Psychological Scale Predicts Shadow Education Decisions Using Machine Learning Models

A newly developed psychological measurement scale can predict family participation in shadow education using supervised machine learning algorithms, according to a three-phase investigation published in SN Social Sciences. Researchers Hanqiang Li and Yan Piaw Chua constructed the instrument within the framework of the theory of planned behavior to address a long-standing gap in educational research. Across multiple analytical evaluations, classification models demonstrated that core psychological dimensions accurately forecast participation behavior without incorporating socioeconomic status variables. In particular, a random forest model reached a classification accuracy of 80.6 percent and an AUC of 0.80. These findings indicate that psychological constructs provide predictive value for participation decisions independently of household financial standing.

The investigation emerged in response to the rapid global expansion of shadow education, which has increasingly evolved into a routine household investment for families striving to preserve a competitive educational advantage. Despite the widespread growth of private supplementary tutoring, the researchers noted that the literature lacked a systematic psychometric instrument designed specifically for this setting. To resolve this empirical limitation, the authors designed a comprehensive instrument intended to capture the underlying cognitive and social factors that govern whether families choose to enroll their children in supplemental academic programs.

To construct and calibrate the measurement tool, the study implemented an exploratory sequential mixed-methods design divided into three consecutive methodological phases. The opening qualitative phase focused on gathering in-depth interview material from 15 participating families to uncover foundational themes surrounding tutoring choices. To process this qualitative narrative data systematically, the researchers deployed an artificial-intelligence-assisted framework analysis. This qualitative framework analysis enabled the researchers to process family perspectives and identify key psychological themes underlying household investment in supplementary tutoring.

Following the qualitative thematic extraction, the authors formulated an initial survey questionnaire directly grounded in the identified concepts. The quantitative validation phase subsequently examined 760 valid responses collected through this structured questionnaire. The researchers applied both exploratory factor analysis and confirmatory factor analysis to the response data to refine the structural properties of the instrument. Through this statistical testing, the authors refined the scale structure and confirmed acceptable psychometric reliability and construct validity across the measured sample.

The resulting instrument encompasses six distinct psychological subdimensions that together define household decision-making regarding shadow education.

In the final evaluative phase of the study, the researchers tested whether these psychological subdimensions could reliably predict actual participation behavior. The authors constructed seven mainstream supervised classification models trained on the survey data, deliberately omitting socioeconomic status variables from the predictive feature sets. Four of the seven machine learning algorithms demonstrated acceptable predictive performance, showing that psychological profiles alone can yield reasonably good classification of whether a family participates in supplementary education. The random forest algorithm proved especially effective, attaining an accuracy of 80.6 percent and an AUC value of 0.80.

The authors emphasize that the study offers complementary evidence by demonstrating the empirical utility of a psychological perspective beyond conventional socioeconomic variables. The researchers noted that psychological dimensions carry independent predictive value for participation behavior when socioeconomic factors are excluded.

Subject of Research: Psychological decision-making scale for shadow education participation

Article Title: Development of​ a Decision-Making Scale for shadow education participation based on the theory of planned behavior: a three-phase mixed-methods study

Article References: Li, H., & Chua, Y. P. (2026). Development of​ a Decision-Making Scale for shadow education participation based on the theory of planned behavior: a three-phase mixed-methods study. SN Social Sciences, 6(11), Article 612. https://doi.org/10.1007/s43545-026-01906-7

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01906-7

Keywords: shadow education, theory of planned behavior, mixed-methods research, decision-making scale, psychometrics, machine learning, random forest, supervised classification, perceived utilitarian benefits, perceived costs and risks, peer pressure, expectations of significant others, objective resource control, internal coping efficacy, factor analysis, socioeconomic status, predictive modeling