Agentic education · Learning science

Actionable companion · Motivation & engagement

From AI evidence to responsible practice: Achievement emotions in AI-assisted learning, academic self-efficacy, and learning engagement and AI dependence among university students in statistics-related programs: a moderated mediation study

AI-assisted editorial implementation guide, separate from the original research. Proposed actions are not findings or recommendations attributed to the source authors. No independent peer review of this guide is asserted.

Read original paperSearch paper agentCC BY 4.0

Evidence and context

Title and abstract screening; stage labels describe reported settings, not verified numerical ages. Design labels require full-text appraisal. Reported stage(s): Higher education. Verify the participants, intervention and comparator before applying this study in a new setting.

Original authors’ abstract, reproduced under the linked source license:

Abstract Background Generative artificial intelligence (AI) tools have become part of everyday coursework for university students, yet the emotional and motivational conditions under which these tools support or undermine learning remain poorly understood. Drawing on the control-value theory of achievement emotions, this study examined whether academic self-efficacy mediates the links of positive achievement emotions in AI-assisted learning with learning engagement and AI dependence, and whether AI literacy moderates the first stage of these pathways. Methods A cross-sectional survey was conducted with 700 undergraduate students (386 female; M_age = 19.81 years, SD = 1.27) majoring in applied statistics and related programs at three universities in Shijiazhuang, China. Participants completed measures of positive and negative achievement emotions, academic self-efficacy, learning engagement, AI dependence, and AI literacy. Confirmatory factor analysis preceded a first-stage moderated mediation analysis with bias-corrected bootstrap confidence intervals (5,000 resamples), controlling for negative emotions, gender, and year of study. Results The six-factor measurement model fitted the data well, χ2(50) = 55.62, p = 0.271, CFI = 0.998, RMSEA = 0.013, SRMR = 0.017. Positive achievement emotions predicted higher self-efficacy (β = 0.374), which in turn predicted stronger engagement (β = 0.395) and lower AI dependence (β = −0.176), all p < 0.001, with AI literacy entered in every equation. AI literacy moderated the emotion-efficacy association (β = 0.152, p < 0.001): the conditional indirect association with engagement rose from 0.088 to 0.208 across ±1 SD of AI literacy, and that with AI dependence fell from −0.039 to −0.093. The direct association with AI dependence was positive (β = 0.447, p < 0.001), and the conditional total association remained positive (0.355 to 0.408). Both indexes of moderated mediation were significant. Conclusion Positive emotional experiences in AI-assisted coursework show beneficial and costly associations simultaneously, and self-efficacy statistically accounts for opposing parts of each. The total association with dependence remained positive at all literacy levels, so the results describe inconsistent mediation rather than protection. AI literacy marks where the emotion-engagement pathway is strongest. Keywords: academic self-efficacy, achievement emotions, AI dependence, AI literacy, higher education, learning engagement, moderated mediation

What kind of evidence is this?

Supporting AI learning science · Survey / observational study

Stage: Higher education. Screening labels describe reported context and study-design language. They are not a risk-of-bias rating or proof of effectiveness for an age band.

Study methods · source passage

Participants were 700 undergraduate students recruited from applied statistics and related programs (Applied Statistics, n = 329; Statistics, n = 204; Data Science and Big Data Technology, n = 167) at three universities in Shijiazhuang, Hebei Province, China. All participants were at least 18 years old (M = 19.81, SD = 1.27, range 18.0 to 23.8). The sample included 386 women (55.1%) and 314 men (44.9%); 351 students (50.1%) reported a rural household registration. Year of study was distributed as follows: first year, n = 206; second year, n = 199; third year, n = 168; fourth year or above, n = 127. Self-reported frequency of AI tool use for coursework spanned the full response range, with 41 students reporting never or almost never, 103 rarely, 226 sometimes, 216 often, and 114 always or almost always.

Limitations and uncertainty · source passage

Several limitations qualify these conclusions. The cross-sectional design cannot establish causal order, and the direction we modeled is not the only defensible one: control-value theory typically treats competence beliefs as antecedents of emotions, and our reverse-direction sensitivity check explained a similar share of variance (R2 = 0.278 vs. 0.258), so the data cannot adjudicate the ordering; longitudinal or experimental work is needed. All measures were self-reported, and although the statistical checks (Harman’s test, full collinearity VIFs, and a latent method factor model) did not indicate pervasive method bias, common method variance cannot be excluded. The AI dependence scale was developed for this study and, although its two-domain structure fitted well and reliability was high, further validation with test–retest data, behavioral criteria, and independent samples is needed; the AI literacy measure was shortened to two content areas and does not cover the ethical and affective dimensions of the original framework. The main measurement model relied on facet parcels, which can conceal item-level misfit; we mitigated this risk with item-level analyses for every construct, but the two strategies answer slightly different questions. In particular, the one-factor item-level model for academic self-efficacy showed suboptimal fit owing to local dependencies among items; although a two-factor solution fitted well and the focal results do not depend on the parcel assignment, the item-level structure of this adapted scale requires further examination in independent samples. The sample came from applied statistics and related programs at three universities in one Chinese city, and students in statistical disciplines may differ from the general undergraduate population in quantitative preparation and tool usage; the findings should not be generalized to other disciplines or educational levels without further evidence. Self-reported AI use frequency, rather than logged usage, underlies our description of the sample. Future studies should include behavioral traces of AI use, metacognitive calibration measures, designs that can separate attraction to AI tools from genuine engagement, and intervention tests of the practical hypotheses outlined above.

Design the system before the pilot

Use this study to inform a learning mechanism or implementation question. It is not direct evidence that autonomous agents improve learning. Keep the proposed agent behavior separate from what the source actually tested.

  • Measure engagement separately from achievement. Check whether novelty-driven participation persists and whether it translates to independent performance.

A 90-day district pilot

  1. Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
  2. Days 16–30 · Design. Choose one disciplinary task. Require a first attempt, a documented AI consultation, a critique of the output and an independent explanation. Align the comparison group and scoring rubric with the same objective. Name an accountable owner and set the baseline, success threshold and budget before implementation.
  3. Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
  4. Days 61–90 · Decide. Blind-score reasoning and accuracy on a new task without AI. Distinguish model examination scores, student perceptions and actual learner gains; only the last directly tests educational benefit. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Huijuan Qin; Kun Liu. Achievement emotions in AI-assisted learning, academic self-efficacy, and learning engagement and AI dependence among university students in statistics-related programs: a moderated mediation study. Frontiers in Psychology (2026-10-07). https://doi.org/10.3389/fpsyg.2026.1994136

Companion prepared 2026-11-01. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.