Scholarlink

Background <p>School aversion is increasingly recognized in adolescents and has been linked to emotional distress, suicide risk, and academic failure. However, studies rely on correlational analysis, making it unclear which risk factors truly cause school aversion. To address this gap, we combined directed acyclic graph (DAG)-based modeling, DoWhy-based causal effect estimation, and counterfactual simulation to explore psychosocial pathways related to school aversion in clinically referred adolescents.</p> Methods <p>Participants were clinically referred adolescents from mental health services who completed standardized psychosocial assessments. Variables were obtained through validated instruments and clinical intake records, covering demographics, adversity, personality, and symptoms. Causal inference was conducted through a combined framework of DAG learning, DoWhy estimation with backdoor propensity-score weighting and logistic-model-based counterfactual simulation. All analyses were performed using Python 3.11.8, with <i>pgmpy</i>, <i>DoWhy,</i> and <i>scikit-learn</i> as core libraries.</p> Results <p>A total of 423 adolescents were assessed, with 57.7% showing school aversion or refusal and 39.7% reporting suicidal thoughts or attempts. DAG modeling identified several school aversion-adjacent relationships, with family structure, bullying, and digital addiction retained as primary school-aversion-centered pathways in downstream analyses, whereas trauma remained a nearby contextual factor. DoWhy analysis showed that the retained associations were SchoolAversion → Bullying [average treatment effect (ATE) = 0.046, <i>p</i> = 0.001] and SchoolAversion → DigitalAdd (ATE = 0.036, <i>p</i> = 0.008), while family structure showed a significant negative effect on school aversion (ATE = -0.162, <i>p</i> = 0.031). Counterfactual simulation showed that removing bullying or digital addiction had relatively small effects on school aversion. In exploratory counterfactual simulations, SchoolAversion → Suicide showed a risk difference of 0.266 (bootstrap&#xa0;95% CI 0.182 to 0.346).</p> Conclusions <p>Among clinically referred adolescents, school aversion appeared not only as an outcome linked to upstream psychosocial context, but also as a central factor associated with downstream bullying, digital addiction, and suicide risk in the present framework. This integrated causal framework may help refine risk screening and intervention prioritization in clinical populations, while further validation in non-clinical and longitudinal samples is needed.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Causal modeling of school aversion in psychiatrically referred adolescents: a DoWhy-based analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Mengmeng Zhang,
  • Xin Ma,
  • Hui Li,
  • Botao Huang,
  • Yuhan Luo,
  • Jie Qian,
  • Meijuan Wang

摘要

Background

School aversion is increasingly recognized in adolescents and has been linked to emotional distress, suicide risk, and academic failure. However, studies rely on correlational analysis, making it unclear which risk factors truly cause school aversion. To address this gap, we combined directed acyclic graph (DAG)-based modeling, DoWhy-based causal effect estimation, and counterfactual simulation to explore psychosocial pathways related to school aversion in clinically referred adolescents.

Methods

Participants were clinically referred adolescents from mental health services who completed standardized psychosocial assessments. Variables were obtained through validated instruments and clinical intake records, covering demographics, adversity, personality, and symptoms. Causal inference was conducted through a combined framework of DAG learning, DoWhy estimation with backdoor propensity-score weighting and logistic-model-based counterfactual simulation. All analyses were performed using Python 3.11.8, with pgmpy, DoWhy, and scikit-learn as core libraries.

Results

A total of 423 adolescents were assessed, with 57.7% showing school aversion or refusal and 39.7% reporting suicidal thoughts or attempts. DAG modeling identified several school aversion-adjacent relationships, with family structure, bullying, and digital addiction retained as primary school-aversion-centered pathways in downstream analyses, whereas trauma remained a nearby contextual factor. DoWhy analysis showed that the retained associations were SchoolAversion → Bullying [average treatment effect (ATE) = 0.046, p = 0.001] and SchoolAversion → DigitalAdd (ATE = 0.036, p = 0.008), while family structure showed a significant negative effect on school aversion (ATE = -0.162, p = 0.031). Counterfactual simulation showed that removing bullying or digital addiction had relatively small effects on school aversion. In exploratory counterfactual simulations, SchoolAversion → Suicide showed a risk difference of 0.266 (bootstrap 95% CI 0.182 to 0.346).

Conclusions

Among clinically referred adolescents, school aversion appeared not only as an outcome linked to upstream psychosocial context, but also as a central factor associated with downstream bullying, digital addiction, and suicide risk in the present framework. This integrated causal framework may help refine risk screening and intervention prioritization in clinical populations, while further validation in non-clinical and longitudinal samples is needed.