Cyberbullying is widely recognised as a form of intentional and repeated aggression that occurs through digital devices such as mobile phones, computers and tablets, targeting individuals who often find it difficult to defend themselves (Smith et al. 2008). Definitions of cyberbullying generally emphasise three key elements of intent to harm, repetition and an imbalance of power between the perpetrator and the victim (Olweus 1993; Smith et al. 2008; Tokunaga 2010). While early studies adapted the definition of traditional bullying to online contexts, later research has expanded the concept to include distinctive features of digital communication such as anonymity, permanence and wide audience reach (Patchin & Hinduja 2006; Tokunaga 2010). Unlike traditional bullying, cyberbullying can occur at any time and in any place, extending beyond the school environment and intensifying the psychological impact on victims. The digital nature of these interactions also means that harmful content can be easily shared or preserved, making escape or resolution more difficult for cybervictims (Alhashmi et al. 2023). Cyberbullying manifests in a wide range of forms, many of which draw from traditional bullying behaviours but are amplified by the unique characteristics of digital platforms. The increasing complexity of online interactions has made it possible for perpetrators to exploit affordances such as anonymity, persistence and audience reach to inflict harm in both direct and indirect ways (Hinduja & Patchin 2014; Kowalski et al. 2014). Recognising the various forms cyberbullying can take is essential for understanding its prevalence and effects (Alhashmi et al. 2023). The typology proposed by Hinduja and Patchin (2014) remains foundational for understanding cyberbullying behaviours, such as harassment, denigration, impersonation and doxing, while the taxonomy proposed by Alhashmi et al. (2023) provides a more structured and multidimensional approach that accounts for evolving technologies and platforms. By categorising cyberbullying according to the nature of the act, delivery mode and type of cyberbullying, the Alhashmi et al. (2023) framework facilitates a deeper understanding of how cyberbullying operates in diverse online contexts. They distinguish between direct cyberbullying as ‘… a direct interaction between the bully and the victim, such as sending threatening messages or emails’ (Alhashmi et al. 2023:6) and indirect cyberbullying as ‘… spreading rumours or harmful content about the victim to others, often without the victim’s knowledge’ (Alhashmi et al. 2023:6). Research on the prevalence of cyberbullying among adolescents has developed significantly over the past two decades. However, inconsistencies in definition, measurement and scope remain a challenge when trying to identify trends or compare studies. While some studies define cyberbullying broadly as any form of online harassment (Ybarra & Mitchell 2004), others emphasise specific characteristics, such as intentionality and power imbalance (Patchin & Hinduja 2006; Smith et al. 2008). These differences in definitions, and their impact on the measures adopted in studies contributes to the observed variance in reported prevalence rates of cyberbullying behaviour, from as low as 4% (Ybarra & Mitchell 2004) to as high as 42.8% (Gohal et al. 2023). Longitudinally, an increase in cyberbullying behaviour is observable, with recent systematic literature reviews also indicating an increase in research interest in cyberbullying globally (Brochado, Soares & Fraga 2017). A trend identified across multiple studies is the strong link between traditional and cyberbullying behaviour. Accordingly, adolescents who experience or engage in traditional bullying are more likely to be involved in cyberbullying (Beran et al. 2015; Olasanmi, Agbaje & Adeyemi 2020; Ybarra & Mitchell 2004). This suggests that cyberbullying does not exist in isolation but rather extends from or amplifies patterns of offline aggression and/or vice versa. Another consistent finding is the relationship between access to and use of ICT (Information and Communication Technologies) and cyberbullying involvement. Studies conducted in technologically advanced countries, such as Canada (Beran et al. 2015) and South Korea (Lee & Shin 2017), indicate that adolescents who spend more time online are more likely to become cybervictims or cyberbullies. This trend is supported by studies from regions with lower ICT accessibility, such as Nigeria (Olasanmi et al. 2020), which found that cyberbullying rates vary based on device and Internet availability. Literature reports inconsistent findings regarding the relationship between gender and cyberbullying involvement. While early studies such as those by Ybarra and Mitchell (2004) and Smith et al. (2008) reported minimal or no significant gender differences in cyberbullying behaviour, more recent research employing detailed instruments and differentiated behavioural categories has revealed consistent gendered patterns. Studies such as Lee and Shin (2017), Bergmann and Baier (2018) and Kyrrestad, Kaiser and Fossum (2023) demonstrate that while overall prevalence rates may appear similar, males and females tend to experience and perpetrate cyberbullying in distinct ways. Males are more likely to be involved in direct forms of aggression, particularly through online gaming or visual media, whereas females are more commonly associated with relational cyberbullying such as exclusion, rumour-spreading and ridicule, which reflects patterns observed in offline traditional bullying (Li 2006). Vo et al. (2026) also found no significant gender differences, but that males were more likely to report involvement as cyberbullies, whereas female students tended to report higher involvement as cybervictims. Similarly, Ma et al. (2024) found that males were more likely to engage as cyberbully victims. These discrepancies may be due in part to developmental differences, as the population of these studies often vary across age groups and school grades (Güllü, Karahan & Akçay 2023). Together, these studies suggest that gender differences in cyberbullying lie not in overall prevalence but in the nature and context of these experiences. Adolescence, and particularly early to middle adolescence, is consistently identified as a period of heightened experiences of cyberbullying. This vulnerability is closely linked to neurodevelopmental changes, as adolescents undergo cognitive, emotional and social transformations that influence how they interact with peers and respond to online conflict (Jaworska & MacQueen 2015). As digital connectivity becomes integral to adolescent social life, especially through social networking sites, the structures of peer influence are reconfigured in ways that may reinforce risk behaviours and online aggression (Huang et al. 2014). Digital autonomy typically increases during early adolescence, with most adolescents beginning to use technology independently between the ages of 10 and 14 (Vaterlaus & Tarabochia 2021). This coincides with mid-adolescence, the period in which research suggests cyberbullying behaviour peaks. Across multiple cultural contexts, quantitative studies consistently identify mid-adolescence as the period where adolescents experience the highest rates of cyberbullying behaviour. In Canada, Beran et al. (2015) found that the highest rates of involvement were among 13–15-year-olds, with lower rates among both younger and older age groups. Smith et al. (2008) and Lee and Shin (2017) similarly observed higher cyberbullying rates among 14–16-year-olds in England and South Korea, respectively. Gohal et al. (2023), who surveyed Saudi Arabian adolescents, found that victimisation was highest among those aged 16–18 years. While these studies differ in specific age groupings, they collectively indicate that cyberbullying involvement intensifies from early into mid-adolescence. As social media platforms grow in popularity and incorporate features that interact with human psychology, they play a central role in cyberbullying patterns (Bayer, Triệu & Ellison 2020). Social media stands out from other online media because of their interactive, dynamic and user-focused nature. Bayer et al. (2020) describe how affordances of social media, such as profiles and streams, encourage both connection and comparison. The accessibility and interactivity of platforms such as Instagram, TikTok, Snapchat and WhatsApp have created digital spaces where social validation and visibility play central roles (Aizenkot 2018; Barlett et al. 2018b; Charteris et al. 2016). Several studies have found that frequent use of social media platforms is associated with higher rates of cyberbullying involvement, both as a victim and as a perpetrator (Abaido 2020; Barlett et al. 2018b; Dredge, Gleeson & De la Piedad Garcia 2014; Lee & Shin 2017; Villanueva-Moya et al. 2023). Cyberbullying behaviours common to social media, which range from spreading rumours and sending hurtful messages to posting embarrassing content, are reinforced by platform features that allow for visibility, permanence and social interaction (Charoenwanit 2019; Villanueva-Moya et al. 2023). As cyberbullying first became a subject of academic study, its effects on adolescents’ mental and physical well-being have been investigated. Well-being is a multidimensional concept encompassing physical health, emotional stability, social relationships and overall quality of life (Bharara et al. 2019). The impacts of cyberbullying on adolescent well-being can be severe and may include the disruption of relationships, decreased optimism and happiness and increased stress (Nixon 2014). Being a cybervictim has been consistently linked to anxiety, depression, decreased self-esteem and, in extreme cases, self-harm and suicidal ideation (Agustiningsih et al. 2024). Mursaleen, Jabeen and Ramzan (2025) highlight adolescence as a crucial developmental stage where such experiences can have lasting effects. Kowalski and Limber (2013) found that those who were both cybervictims and cyberbullies reported higher levels of depression, anxiety and psychosomatic health issues, while Fahy et al. (2016) similarly found long-term associations with depression and social anxiety. Franks (2015) observed declines in self-esteem and confidence and increases in sadness, aggression and paranoia. Although cyberbullying generally harms well-being, Brighi et al. (2019) found that resilience had a small mediating effect, helping to reduce emotional distress. Adolescence is also a critical period for personal and academic development, yet cyberbullying creates challenges that can hinder academic success. Well-being and academic performance are closely related, as emotional distress weakens motivation and engagement (Needham, Crosnoe & Muller 2004). Poor well-being, marked by anxiety, depression and stress, has been linked to concentration difficulties, lower motivation and reduced school engagement (Agustiningsih et al. 2024; Needham et al. 2004). Cybervictims often experience absenteeism, declining grades and school avoidance (Beran & Li 2007; Charoenwanit 2019; Franks 2015). Gohal et al. (2023) found that over a quarter of adolescents reported a negative impact on academic performance because of cyberbullying, with nearly 20% considering leaving school. Theoretical perspectives suggest that cyberbullying disrupts motivation and self-perception, leading to poorer academic outcomes (Pintrich 2000; Ryan & Deci 2000). Over the past two decades, research on cyberbullying among South African adolescents has demonstrated both an overall rise in prevalence rates and notable shifts in how it is conducted. Early studies, such as those by Schaffner (2010) and Oosterwyk and Parker (2010), found that approximately 16% – 32% of adolescents reported experiences as cybervictims. Tustin, Zulu and Basson (2014) reported that 15.2% of adolescents who experienced bullying were specifically cybervictims, while Payne (2015) found that 13.5% of adolescents identified as cybervictims in the previous 2 months. As access to mobile phones and Internet-based communication expanded over the past decade, these rates rose. More recent findings by Els (2024) indicate a significant increase, with 43.5% of adolescents reporting experiences as cybervictim. These patterns suggest that as digital connectivity became more common in adolescents’ daily lives, so too did their exposure to cyber risks. Additionally, the evolution of cyberbullying definitions over time, from early conceptions limited to social networking platforms (Schaffner 2010) and mobile chat services (Oosterwyk & Parker 2010), to broader definitions encompassing instant messaging apps and anonymous platforms (Els 2024; Payne 2015), has likely contributed to variations in reported prevalence.

While Els (2024) provides valuable insight into the relationships between different forms of bullying among South African adolescents, there is a dearth of data about the prevalence of specific forms of cyberbullying, the social media platforms on which they occur and their associations with well-being and academic performance among this cohort.

We adopted a post-positivist research philosophy, which assumes that while absolutes are unattainable when studying human behaviour, it is possible to use empirical data and objective measurement to approximate underlying patterns within social contexts (Creswell & Creswell 2018). This approach aligns with the goal of exploring theoretically grounded questions and examining relationships between defined variables. Accordingly, a quantitative approach was adopted to analyse data collected through cross-sectional surveys of learners from five high schools in South Africa. Invitations to participate in the research project were sent to multiple high schools across South Africa, with five responding positively. While invitations were sent to school across Quintiles 1–5, the schools that agreed to participate were all in Quintile 5, classifying them as fee-paying schools located in higher socioeconomic communities. While this limited our ability to generalise any findings across lower quintile schools in South Africa, it did allow for an exploration of cyberbullying behaviours within digitally advanced school environments. Following each school’s agreement to participate, permission to proceed with data collection procedures was obtained from the relevant provincial departments of education. Each school coordinated the data collection process in a way that minimised disruptions to the school’s schedule, with learners completing questionnaires in a dedicated 30-min time slot in normal class settings. Prior to questionnaire completion, a teacher introduced the project and explained that participation was voluntary and anonymous, and that no personal information was collected. Learners could choose to complete a paper-based version of the questionnaire or use their smartphones to complete an online version. To access the online version, a quick response (QR) code provided to each teacher could be scanned, and all questionnaires were available in both English and Afrikaans.

A combination of different questionnaires was used at each school, addressing different aspects of adolescent online behaviour. For the purpose of the present study, we analysed the data collected through one of these questionnaires, which was completed by 659 learners across four of the participating schools.

In addition to items relating to demographic data (school, gender and grade), the questionnaire included measures of academic performance, social media use, well-being and cyberbullying. The sections that follow describe each of the measures.

To address RQ1, we calculated descriptive statistics for the different forms of cyberbullying based on key demographic indicators (gender, age and school). To address RQ2, we calculated bivariate correlations between the frequencies of use of different social media platforms and involvement with direct and indirect cyberbullying. To address RQ3, we performed linear regressions to predict variance in each of the five different indicators of psychological well-being using demographics and cyberbullying indicators as predictors. Finally, to address RQ4, we conducted a two-step hierarchical linear regression to predict variance in academic performance. In the first model, predictors included demographic and psychological well-being indicators. In the second model, we added cyberbullying indicators.

Ethical clearance to conduct this study was obtained from the Stellenbosch University Social, Behavioural and Education Research Ethics Committee (Ref. No. 29470). Permission to conduct research in collaboration with the participating high schools was also obtained from the relevant provincial education departments, as is required by South African legislation. The data contained no personal identifying information of any of the participants. All participants were informed of the purpose and nature of the research, and informed consent was obtained from the learners’ parents or guardians prior to their participation in the project. No information that would enable personal identification of participants was collected.

Nearly two-thirds of participants (64%) reported involvement as either a cybervictim, cyberbully or both at least once in the last 6 months. 19.9% (n = 131) of the participants were categorised as only cybervictims, while 6.7% (n = 44) were only cyberbullies. The largest category consisted of participants who reported participating in both roles (n = 246, 37.3%). The remaining participants (n = 238, 36.1%) reported no involvement in cyberbullying at all. Over half of the female participants reported some form of cyberbullying involvement (n = 296, 65.1%). Similarly, over half of the male participants also reported some form of cyberbullying involvement (n = 122, 62.6%). A slightly higher proportion of females reported being only a cybervictim at least once (n = 96, 21.1%) compared to males (n = 35, 19.9%). In contrast, a higher percentage of males reported acting as only a cyberbully at least once (n = 15, 7.7%) than females (n = 28, 6.2%). The largest proportion of involvement for both genders was found in the combined category, with 37.8% of females (n = 172) and 36.9% of males (n = 72) reporting dual involvement in cyberbullying.

In general, the proportion of participants who reported some form of cyberbullying involvement decreased as the grade increased. Around three-quarters of the participants in Grade 8 reported some form of cyberbullying involvement (n = 101, 75.4%), while only about half of the participants in Grade 12 reported some form of cyberbullying involvement (n = 54, 52.9%).

The means and standard deviations for each of the cyberbullying behaviours in the RCBI-II are presented in Table 2.

Mann–Whitney independent samples t tests indicated that females reported higher frequencies of being insulted online (U = 39 111, p = 0.003, r = −0.118), and having rumours spread about them (U = 40 074, p = 0.013, r = −0.097), while males reported higher frequencies of threatening others online (U = 42 053, p = 0.011, r = 0.052) and sharing inappropriate photos and/or videos of others without their permission (U = 41 108, p < 0.001, r = 0.073). None of the other differences were statistically significant (p > 0.05).

The Spearman’s rank-order correlations between grade and each of the 10 cyberbullying behaviours, from both the victim and bully perspective, showed that grade had small, negative correlations with someone taking over the password of their account (r_s = −0.115, p = 0.003) and having an inappropriate photo or video of them shared (r_s = −0.08, p = 0.040). This suggests that younger adolescents were slightly more likely to report experiencing these behaviours. None of the other cyberbullying behaviours, either experienced or perpetrated, were significantly associated with grade.

Table 3 presents the means and standard deviations of the frequency of use for social media platforms. WhatsApp was the most frequently used platform (M = 5.77, SD [standard deviation] = 1.01), followed by Instagram (M = 4.50, SD = 1.98), TikTok (M = 3.73, SD = 2.25) and YouTube (M = 3.72, SD = 1.69).

Mann–Whitney U tests indicated that females reported significantly higher usage of WhatsApp (U = 30 103, p < 0.001, r = −0.32), Instagram (U = 33 480, p < 0.001, r = −0.25), TikTok (U = 30 536, p < 0.001, r = −0.31) and Snapchat (U = 31 672, p < 0.001, r = −0.29). In contrast, males reported significantly higher usage of YouTube (U = 35 431, p < 0.001, r = 0.20), Discord (U = 26 253, p < 0.001, r = 0.41), Reddit (U = 36 202, p < 0.001, r = 0.18), Twitch (U = 35 927, p < 0.001, r = 0.19) and Twitter (X) (U = 40 239, p < 0.001, r = 0.09). Only Facebook usage did not significantly differ by gender (U = 43 168, p = 0.418, r = 0.03).

Figure 1 displays the results of the Spearman’s correlation that was computed to assess the relationship between the use of four social media platforms and the different types of cyberbullying, as experienced as a cybervictim. The results showed that TikTok use was significantly associated with multiple types of cyberbullying, including being threatened (r_s = 0.130, p < 0.001), being insulted (r_s = 0.152, p < 0.001), being sent embarrassing or hurtful messages (r_s = 0.142, p < 0.001), having a secret shared without permission (r_s = 0.138, p < 0.001), having inappropriate videos and/or photos shared (r_s = 0.089, p = 0.023) and having rumours spread about them (r_s = 0.164, p < 0.001). Instagram use was positively correlated with being insulted (r_s = 0.124, p = 0.001), having a secret shared without permission (r_s = 0.104, p = 0.008) and having rumours spread about them (r_s = 0.125, p = 0.001). Snapchat showed similar correlations to Instagram, with the addition of it being correlated to being sent embarrassing or hurtful messages (r_s = 0.130, p < 0.001). WhatsApp use had the fewest correlations among the four social media and was only associated with being insulted (r_s = 0.118, p = 0.002) and having rumours spread about them (r_s = 0.186, p < 0.001).

Figure 2 presents the results of Spearman’s correlation that was computed to assess the relationship between the use of four social media platforms and the different types of cyberbullying, as engaged in as a cyberbully. The results showed that the use of Instagram was significantly correlated with four of the types of cyberbullying, including taking over the password of an account (r_s = 0.114, p = 0.003), insulting someone (r_s = 0.089, p = 0.023), sending embarrassing or hurtful messages (r_s = 0.143, p < 0.001) and sharing secrets (r_s = 0.134, p < 0.001). TikTok use was also associated with multiple types, such as sending embarrassing or hurtful messages (r_s = 0.163, p < 0.001), sharing secrets (r_s = 0.165, p < 0.001) and insulting someone (r_s = 0.090, p = 0.021). WhatsApp use was significantly positively correlated with two types of cyberbullying, insulting someone (r_s = 0.125, p = 0.001) and sharing secrets (r_s = 0.080, p = 0.040). Snapchat usage showed similar correlations to WhatsApp, with the addition of sending embarrassing or hurtful messages (r_s = 0.089, p = 0.022).

We classified each of the cyberbullying behaviours as either direct or indirect based on Alhashmi et al.’s (2023) taxonomy. This produced four scales: Indirect cybervictim (α = 0.629), indirect cyberbully (α = 0.628), direct cybervictim (α = 0.713) and direct cyberbully (α = 0.611). While the alpha values for these subscales fell below the 0.7 threshold, each displayed good mean inter-item correlation (MIC > 0.35), making them acceptable for use in further analyses (Briggs & Cheek 1986). Table 4 presents the Spearman’s correlation coefficients between the four forms of cyberbullying involvement, the five well-being dimensions and academic performance. Academic performance was significantly negatively associated with being a direct (r_s = −0.173, p < 0.001) and indirect cybervictim (r_s = −0.186, p < 0.001). Conversely, four of the well-being dimensions were positively associated with academic performance, including perseverance (r_s = 0.256, p < 0.001), optimism (r_s = 0.155, p < 0.001), connectedness (r_s = 0.165, p < 0.001) and happiness (r_s = 0.146, p < 0.001). Perseverance, optimism and happiness all displayed significant associations with multiple cyberbullying variables.

This study is subject to several limitations that should be acknowledged and was considered when interpreting the findings. Firstly, this study made use of a cross-sectional survey design, where the data were collected at a single point in time. This approach does not allow causal inferences to be made between any of the study’s key variables. The temporal order of the variables is unknown, and therefore, the directionality cannot be determined. The cross-sectional nature also introduces the risk of temporal bias, as responses at a single point in time may not accurately reflect long-term effects or patterns. Additionally, the survey data were entirely quantitative, consisting of fixed-response items. While this method facilitated comprehensive statistical analysis, it limited the depth of the insights and interpretations of the data. As no open-ended questions were included, there was also no opportunity to explore the motivation or context behind the reported experiences. The data used in this study were collected via self-report survey items, the reliability and validity of which are often limited as they require the participant to be highly self-aware. The approach also introduces potential biases, such as recall bias or social desirability bias. Adolescents may have misremembered past cyberbullying involvement or inaccurately reported the frequency of their involvement. Additionally, they may have underrepresented their involvement for fear of judgement or social stigma. Furthermore, the participants’ mood, attention or understanding of the questions at the time of the survey may have influenced their responses. Another limitation of the study was that a portion of the data from one school, HS1, was unusable. HS1 was an all-boys school, which impacted the gender distribution and resulted in the usable sample consisting of more females than males. While the analysis controlled for this in gender-based comparisons, the skewed distribution may influence the generalisability of the findings. Some of the gender-related trends may have been masked as a result. In addition, the final sample for the study consisted of learners from only Quintile 5 schools, which represents the highest socioeconomic areas in South Africa. These schools are typically more digitally advanced and were therefore relevant for this study. However, this did limit the generalisability of the findings, as South Africa is a digitally divided and diverse country. Secondly, while the survey made use of validated scales to assess constructs such as cyberbullying involvement and well-being, these complex concepts were measured and collected via structured survey items with fixed-response options. While these measures increase consistency and enable statistical analysis, they can only approximate the true experience lived by adolescents. For instance, while the EPOCH scale offers a robust method for well-being measurement, it may not capture the personal or cultural contexts that shape an adolescent’s actual well-being. Therefore, these responses should be viewed as a general indicator of the constructs being measured.