Gossip is defined as a type of evaluative conversation about somone else who is not present (Eder and Enke, 1991; Foster, 2004). These conversations have a situational nature, as the message emission and interpretation depend on the context in which they are produced (Wert and Salovey, 2004). Gossip is a widespread and repeated practice, to which a great part of daily conversations is usually devoted (Dunbar et al., 1997; Levin and Arluke, 2013). Gossip has and had a bad reputation throughout time and across cultures because it is considered as self-serving and seen to be carried out to maliciously manipulate the impressions the hearers have about a third person (Goodman and Ben-Ze'ev, 1994). Nevertheless, some evidence exists of the positive effects of gossip (Baumeister et al., 2004; Dunbar, 2004; Alexandrov et al., 2013).

Research on the positive side of gossip has mainly focused on its group-serving functions. It has been said that it can be useful to inform members of the group about its norms and sanctions (Grosser et al., 2010), to protect the group from those who break the rules (Beersma and Van Kleef, 2011; Feinberg et al., 2014), to influence its members' reputation (Giardini, 2012) and to strengthen social links within the group (Dunbar, 2004). Seen this way, gossip promotes the common good (Beersma and Van Kleef, 2012), and constrains self-serving behaviors that could damage the group given the influence it may have on members' reputation and inclusion (Beersma and Van Kleef, 2011). Individual benefits have also been ascribed to those who receive gossip, as it allows obtaining evaluative information about others that can serve as a reference with which to evaluate oneself (Wert and Salovey, 2004). In particular, gossip has been shown to increase one's self-improvement value, an important dimension of self-evaluation (Martinescu et al., 2014). This benefit reflects the learning function of gossip, promoting important social learning about how to achieve success or avoid failure in specific social situations (Baumeister et al., 2004). In this context, positive gossip is valuable to the individual because it provides a model to imitate (Litman and Pezzo, 2005). The advances in this line direct toward the necessity of making the gossipers' moral deliberation explicit to regulate the relationships and generate a greater confidence (Peters and Kashima, 2015).

In addition to providing information, influence, and strengthen group bonds, gossip also has a gratifying individual function: entertainment (Foster, 2004). What differentiates gossip from social curiosity—a related type of behavior—is its objective. While gossip predominantly has an entertainment purpose, social curiosity is mainly directed at obtaining information about other people (Hartung and Renner, 2013).

A recent study with more than 55,000 adolescents from 20 different countries shows the widespread use of ICT in social communication and its importance in social and personal development (Areepattamannil and Khine, 2017). Although there is general awareness of ICT's importance in adolescent relationships, the number of studies on cybergossip is limited (Oluwole, 2009; Laghi et al., 2013; Subramanian, 2013). Moreover, most of the studies that do exist have taken place within the field of computer science (e.g., Apolloni et al., 2013), and only a few have come forth from the psychology sphere (e.g., Gabriels and De Backer, 2016).

Cybergossip is the equivalent of gossip in the virtual world, but both phenomena must be studied in different ways, as the contexts in which they take place differ and matter. First, face-to-face gossip makes use of verbal language, whilst cybergossip can take the form of written messages, images, and videos. Second, cybergossip messages are recorded and can be reproduced exactly as originally created, which may influence what is shared in the first place (Bertolotti and Magnani, 2013). Third, cybergossip has the potential to reach a wide audience, at any time, and instantaneously. Fourth, given the limited prosodic, paralinguistic, and extralinguistic elements that can be used in virtual communication, it is harder for both the sender to convey the social intention behind the message and for the recipient to recognize it (Carrera and Pelayo, 2002). This increases the possibility of misunderstandings, which can lead to cyberaggression and cybervictimizing behaviors and therefore to cyberbullying (Del Rey et al., 2015).

It is also important to differentiate between cybergossip and cyberbullying. Gossip's negative potential has led numerous studies to identify it as an indirect form of peer group aggression that has the aim to manipulate another's reputation or achieve his/her exclusion from the group (Archer and Coyne, 2005). Nevertheless, a clear intention of damaging the other is not necessarily present in cybergossip, while it is in cyberbullying (Tokunaga, 2010). Therefore, while related, gossip, cybergossip, and cyberbullying are not directly comparable and must be studied with tools appropriately designed for each of these phenomena.

Despite the differences between the phenomena, work on cybergossip should take research into its off-line version as a starting point. Gossip has been widely studied and with different research methodologies. As anthropology is the dominant field for the study of gossip, observational techniques have been most commonly used (Foster, 2004), but this is problematic. Confidence and discretion are two factors that influence gossip (Gluckman, 1963) and these conditions are difficult to guarantee in experimental and observational studies. Self-report studies do not violate these conditions due to their anonymous and retrospective nature, and therefore, despite drawbacks which are mainly related to the risk of obtaining socially desirable answers, form the ideal methodology for the measurement of gossip (Brady et al., 2017). While the need for such quantitative studies has been identified (Noon and Delbridge, 1993), to date, very few exist, and the scarce scales that are available have a number of limitations. Several scales too narrowly define gossip as a negative behavior and even as an indirect mode of aggression. Examples are the Direct and Indirect Aggression Scale (DIAS) (Björkqvist et al., 1992) and the Indirect/Social/Relational Aggression scale (ISRA) (Coyne et al., 2006). In these examples gossip is equated with expanding rumors and measured using only one item (commenting badly about others). Other instruments that do define gossip more widely, including both the positive and the negative potential, focus on specific aspects, such as the content of the message (Nevo et al., 1993) attitudes toward gossip (Litman and Pezzo, 2005), motivations behind gossip (Beersma and Van Kleef, 2012), and functions of gossip (Foster, 2004), but they do not measure the frequency of the behavior itself (Brady et al., 2017). A final limitation of the latter group of scales is the lack of European and Latin American samples used in their validation. Below, several existing instruments are described in a bit more detail.

The widely known Tendency Gossip Questionnaire (Nevo et al., 1993) was validated in a study with 120 Israeli college students. It consists of a 7-point frequency scale of 19 items that assess the content of conversations (physical appearance, achievement, social information, and sublimated gossip). This scale has been criticized for having too few items with which to measure the wide content variability possible in gossip (Litman and Pezzo, 2005).

The Gossip Functions Questionnaire (Foster, 2004) measures agreement with 24 statements related to the four basic functions that gossip is said to have: information provision, creation of friendship, attainment of influence, and entertainment. More research focussing on these functions has subsequently taken place with English speaking countries (USA, Canada, Australia, United Kingdom, Ireland and New Zealand) (Watson, 2011; Hartung and Renner, 2013), but both the original scale and the subsequent work fail to consider that gossip is subject to the context in which it happens (Paine, 1967) and that it is not a state (trait) or aptitude.

The Motives to Gossip Questionnaire (Beersma and Van Kleef, 2012), validated with university students in the Netherlands, does include context to some extent: 22 statements referring to specific behaviors and grouped in four dimensions (inform, entertain, influence, and meeting the norms of the group) were presented to undergraduate students, and the reasons for getting involved in their latest gossip experience were assessed. The drawback of this instrument is that it only assesses the respondents' latest gossip event and not the frequency of the behavior. Another criticism is that its influence dimension is solely negative, being related to indirect aggression. Stirling's (1956) and Foster's (2004) original studies recognize there can also be a positive side to the influence dimension, which is related to the learning of the group's culture and norms.

Brady et al. (2017) recently validated the Workplace Gossip scale with work-experienced Canadian students, which does include the frequency with which gossip is carried out and thereby claim this study helps resolve the lack of generality of the construct and the possible measurement bias from which previous studies suffered. While this is indeed a good step forward, its focus on the work environment, with its specific issues and dimensions, mean it might not be as relevant for other contexts.

The scales discussed above have all been designed and validated for (young) adults. For children and adolescents, gossip research exists mainly in the form of narrative and observational studies, and these approach gossip from the same narrow negative angle of indirect agression aimed at damaging others (Xie et al., 2005) as some of the studies that focus on adults. Results from evolutive studies show that as boys and girls grow up, they increasinly demonstrate indirect aggressive behaviors, such as gossip, to victimize their peers (Archer and Coyne, 2005). In ethnographic studies, adolescent girls are said to gossip more than boys with the aim to keep their social status (Merten, 1997). This view on the relationship between gender and gossip—understood in its malignant version—is confirmed in a review by McAndrew (2014). On the other hand, meta-analyses on gender differences in indirect aggression, including gossip, do not find a consistent relationship (Archer, 2004; Card et al., 2008).

Research into children and adolescents needs to adopt the methodological advances that were made in the study of gossip among adults, and include its positive potential to transmit understanding of group norms and see it as part of the set of socially competent behaviors (McDonald et al., 2007). It also needs to reflect the changes in interaction and social relationships brought about by ICT (Lee et al., 2016). Examples of scales that have focussed on online behaviors among adolescents and that cover cybergossip are the Cyber-aggression Questionnaire for Adolescents (CYBA) (Álvarez-García et al., 2016) and the European Cyberbullying Intervention Project Questionnaire (ECIPQ) (Del Rey et al., 2015). However, these suffer from the same limitation as some of the off-line gossip scales, in that they treat cybergossip as a form of cyberaggression and do not include positive forms of gossip. A focus on gossip in its wide sense is especially relevant in adolescence, because at this stage relationship styles are consolidated. Identifying and describing the behaviors that characterize these relationship styles is of great research interest because they predict important aspects of social and personal functioning in adulthood (Perry and Pauletti, 2011). This argument is strengthened by findings from Engelmann et al. (2016), who observed positive effects of gossip already in childhood.

To provide in the need for a cybergossip instrument for children the Cybergossip Questionnaire-Primary scale was designed and validated for schoolchildren aged 10–12 (López-Pradas et al., 2017). It was designed to meet the abovementioned needs relating to measurement, definition and the role of ICT. The instrument includes nine types of behaviors covering the four functions of gossip (influence, entertainment, information, and friendship). Validation showed the scale has the optimal psychometric properties to study cybergossip among primary school children, but, to date, no such scale exists for adolescents.

The first objective of this study was to validate the Cybergossip Questionnaire-Primary scale (López-Pradas et al., 2017) for use among adolescents as well. A separate cyberbullying scale was used to analyse the discriminant validation of the cybergossip scale. The study used samples from two countries in different regions, Spain and Colombia, that have cultural similarities, a shared history and language. This ties it into the current trend of carrying out cross-cultural comparative studies, but adds diversity as, to date, these comparisons have most commonly taken place between European countries or between the United States and different Asian countries (Romera et al., 2017). The second objective was to measure the frequency of adolescents' use of cybergossip, and to analyze the possible differences by country of origin and gender. This validated scale will allow future research to gain further insights into de norms, values, conventions and behaviors of this age group when using ICT (Romera et al., 2016).

Our hypotheses were: (1) the instrument will show a one-factor structure with optimal psychometric properties, homogeneity of measure across gender and country and will have a low association with cyberbullying; (2) Colombian adolescents will be less involved in cybergossip, due to their lower use of ICT (Said-Hung, 2014); and (3) there will be no gender differences in the frequency of cybergossip (Card et al., 2008).

The overall sample comprised 3,747 adolescents (48.5% boys) aged 10 to 19 (M = 13.98; SD = 1.69). The Colombian subsample comprised 1,931 students (46.9% boys), aged 10–19 (M = 14.22; SD = 1.89), and the Spanish subsample comprised 1,816 students (50.2% boys) aged 12–19 (M = 13.69; SD = 1.35). The study used a convenience sample based on accessibility.

The research had a transversal, restrospective, ex post facto design, with one group and multiple measures (Montero and León, 2007). The study was approved by the Comité Coodinador de Ética de la Investigación Biomédica de Andalucía (Coordinating Ethics Committee of Biomedical Research of Andalusia) and was in accordance with all regulations concerning profesional ethics as stated in the International Conference on Harmonization Good Clinical Practice Guideline. The study was approved by the Spanish school boards and the Colombian schools' management, and written parental consent was obtained. In both countries, the students were visited and the anonymous, confidential and voluntary nature and the objective of the study were explained before the survey was taken. The authors were responsible for the data collection. Students had 20 min to fill in the questionnaire. The researchers were present during this time to answer any questions.

Mardia's coefficient was calculated to assess the multivariate normality of the dataset using the program “R” (R Development Core Team, 2008) and package “MVN” (Kormaz et al., 2015).

The psychometric properties of the CGQ-A scale were verified through item response theory (IRT) analysis, calculating a three-parameter model (3PL) fitted to polytomous scales (Muraki, 1990). This approach gives values for each item on the parameters discrimation (a), difficulty (b) and probability of success or failure (c). This analysis was carried out using “eRm; colospace and mirt” libraries (version 3.3.2) in “R” (R Development Core Team, 2008).

To assess construct validity, confirmatory factorial analyses (CFA) were carried out using the EQS 6.2 program (Bentler and Wu, 2012). The maximum likelihood estimation method (MLE), robust scaling (Bryant and Satorra, 2012), and polychoric correlations (Morata-Ramírez and Holgado-Tello, 2013) were used, which is the recommended approach for categorical variables and under the absence of multivariate normality. The fit of the models was tested with the following indexes: Satorra-Bentler scaled chi square (χS-B2) (Satorra and Bentler, 2001); the comparative fit index (CFI) and the non-normality fit index (NNFI) (≥0.90 is adequate; ≥0.95 is optimal); the root mean square error of approximation (RMSEA) and the root mean square residual (SRMR) (≤0.08 is adequate; ≤0.05 is optimal) (Hu and Bentler, 1999). The Akaike information criterion (AIC) was used to compare the obtained models, where the best model has the lowest value.

The generalization of the model, that is, the degree of robustness of the factorial structure or invariance, was tested through multi-group analysis, with country and gender as analysis criteria. This analysis consists of comparing a set of increasingly restrictive models. In Model 1, configural invariance was tested by imposing the same factorial structure on both subsamples, and checking whether the fit indexes of the combined model indicated good model fit. Subsequently, in Model 2 the factorial loads were restricted and the fit indexes of Models 1 and 2 were compared. Changes (Δ) between the models of NNFI, CFI, RMSEA, and SRMR ≤0.01 were accepted as evidence of measurement invariance (Dimitrov, 2010). As a further test of invariance, the chi square difference test (ΔχS-B2) was used, where non-significant differences demonstrate invariance in the models (Bollen, 1989; Satorra and Bentler, 2010). The multi-group analysis was carried out in EQS 6.2 (Bentler and Wu, 2012).

The discriminant validity was tested through Spearman's Rho correlations between the CGQ-A scale and the ECIPQ dimensions (cybervictimization and cyberaggression).

Internal consistency was analyzed with McDonald's Omega index (Ω), recommended for categorical variables under the absence of multivariate normality (Elousa-Oliden and Zumbo, 2008). For this we used the 9.2 Factor program (Lorenzo-Seva and Ferrando, 2006). The composite reliability (CR) was determined to test general reliability of the set of items, and we looked at the average variance extracted (AVE) to test the accuracy with which the construct is measured. The used cut-off values were 0.70 for CR and 0.50 for AVE (Hair et al., 2005).

The level of involvement in cybergossip was calculated by taking the average of the nine items. The adopted statistical significance level was 0.05.

Mardia's analysis generated a coefficient of skewness of 63.027; p > 0.05 and a coefficient of kurtosis of 279.537; p > 0.05, indicating the data fail to meet the assumption of multivariate normality.

The 3PL (IRT) analysis provided discrimination values above 1, which are considered good values; the difficulty degree of the items ranged from 0.05 to 1.60, which is acceptable (acceptable values range from −3 to 3); and the probability of failure values were low, indicative of high quality items (Baker, 1992) (see Table 1).

The inter-item correlation analysis showed values in the upper-middle range. The correlations between the CGQ-A scale and the ECIPQ cyberaggression and cybervictimization dimensions were in the medium range, suggesting a low collinearity and discriminant validity between the instruments (see Tables 2, 3).

The CFA of the one-dimension structure of the CGQ-A scale showed an adequate fit: χS-B2 = 286.581; df = 27; p < 0.001; NNFI = 0.979; CFI = 0.984; RMSEA = 0.055 (90% CI [0.050, 0.061]); SRMR = 0.048; AIC = 232.581 (see Figure 1).

The obtained value of the internal consistency was within the optimal range (Ω_CGQ−A = 0.839).

The values of the CR and AVE indexes were also found within the optimal range: CR = 0.904; AVE = 0.513, over 0.70 y 0.50, respectively (Hair et al., 2005).

The results of the multi-group analysis indicated proof of configural and measurement invariance for both the country and gender tests. For both variables, the chi-square differences (χS-B2) between Models 1 and 2 were non-significant and the deltas (Δ) of the CFI, NNFI, RMSEA and SRMR indexes in all comparisons were below the 0.01 cut-off point (see Table 4).

The adolescent participants in the study had an overall cybergossip involvement level of 1.36 (SD = 0.49). Cybergossip occurred more often among the Spanish participants (M = 1.57; SD = 0.59) than the Colombian ones (M = 1.24; SD = 0.37) (t = 17.48; p > 0.001). As also hypothesized, no differences were found between girls and boys (t = 1.75; p = 0.08; M = 1.35; SD = 0.47 vs. M = 1.38; SD = 0.53, respectively).

A number of limitations of this study must be mentioned. No explicit tests of convergent validity have been included. These were omitted as this study's particular aim was to observe its divergent validity, which is why we focussed on the correlation between the concepts. Another potential limitation is that no test-retest measures have been employed to confirm the reliability of the results, although different samples have been used and the optimal psychometric properties of the CGQ-A scale have been established.

It is necessary to continue advancing our understanding of gossip, especially of its effect on group characteristics, following recent evolutionary studies that highlight the influence of social networking on social behavior (Wu et al., 2016). The role of gender in cybergossip also requires further attention, for example with respect to the possible nuances that relational variables such as friendship may introduce (Watson, 2012). Another interesting direction of research would be to follow up on the findings by Areepattamannil and Khine (2017) that personal (availability, ability, habit, uses) and motivational (self-efficacy, interest, enjoyment) factors correlate with the use of ICT for gossip. It is also necessary to generate new research of a developmental nature to better understand what is behind differences in involvement in gossip and cybergossip.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.