Participants

The sample was convenient and included 799 social media users (338 men, 42.3 %, and 461 women, 57.7 %), with a mean age of 39.7 years (SD = 13.84). A total of 7.1 % (n = 57) were foreign nationals residing in Argentina. Most participants were employed (n = 629, 78.8 %). With respect to education, 38 % (n = 303) reported having completed university or tertiary studies, among whom 12.5 % (n = 38) had completed postgraduate studies. Fourteen percent (n = 112) reported having completed secondary school. Most participants self-identified as belonging to middle (n = 474; 59.4 %) or upper-middle (n = 138; 17.3 %) socioeconomic levels. The participants were also asked about the number of hours per day they engaged in online activities requiring connection (1: I do not use the internet; 2: Less than one hour; 3: 1 hour; 4: 2 hours; and so forth up to 24 hours). The average daily connection time was 8.12 hours (SD = 4.75).

Materials

Negative Comments on Social Media Survey. A self-developed survey, which is based on the content and format of a previous instrument on negative comments and trolling experiences (^{Akhtar & Morrison, 2019}), was adapted to the local population in terms of terminology and commonly used social networks. As the questionnaire includes both closed- and open-ended questions oriented toward specific experiences, it is not a standardized scale for measuring latent constructs. For this reason, reliability and validity studies, either in their original or adapted versions, are lacking, as their purpose is exploratory and descriptive, with a focus on the collection of factual self-reported information regarding social media use and the reception of negative comments.

The first section of the survey explored the type of social network used, time spent, and primary activities performed on each network, followed by the negative comments received. Each social network is treated individually, given that most studies focus on the use of a specific network, especially Facebook (e.g., ^{Ellison et al., 2007}). For each listed network, participants indicated on a Likert scale: time of use (1: I do not use it to 8: I use it more than 4 hours per day); whether they received negative comments (1: Never to 5: All the time); whether they engaged in live streaming, stories, or reels (1: Never to 7: Almost all the time); the perceived negative impact of comments received for those activities (1: Not at all to 5: Very much); whether they posted on others’ accounts (1: Never to 5: Always); and the perceived impact of receiving negative comments on those posts (1: Not at all to 5: Very much). Finally, the participants rated the overall perceived impact of having received negative comments on social media (1: Not at all to 5: Very much).

In the second section, the participants indicated whether the negative comment came from an identifiable or nonidentifiable account, the content of the comment (e.g., false information, political, physical, racial, or sexual), and the main action they took in response (e.g., I read it and do not respond, I read it and sometimes respond, I read it fully, I always read and respond).

In the third section, three open-ended questions were included: why they believed they received negative comments; what action they usually took after receiving them; and an optional space to provide any further comments on the issue. Finally, sociodemographic data were collected (e.g., sex, age, place of residence, educational level, occupation, and socioeconomic level).

Big Five Inventory (BFI; ^{John et al., 1991}; Argentine adaptation by ^{Castro Solano & Casullo, 2001}). This instrument consists of 44 items rated on a five-point Likert scale (1: Strongly disagree to 5:Strongly agree). It assesses five personality traits (extraversion, agreeableness, conscientiousness, neuroticism, and openness to experience). The original authors demonstrated its validity and reliability in adult general populations in the United States. Those studies confirmed the concurrent validity with other recognized personality measures. Studies conducted in Argentina have verified factorial validity for adolescent populations, nonclinical adult populations, and military populations (^{Castro Solano & Casullo, 2001}). In all the cases, a five-factor model explained approximately 50 % of the variance in the scores. For the present sample, internal consistency was as follows: extraversion: α = .79, ω = .79; agreeableness: α = .70, ω = .72; conscientiousness: α = .81, ω = .82; neuroticism: α = .85, ω = .83; and openness to experience: α = .79, ω = .82.

Inventory of the Five Personality Continuums-Short Form (ICCP-SF; ^{de la Iglesia & Castro Solano, 2023}). This instrument was developed for use in the Argentine population and operationalizes the Dual Personality Model, which measures five positive traits (serenity, humanity, integrity, moderation, and sprightliness) and five pathological traits (negative affect, detachment, antagonism, disinhibition, and psychoticism). It contains 55 items rated on a six-point Likert scale (0: Strongly disagree to 5: Strongly agree). Psychometric studies included exploratory and confirmatory factor analyses; internal consistency analyses (Cronbach’s alpha and McDonald’s omega); convergent validity studies with measures of positive, pathological, and normal traits; and external validity analyses with relevant indicators (well-being and psychological symptoms). For the present sample, the α values for the trait scales ranged between .76 and .90, and the ω values ranged between .87 and .93.

Mental Health Continuum-Short Form (MHC-SF; ^{Keyes, 2005}; Argentine adaptation by ^{Lupano Perugini et al., 2017}). This 14-item instrument assesses (a) emotional well-being, defined in terms of positive affect and life satisfaction (hedonic well-being); (b) social well-being (including acceptance, social actualization, contribution, coherence, and integration); and (c) psychological well-being, based on Ryff’s theory (1989) (autonomy, mastery, personal growth, positive relations, self-acceptance, and purpose). The items ask how often respondents experienced certain emotions on a Likert scale ranging from 0: Never to 5: Every day. Validation studies in Argentina confirmed the three-factor structure and its invariance by sex and age. Evidence of convergent and divergent validity has also been obtained (^{Lupano Perugini et al., 2017}). The reliability coefficients for the present sample were as follows: emotional well-being: α = .84, ω = .85; social well-being: α = .75, ω = .77; and psychological well-being: α = .85, ω = .85.

Procedure

Data were collected by advanced students completing a research practicum at a private university in Buenos Aires, Argentina. Participation was voluntary, and no compensation was provided. Surveys were administered online via the SurveyMonkey platform. On the survey’s introductory page, participants were asked to provide informed consent, the anonymity of the data was assured, and it was stated that the information would be used exclusively for research purposes. Data collection was supervised by a faculty researcher.

The study adhered to international ethical guidelines (APA and NC3R) as well as the standards of the National Scientific and Technical Research Council of Argentina (CONICET) for ethical conduct in the social sciences and humanities (Resolution No. 2857/2006).

Data analysis

Descriptive statistics and Spearman’s correlations were calculated to examine associations between variables, along with z tests with Holm-Bonferroni correction for comparing proportions. The predictive capacity of normal, pathological, and positive personality traits for perceived impact was evaluated via hierarchical multiple regressions, verifying the basic statistical assumption. Open-ended responses were processed by preprocessing techniques and lexical frequency analysis.

For descriptive statistics, frequencies, correlations, and regression analyses, we used Jamovi software, version 2.2.5 solid, which operates through the R environment.

For automated text analysis, we employed the Quanteda package (version 3.3.0) in R (version 4.3.2) via RStudio (version 2024.04.2+764). Preprocessing included conversion to lowercase; removal of Spanish stop words (Quanteda stopword dictionary); elimination of punctuation, numbers, and special characters; and lexical normalization and lemmatization. Stemming was also applied to group morphological variants of the same word, and unigram tokens were created, excluding higher-order n-grams. No minimum frequency threshold for tokens was applied beyond the preprocessing steps described. Intercoder reliability was not employed, as the analysis was automated.

Analysis of Negative Comments Received, by Platform, Comment Type, and Action Taken

First, the frequency of negative comments received was calculated according to the social network, the origin of the comment (identifiable vs. nonidentifiable accounts), the type of comment (e.g., political, physical appearance, or sexual), and the main action taken in response.

Table 1 presents the proportion of users who reported having received negative comments on each social network, in descending order, along with their corresponding 95 % confidence intervals. Only the percentages and confidence intervals for this group are reported, as they constitute the population of interest for subsequent analyses. Overall, the frequency was low across all platforms, with higher rates on Twitter and Instagram, followed by Facebook and WhatsApp, and levels close to 1 % on YouTube and Twitch.

Table 1: Reception of negative comments by the social media network 

Table 2 displays pairwise comparisons via the z test for differences in proportions, applying the Holm-Bonferroni correction for multiple comparisons. The results indicate that Twitter presented significantly higher percentages than did Facebook and WhatsApp but did not differ from Instagram. In turn, Instagram registered higher values than YouTube and Twitch did but no differences compared with Facebook or WhatsApp did. Both Facebook and WhatsApp presented higher proportions than YouTube and Twitch did, with no differences between them. YouTube and Twitch did not differ significantly.

Table 2: z tests for differences in the proportions of negative comment reception across social media networks 

Note. The exact significance values are reported in parentheses.

With respect to the sources of negative comments received, they were distributed similarly between identifiable accounts (n = 368; 53.48 % (95 % CI: 49.68-57.24)) and nonidentifiable accounts (n = 321; 46.52 % (95 % CI: 42.76-50.32)), with no statistically significant differences (z test for difference in proportions: z = 1.15, df = 1, p > .05).

With respect to content type, more than half of the participants who reported having received negative comments indicated that these comments were related to false information and/or political content. Notably, content categories are not mutually exclusive, meaning that the same individual may have received more than one type of comment. Similarly, a substantial proportion of participants reported not having received negative comments.

For actions taken in response to such comments, among the 319 participants who answered this question, 53.3 % reported that they usually read them without responding, whereas 33.2 % stated that they read them and sometimes respond. The remaining participants reported other actions with lower frequency.

Table 3 presents comparisons by sex. Compared with men, women were significantly more likely to receive comments from nonidentifiable accounts. In addition, women tended to receive negative comments related to false information, sexual aspects, or physical appearance more frequently, whereas men were more likely to receive negative comments related to racial aspects. With respect to actions taken in response to negative comments, women tended to read them without responding at higher rates than men did. No significant gender differences were found in the other types of responses.

Table 3: Source, Type of Comment, and Action Taken, by Gender 

Automated test analysis of negative comments

The participants were asked three open-ended questions aimed at exploring, first, why they believed they received negative comments; second, what actions they took after receiving such comments; and third, they were given the opportunity to provide an open comment on the topic. The data are presented in Table 4.

A content analysis was conducted on the participants’ responses. First, sentences are tokenized into words or units of meaning. Preprocessing included lexical normalization, removal of stop words, and semantic grouping of similar terms. If two words had a similar meaning (e.g., ignore, not respond), they were grouped into the same word category. Word frequencies were then calculated for the most frequent tokens. The same procedure was applied to the three open-ended questions.

Notably, a single participant could contribute multiple tokens within the same response. Therefore, the percentages presented in Table 4 correspond to the relative frequency of each token or lexical category over the total number of tokens identified rather than the number of participants. Categories were constructed by grouping tokens with closely related meanings.

Regarding why people believed they received such comments, the analysis of frequent words suggests that this was due to having different opinions or thoughts about people, society, or specific topics (e.g., due to discrimination, an attempt to bother others, differing opinions, or because what I write bothers them).

Second, in response to the question about the actions most frequently taken upon receiving negative comments on social media, 70 % of the actions consisted of blocking and reporting (e.g., block, report, delete). Twenty percent reported responding and/or contacting the commenter, and the remaining 10 % took no action.

Finally, in the open comments, although the words were less interpretable than in the previous questions, they indicated that such comments impact, affect, or harm people (e.g., they make me angry, they frustrate me and make me feel powerless, I try not to get into a ridiculous fight, they affect me when they are aggressive or violent).

Table 4: Content analysis of open-ended responses on negative comments received 

Relationships between the perceived impact of negative comments and personality traits (normal, pathological, and positive)

First, significant positive associations were found with the personality traits of neuroticism and negative affect. Likewise, significant negative associations were observed with agreeableness, serenity, and sprightliness (Table 6).

Next, a hierarchical multiple regression analysis was conducted to examine whether personality variables (normal, pathological, and positive) contributed to explaining the variance in the perceived impact of negative comments received (Table 7). Normal personality traits were included in the first step, pathological traits were added in the second step, and positive personality traits were incorporated in the third step. No control variables were included in the previous steps.

Before the models were interpreted, the assumptions of the analysis were tested. Graphical inspection of standardized residuals and normality tests indicated an approximately normal distribution without relevant skewness. Homoscedasticity analyses revealed that the variance of the residuals remained constant across the predicted values. In each model, the absence of multicollinearity was verified through the examination of variance inflation factors (VIFs), tolerances, eigenvalues, condition indices, and variance proportions. None of the models had VIF values greater than 5, tolerances less than .10, or condition indices above 30 accompanied by high variance proportions (> .50) in more than two predictors, indicating statistical independence among the variables.

Overall, personality traits explained 19 % of the variance in the perceived impact of the negative comments received. A comparison of the models revealed that including pathological and positive variables improved prediction beyond that provided by normal personality traits alone, which accounted for only 9 % of the variance. Pathology and positive personality traits together contributed an additional 10 % (5 % each).

Table 6: Correlations between the perceived impact of comments and personality variables 

Note. Significant correlations with a small effect size are shown in bold.

Table 7: Hierarchical multiple regression to predict the perceived impact of negative comments 

Note. CI: confidence interval; LL: lower limit; UL: upper limit. The β values represent standardized coefficients.

*p < .05. **p < .01. ***p < .001.

Limitations, future research directions, and practical implications of the findings

First, the main limitation of this study lies in the fact that trolling behaviors were analyzed in a general way on the basis of participants’ perceptions of comments that they considered negative. No distinction was made between different types of trolling behavior-whether oriented more toward the perpetrator’s simple enjoyment or toward a more aggressive connotation. Future studies could focus on differentiating the types of trolling behaviors, their characteristics, and their consequences.

Another limitation concerns the representativeness of the sample used, not only in terms of size but also in terms of the fact that most participants were residents of urban centers in Buenos Aires, with limited participation from other regions of the country. This sample was also highly educated, which may influence the way individuals interpret receiving negative comments on social media.

Regarding the data collection instruments, most were self-reported inventories, which may have led participants to orient their responses toward socially desirable patterns and avoid disclosing that they had been victims of attacks or ridicule online. These aspects may affect the generalizability of the findings; therefore, it would be desirable to replicate the analyses with more diverse and larger samples.

For future research directions, it would be useful to conduct international comparative studies since, to date, research has focused primarily on individual variables or characteristics of virtual environments without exploring whether trolling has the same prevalence and impact across different sociocultural contexts. Another relevant line of inquiry would be to analyze the frequency and impact of trolling by differentiating the type of social network (e.g., X, Instagram, TikTok), type of user (e.g., politician, artist, influencer, ordinary user), or type of content targeted by aggressive comments (e.g., political, artistic, social posts).

In addition, future studies could examine the relationships between personality traits and the aspects analyzed in the first objective. This would allow researchers to study whether users with certain personality characteristics are more likely to perceive themselves as victims of trolling than other profiles are and how they tend to respond and react to such attacks. Relatedly, it is also important to analyze the role of potential mediating or moderating variables in the relationship between personality factors and the well-being effects of both being a victim of trolling and engaging in trolling behaviors. For example, variables such as exposure time (^{Alavi et al., 2025}) or self-esteem levels may play important roles in this relationship (^{Zhou et al., 2023}).

The findings obtained provide insights that may inform practical implications at different levels and for different stakeholders. At the individual level, it seems important to strengthen factors that can serve as protective agents against disruptive online behaviors, such as self-regulation, moderation, and clarity of goals. This highlights the importance of developing informational campaigns, whether in educational or community settings, that inform users about both the benefits and risks of internet and social media use, as well as how to respond if one becomes a victim of trolling or other forms of online antisocial behavior (e.g., not responding, blocking). The aim is to foster what some authors call digital well-being, understood as a state in which subjective well-being is preserved in an environment characterized by abundant digital communication (^{Vanden Abeele & Nguyen, 2022}). The goal is for individuals to be able to use digital media in ways that promote a sense of comfort, safety, satisfaction, and personal fulfillment.

Finally, at the technological and public policy levels, it is important to continue refining tools that detect troll accounts to take appropriate measures on the basis of the harm they may cause to users. In summary, it is necessary to work toward the development of healthy virtual environments that allow people to access content and interact positively with others-fulfilling the original purpose of internet development.