Study Maps How Soft Skills Shape Students’ Peer Relationships at School

A student’s classroom friendships may do more than provide companionship: they may help shape confidence, empathy, curiosity and the ability to solve problems. A network-based study of more than 1,500 final-year high-school students in southern Italy has found that the informal relationships adolescents form with classmates are associated with distinct patterns of “soft skills”—transferable personal, social and practical abilities that influence learning, cooperation and future work. The findings suggest that classrooms function as social ecosystems in which competencies are reinforced through everyday conversations, emotional support, advice and shared tasks. The study does not show that friendship directly causes these skills to develop, but it offers evidence that the social structure of a class is closely linked to how students report their own abilities.

The research, led by Nunzia Brancaccio, Valeria Policastro, Giancarlo Ragozini and Maria Prosperina Vitale, surveyed students in 28 high schools across Italy’s Campania region. The participating schools represented lyceums, technical institutes, vocational schools and multi-track institutions across the provinces of Avellino, Benevento, Caserta, Naples and Salerno. In total, 1,532 students from 91 classes took part in the broader survey, which was administered in February 2025 during class time. Most participants were in their final year of high school, a period when decisions about university, employment and adult life make self-confidence, commitment and problem solving especially consequential. The sample was selected through a non-probabilistic, two-stage design and was not fully representative of all Italian high-school students, a limitation the researchers emphasise.

To measure soft skills, the investigators used the 3SQ Soft Skills Self-evaluation Questionnaire, a validated instrument containing 41 items rated on five-point scales ranging from “almost never/never” to “almost always/always.” The questionnaire assesses ten dimensions: self-confidence, openness, resilience, curiosity, collaboration, empathy, leadership, commitment, autonomy and problem solving. These dimensions are grouped into broader areas involving personal effectiveness, relationships and service, influence on others, and achievement-oriented capabilities. Students responded to statements such as “I believe in myself,” “I listen carefully to others’ opinions,” “I try to find alternative solutions when something does not work,” and “I keep my commitments even when it is difficult.” The scale showed high internal consistency, with reliability values ranging from 0.80 to 0.90 across dimensions and an overall Cronbach’s alpha of 0.92.

The researchers also converted each classroom into a social network. Rather than asking only whether students belonged to a group, the survey asked them to nominate up to five classmates for four types of connection: close friendship, conversations about personal or emotional matters, help with schoolwork, and discussions about future plans or educational choices. Friendship and emotional support were combined into a relational network, while scholastic and appraisal support formed a task-oriented network. Because students could identify the same classmate in more than one category, each connection could carry a weight of zero, one or two. The data were anonymised so that researchers could reconstruct patterns of relationships without exposing individual identities.

The analytical strategy combined psychometrics with network statistics. First, the team applied a graded response model, a form of item response theory, to estimate each student’s position on an underlying, or latent, soft-skill trait. This method does not simply add up questionnaire responses. It estimates how strongly each item discriminates among students and where the thresholds lie between response categories. In technical terms, the probability that a student selects a response category or higher is modelled as a logistic function of the student’s latent trait, the item’s threshold and its discrimination parameter. An item with high discrimination more effectively distinguishes students at different levels of a competency. The approach also lets researchers determine where a questionnaire is most precise: some items may provide more information for students with relatively low levels of a skill, while others are more useful near the middle or upper end of the scale.

In the second step, the estimated skill scores were entered into network autocorrelation models. These models examine whether connected students tend to have similar outcomes while accounting for characteristics such as gender, parents’ educational attainment and school type. Mathematically, the model includes a network-weighted average of classmates’ scores, multiplied by an autocorrelation coefficient called rho. A positive rho indicates that students linked within a network tend to have similar skill levels; a negative coefficient indicates an inverse pattern. Unlike a simple comparison of classroom averages, the method preserves the direction and structure of specific ties. The analysis therefore asks whether students connected through friendship, emotional support or school-related advice show patterns of similarity that cannot be reduced to belonging to the same class.

For the network analysis, however, the researchers used only 20 classes with the highest participation and the least missing data, producing an effective sample of 453 students after excluding absentees. This was approximately one-quarter of the full survey sample, and the authors describe the results as exploratory rather than population-level estimates. Within the relational network, the strongest positive associations appeared for openness, with a rho of 0.246, and self-confidence, with a rho of 0.229. Positive associations were also found for problem solving, 0.190; collaboration, 0.155; commitment, 0.141; and empathy, 0.136. In practical terms, students who named classmates as friends or sources of emotional support tended to resemble those peers in several social and personal competencies. The pattern is consistent with a process in which encouragement, feedback and shared emotional experiences reinforce behaviours such as listening, cooperation and confidence.

Task-oriented ties were associated with a somewhat different but overlapping profile. Positive network associations were observed for openness, 0.213; problem solving, 0.203; commitment, 0.175; collaboration, 0.156; empathy, 0.152; self-confidence, 0.183; and curiosity, 0.123. The link with curiosity may reflect an information-spillover process: students approached for academic help or advice about future choices may influence how their classmates think about learning and exploration. The only significant negative association in this network involved autonomy, with a rho of -0.145. The researchers interpret this pattern as possible functional complementarity rather than a deficit: highly independent students may interact with classmates who are more likely to seek assistance, creating a network in which different roles coexist rather than converge on the same trait.

The results also reveal which abilities appear less connected to classmates’ reported skills. Leadership and resilience showed no significant network association in either the relational or task-oriented domain. Those traits may depend more heavily on individual characteristics or experiences outside the measured peer networks, although the study cannot determine why. The researchers caution that cross-sectional data provide only a snapshot, making it impossible to distinguish social influence from selection. Students with similar interests or personalities may choose one another as friends, rather than becoming similar because they interact. The models also explained only part of the variation in soft skills, leaving potential roles for personality, intelligence, technical abilities and other unmeasured factors.

Even with those caveats, the study offers a detailed framework for understanding how social relationships may become educational resources. Its findings support the idea of social scaffolding, in which students’ confidence and competence are sustained through dialogue, modelling and assistance from others. A classmate who explains a difficult concept, listens to anxiety about the future or invites a peer into a group activity may be doing more than solving an immediate problem; such interactions can create repeated opportunities to practise communication, empathy, persistence and collaborative decision-making. The researchers argue that schools could use cooperative projects, peer tutoring and mentoring to make these interactions a deliberate part of learning, while avoiding the assumption that informal networks are automatically inclusive or beneficial.

Longitudinal studies will be needed to test whether relationships genuinely drive changes in soft skills over time. Repeatedly mapping the same students and their networks could help separate peer influence from friendship selection and reveal whether changes in one student’s skills spread through connected classmates. Future research could also examine how classroom composition, teacher practices and school cultures shape these processes. For now, the Italian study’s central message is that soft skills do not develop solely inside individual minds or through formal instruction. They are also associated with the invisible web of trust, advice and cooperation connecting students at their desks—a web that may be one of the classroom’s most underused educational resources.