Cyberbullying: a systematic review of research, its prevalence and assessment issues in Spanish studies

Cyberbullying is a very complex phenomenon influenced by many different factors ( Baldry, Farrington, & Sorrentino, 2015 ) and, as bullying, it is very difficult to define or measure ( Patchin & Hinduja, 2015). Evans, Fraser, and Cotter (2014) conducted a systematic review of anti-bullying interventions, and found that about one fourth of the studies d used one-item evaluation whereas about three fourths used multi-item assessment. The results of these two forms of measuring bullying differed substantially. Among one-item evaluations, desirable effects on victimization and perpetration were found in 83% and 67% of the studies, respectively. On the other hand, in multi-item evaluations, this was found in 57% and 44%, respectively. Bullying and cyberbullying assessment strategies were also systematically reviewed by Vivolo-Kantor, Martell, Holland, and Westby (2014) . The authors found that about one third of the studies used the term “bullying” and also about one third d a definition of the phenomenon. Informants also varied among the studies, with the vast majority using self-reports, only about ten percent using peer-nominations, and about five percent using both. Rating scales were used in most of the studies, although about one fourth used dichotomous response. In about forty percent of the studies, timeframes for measuring the phenomena were unknown.

A systematic review of instruments used for evaluating cyberbullying was conducted by Berne et al. (2013) . It was found that definitions of cyberbullying and cybervictimization varied among the studies, although both were present in about one half of the instruments. About one half of the studies d devices such as cell phones or e-mail. Internal consistency or validity was reported in about one-half. About one fourth of the studies d a confirmatory factor analysis and almost all the instruments were self-reports.

Taking into account the difficulties in evaluating cyberbullying, together with great differences among the studies, it is still necessary to establish some common standards for assessment. Among other possible difficulties, these big differences among the studies make it very difficult to report or compare the prevalence among different geographic areas or time points. When the same instruments are used before and after interventions, it is possible to measure decrease or increase in the phenomena in the participants of each study (see Farrington & Ttofi, 2009 ). But it is worth mentioning that many governments provide great amounts of resources and researchers make a great effort with the objective of decreasing bullying or cyberbullying in the whole society, not only in the participants of their studies.

Cyberbullying is a new phenomenon, but some interventions against this kind of violence have already been conducted in Spain ( Garaigordobil & Martínez-Valderrey, 2015; Ortega-Ruiz, Del Rey, & Casas, 2012 ) and all over the world (see reviews conducted by Cross et al., 2015; Della Cioppa, O’Neil, & Craig, 2015 ). At the same time, anti-bullying campaigns have been conducted for decades and started with the suicides of adolescents in Norway attributed to school bullying in the early 1980s. The first international seminar on bullying was carried out by the Council of Europe in 1987. Also in the 1980s, the Norwegian government supported the first anti-bullying national campaign ( Roland, 2010 ). Afterwards, Olweus Bullying Prevention Program was implemented, supported by the Norwegian Ministry of Education and later became an example followed all over the world ( Olweus & Limber, 2009 ). Inspired by the Norwegian program, the Sheffield Anti-Bullying project was conducted in the UK ( Smith, 1997 ) with 23 schools and information packs distributed to 19,000 schools. In Spain, Sevilla Anti-Violencia Escolar (SAVE) (Ortega, 1997) and Andalucía Anti-Violencia Escolar (ANDAVE) projects were also conducted in the 1990s ( Ortega & Del Rey, 2003 ). Since then, hundreds of interventions have been conducted all over the world ( Farrington & Ttofi, 2009 ), many of them supported by the European Research Council and different national and international calls, some of them even specific to the topic.

But are these intense efforts bearing fruit? Are bullying or cyberbullying rates in Spain, after twenty years of intervention, increasing, decreasing, or unchanging? These questions can be answered by analyzing and comparing the studies on the prevalence of the phenomena. Therefore, the first objective of this systematic review is to describe the studies on cyberbullying in Spain, providing a global vision of the field. General information on the topic, taking the number of studies in each year, geographic areas, participants of the studies, sampling methods and the roles studied in each article, is going to be reported.

A label held by a concept is not meaningful unless it is clearly defined and operationalized. Studies that use the same label but with concepts defined and operationalized in different ways can even evaluate completely different constructs. Therefore, another objective of the study is to the concept that is being studied in the Spanish research on cyberbullying, taking into account the definition used in each study and its evaluation. Also, the psychometric properties of the instruments are d, with a specific attention to the election of one-item or multi-item evaluations in relation to the cut-off scores and timeframes. Timeframes are important since prevalence might differ when the participants give their answers on what happened, for example, in the past month or during their lifetime. The cut-off points are important since they indicate the frequency of each behavior to be classified as cyberbullying. Moreover, the concept measured changes depending on whether the cut-off is applied to a global one-item measure (e.g., How many times have you been cyberbullied?), to the total score in a scale (e.g., a summation of scores in items such as having insulted, having stolen the identity, having uploaded pictures, etc.) or to any item in the scale (e.g., classifying a participant who reported having insulted somebody on the Internet once as a cyberbully). Finally, prevalence of cyberbullying in relation to its evaluation is going to be reported.

The first study on cyberbullying in Spain published in a high impact journal was conducted by Ortega, Calmaestra et al. (2008) . Since then, the number of publications has been growing very rapidly (see Figure 1).

Figure 1

The Number of Articles on Cyberbullying Published by Spanish Authors in High Impact Journals.

Five studies have been conducted in Andalusia ( Del Rey et al., 2015; Elipe, Ortega, Hunter, & Del Rey, 2012; García-Fernández et al., 2015; Ortega, Calmaestra et al., 2008; Ortega, Elipe, & Calmaestra, 2009; Ortega, Elipe, Mora-Merchán et al., 2009 ), two in Asturias ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011; Álvarez-García, Núñez et al., 2015 ), three in the Basque Country ( Calvete, Orue, Estévez, Villardón, & Padilla, 2010; Gámez-Guadix, Gini, & Calvete, 2015; Garaigordobil, 2015a ), three in Castilla La Mancha ( Navarro, Ruiz-Oliva, Larrañaga, & Yubero, 2015; Navarro, Serna, Martínez, & Ruiz-Oliva, 2013; Navarro, Yubero, Larrañaga, & Martínez, 2012 ), two in Extremadura ( Cuadrado-Gordillo & Fernández-Antelo, 2014; León del Barco et al., 2012 ), one in Murcia (Giménez-Gualdo et al., 2015 ), four in the Valencian Community ( Buelga et al., 2010; Buelga, Cava et al., 2015; Buelga, Iranzo et al., 2015; Buelga & Pons, 2012 ) and one was conducted in the whole country ( García-Moya, Suominen, & Moreno, 2014 ).

The total number of students participating in the studies was 41,013. The mean number of participants by study was of 1,953 ( SD = 1720.42), ranging from 638 ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011 ) to 7,580 (García-Moya et al., 2014 ). The total number of schools that participated in the studies was 249 ( M = 12.45, SD = 5.97), ranging from 3 (Del Rey et al., 2015) to 24 (Elipe et al., 2012 ). Adolescents were the most studied age group with 16 studies conducted in Compulsory Secondary Education ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011; Álvarez-García, Núñez et al., 2015; Buelga et al., 2010; Buelga, Cava et al., 2015; Buelga, Iranzo et al., 2015; Buelga & Pons, 2012; Calvete et al., 2010; Cuadrado-Gordillo & Fernández-Antelo, 2014; Del Rey et al., 2015; Elipe et al., 2012; Gámez-Guadix et al., 2015; Garaigordobil, 2015a; Giménez-Gualdo et al., 2015; León del Barco et al., 2012; Ortega, Calmaestra et al., 2008; Ortega, Elipe, Mora-Merchán et al., 2009 ), of which three ( Elipe et al., 2012; Garaigordobil, 2015a; Giménez-Gualdo et al., 2015 ) d also upper secondary schools (Bachillerato). Primary schools were aimed in 4 studies ( García-Fernández et al., 2015; Navarro et al., 2012; Navarro et al., 2013; Navarro et al., 2015 ) and one was conducted online with adolescents without specifying the school level ( García-Moya et al., 2014).

In 10 studies ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011; Álvarez-García, Núñez et al., 2015; Buelga et al., 2010; Buelga, Iranzo et al., 2015; Buelga & Pons, 2012; Elipe et al., 2012; Garaigordobil, 2015a; García-Fernández et al., 2015; León del Barco et al., 2012; Navarro et al., 2015 ), participants were randomly ed within the autonomous community (regions), among which, in 7 studies, the ion was stratified ( Buelga et al., 2010; Buelga, Iranzo et al., 2015; Buelga & Pons, 2012; Elipe et al., 2012; Garaigordobil, 2015a; García-Fernández et al., 2015; León del Barco et al., 2012 ). In 7 studies, participants were randomly ed within one province 1 1

Please note that Murcia is an autonomous community that only has one province and, therefore, it was d among the studies in “one province”.

The most popular definition cited in the Spanish studies was the one proposed by Smith et al. (2008) , in which cyberbullying is defined as repeated intentional aggression, perpetrated by a group or an individual, using electronic devices, on a victim who cannot easily defend him/herself ( Buelga et al., 2010; Buelga, Iranzo et al., 2015; Buelga & Pons, 2012; Calvete et al., 2010; Garaigordobil, 2015a; León del Barco et al., 2012; Ortega, Calmaestra et al., 2008 ). Also the definition proposed by Tokunaga (2010) , according to which cyberbullying is an intentional, repeated, and harmful aggression perpetrated through the electronic devices characterized by an imbalance of power, was frequently used ( Del Rey et al., 2015; Navarro et al., 2012; Navarro et al., 2013 ). In general, the vast majority of the definitions stated that cyberbullying is perpetrated through the electronic devices, that it is intentional, repeated and that there is an imbalance of power between the perpetrator and the victim.

All the studies utilized self-reports to evaluate cyberbullying. Multi-item questionnaires with Likert response scales were the most popular, present in 11 studies ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011; Álvarez-García, Núñez et al., 2015; Buelga, Iranzo et al., 2015; Calvete et al., 2010; Cuadrado-Gordillo & Fernández-Antelo, 2014; Del Rey et al., 2015; Gámez-Guadix et al., 2015; Garaigordobil, 2015a; Navarro et al., 2012; Navarro et al., 2013; Navarro et al., 2015 ) followed by one-item evaluations also with Likert response scales in 8 studies ( Buelga et al., 2010; Buelga, Cava et al., 2015; Buelga & Pons, 2012; Elipe et al., 2012; García-Fernández et al., 2015; García-Moya et al., 2014; Ortega, Calmaestra et al., 2008; Ortega, Elipe, Mora-Merchán et al., 2009 ).

Timeframes for reporting different forms of involvement varied among the studies, with three asking about behaviors that “ever” happened ( Calvete et al., 2010; Gámez-Guadix et al., 2015; Giménez-Gualdo et al., 2015 ), two that did not specify any timeframe ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011; León del Barco et al., 2012 ), five focusing on the past year ( Buelga et al., 2010; Buelga, Cava et al., 2015; Buelga, Iranzo et al., 2015; Buelga & Pons, 2012; Garaigordobil, 2015a ), two on the past six months ( Navarro et al., 2012; Navarro et al., 2013 ), one on the past couple of months ( García-Moya et al., 2014 ), four on the past three months ( Álvarez-García, Núñez et al., 2015; Elipe et al., 2012; García-Fernández et al., 2015; Navarro et al., 2015 ), and four on the past two months ( Cuadrado-Gordillo & Fernández-Antelo, 2014; Del Rey et al., 2015; Ortega, Calmaestra et al., 2008; Ortega, Elipe, Mora-Merchán et al., 2009 ).

The cut-off points used to classify participants as cyberbullies, cybervictims, or cyber bully/victims also differed among the studies. Among evaluations using one item, some classified students as involved when their response was above the lowest possible point (e.g., if a scale ranged from 0 to 3, answers 1, 2, and 3 would be classified as involved; 1 occasional, 2 and 3 severe) ( Elipe et al., 2012; García-Fernández et al., 2015; Ortega, Calmaestra et al., 2008; Ortega, Elipe, Mora-Merchán et al., 2009 ); others above the second lowest possible response (e.g., if a scale ranged from 0 to 5, answers 0 and 1 would be classified as uninvolved ; 2, 3, 4, and 5 would be classified as involved; 2 and 3 moderate; 4 and 5 severe) ( Buelga et al., 2010; Buelga, Cava et al., 2015; Buelga & Pons, 2012 ) or using a dichotomous yes/no answer ( Giménez-Gualdo et al., 2015 ). These cut-off points were even more complex in case of scales. In most of the studies, prevalence was calculated taking into account affirmative answers to any item and not to the total scores on the scale. Children were classified as uninvolved in they answered “never” to all the items and the rest was considered to be involved ( Álvarez-García, Núñez et al., 2015; Buelga, Cava et al., 2015; Buelga, Iranzo et al., 2015; Calvete et al., 2010; Garaigordobil & Martínez-Valderrey, 2015 ). There was a study that used a cluster analysis instead of a cut-off point ( Gámez-Guadix et al., 2015 ), studies that did not specify a cut-off point ( Cuadrado-Gordillo & Fernández-Antelo, 2014; León del Barco et al., 2012 ), or overall prevalence ( Álvarez-García, Núñez, Álvarez-Pérez et al., 2011 ). There were also studies that classified students as involved when their response was above the second lowest possible point in at least one item (e.g., if a scale ranged from 0 to 4, students who answered 2, 3, or 4 in at least one item would be classified as involved; 0 and 1 uninvolved) ( Del Rey et al., 2015; Navarro et al., 2012; Navarro et al., 2013 ). Finally, there was also a study that d the highest possible answer in at least one item for classifying a participant as involved (e.g., if a scale ranged from 0 to 4, students who answered 4 in at least one item would be classified as involved) (Navarro et al., 2015 ). Two of the studies using scales also differentiated the moderate and the severe involvement, based on the percentile 95 ( Álvarez-García, Núñez et al., 2015 ) and one standard deviation above the mean ( Buelga, Iranzo et al., 2015).

Cyberbullying victimization was evaluated by means of scales and one-item global questions. Among the scales, the mean prevalence was of 26.65% ( SD = 23.23), the median prevalence was of 24.4% and the range was from 78.31% (including occasional - 72.74% and frequent - 5.57%; Álvarez-García, Núñez et al., 2015 ) to 4.6% (Navarro et al., 2015 ). In almost all the cases (except Gámez-Guadix et al., 2015 , where the prevalence was based on cluster analysis), the prevalence was calculated taking into account answers above the established cut-off point to at least one item from the scale (independently of the total number of the items). The highest prevalence was found among the studies that used a cut-off in which an answer different from “never” to any item would be classified as victims (including occasional and frequent) – 78.31% ( Álvarez-García, Núñez et al., 2015 ), 30.1% (Calvete et al., 2010), 30.2% (Garaigordobil, 2015a ). The prevalence was lower when the cut-off point was established in the middle of the scale (i.e., on a 5-point Likert scale ranging from 1 to 5, the cut-off is having answered 3 in any item) – 4.65% ( Del Rey et al., 2015), 24.2% (Navarro et al., 2012), 24.6% (Navarro et al., 2013 ). Finally, the prevalence of 4.6% was found in a study that uses the highest possible answer to at least one item as the cut-off point ( Navarro et al., 2015 ). It is worth mentioning that besides the cut-off scores, there are also other important differences among the studies. For example, the studies conducted by Navarro et al. (2013, 2014, 2015) focused on primary education whereas the rest focused on adolescents. There were also different timeframes for bullying (e.g., Calvete et al., 2010 asked about having “ever” been victimized while Del Rey et al., 2015 asked about the past two months). In the study conducted by Álvarez-García, Núñez et al. (2015) , victims were further classified into occasional (72.74%) and frequent (5.57%).

One-item evaluations yielded different results with mean prevalence of 10.70% ( SD = 8.26), median prevalence of 7.98%, ranging from 26.8% ( Buelga et al., 2010) to 5% (García-Moya et al., 2014 ). Also in this case, the highest prevalence was reported when students were classified as victims when their answer was different from “never” – 10.1% ( García-Fernández et al., 2015 ), 10.8% ( Ortega, Calmaestra et al., 2008 ), 5.85% ( Ortega, Elipe, Mora-Merchán et al., 2009 ); although the highest prevalence (26.8%) was reported in a study that used the second lowest possible score as a cut-off point (i.e., students who answered never and once were classified as non-victims and students who answered 2 or 3 times or more as victims, Buelga et al., 2010 ). It is worth mentioning that most of these studies classified victims into occasional and severe.

Perpetration was also measured by means of one-item or multi-item instruments. The mean prevalence using multi-item scale was of 24.64% ( SD = 24.35) and the median prevalence of 15.5%, ranging from 56.5% ( Buelga, Iranzo et al., 2015) to 2% (Navarro et al., 2015 ). All the cut-off scores were established taking into account an answer given by a participant in at least one of the items (independently of the number of the items in the scale). Again, the highest prevalence was found when participants were classified as involved after giving at least one answer different from “never” – 56.5% (46% occasional and 10.5% severe) ( Buelga, Iranzo et al., 2015), 44.1% (Calvete et al., 2010), 15.8% (Garaigordobil, 2015a ) and the prevalence was lower when the cut-off was established in the middle of the scale – 5.1% ( Del Rey et al., 2015 ) or requiring the highest possible answer – 2% ( Navarro et al., 2013).

One-item evaluations yielded different results also in this case, with the mean prevalence of perpetration of 14.06% ( SD = 13.82), median perpetration of 6.9%, ranging from 32% ( Buelga, Cava et al., 2015) to 1.37% (Giménez-Gualdo et al., 2015 ). The lowest prevalence was reported with a dichotomous question about the involvement ( Giménez-Gualdo et al., 2015 ), and the highest prevalence in studies that used the second lowest possible score as a cut-off point (i.e., students who answered never and once were classified as non-victims and students who answered 2 or 3 times or more as victims) – 32% in Buelga, Cava et al., 2015 and 31.4% in Buelga and Pons (2012) . In many studies, victims were classified as occasional and severe.

Prevalence of by-standing was reported in two studies. Álvarez-García, Núñez, Álvarez-Pérez et al. (2011) studied the prevalence of by-standing in violence through the information and communication technologies (not labeled as cyberbullying) finding that different behaviors were witnessed by 35.4% to 51.9% of the participants (no overall prevalence). Garaigordobil (2015a) reported that 65.1% of the participants responded having witnessed at least one or more behaviors d in a multi-item scale.

Three studies focused on the bully/victim status evaluated on a scale. Del Rey et al. (2015) found a prevalence of 2.09% using a cut-off established in the middle of the scale in at least one of the items for both roles (i.e., in a 5-point Likert scale ranging from 1 to 5, having answered at least 3 to at least one item on victimization and at least one on perpetration). Cuadrado-Gordillo and Fernández-Antelo (2014) reported 3.22% of bully/victims but did not specify a cut-off point. Gámez-Guadix et al. (2015) reported a prevalence of 46.61% of bully/victims in time 1 of their longitudinal study, based on a cluster analysis.

Also three studies reported the prevalence of the bully/victim status based on one-item evaluation. García-Fernández et al. (2015) found 3.6% based on answers different from “never” to one item on perpetration and one on victimization. Giménez-Gualdo et al. (2015) found 1.31% based on a dichotomous yes/no answer and Ortega, Calmaestra et al. (2008) found 8.4% also based on answers different from “never”.