A Comparative Study on Cyberbullying Behaviors Among Korean and American College Students: Insights from Social Learning Theory and General Strain Theory

Abstract

Cyberbullying has emerged as a prominent social issue in recent years, affecting individuals across various age groups, including college students. This study aims to shed light on cyberbullying behaviors among American and Korean college students, drawing on the lenses of social learning theory and general strain theory as theoretical frameworks. Two survey data sets of 1067 college students (686 from South Korea and 381 from the U.S.) revealed that the social learning variable of definitions was the key predictor of cyberbullying behaviors for American students, while differential association was the significant predictor for Korean students. General strain variables were found to be not strong predictors of cyberbullying behaviors for American students, while these variables were significant predictors for Korean students. The differences between two college student populations in terms of the effects of theoretical factors on cyberbullying behaviors suggests the necessity of diverse approaches toward cyberbullying preventive strategies.

1. Introduction

As technology rapidly evolves, cyberbullying has emerged as a growing global issue that demands deeper investigation to understand its underlying causes and far-reaching impacts. Cyberbullying is typically defined as “willful and repeated harm inflicted through the use of computers, cell phones, and other electronic devices” (Hinduja and Patchin 2014, p. 11). A 2021 national survey of adolescents in the United States found that 55% of respondents had experienced cyberbullying victimization at some point in their lives, marking a substantial increase from the 36% reported in 2021 and 37% in 2019 (Patchin and Hinduja 2022, 2024). Cyberbullying is also prevalent among university students. Studies focusing specifically on college and university populations have found that approximately 20–30% of participants reported either perpetrating or experiencing cyberbullying victimization (Arslan et al. 2012; Floros et al. 2013; MacDonald and Roberts-Pittman 2010; Qudah et al. 2019; Zalaquett and Chatters 2014).

Cyberbullying is also a growing problem on international campuses. For example, a Portuguese study (Francisco et al. 2015) reported that about 8% of the college students had engaged in cyberbullying at some point in their lives, while a study in Qatar found that approximately 7% of students had cyberbullied others (Alrajeh et al. 2021). Differences in cyberbullying prevalence across these countries could be related to cultural differences as well as variations in knowledge, ability, and access to technology (Qudah et al. 2019). Cyberbullying perpetration among Korean college students was influenced by factors such as academic performance and peer associations with others involved in cyberbullying (G. Lee et al. 2022; Lee et al. 2021a). Additionally, holding a higher moral stance against cyberbullying was identified as a protective factor. Similarly, a study conducted by Chen et al. (2020) on attitudes toward cyberbullying among Chinese college students found that emotional warmth served as a protective factor against engaging in cyberbullying.

In a smaller study, most university-level victims of cyberbullying reported experiencing victimization during grade school as well, although victimization appeared to decrease as students approached college (Webber and Ovedovitz 2018). The same study also identified potential gender differences, suggesting that female students at the state university were more likely to experience cyberbullying. This comparison of a similar demographic internationally plays an important role in understanding the development of such behaviors, as well as highlights areas for future research and intervention.

Cyberbullying perpetration reflects behaviors commonly associated with traditional bullying, such as emotional, psychological, or deceptive aggression (Espelage et al. 2013; Giumetti et al. 2022; Kırcaburun et al. 2019; Lucas 2018). Because cyberbullying and traditional bullying share similar risk factors and victims often experience both forms simultaneously, it can be challenging to determine which consequences result specifically from cyberbullying (Olweus and Limber 2017). The primary distinction of cyberbullying is the element of anonymity, which notably influences aggressive behavior online (Lowry et al. 2016). Factors such as family incivility, emotional instability, and limited empathy also strongly contribute to cyberbullying perpetration, reflecting deficits in social interaction and emotional development (Çelik et al. 2012; Garifullin 2021; Jin and Miao 2021). These factors align with Akers’ (1985) social learning theory, highlighting the role of familial and peer influences in the development of deviant behavior.

Due to the similarities between traditional bullying and cyberbullying, prior research employed conventional criminological theories to understand cyberbullying behaviors, such as social learning theory (Espelage et al. 2016; G. Lee et al. 2022; Li et al. 2016; Lucas 2018; Shadmanfaat et al. 2020) and general strain theory (G. Lee and Sanchez 2018; Lee et al. 2021b; Lianos and McGrath 2018; Paez 2018). Social learning theory (SLT) asserts that there are four elements contributing to the establishment of deviant behaviors—differential association, differential reinforcement, imitation, and definitions. Research employing SLT to explain cyberbullying has consistently found differential association (interaction with peers who engage in similar behaviors) to be a particularly strong predictor of cyberbullying perpetration (Li et al. 2016; Shadmanfaat et al. 2020). Even when accounting for all four components of Akers’ SLT (Akers 1985, 1998), differential association remains highly influential. Furthermore, different elements of SLT help explain various dimensions of cyberbullying, including perpetration, victimization, and observation. For example, cyberbullying victimization has been significantly linked to SLT’s definitions component, prior victimization experiences, and the victim’s age (Espelage et al. 2016; Lucas 2018). Perpetration has also been associated with demographic factors such as race/ethnicity and socioeconomic status (SES), with findings suggesting individuals from higher SES backgrounds may be more likely to engage in cyberbullying.

Research utilizing Agnew’s general strain theory (GST; Agnew 1985; Agnew and White 1992), specifically adapted for cyberbullying contexts, has identified prior cyberbullying victimization, negative emotions (such as anger or desire for revenge), perceived social support, academic strain, and financial strain as critical factors predicting cyberbullying behaviors (Lee et al. 2021b; Lianos and McGrath 2018; Paez 2018). Previous studies have found that experiencing cyber victimization or engaging in cyber perpetration are significant sources of strain contributing to cyberbullying. Moreover, the relationship between perpetration and cyberbullying is mediated primarily by anger, aligning with the established GST literature (Lee et al. 2021b; Lianos and McGrath 2018). The relevance and influence of GST are further examined and explained in this study.

In this study, we investigated the applicability of SLT and GST in explaining cyberbullying behaviors across different cultural contexts, specifically the United States and South Korea. Examining the cross-cultural applicability is essential for testing the validity of research findings beyond the confines of American research. Testing the theoretical models with a South Korean sample may have a particular implication due to South Korea’s unique cultural context, which embodies a blend of Western and traditional non-Western influences.

Historically characterized as a traditional, non-Western society with strong collectivist values, South Korea has experienced rapid industrialization and Westernization since the 1970s. This transformation has resulted in the coexistence of traditional (e.g., collectivism) and Western cultural elements (e.g., individualism; Kim et al. 2010). Such a cultural duality positions South Korea as neither too divergent nor too similar to Western contexts, offering an ideal environment to examine the generalizability of models initially developed in the U.S. The current study aims to assess the applicability of the model of cyberbullying among American and Korean college students. Additionally, it explores the prevalence and characteristics of cyberbullying behaviors, analyzing demographic and contextual factors—such as gender, grade level, and anonymity—that influence cyberbullying perpetration in both countries. Finally, this paper provides practical implications for policy development and intervention strategies designed to reduce cyberbullying among college students.

2. Theoretical Framework

2.1. Social Learning Theory (SLT)

SLT is a theory that can be applied to many different disciplines in many ways. When discussing social learning in the context of crime and deviant behavior, Akers (1985, 1998) provides four defining principles that are involved in the process for establishing criminal behavior. He argued that differential association and conforming with deviant peers, as well as the imitation of the behavior of these peers, leads to the creation of individual definitions and tendencies toward crime, which are then differentially reinforced, or encouraged, typically by said peers, continuing the cycle of crime and deviance. This association with peers is emphasized by SLT, specifically in close peer and family groups (Akers 1985, 1998). Akers and Lee (1999) found that age and adolescent substance use—specifically marijuana use—supports the theory that the variables of association, definitions favorable to marijuana use, and reinforcement all mediate the relationship between age and substance use in adolescents. Prior research stated that nothing could explain the age curve of crime and deviance (Akers and Lee 1999). However, Aker’s study specifically showed that adolescents between the 7th and 12th grade were the most likely to be susceptible to differential association in general, while also being most likely to be exposed to the definitions for marijuana and substance use, as well as impacted by reinforcement directly related to SLT (Akers and Lee 1996, 1999; G. Lee et al. 2004). From a demographic perspective, males were more likely to be influenced by peer associations and engage in deviant behaviors and activities, whereas females were more likely to become victims (Bao et al. 2016; Miller and Morris 2014; Webber and Ovedovitz 2018). This emphasizes that males are more likely to form deeper attachments with peer groups, compared to females in general, as well as being more likely to engage in delinquent acts due to social bonds (Cullen et al. 2008).

When examining cyberbullying through the lens of SLT, research has shown that cyberbullies often associate with peers who engage in and justify abusive behaviors (Curry and Zavala 2020; Li et al. 2016), whereas victims are less likely to associate with peers who have experienced cyber victimization (Curry and Zavala 2020). Multiple studies have moderately supported applying SLT to various forms of cyberbullying and cyberdeviance (Hinduja and Patchin 2013; Li et al. 2016; Marcum et al. 2014; Miller and Morris 2014; Shadmanfaat et al. 2020). Specifically, differential association with cyber-deviant peers has consistently emerged as a strong predictor of cyberbullying behaviors (Hinduja and Patchin 2013; Miller and Morris 2014; Shadmanfaat et al. 2020). Shadmanfaat et al. (2020) emphasized that all SLT components significantly predicted cyberbullying perpetration. Additionally, a recent study by G. Lee et al. (2022) found that sociodemographic factors, such as age and gender, influenced cyberbullying behavior primarily through mediation by core SLT elements.

2.2. General Strain Theory (GST)

GST, first coined by Agnew (1985) and Agnew and White (1992), differs from other theories of crime, like SLT, in two distinct ways. The first is the type of relationship that leads to delinquency, and the other is the motivation for delinquency. According to Agnew, GST focuses on the negative relationship with others, and adolescents are then pressured into deviant behaviors and delinquency through negative emotions, specifically anger, that result from these strained interactions (Agnew and White 1992; Agnew 2001). He also identifies three main types of strain: the inability to achieve positively valued goals, the removal of positive stimuli, and the introduction of negative stimuli (Agnew 1985; Agnew and White 1992). The combination of the three types of strain and negative emotions is the key to the development of deviant behavior. GST also differs from other criminological theories in that it explains why individuals who experience major strains may not necessarily engage in delinquent behavior. This is often attributed to adolescents learning appropriate, non-delinquent coping strategies when faced with strain (Agnew 1985, 2001, 2012; Agnew and White 1992).

Studies examining the effects of strain on cyberbullying have found that strain experienced by college students significantly influences both cyberbullying victimization and perpetration (G. Lee and Sanchez 2018; Lee et al. 2021b; Lianos and McGrath 2018; Paez 2018). The previous literature has also shown that most measures of general strain—in this case anger and depression—are positively associated with both cyber dating abuse victimization and perpetration (Curry and Zavala 2020; Lee et al. 2021b; Lianos and McGrath 2018). Another study reported that the impact of strain on adolescents who were both victims and perpetrators of cyberbullying played a significant role in their individual responses to violence associated with the strain experienced (Guo 2021).

2.3. Theoretical Comparison of GST and SLT

Theoretical comparisons across multiple frameworks have become increasingly common in the study of crime and deviance. The comparison between GST and SLT on deviant and delinquent behavior has become popular in recent years. Although cyberbullying is a relatively new form of deviant behavior, it can be effectively examined through the lenses of both GST and SLT. Notably, both theories have been found to be positively associated with cyber dating abuse victimization and perpetration. For instance, Curry and Zavala (2020) found that two key measures of general strain—physical abuse victimization and feelings of anger—were linked to higher levels of cyber dating abuse victimization. Past experiences of physical abuse may increase individuals’ vulnerability to future victimization, as perpetrators might perceive them as easier targets. In terms of SLT, this study also found that greater association with individuals who are either perpetrators or victims of abuse was weakly related to cyber dating abuse victimization. However, general strain appeared to have a stronger overall influence on both victimization and perpetration.

Secondly, general strain variables—such as academic shortcomings, perceived injustice, academic failure, unengaging coursework, and the loss of scholarships—have been shown to significantly influence cyberbullying behaviors (G. Lee and Sanchez 2018; Smith et al. 2013). College students who experienced the strain of losing a scholarship or academic eligibility were significantly more likely to be victims of cyberbullying compared to those who had not faced such challenges. Additionally, academic difficulties and the loss of scholarships were also positively associated with prior engagement in cyberbullying perpetration.

Third, bullying has been found to be significantly associated with delinquency (Cullen et al. 2008). Males, in particular, are more likely to externalize their responses to bullying, which is consistent with the previous literature (Bao et al. 2016; Webber and Ovedovitz 2018). Furthermore, adolescents involved in cyberbullying—whether as perpetrators or victims—were more likely to engage in physical aggression and substance use compared to those not involved in cyberbullying (Guo 2021; G. Lee et al. 2004). Studies examining the influence of peer associations and negative emotions on cyberbullying roles (Çelik et al. 2012; Guo 2021) have shown that both boys and girls involved in cyberbullying are more vulnerable to responding to strain with violence and substance use. These behaviors appear to be driven by the combined effects of emotional strain (Guo 2021) and social learning influences (G. Lee et al. 2004) associated with cyberbullying experiences.

The effects of cyberbullying are significantly reflected in both GST and SLT. While strain appears to have a stronger influence on both the perpetration and victimization of cyberbullying, elements of SLT, specifically peer associations, also play a critical role in shaping these behaviors. Although both theoretical frameworks contribute meaningfully to understanding cyberbullying, GST may offer a more robust explanation in the cultural contexts studied, particularly in relation to the emotional and interpersonal stressors that drive aggressive online behavior. However, SLT remains important, as social learning mechanisms, such as peer reinforcement and exposure to deviant models, are also evident predictors. Based on the literature review, we propose the following three hypotheses:

Hypothesis 1.

The social learning variables of differential association, differential reinforcement, imitation, and definitions favorable to cyberbullying will have significant effects on cyberbullying perpetration among U.S. and Korean students.

Hypothesis 2.

The general strain variables of academic shortcomings, perceived injustice, academic failure, insipid classes, and losing scholarship will have significant effects on cyberbullying perpetration among U.S. and Korean students.

Hypothesis 3.

College Students who are male, spent more time on the internet, and have higher online anonymity and lower GPA are more likely to commit cyberbullying behaviors among U.S. and Korean students.

3. Materials and Methods

3.1. Sample and Procedure

Two data sets from undergraduate students in South Korea and the U.S. were analyzed. The American college students’ sample was collected from a university located in the southeastern area of the U.S. during the fall 2015 semester. Korean college students’ sample was collected from two universities in the Seoul Metropolitan Area of South Korea during the fall 2016 semester. A multistage cluster sampling method was used; we randomly selected the sample courses from a catalog of sections offered by each college to ensure that our participants represented the entire student body. Students in the sample classes were asked to provide informed consent before completing survey questionnaires. We ensured that no personal identifying information was collected on the questionnaire. As a result, the total number of completed questionnaires was 381 from the U.S. (83% response rate) and 686 from South Korea (98%). This research was approved by the appropriate universities’ Institutional Review Boards (Study #16-101 and #17-007).

3.2. Measures

3.2.1. Cyberbullying

We used six of the twenty-eight items on the revised Cyber Bullying Inventory (Brack and Caltabiano 2014) to measure cyberbullying behaviors. The students were asked to answer the following questions: how often have you performed the six instances described to others: (1) threatening in online forums (e.g., chatrooms, Facebook, or Twitter), (2) insulting in online forums (e.g., chat rooms, Facebook, or Twitter), (3) sharing private internet conversations without the other’s knowledge (e.g., chatting with a friend on Skype with other(s) in the room), (4) making fun of comments in online forums (e.g., Facebook), (5) sending threatening or hurtful comments through email or text messages, and (6) published online an embarrassing photo without permission. All items were assessed on a 4-point scale of responses: (0) never, (1) once, (2) two or three times, and (3) more than three times. The scores for all six items were summed and divided by the number of items to calculate the final value (Cronbach’s alpha = 0.56 for American students, and 0.59 for Korean students). For American students, the mean was 1.61 (SD = 2.26), with 50.2% reporting that they have not engaged in any cyberbullying activities. However, 70.1% of Korean students reported that they have not engaged in any cyberbullying activities. The mean was 0.91 (SD = 1.83) for Korean students.

3.2.2. Social Learning Variables

Differential Association. To measure differential association, respondents were asked the following: “To the best of your knowledge, about how many of your face-to-face (or online) friends have engaged in the following activities within the last year?” The activities were the same six instances asked in the previous cyberbullying question. Each item was measured on a 4-point scale of responses: (0) none, (1) some, (2) most, and (3) all. We first summed the scores for all twelve items (six items for face-to-face friends and six items for online friends) and divided them by the number of items to calculate the final value (Cronbach’s alpha = 0.86 for American students and 0.89 for Korean students). The mean was 0.33 (SD = 0.36) for American students and 0.22 (SD = 0.34) for Korean students.

Differential Reinforcement. This element was also measured by the same six cyberbullying items. The respondents were asked the following: “To what extent would you expect that your face-to-face friends (or online friends) would approve or disapprove of you participating in each of the below computer-related activities?” Each item was measured on a 4-point scale of responses: (1) strongly disapprove, (2) disapprove, (3) approve, and (4) strongly approve. We excluded the neutral category of the Likert scale to avoid the potential misuse of a midpoint (Chyung et al. 2017). We summed the scores for all eight items (four items for face-to-face friends and four items for online friends) and divided them by the total number of items to calculate the final value (Cronbach’s alpha = 0.92 for American students and 0.94 for Korean students). The mean was 1.37 (SD = 0.51) for American students and 1.32 (SD = 0.53) for Korean students.

Imitation. To assess imitation, participants were asked to evaluate the extent to which they had learned about any or all the six cyberbullying activities from family members, friends, internet searches, internet message boards, and chat rooms. The participants rated their learning experiences on a scale ranging from (0) nothing to (3) everything (Cronbach’s alpha = 0.64 for American students and 0.74 for Korean students). The mean was 1.14 (SD = 0.57) for American students and 0.35 (SD = 0.47) for Korean students.

Definitions Favorable to Cyberbullying. To measure definitions favorable to cyberbullying, respondents were asked to assess the extent to which they perceived the six cyberbullying behaviors in the inventory to be wrong. Each item was measured on a 4-point scale of responses: (0) not wrong at all, (1) slightly wrong, (2) mostly wrong, and (3) very wrong. It was reverse coding; the higher the score is, the stronger (i.e., more favorable) the definition was to cyberbullying behaviors. We also summed all six items and divided them by the number of items (Cronbach’s alpha = 0.84 for American students and 0.93 for Korean students). The mean was 0.45 (SD = 0.52) for American students and 0.54 (SD = 0.71) for Korean students.

3.2.3. General Strain Variables

Academic Shortcoming. We developed six strain variables based on the study conducted by Smith et al. (2013), who examined college student cheating and plagiarism. To measure personal academic shortcomings, students were asked to rate the four items: (1) I am a poor test taker; (2) I tend to procrastinate when it comes to schoolwork; (3) for some reason, I have a problem with class attendance; and (4) I have a short attention span, which interferes with my academic life. Each item was measured on a 4-point scale of responses ranging from (1) strongly disagree to (4) strongly agree. The scores for all four items were summed and divided by the number of items (Cronbach’s alpha = 0.52 for American students and 0.60 for Korean students). The mean was 2.31 (SD = 0.59) for American students and 2.12 (SD = 0.52) for Korean students.

Perceived Injustice. This variable is to assess the extent to which students feel injustice during their college careers. Students were asked to rate the following two items: (1) students who cheat have an unfair advantage for getting a good job following graduation and (2) students who cheat have an unfair advantage for getting into a graduate or professional school following graduation. Each item was measured on a 4-point scale of responses ranging from (1) strongly disagree to (4) strongly agree. The scores for these two items were summed up and divided by the number of items. The mean was 2.83 (SD = 0.85) for American students and 1.83 (SD = 0.83) for Korean students.

Academic Failure. Students were asked (1) whether they have ever failed a class and (2) ever been placed on academic probation while in college (0 = no and 1= yes). The scores for these two items were summed up and divided by the number of items. The mean was 0.19 (SD = 0.28) for American students and 0.15 (SD = 0.30) for the Korean students.

Insipid Classes. Students were asked to rate whether students felt like they had to sit through insipid classes—classes lacking meaning or interesting content for the respondent—on a 4-point scale of responses ranging from (1) strongly disagree to (4) strongly agree. The mean was 2.17 (SD = 0.76) for American students and 2.40 (SD = 0.71) for the Korean students.

Lose Scholarship. This binary variable asks how many times students have been threatened with losing or have actually lost a scholarship (0 = no and 1 = yes). The mean was 0.21 (SD = 0.41) for American students and 0.23 (SD = 0.42) for Korean students.

3.2.4. Control Variables

Time Spent on the Internet. Multiple studies reported that cyberbullying behaviors are positively associated with time spent online (Holt et al. 2012; Lee and Shin 2017; Li et al. 2016; Sasson and Mesch 2017). Students were asked to respond to the question—“How many hours a day do you spend on the internet?”—on a 6-point scale: (1) 1–2 h, (2) 3–4 h, (3) 5–6 h, (4) 7–8 h, (5) 9–10 h, and (6) more than 10 h. The mean was 2.56 (SD = 1.19) for American students and 1.95 (SD = 0.97) for Korean students.

Anonymity. This binary variable measures each student’s overall online anonymity (0 = low anonymity and 1 = high anonymity). Students were asked to respond to whether they disclosed eight separate items of personal information anywhere online: age, gender, pictures of themselves, their telephone number, goals/aspirations, sexual information, emotional/mental distresses, and family conflicts. Students’ responses were coded as low anonymity when disclosing more than three items. The mean was 0.19 (SD = 0.39) for American students and 0.42 (SD = 0.49) for Korean students.

Gender. This binary variable was coded as (0) male and (1) female. The mean was 0.58 (58% female; SD = 0.49) for American students and 0.53 (53% female; SD = 0.50) for Korean students.

Classification. This ordinal variable was coded as (1) freshman, (2) sophomore, (3) junior, and (4) senior. The mean was 2.84 (SD = 1.08) for American students and 2.24 (SD = 1.09) for Korean students.

Grade Point Average (GPA). The average GPA was 3.27 (SD = 0.43) for American students and 3.47 (SD = 0.50) for Korean students.

3.3. Data Analysis

The data analysis was designed to examine the effects of the social learning variables and the general strain variables on the cyberbullying behaviors among two samples of American and Korean college students. First, zero-order correlation analyses were conducted to assess the relationships between the key variables and to identify preliminary associations between cyberbullying behaviors and the theoretical constructs. Second, negative binomial regression analyses were employed to analyze the multivariate effects on cyberbullying behaviors. The use of negative binomial regression is methodologically justified given the nature of the data. A significant proportion of respondents in both samples reported no engagement in cyberbullying, with 50.2 percent of American and 70.1 percent of Korean college students indicating no such behaviors. This high frequency of zero values highlights the need for a statistical model that can account for over-dispersion and excess zeros. Traditional regression methods, such as ordinary least squares (OLS), are ill-suited for this type of data, as they may produce biased parameter estimates and lead to inaccurate inferences (Long 1997). In contrast, negative binomial regression accommodates over-dispersion by incorporating a dispersion parameter that relaxes the restrictive assumptions of Poisson regression. Furthermore, this approach allows for a more accurate estimation of the relationships between predictors and infrequent behaviors, such as cyberbullying, especially when the dependent variable consists of count data with a right-skewed distribution. The negative binomial model offers a statistically sound and substantively meaningful analytic strategy in this context.

4. Results

Table 1. Descriptive statistics and correlations for study variables.

						American College Students Sample (N = 381)
	Variables	1		2		3		4		5		6		7		8		9		10		11		12		13		14		15
1	Cyberbullying	1
2	Differential Association	0.50	**	1
3	Differential Reinforcement	0.27	**	0.36	**	1
4	Imitation	0.21	**	0.22	**	0.12	*	1
5	Definitions	0.37	**	0.23	**	0.29	**	0.15	**	1
6	Academic Shortcoming	0.14	**	0.10		0.02		0.04		0.08		1
7	Perceived Injustice	−0.03		−0.01		−0.06		0.02		−0.10	*	0.12	*	1
8	Academic Failure	0.10	*	0.01		0.05		0.02		0.13	*	0.15	**	−0.01		1
9	Insipid Classes	0.10		0.07		0.08		0.00		0.15	**	0.25	**	−0.05		0.21	*	1
10	Lose Scholarship	0.15	**	0.06		0.01		0.01		0.16	**	0.17	**	0.03		0.30	**	0.19	**	1
11	Time Spent on the Internet	0.27	**	0.16	**	0.09		0.25	**	0.21	**	0.11	*	−0.08		0.18	**	0.15	**	0.16	**	1
12	Anonymity (=1)	−0.16	**	−0.15	**	−0.04		−0.09		−0.02		−0.03		0.01		0.02		0.04		−0.05		−0.07		1
13	Gender (female=1)	−0.16	**	0.02		−0.12	*	0.03		−0.21	**	−0.01		0.08		−0.22	**	−0.06		−0.16	**	−0.12	*	−0.03		1
14	Classification	−0.02		0.02		−0.05		−0.07		0.02		0.01		−0.05		0.22	**	0.00		0.11	*	0.03		0.07		−0.11	*	1
15	GPA	−0.14	**	−0.07		−0.12	**	0.01		−0.18	**	−0.28	**	0.08		−0.53	**	−0.23	**	−0.32	**	−0.16	**	0.04		0.17	**	−0.14	*	1
	Mean	1.61		0.33		1.37		1.14		0.45		2.31		2.84		0.19		2.17		0.21		2.56		0.19		0.58		2.84		3.27
	Std. Deviation	2.26		0.36		0.51		0.57		0.52		0.58		0.85		0.28		0.76		0.41		1.19		0.39		0.49		1.08		0.43
						Korean College Students Sample (N = 686)
	Variables	1		2		3		4		5		6		7		8		9		10		11		12		13		14		15
1	Cyberbullying	1
2	Differential Association	0.55	**	1
3	Differential Reinforcement	0.30	**	0.31	**	1
4	Imitation	0.51	**	0.62	**	0.28	**	1
5	Definitions	0.17	**	0.07		0.07		0.04		1
6	Academic Shortcoming	0.11	**	0.10	**	0.02		0.10	**	0.03		1
7	Perceived Injustice	0.15	**	0.14	**	0.07		0.09	*	0.07		0.28	**	1
8	Academic Failure	0.18	**	0.15	**	0.07		0.17	**	0.08	*	0.16	**	0.04		1
9	Insipid Classes	0.07		0.01		0.05		0.04		0.05		0.28	**	0.18	**	0.21	**	1
10	Lose Scholarship	0.13	**	0.08	*	0.07		0.05		0.09	*	0.17	**	0.06		0.25	**	0.15	**	1
11	Time Spent on the Internet	0.09	*	0.07		0.01		0.11	**	0.02		0.14	**	0.07		0.05		0.04		0.01		1
12	Anonymity (=1)	−0.11	**	−0.09	*	−0.04		−0.12	**	−0.03		−0.05		0.05		0.06		0.04		−0.05		−0.02		1
13	Gender (female=1)	−0.24	**	−0.16	**	−0.10	**	−0.19	**	−0.13	**	0.04		−0.04		−0.20	**	0.09	*	−0.05		0.06		−0.02		1
14	Classification	0.04		0.11	**	0.03		0.06		0.02		−0.01		−0.01		0.16	**	−0.02		0.13	**	0.02		0.03		−0.10	*	1
15	GPA	−0.08	*	−0.02		−0.05		−0.06		−0.04		−0.36	**	−0.06		−0.18	**	−0.20	*	−0.20	**	−0.06		0.08	*	0.05		0.17	**	1
	Mean	0.91		0.22		1.32		0.35		0.54		8.47		3.66		0.29		2.40		0.23		1.95		0.42		0.53		2.24		3.47
	Std. Deviation	1.32		0.53		0.53		0.47		0.71		2.07		1.65		0.59		0.71		0.42		0.97		0.49		0.50		1.09		0.50

Note: * p < 0.05; ** p < 0.01.

presents the zero-order correlations among the variables used in the current study for both Korean and American college student samples. Among American students, cyberbullying exhibited significant correlations with all four social learning variables: differential association (r = 0.50, p < 0.01), differential reinforcement (r = 0.27, p < 0.01), imitation (r = 0.21, p < 0.01), and definitions (r = 0.37, p < 0.01), respectively. In comparison, Korean college students also showed significant correlations between cyberbullying and these same variables: differential association (r = 0.55, p < 0.01), differential reinforcement (r = 0.30, p < 0.01), imitation (r = 0.51, p < 0.01), and definitions (r = 0.17, p < 0.01). The most notable differences between the two groups emerged in the variables of imitation and definitions favorable to cyberbullying, with imitation having a stronger association among Korean students and definitions showing a stronger association among American students. Nevertheless, all correlations among these college populations with social learning variables demonstrate substantial effects on cyberbullying behaviors.

Table 2. Negative binomial regression on cyberbullying.

	American College Students (N = 381)
	Baseline Model				Social Learning Model				General Strain Model				Full Model
Variables	b	S.E	Wald		b	S.E	Wald		b	S.E	Wald		b	S.E	Wald		VIF 1
Differential Association					0.68	0.40	2.84						0.59	0.41	2.12		1.27
Differential Reinforcement					0.36	0.28	1.65						0.39	0.29	1.73		1.25
Imitation					−0.06	0.27	0.06						−0.04	0.27	0.02		1.13
Definitions					0.81	0.25	10.72	**					0.77	0.25	9.18	**	1.24
Academic Shortcoming									0.57	0.26	4.76	*	0.53	0.29	3.44		1.19
Perceived Injustice									−0.22	0.16	1.86		−0.12	0.18	0.42		1.06
Academic Failure									0.32	0.57	0.32		0.21	0.61	0.12		1.54
Insipid Classes									0.17	0.18	0.86		0.12	0.20	0.37		1.16
Lose Scholarship									0.01	0.34	0.00		0.04	0.38	0.01		1.21
Time Spent on the Internet	0.19	0.11	3.21		0.13	0.11	1.20		0.14	0.11	1.55		0.08	0.12	0.42		1.18
Anonymity (=1)	−1.01	0.49	4.22	*	−0.85	0.52	2.72		−0.92	0.50	3.45		−0.84	0.53	2.57		1.09
Gender (female = 1)	−0.71	0.29	6.08	**	−0.49	0.31	2.50		−0.65	0.30	4.81	*	−0.47	0.32	2.19		1.13
Classification	0.05	0.13	0.12		0.07	0.14	0.27		0.05	0.14	0.14		0.10	0.15	0.42		1.08
GPA	−0.40	0.32	1.57		−0.02	0.35	0.00		0.05	0.39	0.01		0.37	0.43	0.73		1.65
Constant	−0.75	1.18	0.41		−3.25	1.47	4.86	*	−3.29	1.71	3.69		−5.71	2.05	7.76	**
Likelihood Ratio Chi-Square (df)	21.63 (5)				42.20 (9)				29.15 (10)				46.68 (14)
	Korean College Students (N = 686)
	Baseline Model				Social Learning Model				General Strain Model				Full Model
Variables	b	S.E	Wald		b	S.E	Wald		b	S.E	Wald		b	S.E	Wald		VIF 1
Differential Association					1.37	0.40	11.56	***					1.22	0.41	8.68	**	1.72
Differential Reinforcement					0.35	0.24	2.09						0.30	0.25	1.44		1.14
Imitation					0.11	0.36	0.09						0.15	0.37	0.16		1.70
Definitions					0.26	0.20	1.76						0.16	0.22	0.55		1.04
Academic Shortcoming									0.05	0.30	0.02		−0.03	0.08	0.18		1.34
Perceived Injustice									0.58	0.16	12.36	***	0.21	0.09	5.83	*	1.14
Academic Failure									−0.12	0.49	0.06		−0.12	0.26	0.21		1.23
Insipid Classes									0.44	0.22	4.17	*	0.48	0.22	4.68	*	1.18
Lose Scholarship									0.40	0.31	1.68		0.34	0.32	1.12		1.14
Time Spent on the Internet	0.05	0.14	0.13		−0.02	0.15	0.02		0.02	0.15	0.01		−0.03	0.15	0.04		1.04
Anonymity (=1)	−1.23	0.36	11.93	***	−1.10	0.37	8.94	**	−1.33	0.37	13.19	***	−1.18	0.37	9.86	**	1.04
Gender (female = 1)	−0.29	0.28	1.10		0.16	0.31	0.27		−0.35	0.30	1.37		0.11	0.33	0.11		1.13
Classification	0.19	0.13	2.18		0.15	0.14	1.21		0.16	0.14	1.26		0.14	0.15	0.91		1.11
GPA	−0.56	0.26	4.60	*	−0.47	0.28	2.90		−0.39	0.30	1.63		−0.38	0.32	1.46		1.28
Constant	−0.61	0.90	0.46		−2.18	1.08	4.06	*	−3.51	1.43	6.01	*	−4.05	1.55	6.80	*
Likelihood Ratio Chi-Square (df)	23.43 (5)				55.94 (9)				46.82 (10)				69.71 (14)

Note: * p < 0.05; ** p < 0.01; *** p < 0.001; VIF ¹ statistics are calculated using a linear regression model.

presents the results of negative binominal regression analyses. The first baseline model examined the effect of demographic and contextual factors on cyberbullying perpetration. Anonymity had negative associations with cyberbullying behaviors for both American (b = −1.01, Wald = 4.22, p < 0.05) and Korean students (b = −1.23, Wald = 11.93, p < 0.001). Female students exhibited a decreased likelihood of cyberbullying perpetration for the American population (b = −0.71, Wald = 6.08, p < 0.01), while GPA showed a negative effect on cyberbullying perpetration for Korean students (b = −0.56, Wald = 4.60, p < 0.05).

The second model introduced social learning variables into the baseline model. Among these, definitions favorable to cyberbullying emerged as a significant predictor for American students (b = 0.81, Wald = 10.72, p < 0.01), while differential association was significant for Korean students (b = 1.37, Wald = 11.56, p < 0.001). The third model examined the effects of general strain variables. For American students, academic shortcomings significantly predicted cyberbullying behavior (b = 0.57, Wald = 4.76, p < 0.05). In contrast, for Korean students, perceived injustice (b = 0.56, Wald = 12.36, p < 0.001) and insipid classes (b = 0.44, Wald = 4.17, p < 0.05) were significant predictors.

The final full model included all the variables of social learning, general strain, and control variables. For American college students, only definitions favorable to cyberbullying remained a significant predictor of cyberbullying perpetration (b = 0.77, Wald = 9.18, p < 0.01). For Korean college students, one social learning variable—differential association (b = 1.22, Wald = 8.68, p < 0.01)—and two general strain variables—perceived injustice (b = 0.21, Wald = 5.83, p < 0.05) and insipid classes (b = 0.48, Wald= 4.68, p < 0.05)—were significantly associated with increased likelihood of cyberbullying perpetration. Additionally, anonymity emerged as a protective factor against cyberbullying across the models for Korean students. Importantly, variance inflation factor (VIF) values ranged from 1.04 to 1.72, indicating no evidence of multicollinearity among the social learning and general strain variables included in the regression models.

5. Discussion and Conclusions

This study found significant impacts of the social learning variables (i.e., differential association and definitions) on cyberbullying perpetration for both sets of university participants, showing support for the existing literature (G. Lee et al. 2022; Li et al. 2016; Shadmanfaat et al. 2020). This suggests the profound impact of social learning processes on the propensity to engage in cyberbullying. As seen in the social learning model and the full model (see Table 2), definitions favorable to cyberbullying were a positive predictor for American students, while Korean students who exhibited differential association was a positive predictor of cyberbullying perpetration. This can be explained by cultural and contextual differences in the social learning process. In American culture, where individualism is emphasized, the concept of definition may play a significant role in justifying cyberbullying behavior. The emphasis on personal autonomy and self-expression among American college students may contribute to the development of individualized justifications for engaging in cyberbullying. This can include viewing such behavior as harmless teasing or as a form of retaliation against perceived injustices, which helps explain the positive association between definitions favorable to cyberbullying and actual cyberbullying behavior. In contrast, Korean culture places a stronger emphasis on collectivist values and social harmony. In this context, differential association may be a more salient predictor of cyberbullying behavior. Korean college students may be influenced by their peer groups and social circles, adopting cyberbullying behaviors that are normalized or encouraged within their social networks. The pressure to conform to group norms and maintain social cohesion may lead individuals to engage in cyberbullying behavior as a means of fitting in or gaining acceptance within their social groups.

The examination of strain variables on cyberbullying perpetration also showed differing results between the two populations. For American students, academic shortcoming was a positive predictor of cyberbullying behavior in general strain model. However, no strain variable was significantly related to cyberbullying perpetration in the full model. Korean students displayed significant positive effects of perceived injustice and insipid classes on cyberbullying. These findings may be attributed to cultural and contextual differences in the sources of strain experienced by students in each culture. In American culture, where individual achievement and success are highly valued, college students who perceive themselves as academically inadequate or experiencing academic shortcoming may feel a sense of frustration, inadequacy, or low self-esteem. As a result, they may engage in cyberbullying behavior as a means of asserting dominance, seeking revenge, or deflecting attention away from their own academic shortcomings. The pressure to excel academically and the competitive nature of the educational environment may exacerbate feelings of strain, leading to maladaptive coping mechanisms such as cyberbullying. On the other hand, in Korea, where social hierarchy and academic achievement are emphasized, college students who experience strain from perceived injustices within the societal structure or frustration with their academic environment injustice may experience feelings of anger, resentment, or powerlessness. As a result, they may engage in cyberbullying behavior as a means of seeking retribution, restoring their sense of justice, or exerting power over others perceived as responsible for the injustices they experience.

While elements of both SLT and GST were statistically significant across the U.S. and South Korean samples, their relative explanatory power differed by context. In particular, a social learning variable—definitions—was more consistently predictive of cyberbullying perpetration, especially for American students, suggesting that learned behaviors play a dominant role in cyber-offending within more individualistic cultures. Conversely, general strain variables—especially indicators of perceived injustice and insipid classes—showed stronger associations for South Korean students, aligning with a collectivist context where social pressure and emotional strain may have more influence. These results suggest that while each theory offers valuable insights independently, their combination provides a more holistic understanding of cyberbullying behavior across different cultural settings.

This study also revealed that Korean college students were less likely to engage in cyberbullying behaviors when they reported higher levels of perceived anonymity. In other words, students who disclosed less personal information online were less likely to perpetrate cyberbullying. This finding is particularly noteworthy when compared to previous research such as Barlett et al. (2021) and Moore et al. (2012), which examined perceived anonymity at the moment of potential cyberaggression, often emphasizing whether users believe they can act without being identified or held accountable. Although the current finding may appear to contradict prior research, it reflects a different conceptual approach: general anonymity may be linked to more cautious or reserved online behavior overall, while perceived anonymity may embolden aggressive acts in specific contexts.

In summary, our findings suggest that social learning variables collectively play predictive roles in cyberbullying behaviors among both American and Korean students. Meanwhile, some general strain variables displayed moderate effects on cyberbullying perpetration in both American and Korean college students compared to social learning counterparts, which is consistent with the existing literature (Lianos and McGrath 2018; Patchin and Hinduja 2011; Shadmanfaat et al. 2020). We suggest that the cultural and contextual differences between the U.S. and South Korea might shape the underlying mechanisms of social learning and general strain, thereby shaping the predictors of cyberbullying behavior differently among American and Korean college students.

Future research could help expand the literature and open up possible explanations for specific conclusions. With only a few similarities such as the impact of differential associations with peers, the clear differences between two college student populations emphasize the need for tailored interventions. Universities could consider offering counseling to academically struggling students who may seek refuge in online space and social media. Furthermore, informing students about safe reporting mechanisms for cyberbullying incidents could contribute to addressing the issue at its source, providing guidance and support to those involved.

6. Limitations

There are several noteworthy limitations of this study. One limitation to this study and its results stems from the reliability of the scales that were used for measuring each variable. Future research should verify that all questions, measures, and scales are internally valid and reliable in measuring what they are seeking to measure. This could greatly impact results, especially on strain variables, where fewer significant results were seen.

Another limitation within this study could be selection bias. Students who participate in deviant behaviors, such as cyberbullying, may be less likely to be truthful in their experiences and potential struggles, as well as less likely to participate in such a survey in the first place. Individuals who are less likely to engage in or endorse deviant behaviors tend to provide more honest responses, as they have little to hide and minimal fear of negative consequences. In contrast, students who may participate in cyberbullying are more likely to do so under conditions of strict anonymity, driven by the fear that their actions or survey responses could be traced back to them. This concern about being identified may influence both their behavior and their willingness to report it accurately (David-Ferdon and Hertz 2009).

Limitations associated with the use of retrospective cross-sectional data must be acknowledged, particularly in establishing temporal order among variables. Such data may be subject to memory errors, recall bias, or even intentional distortion, all of which introduce uncertainty when attempting to infer causal relationships. As a result, the ability to draw strong causal conclusions from the findings is inherently limited. The last limitation is associated with the challenge of adequately considering structural and cultural differences between the two countries. Both structural and socioeconomic differences between countries, as well as variations in internet access and digital literacy, could impact the prevalence and nature of cyberbullying. Additionally, two countries may have unique cultural norms, values, and attitudes that shape the understanding and perception of cyberbullying.

By broadening the existing empirical evidence and literature on cyberbullying, this study reinforces the notion that, like other forms of deviance, cyberbullying is significantly associated with social learning variables. Although preventing associations with deviant peers may be challenging, recognizing these influences offers valuable insight into the origins and persistence of such behaviors. This understanding not only sheds light on why individuals continue to engage in cyberbullying but also contributes to a broader effort to uncover the underlying causes of deviant behavior more generally.