The study, published in Frontiers in Psychology, treated self-efficacy as a dynamic process rather than a trait. Participants were recruited through youth sport training programs and stratified by sport discipline, competitive tier, and sex. They ranged in age from 13 to 18, with 168 male and 152 female athletes, and had an average of 4.6 years of training history. Competitive tiers spanned recreational or school programs, regional or provincial squads, and national teams, with individual disciplines slightly outnumbering team sports.

Over eight consecutive weeks, each athlete answered four signal-contingent prompts per day through a smartphone questionnaire. The day was divided into four fixed windows, from a late-morning slot through an evening period when platform use peaks, and within each window the signal arrived at a random minute. Participants had 30 minutes to respond. Weekly engagement assessments and a baseline trait battery anchored the momentary readings to slower-moving characteristics. The researchers screened compliance before computing any index, requiring at least half of the 224 scheduled prompts; six athletes fell below that threshold and 26 withdrew, leaving the final sample of 320. Among those retained, the overall response rate was 79.1 percent, with a median of 184 valid occasions per athlete, and compliance did not decline over the study window.

At each prompt, participants reported counts of posting, received reactions, comments, and comparison-relevant browsing over the preceding interval. The researchers formed two distinct quantities: feedback volume, the standardized sum of these counts regardless of tone, and negative comparison intensity, the standardized count of events the participant flagged as unfavorable, such as encountering a peer performing better or receiving criticism. Valence was coded by the participants themselves at the moment of report through a single forced-choice item. To check consistency, two trained assistants independently recoded free-text descriptions from a random 15 percent of flagged events, achieving high agreement with the participant classifications.

From the repeated efficacy scores, the team built two dispersion measures. The intraindividual standard deviation captures the overall spread of an athlete’s readings around their own mean, while the root mean square of successive differences weights how sharply consecutive readings change and was treated as the primary index of fluctuation amplitude in the structural models. Because dispersion on a bounded response scale is arithmetically entangled with the person mean, the mean was retained as a covariate in every amplitude model, and a residualized amplitude score was also constructed by partialling out both the person mean and the person-specific autoregressive parameter.

The first substantive finding concerned the shape of the signal itself. A two-level dynamic structural equation model estimated a clearly non-zero lag-1 autoregressive parameter of 0.44, meaning momentary efficacy at one occasion carried substantial information about the next, and this parameter varied appreciably across athletes. An intraclass correlation of 0.53 indicated that roughly half the variance in momentary readings lay between athletes and half within them. The group-mean trajectory traced a gently undulating course with a shallow mid-period dip, and that trough aligned with a high-load training block preceding regional fixtures. Athletes who shared almost the same average confidence followed visibly different paths, some nearly level and others swinging through wide arcs.

The event-level models then showed a marked valence asymmetry. Negatively valenced comparison produced the strongest immediate increase in momentary instability, and its lagged term remained reliable, so the disturbance carried into the following interval rather than dissipating. Positive feedback’s contemporaneous coefficient was smaller, and its lagged term fell short of conventional significance, suggesting affirming signals briefly boosted confidence without leaving a durable trace. A cross-level interaction showed that social comparison orientation, the stable tendency to define oneself against others’ standing, amplified the response to negative feedback alone; predicted instability was highest where high negative feedback met high orientation, while the positive-feedback surface stayed comparatively flat across all orientation levels.

At the person level, the results qualified the exposure claim in an instructive way. Feedback volume was not reliably associated with fluctuation amplitude once valence was in the model, so the sheer density of evaluative input did not matter. Negative comparison intensity, by contrast, accounted for the bulk of the explained variance, and the product of negative intensity and comparison orientation was reliable for both amplitude indices. In the final multilevel structural equation model, fluctuation amplitude was negatively associated with all three engagement facets beyond mean efficacy, with the largest coefficient for absorption, an intermediate one for vigor, and the smallest for dedication. A formal equality test, however, narrowed that conclusion: constraining the three amplitude-to-facet paths to equality worsened fit, and pairwise comparisons showed absorption differed reliably from dedication, but absorption could not be distinguished from vigor, nor vigor from dedication. The direct effect of negative comparison reached vigor and absorption but not dedication, so the facet-by-facet hypothesis held only in part.

The mediation was partial rather than full. The indirect path from negative comparison through amplitude to engagement was reliable, with a Monte Carlo confidence interval excluding zero, and the standardized mediation effect size was 0.09, a small-to-moderate figure. The proportion of the total effect carried indirectly was 0.46, but the authors caution that this ratio is unstable in partial-mediation models, with a wide sampling interval. Mean efficacy retained its positive association with engagement even as amplitude carried a separate negative one, addressing the worry that instability might be nothing more than disguised low confidence. Indices of moderated mediation excluded zero on the negative arm of feedback but not the positive arm, for both the mean path and the amplitude path.

The authors are explicit that the design is observational and the language associational. No objective indicators of training load, competition outcome, or recovery status were collected, so the models cannot separate the psychological consequences of unfavorable comparison from the physiological and competitive stress arriving in the same weeks. Calendar phase, entered as a time-varying covariate and used to split the series, is a coarse proxy, though the lagged effect survived its addition and held within each phase separately. Competitive tier neither predicted amplitude nor moderated the negative-comparison effect, though the national subgroup was small and that test correspondingly weak. Substituting the residualized amplitude score attenuated the amplitude-to-facet paths by roughly a tenth and left every one reliable, and automated sentiment polarity agreed substantially with participant-coded valence, with that agreement not varying by comparison orientation, arguing against a coding artifact. A reverse within-person path, from momentary efficacy to subsequent comparison-relevant browsing, was indistinguishable from zero, which supports the directional reading but is weaker evidence than a positive forward finding. The sample came from one cultural and competitive setting, all measures were self-reported, and signal-contingent sampling within fixed windows may miss the most charged moments. The authors propose that a brief structured efficacy-reflection conversation within 24 hours of competition, and simple self-monitoring of confidence before and after feed use, are inexpensive steps grounded in the lagged estimates, while noting that whether such routines actually narrow the swings remains an open and testable question.

Subject of Research: Self-efficacy fluctuation and training engagement among adolescent athletes in relation to online social feedback and comparison orientation

Article Title: Self-efficacy fluctuation and training engagement in adolescent athletes: online social feedback and the moderating role of comparison orientation

Article References: Xu, Y., Zhou, B., Zhu, Z., & Li, M. (2026). Self-efficacy fluctuation and training engagement in adolescent athletes: online social feedback and the moderating role of comparison orientation. Frontiers in Psychology, 17, Article 1908710. https://doi.org/10.3389/fpsyg.2026.1908710

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1908710

Keywords: adolescent athletes, self-efficacy, ecological momentary assessment, online social feedback, social comparison orientation, training engagement, social media, sport psychology, negative comparison, fluctuation amplitude, longitudinal study, fluctuation