club^a		.195	-
Student Government^a		.054	-
Volunteer Club^a		.151	-
Less than 19k^a		.103	-
20k to 39k^a		.175	-
40k to 74k^a		.216	-
75k and above^a		.321	-
White^a		.816	-
Black^a		.123	-
Asian^a		.049	-
Hispanic^a		.236	-
Env. 4-yr degree^a		.962	-
Age		14.732	1.857
Sex^a		.512	-

Note. N = 4,610. Std. Dev. = Standard deviation. Env. 4-yr degree = Student Envisions a four year degree. ^aMeans for variables that are dichotomously coded can be interpreted as proportions.

Analytical strategy

We use a cumulative ordered logit model, more commonly known as ordinal regression, to assess the association between the unique types of bullying victimization and academic achievement. While Ordinary Least Squares (OLS) regression is often used to measure the differences in ordered categorical data, such analyses must ensure that the outcome variable falls within a Gaussian distribution so as not to violate the assumption of normality. When the outcome variable has seven or more ordered categories and a symmetrical distribution, the bias generated by treating categorical outcomes as continuous remains relatively small (Bollen & Barb, 1981; Hox, 2010; Johnson & Creech, 1983). In the current analysis, however, our outcome variable is composed of five categories, and the distribution is not normal. To adjust for this nonlinearity, ordinal regression undergoes a unique iterative process wherein each category of the outcome variable generates a unique intercept called a threshold. Each threshold represents a cutoff point between every level of academic achievement, which allows for the modeling of consistent effects between each independent variable and the outcome variable in the regression equation. Much like multinomial and binary logistic regression, the intercepts generated in ordinal regression require a comparison category to generate parameter estimates and meaningful interpretations of the coefficients (k-1).

Due to the encoding of the outcome variable, the highest value (5 – which represents mostly As) is held for comparison to the other thresholds and no intercept is generated for this category. Additionally, threshold 1 represents the intercept for mostly Fs, threshold 2 – mostly Ds, threshold 3 – mostly Cs, and threshold 4 – mostly Bs. Given that the cumulative ordered logit model assumes the slope of each explanatory variable remains parallel across all thresholds, holding the grade of mostly As is an intuitive reference category for assessing the effects of bullying victimization on academic achievement. To ensure that our models meet this proportional odds assumption, a test of parallel lines is conducted for each ordinal regression equation estimated.

A stepwise procedure is used to examine the relationship between bullying victimization and academic achievement. As previously mentioned, prior research suggests that the direct relationship between bullying and lower levels of academic achievement is relatively weak (Kowalski et al., 2014) and that demographic covariates may hold more explanatory power. The stepwise method will allow for the modeling of a pseudo coefficient of determination and will test whether our final model specifications remain robust, notwithstanding the addition of more influential explanatory variables. Additionally, goodness-of-fit for each ordinal regression model in the stepwise procedure will include a Pearson statistic and the likelihood ratio (deviance) statistic to determine whether the covariates properly fit the specified model.

Model 1 represents the baseline equation in the stepwise ordinal regression analysis and includes the stratified types of bullying victimization. Model 2 then adds the covariates for extracurricular activities to capture whether school engagement/involvement affects the bullying variables and their relationship with academic achievement. Model 3 includes the covariates for household income to measure the influence of financial disadvantage. Model 4 then completes the ordinal regression equation by adding the demographic covariates and represents the equation used for the main discussion throughout the article. It is important to note that collinearity diagnostics reveal that no variance inflation factors exceed 2 and tolerance levels do not fall below .20, which indicates multicollinearity is not problematic in Model 4. Models 5, 6, and 7 display the direct observation of whether the types of bullying victimization were experienced monthly, weekly, or daily.

Finally, a supplemental ordinal regression analysis is conducted to ensure that our original findings remain robust across more complex specifications.¹ Our initial analysis assumes that each form of bullying victimization had a unique effect on academic achievement (as depicted in Model 4). Nevertheless, it is also plausible that interaction effects exist among the different types of bullying victimization (Holt et al., 2007). A failure to consider such an effect could lead to a misspecified model and biased parameter estimates. Model 8 includes six unique two-way interaction terms, all the original variables of theoretical interest, and the control variables. The interaction terms based on the types of bullying victimization were (1) Verbal × Social; (2) Verbal × Physical; (3) Verbal × Cyber; (4) Social × Physical; (5) Social × Cyber; and (6) Physical × Cyber. There was no indication that multicollinearity vitiated our results because none of the variance inflation factor scores exceeded a value of four in Model 8.

It is important to note that the ordinal regression coefficients generated in Models 1–8 maintain four unique interpretations by holding the reference category constant. Specifically, the logit estimate for each independent variable can be interpreted exclusively for every level of academic achievement. These initial estimates, however, do not provide clear interpretive values. For a meaningful interpretation, we calculate the exponentiated beta coefficient for each logit estimate, which transforms them into an odds ratio. These odds ratios are then converted into a percentage using the following formula: 100(e^b–1). To clarify, the logit estimate for social bullying in Model 4 is −.365, and its exponentiated function is .694. Implementing the formula listed above: 100(.694–1) = −31 percent.

Results

Table 2 reports the results of the four regression equations that compose the stepwise procedure. Model 1, which includes only the bullying victimization variables, reveals that the overall impact is relatively low as it explains approximately four percent of the variability in the dependent variable (pseudo R² = .04). This preliminary finding aligns with prior research (Hanish & Guerra, 2002; Kowalski et al., 2014) that suggests demographic covariates maintain more influence over a student’s academic performance than bullying victimization. A visual inspection of Table 2 reveals that only physical bullying is statistically significant and has a negative relationship with academic achievement.

Table 2. Stepwise ordinal regression equations

	Model 1 bullying victimization		Model 2 extracurricular activities		Model 3 household income		Model 4 demographics
Variable	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Verbal bullying	−.141	.141	−.118	.147	−.080	.145	−.097	.147
Social bullying	−.062	.140	−.307*	.146	−.306*	.145	−.367**	.147
Physical bullying	−.442*	.176	−.311	.183	−.295	.180	−.264	.182
Cyber bullying	−.010	.218	−.225	.222	−.223	.225	−.338	.228
Athletic clubs			.423***	.060	.353***	.061	.439***	.062
Spirit groups			.178	.109	.190	.111	.034	.114
Performing arts			.555***	.066	.512***	.067	.368***	.069
Academic clubs			1.057***	.082	1.010***	.082	1.013***	.084
Student government			.754***	.153	.707***	.155	.714***	.157
Volunteer clubs			.502***	.087	.478***	.088	.475***	.090
20 k to 39 k					−.052	.085	.023	.087
40 k to 74 k					.214**	.081	.218**	.082
75k and above					.593***	.076	.505***	.078
White							.404***	.087
Asian							1.544***	.172
Hispanic							−.330***	.072
Env. 4-yr degree							.893***	.149
Age							−.080***	.016
Sex							−.612***	.061
Threshold 1 – F	−5.289***	.205	−4.844***	.206	−4.701***	.210	−5.183***	.372
Threshold 2 – D	−3.929***	.105	−3.479***	.108	−3.335***	.116	−3.808***	.328
Threshold 3 – C	−1.714***	.041	−1.222***	.049	−1.069***	.064	−1.495***	.314
Threshold 4 – B	.303***	.031	.953***	.046	1.135***	.063	.813*	.313
Pseudo R^2 Naglekerke	.040		.130		.146		.204
Pearson Chi-Square	95.249***		1182.414***		2147.146***		7982.191
Deviance Chi-Square	56.810		607.916		1195.494		4595.137

N = 4610. SE = Standard error. Env. 4-yr degree = Student Envisions a four year degree. All models tested for robust standard errors.

*p < .05. **p < .01. ***p < .001 (Two-tailed test).

This model, however, must be viewed with a healthy skepticism for two reasons. First, the goodness-of-fit statistics reveal that the model is only partially appropriate, and the results may be mixed. Specifically, the Pearson statistic returns statistically significant, and the deviance statistic is not. This is an indication that more variables are needed to understand the relationship between bullying victimization and academic achievement adequately. Second, this baseline model does not pass the test of parallel lines, which means that the interpretations of physical bullying do not remain the same across all ordered categories of the dependent variable.

Model 2 then adds the covariates for extracurricular activities. By adding these variables to the equation, the pseudo R² drastically increases, and 13 percent of the variation in academic achievement is explained. As expected, the bullying victimization variables alter in levels of significance due to the increase of explanatory power in the equation. In Model 2, social bullying yields a statistically significant negative relationship with academic achievement instead of physical bullying while verbal and cyberbullying remain non-significant. Additionally, five out of the six covariates added to the equation to measure school engagement/involvement return a statistically significant positive relationship with academic achievement. The fit statistics display that Model 2 passes the test of parallel lines and is a better fit than Model 1, but more variables are still needed.

Model 3 adds the covariates representative of household income that measures financial disadvantage. The addition of these variables increases the amount of variation explained from 13 percent in Model 2 to approximately 15 percent, and the model passes the test of parallel lines. Social bullying remains statistically significant, along with the covariates added in Model 2 and the direction of the relationships with academic achievement remain the same. Additionally, two of the three household income covariates return with a salient relationship. Students who belong to a household that earns between 40,000 and 74,000 dollars a year and those who belong to a household that earns over 75,000 dollars a year maintain a positive relationship with academic achievement. Again, the fit statistics dictate that Model 3 yielded a substantially improved equation over Model 2, but results still may remain mixed, and more variables are needed to capture the full outcome.

Finally, Model 4 includes the demographic covariates which complete the full ordinal regression equation intended to estimate the impact of the control variables and the specific types of bullying victimization on academic achievement. An examination of the fit statistics reveals that both the Pearson and deviance statistics return nonsignificant, which suggests that the observed data adequately fits the model. The pseudo R² of the final model increases substantially and dictates that approximately 21 percent of the variation in the dependent variable is explained. Additionally, Model 4 passes the test of parallel lines. Notwithstanding the addition of more powerful explanatory covariates, social bullying maintains a statistically significant negative relationship with a student’s academic performance. Specifically, when students report that social bullies perpetrated their victimization, the log odds of them earning any grade other than mostly Fs is 31% lower. Somewhat surprisingly, the effects of the other forms of bullying victimization are trivial in magnitude and not of substantive importance. The strong effect of social bullying on academic achievement is consistent with the results reported in previous studies (Buhs, Ladd, & Herald, 2006; Morrow, Hubbard, & Swift, 2014; Woods & Wolke, 2004).

In reference to the control variables, all the extracurricular activity variables have a substantive effect on academic achievement with the exception of spirit groups. The log odds of a youth earning above-average grades is 64% higher if he or she participates in an athletic club, 44% higher for participation in performing arts clubs, 2.7 times higher for partaking in academic clubs, two times higher for belonging to student government associations, and 61% higher for participating in community service/volunteer clubs. In comparison to students who live in a household that earns under 19,000 dollars a year, the log odds of a student earning above-average grades is 26% higher when they live in a household that earns between 40,000 and 74,000 dollars a year and 66% higher when they belong to a household earning over 75,000 dollars a year.

Examination of the demographic variables shows that females, younger students, Asian, and White students perform better academically. The log odds of a male student earning any other grade other than mostly Fs are 46% lower than for a female student. As the age of the student increases, the likelihood of performing poorly in school also rises. For every one-year increase in age, the log odds of a student earning above-average grades decrease by approximately 8%. The log odds of an Asian student earning above-average grades are 4.6 times higher than for Black students. The log odds of a White student earning above-average grades are 50% higher than for Black students. The log odds of a Hispanic student earning any other grade other than mostly Fs are 28% lower than for those who are not Hispanic.

Finally, for students who envision themselves graduating from a four-year college degree program, the log odds of them earning above-average grades are 2.4 times higher. It is important to recognize that because 96% of the sample answered the question in the affirmative, this finding may be due in part to students over-reporting higher grades. Nevertheless, because an analysis without this variable produced no meaningful changes in the results, we felt it appropriate to keep the variable in the analysis because of theoretical considerations.

Models 5 through 7 in Table 3 examine whether the frequency of a specific type of bullying victimization associated with academic achievement. Frequency is measured as a direct observation of whether the victim experienced monthly, weekly, or daily instances of bullying. The results for these three models are virtually identical to those reported in Model 4. The social bullying variable again achieves statistical significance in all three equations. The log odds of a youth earning any other grade other than mostly Fs is 30% lower if he or she experiences monthly social bullying, 30% lower if he or she experiences weekly social bullying, and 32% lower if he or she experiences daily social bullying. The effects of the control variables displayed in Models 5, 6, and 7 are nearly identical to those reported in Model 4.

Table 3. Ordinal regression results for frequency of occurrence and interaction effects

	Model 5 monthly occurrences		Model 6 weekly occurrences		Model 7 daily occurrences		Model 8 interaction effects
Variable	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Verbal bullying	.018	.160	−.092	.153	−.113	.148	−.283	.193
Social bullying	−.360**	.158	−.361*	.152	−.384**	.148	−.573**	.195
Physical bullying	−.318	.207	−.299	.192	−.108	.188	−.351	.325
Cyber bullying	−.453	.272	−.381	.246	−.238	.238	−.600	.345
Athletic clubs	.441***	.062	.439***	.062	.434***	.062	.439***	.063
Spirit groups	.037	.114	.036	.114	.035	.114	.025	.114
Performing arts	.367***	.069	.368***	.069	.371***	.069	.373***	.069
Academic clubs	1.018***	.084	1.015***	.084	1.009***	.084	1.019***	.085
Student government	.710***	.157	.713***	.157	.723***	.157	.723***	.157
Volunteer clubs	.475***	.090	.473***	.090	.472***	.090	.478***	.090
20 k to 39 k	.022	.087	.022	.087	.019	.087	.023	.087
40 k to 74 k	.216**	.082	.218**	.082	.215**	.082	.213**	.082
75k and above	.508***	.078	.505***	.078	.506***	.078	.503***	.078
White	.405***	.087	.404***	.087	.408***	.087	.401***	.087
Asian	1.545***	.172	1.540***	.172	1.546***	.172	1.545***	.172
Hispanic	−.330***	.072	−.331***	.072	−.329***	.072	−.330***	.072
Env. 4-yr degree	.888***	.149	.894***	.149	.882***	.149	.886***	.149
Age	−.080***	.016	−.079***	.016	−.080***	.016	−.080***	.016
Sex	−.610***	.061	−.612***	.061	−.611***	.061	−.614***	.061
Verbal × Social							.581	.314
Verbal × Physical							−.116	.395
Verbal × Cyber							.014	.565
Social × Physical							−.026	.404
Social × Cyber							−.081	.570
Physical × Cyber							1.595*	.654
Threshold 1 – F	−5.192***	.372	−5.179***	.372	−5.234***	.373	−5.217***	.373
Threshold 2 – D	−3.817***	.328	−3.804***	.328	−3.837***	.328	−3.841***	.329
Threshold 3 – C	−1.503***	.314	−1.491***	.314	−1.504***	.314	−1.525***	.315
Threshold 4 – B	.807**	.314	.818**	.314	.808*	.314	.787	.315
Pseudo R^2 Naglekerke	.204		.204		.208		.206
Pearson Chi-Square	7892.320		8041.660		7859.916		8073.132
Deviance Chi-Square	4593.861		4600.173		4583.571		4583.416

N = 4610. SE = Standard error. Env. 4-year degree = Student Envisions a four year degree. All models tested for robust standard errors.

*p < .05. **p < .01. ***p < .001 (Two-tailed test).

Model 8 displays the results for the supplemental ordinal regression analysis used to measure the degree to which multiple bullying victimization experiences affected levels of academic achievement. With the addition of all six interaction terms, social bullying still maintained a statistically significant negative association with academic achievement. The only interaction term that was statistically significant was PhysicalXCyber, but the logit estimate was in the opposite hypothesized direction. The addition of these interaction effects was intended to not only measure the effects of multiple forms of victimization but also to examine whether these interactions made a substantial increase to the amount of variation explained in the outcome variable. The pseudo R² for Model 8 only increases by .001 when compared to the general equation displayed in Model 4. Therefore, the supplementary ordinal regression analysis provides little evidence that there are interaction effects among the different forms of bullying victimization within the sample. It does, however, provide strong support for the negative relationship that social bullying maintains across seven different equations and that it is not a product of model misspecification.

Discussion

The current study was designed with two specific goals: (1) to explore the unique relationship between verbal, social, physical, and cyberbullying victimization and self-reported grades; and (2) to explore whether relational strain induced by forms of social bullying associate with academic performance. The findings generated in this analysis show that the specific type of bullying experienced by a victim does matter in predicting academic performance. All of the effect of bullying victimization on academic achievement is engendered by social bullying rather than by verbal, physical, or cyberbullying. The log odds of a student earning any other grade other than mostly Fs are 31% lower when he or she is a victim of social bullying. GST helps explain this relationship because the experience of social bullying introduces harmful stimuli into a victims life which must be placated in some manner. Moreover, the behaviors in which victims cope with the relational strain brought on by social bullying can decrease their overall academic performance.

This is intuitively sensible as prior research shows that social bullying tends to be more widespread than verbal, physical, or cyberbullying for the age range of the youth analyzed in this study (Scheithaur et al., 2006). Also, social bullying is more challenging to control than other forms of bullying because it often occurs in group settings where the opportunity to harass a victim is enhanced (Owens et al., 2000). Given the subtle nature of social bullying, the results produced from our ordinal regression analysis have germane implications for policy, theory, and future research on bullying victimization.

To the extent of our knowledge, the current study is the first to look into the association between relational strain and adverse academic outcomes induced by social bullying victimization. Given that social bullying maintains a robust relationship with academic achievement net of a variety of specifications to our ordinal regression models, it supports the idea that the victims’ coping mechanisms that affect academic performance stem from relational strain. Our exploration of GST under this lens provides additional insight into the theoretical assertations of relational strain as it relates to bullying victimization. While not an explicit test of the theory, our findings suggest that relational forms of victimization may yield unique coping methods that physical or verbal bullying victimization otherwise would not. Future research should focus on discovering which coping mechanisms students are most likely to engage in as a result of experiencing relational strain in comparison to other forms of victimization. Further understanding of this relationship will allow for a more precise explanation for the incongruous findings found in the literature between bullying victimization and academic achievement.

The acknowledgment that relational strain is ubiquitous among adolescents and that it adversely associates with academic achievement can serve as a theoretical conduit to enhance practical school intervention programs that combat social/relational bullying. Specifically, researchers Moon and Morash (2013) conducted an analysis where the focus was on GST as a theoretical foundation for the design of school bullying interventions. Their main argument dictated that even though GST was well established, and that Agnew (2010) laid out detailed programmatic implications, they were not functionally applied in the school environment. To briefly expand on this perspective, prior research suggests that most bullying reduction programs implemented within schools do not focus on a specific type of bullying (Ttofi & Farrington, 2011). Instead, a single program addresses several different types of bullying simultaneously to reduce the overall occurrence of victimization. Such a situation is problematic not only because each specific form of bullying requires individual attention to prevent (Ttofi & Farrington, 2011), but a general anti-bullying program might reduce some forms of bullying while amplifying others (Ortega, Del Ray, & Mora-Merchan, 2004).

Our findings garner support for the claim made by Moon and Morash (2013) wherein theory and practice must work together to develop practical school-based policy interventions that reduce the harmful effects of relational strain brought on by social bullying victimization. For example, of the many broad programmatic applications suggested by Agnew (2010), two seem particularly well suited to intervene with social bullying. The first is to equip individuals with the traits and skills to avoid strains in a proactive manner. Since victims of social bullying are more likely to have poor social skills (Juvonen & Graham, 2001), it would benefit students to strengthen interpersonal traits and social problem-solving behaviors. The second intervention strategy is to provide coping skills and resources to victims of social bullying in a reactive manner. Negative emotions such as anger, frustration, and anxiety induced by social exclusion/isolation often engender depression and school-avoidance behaviors that lower academic achievement (Agnew, 2010)

Consequently, more research is needed on school contexts before these anti-bullying intervention programs aimed at decreasing social bullying can be effectively applied. Contextual analyses are relevant to the current discussion because the impact of social bullying on a victim’s academic performance may vary from school to school. Specifically, it is argued that school context influences the type of bullying victimization (Hanish & Guerra, 2000; Morrow et al., 2014), the frequency of bullying victimization (Kochenderfer-Ladd & Wardrop, 2001; Muller, 2015), and the academic achievement of students (Akey, 2006; Hallinger, Bickman, & Davis, 1996). Based on our research, we believe that future investigations should concentrate on multilevel analyses in which specific forms of bullying victimization on academic achievement are nested within differing school contexts. A better understanding of contextual outcomes will allow for more practical approaches to reducing the occurrence of social bullying victimization in schools.

We began this study by noting that most research on the effect of bullying victimization on academic performance has relied on a relatively broad measure of bullying. We argued here that prior research has, for the most part, neglected the possibility that the effect of bullying victimization on academic performance might vary to some degree by the specific type of bullying experienced by a youth. If our claim has merit, the combining of different forms of bullying into one inclusive measure will engender weak empirical tests of the relationship between bullying victimization and academic performance. Such a situation would also furnish an explanation for why some studies report an inverse relationship between bullying victimization and academic performance, whereas others do not.

This study, however, does have a few methodological limitations that warrant some consideration. First, there always remains the chance that our model was incorrectly specified and that the exclusion of an influential variable from the analysis might have impacted our findings and conclusions adversely. Second, our measures of bullying victimization and academic achievement are self-reported. Self-report data allow us to study victims and offenders because these data reveal victimization experiences that would otherwise remain unreported. Self-report data on academic achievement often suffer from misclassification as students are more likely to report favorable grades. Nonetheless, because self-report grades and actual grades derived from administrative sources are reported to predict similar outcomes (Cole, Rocconi, & Gonyea, 2012; Kuncel, Crede, & Thomas, 2005), we believe that our model is accurate given the current state of knowledge on this topic.

Third, while sometimes disagreeing on the underlying causal processes responsible for the relationship, most interpretations of the observed association between bullying victimization and academic performance do agree on the direction of the effect. Assuming that a recursive causal structure underlies the relationship between bullying victimization and school performance, we regressed academic achievement on the various forms of bullying victimization in the current analysis. This assumption of a recursive relationship, however, may be theoretically problematic because it is plausible that school performance affects bullying victimization as well as being impacted by it. To our knowledge, no study published to date tests for the possibility of a reciprocal relationship between bullying victimization and academic performance. Thus, to the extent to which the causal structure is nonrecursive, estimates of the effect of bullying victimization on school performance yielded by prior research and by the current study may be biased to some degree. Longitudinal data are needed to establish causal direction more adequately. Future research might wish to investigate this issue further.

Fourth, while we differentiated between the significant types of bullying victimization, all forms of cyberbullying were combined as a result of data limitations. Research on the specific types of bullying is becoming increasingly popular (Olweus, Limber, & Breivik, 2019), but the collection of secondary data on cyberbullying victimization still consists of broad aggregate measures. The further disaggregation of cyberbullying victimization (direct and indirect forms) may provide a more nuanced understanding of its relationship with academic achievement. At this point, however, composite measures of cyberbullying remain the norm.

Last, there is a possibility that other forms of victimization, aside from bullying, have more influence on various negative outcomes. This is an integral piece of information for future studies to investigate as other forms of victimization may hold more explanatory power in the relationship between bullying and academic achievement. Increasing empirical knowledge on more severe crime victimization may furnish an explanation for the relatively low level of variability between bullying victimization and academic achievement reported in our results and prior research (Hanish & Guerra, 2002; Kowalski et al., 2014).

Despite these limitations, the work presented here has taken us considerably closer to developing a more nuanced understanding of the relationship between bullying victimization and academic performance. The noteworthy association of social bullying suggests that more considerable attention should focus on explaining why social bullying is more salient versus verbal, physical, or cyberbullying. The theoretical and practical implications of our study may help to create a healthier learning environment for students overall while our methodological approach may help advance future research.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The data that support the findings of this study are openly available in the Interuniversity Consortium for Political and Social Research at DOI: http//:doi.10.3886/ICPSR36354.v1, reference number: 36354.

Additional information

Funding

This research did not receive any funding support from agencies in the public, commercial, or not-for-profit sectors.

Notes

1. We would like to thank an anonymous reviewer for suggesting the analysis of interaction effects.