Introduction

Cyber aggression is unfortunately very common in the everyday lives of many adolescents, and it can be very harmful -even more so than face-to-face aggression- to those subject to it (^{Sticca & Perren, 2013}). Observation and measurement are crucial to understanding cyber aggression and to design better prevention strategies. However, to measure aggression (any action intended to harm others; ^{Parke & Slaby, 1983}), or any reaction to it, be it as a victim or a bystander (^{Macháčková, Dedkova, Sevcikova, & Cerna, 2018}), is a great challenge. Many scientific studies of aggression and cyber aggression are based on self-reports (e.g., ^{Burt & Alhabash, 2018}; ^{Kim et al., 2020}; ^{Kokko et al., 2014}; ^{Martínez, Rodríguez-Hidalgo, & Zych, 2020}; ^{Menin, Guarini, Mameli, Skrzypiec & Brighi, 2021}; ^{Runions, Bak, & Shaw, 2016}; ^{Wright, 2017}). However, a great limitation of self-reports is the possible social desirability by which participants often do not respond with absolute honesty as they wish to project a positive image of themselves. Another alternative is peer reporting (e.g., ^{Hooijsma, Huitsing, Dijkstra, Flache & Veenstra, 2020}), that, although possibly more reliable that self-reporting, can also be biased given that peers may have limited perceptions of the situations, or preferences, and particular relations of friendship with their peers. Reports by teachers (e.g., ^{Dickson et al., 2015}) and parents (e.g., ^{Bartels et al., 2018}) are common, but they too can be limited, in part because adults often do not find out about situations of aggression that occur between children or adolescents (e.g., ^{Atlas & Pepler, 1998}). An important alternative is direct observation of aggressive behavior. Pepler and her group designed a system to observe aggressive situations using wireless cameras and microphones (Pepler & Craig, 1995). By doing so, they were able to measure the frequency of this behaviour very precisely and, as well, identify the social dynamics involved in many of these situations (e.g., Atlas & Pepler, 1998; Craig & Pepler, 1997). However, this methodology is technically complex and requires a great amount of time, in part, because aggression is usually not so frequent (except during the first years of life; ^{Tremblay, 2002}).

Aggression has also been observed in experimental contexts. For example, there are a number of experimental paradigms in which participants can decide on the intensity of a negative stimulus (e.g., noise, electric shock, hot sauce) received by alleged opponents in a competition (e.g., ^{Anderson & Bushman, 1997}; ^{Lieberman, Solomon, Greenberg & McGregor, 1999}). However, these experiments present limitations too. For example, they differ from common situations where people exercise aggression daily, and this affects the external validity of the experiments, that is, the possibility to generalize the results to other contexts. ^{Anderson and Bushman (1997)} argue that this effect is often necessary to guarantee high internal validity; that is, to be sure that the causal relationships between the variables are not due to uncontrolled variables.

To achieve both internal and external validity, it would ideally be necessary to control variables in an experimental context that is very similar to the context where the daily interactions occur. Electronic media present an opportunity to do just this as they provide a specific environment to directly observe aggressive behaviour, and/or reactions to it, in contexts that are similar to how this behaviour occurs daily in current types of aggression, at least, among adolescents. Electronic contexts enable us to simulate very similar interactions to those that occur on social networks and to plan credible aggressive interaction to be able to systematically study participants’ reactions to aggressive situations.

In this article, we report the design and early results given by SIMA, a system that simulates online chats and that enables direct observation of adolescents’ bystander responses to what they believe is real cyber conflict. We report the reactions of the participants to the SIMA simulator, as well as the evolution over time of their different types of responses and some preliminary findings related to differences by sex, type of school, and topic of the argument.

Given the importance of the role of bystanders in the dynamics of aggression, we focus on their bystander responses to these situations. Observational research as that conducted by Pepler and her group has shown that what peer bystanders do has a great impact on whether the aggression increases or reduces (^{Atlas & Pepler, 1998}; Craig & Pepler, 1998). In fact, several prevention programs have been aimed at promoting assertive (i.e., emphatically defending their own and others’ rights but without harming others; ^{Lange & Jakubowski, 1980}) and empathetic (i.e., feeling something similar or compatible to what others are experiencing; ^{Hoffman, 2000}) bystander intervention (^{Chaux et al., 2017}; ^{Kärnä et al., 2011}; ^{Wölfer et al., 2013}). We also analyzed responses to situations of cyber conflict given that almost all the studies conducted on the topic have focused on cyberbullying; that is, repeated and systematic cyber aggression against someone, usually in a situation of vulnerability and power imbalance (^{David-Ferdon & Feldman, 2007}; ^{Del Rey, Mora-Merchán, Casas, Ortega-Ruiz & Jaén, 2018}; ^{Grigg, 2010}; ^{Redondo, Luzardo & Rangel, 2016}; ^{Wright, 2015}; ^{Wright, et al., 2015}). We found no studies on cyber aggression where there is no clear power imbalance among those involved and where aggression could go in both directions, despite that this type of cyber aggression is also very common in the lives of today’s adolescents, as pointed out by ^{Corcoran, Mc Guckin & Prentice (2015)}.

Some studies have shown that it is possible to take advantage of digital contexts to study the way that different variables can affect behaviour. ^{Cheng, Bernstein, Danescu-Niculescu-Mizil & Lezkovec (2017)}, for example, conducted an experiment where participants interacted on a video chat on controversial political topics (e.g., the 2016 US presidential camping). During the experiment, the researchers sought to manipulate: 1) the participants’ mood (before interacting on the chat, half of the participants had to take a very easy quiz, and the other half, a very difficult one); and 2) the context prior to the discussion (half had been subjected to prior insulting comments, and the other half to neutral comments). ^{Cheng et al. (2017)} found that those that began the chat with a negative affect or in a context of having received prior offenses resorted to offenses more frequently than those who received prior positive affect in a neutral context. Therefore they showed that it is possible to influence the frequency of cyber aggression by manipulating the affective and contextual variables.

Similarly, ^{Burt, Kim & Alhabash (2020)} developed a simulated chat to measure digital aggression. In a reaction-time game adapted from the Taylor Aggression Paradigm (TAP; ^{Chester & Lasko, 2019}), participants received offensive chats from fictitious co-players who they thought were competitors, and their responses in the chat were recorded and analysed. Aggressive responses in their chats were found to correlate significantly to other measures of self-reported aggression and aggressive content in their Twitter posts (^{Burt et al., 2020}; ^{Kim, Clark, Donnellan & Burt 2020}).

With this study, we sought to contribute to our understanding of adolescents’ behavior in online contexts by piloting a simulator that allows the direct manipulation of the content of the interactions. Specifically, we intended to conduct a pilot implementation of the SIMA simulator and answer the following research questions:

Method

Results

Participants’ offenses against the alleged members of the chat group were very frequent. On average, each participant used 4.3 offenses during the simulation. The frequency of aggressive interventions increased continually during the interaction. In fact, in the last 10 interactions, there were approximately the same number of offenses as there were in the first 35 interactions (see Figure 1). There were no significant differences between public and private schools (see Table 1), or between male and female (see Table 2). But differences were observed for topics: those who were assigned to discussion about tattoos made more offensive comments than those who had to discuss environmental topics (see Table 3).

Figure 1 Average cumulative counts for offenses directed at those supposedly in the chat group and others not present in the chat group. 

Offenses against those who were not part of the chat were less frequent (1.3 on average) and they were concentrated in the second quarter of the time for the interaction (see Figure 1); that is, when the planned offensive comments began (e.g., 12: “and, they are ignorant, for example, the bullfighters…”; 13: “and what about those who go to see bullfighting? Are they ignorant too?”). There were no differences between types of school or between sexes (see Tables 1 and 2), but differences were observed for topics: those who discussed environmental topics or animal maltreatment offended others to a significantly greater extent than those who argued about tattoos (see Table 3).

As can be seen in figures 2 and 3, defending assertively him/herself and the other members of the chat increased substantially in the second half of the simulation. Assertively defending chat members was two times more frequent than assertively defending themselves. In contrast, aggressively defending themselves was more than three times more frequent than aggressively defending chat members. Significant and substantial differences were found between the types of school in three categories related to the defence. In particular, interventions that assertively defended members of the chat session, assertively defending others who were not in the chat groups, and the participants aggressively defending themselves were much more common among participants from public schools (see Table 1). On the other hand, assertively defending others who were not in the chat group and aggressively defending themselves were much more common among female than male participants (see Table 2).

Figure 2 Average cumulative counts for assertively defending themselves, defending those supposedly in the chat group, and defending others not present in the chat group. 

Figure 3 Average cumulative counts for aggressively defending her/himself, aggressively defending other members of the chat group, and aggressively defending others who are not members of the chat group. 

Both conventional social interactions, and thematic interventions were more frequent among public school students than among private school students (see Table 1). We found no significant differences by sex (see Table 2). Social interactions occurred almost exclusively in the first 6 and last 3 iterations, whereas the thematic interventions were more frequent in the first interactions than in the last. Less than 10 % of the thematic interventions took place in the last quarter of the time allocated to the interaction (see Figure 4).

Figure 4 Average cumulative counts for social, thematic, and other unclassified interventions. 

After finishing the chat, the participants were asked to answer open questions. An analysis of their answers to the questions “What did you think of the interaction?” and “What did you think of the other people in the chat group?” indicated that the majority seemed to have believed that the chat session was real. None of the answers indicated that they had not believed that the interaction was real, and only 8.3 % of the participants responded in a way that suggests that they may have suspected that the interaction was not real (e.g., “it was a bit strange”; “very strange”). In contrast, 69 % responded in ways that suggest that they did think that the interaction was real (e.g., “they are very disrespectful”; “some of them are respectful but they become aggressive when defending their point of view”).

Discussion

Electronic media provide a great opportunity to directly measure behaviour in digital contexts like those where a large part of today’s interactions among adolescents take place. These measurements can be much more reliable than some methods currently used to measure behaviour in contexts of aggression, as self-reporting or reports by adults. They allow this analysis to be conducted in a context where variables can be manipulated to directly observe the causal effect of certain variables on behaviour. For example, specific variables could be manipulated to see whether this influences the participants' behaviour (e.g., whether those who received an intervention to promote empathy intervene more when it comes to deescalating conflict). ^{Anderson and Bushman (1997)} suggest that when seeking to increase external validity (i.e., results that can be generalized to real contexts) conducting studies in real contexts often ends up affecting internal validity (i.e., being confident that the variables are, in fact, casually related) given that in real contexts it is impossible to control the variables that can affect the results (e.g., confound variables). Computer simulations like the ones presented here simultaneously allow high internal and external validity, given that they can be conducted in experimental contexts where the variables are well-controlled and, at the same time, in context similar to real ones where online interactions take place on a daily basis. The fact that most of the participants seemed to have believed that they were taking part in a real online conversation indicates that simulators like SIMA have great potential to simultaneously achieve both types of validity.

In this study, we found that cyber aggression was very common. Following the thematic interventions, the offenses against the other (alleged) participants in the chat group were the most common interventions. On average, each participant wrote more than 6 offenses, 4 of which were directed at other members of the chat group. Also, 85 % of the participants wrote at least one offence. This highlights a worrisomely high frequency of aggression, being consistent with studies that suggest that the distance to the victim and anonymity on social networks favour the use of aggression (^{Arab & Díaz, 2015}), in part, because online contexts seem to be associated with lower levels of empathy (^{Carrier, Spradlin, Bunce & Rosen, 2015}).

No differences by sex were found in terms of the total amount of cyber aggression. This is consistent with studies carried out in different cultures that show that there are few differences between men and women in the frequency of indirect aggression (i.e., when the victim does not know who began the aggression), in contrast to direct physical and verbal aggression that is more common among men than it is among women (^{Card, Stucky, Sawalani & Little, 2008}; ^{Österman et al., 1998}; ^{Salmivalli & Kaukiainen, 2004}). In this case, the anonymity of chats allows people to offend others assuming that victims will never know who offended them. More studies are needed to verify whether this lack of difference between men and women also occurs in other cultures.

Assertive bystander intervention is fundamental to stopping situations of aggression in face-to-face contexts (^{Atlas & Pepler, 1998}; ^{Craig & Pepler, 1997}; ^{Salmivalli, 2010}, Salmivalli, 2014) and probably also in online contexts (^{Constantino & Álvarez, 2010}). In this study, however, only 38 % of the participants intervened assertively to defend those who they thought were other adolescents being victims of cyber aggression. Also, these assertive interventions were significantly more frequent between students from the private school than those from the public school. It is not clear whether this difference is associated with more empathy or assertiveness among students from the private school, with a different perception of the seriousness of the offenses, or something else. In fact, many comments from the students from the public school suggest that some of them did not consider that the attacks were particularly alarming. For example, when asked what happened during the interaction, one of the participants answered, “nothing important”, and another, “a few fights but it was very entertaining”. It would be interesting to conduct a qualitative exploration of whether the students from different contexts could be interpreting the level of seriousness of the offences differently and, therefore, the need to intervene to defend those being attacked. The results also showed a greater frequency of assertive interventions among female adolescents than male adolescents, when assertively defending other people that were not part of the chat group (e.g., youngsters with tattoos). Another important element to explore in future studies is whether this difference is associated to greater sensitivity or empathy, a greater capacity to intervene assertively or some other difference.

It is important to recognize that the study had several limitations that could be corrected in future versions. On the one hand, answers to the open questions suggest that the great majority of participants believed that the chat session was real, but we cannot rule out that some may have realized that it was a simulation. For example, some students may have seen their classmates’ chats and discovered that they had very similar interactions in the public school where there was very little space from one computer to another. We tried to avoid this by staggering the topics between the computers and changing most of the aliases of the fictitious characters, but we cannot rule out that they may have seen each other’s interactions anyway. Future technological adjustments will help to minimize the possibility of these risks occurring.

Another limitation of this study is that the classification of the participants’ interventions depended on our interpretations of them. We may have misunderstood what the participants might have wanted to say. This is particularly possible given the ambiguity in short interactions typical of online chat sessions, as well as the generational difference between participants (adolescents) and the assessors (adults). Even though the KAPPAs obtained were high, we cannot ensure that what coders interpreted in a coherent manner correspond to what the participants intended to say. Additionally, coders were conscious of the hypotheses of the study, which could have biased their interpretations. Furthermore, it was necessary that coders reached agreements related to some of the entries since students used local expressions that were not known by both researchers. For this reason, coding was not totally independent and KAPPAs may be overestimated. Future studies might need independent coders, closer in age to the participants, from their local contexts, and unaware of the hypotheses of the study. Interviewing the participants associated to their interventions may help to confirm the validity of the interpretations. Also, future studies could include emoticons to highlight the participants’ emotional experiences during the chat session.

We also found significant differences in some of the categories among the groups assigned to different topics of interaction. For example, there were more assertive interventions amongst those assigned the topic of tattoos than among those assigned the topic of animal maltreatment and the environment. This is surprising given that the interventions of the alleged peers were planned to be equivalent in terms of the type of intervention (e.g., topic), their level of aggression, and when these took place. It is possible that the different topics may have aroused different levels of empathy and, therefore, different dispositions to intervening assertively to defend the alleged participants of the chat group. Future research should determine whether it is possible to obtain greater coherence among the discussion topics, or whether the simulation should be limited to single topics to guarantee comparability between the interventions by the different participants.

Another important aspect to explore in future research is a possible adaptation of the system to simulate other types of situations. In fact, we have already constructed a variation to the system that corresponds more closely to a situation of cyberbullying than one of cyber-conflict. In the situation reported here, the aggression is bilateral and there is no clear power imbalance between the two mutually aggressive alleged peers. This corresponds more to a situation of conflict than it does one of bullying (^{Chaux, 2012}). In contrast, in cyberbullying, all the aggression of the alleged peers is directed at one specific peer who, in turn, does not seem to be able to defend him/herself in a clear show of power imbalance. We do not know yet whether cyberbullying or cyber-conflict generates more assertive interventions by the participant-bystanders.

Despite the limitations, SIMA provides a great opportunity to directly explore adolescents’ behaviour in digital spaces very similar to those where situations of aggression occur on a daily basis. These simulators could lead to a new drive towards research in social psychology that could not previously be conducted for ethical reasons (e.g., ^{Burger, 2009}). In addition to its usefulness as a methodological tool through which to observe behaviour, the simulator has many practical implications. On the one hand, it could be used to assess the impact of programs aimed at promoting assertive bystander intervention in situations of online aggression. On the other, it can serve as a pedagogical strategy; for example, for students to practice how they could assertively intervene in situations of online aggression, as the system would allow them to analyse their own reactions and identify what they could improve on and how. Finally, identifying variables that can explain the bystander behaviour in situations of cyber aggression can inform future programs where variables should be taken into account to further promote assertive bystander intervention. All this can contribute not only to a better understanding of aggression but also to promoting more peaceful and constructive online environments.