How Education Subsidies Affect Junior High School Students’ Noncognitive Ability Development: Evidence from China

Abstract

Education subsidies play a crucial role in allocating educational resources and significantly impact students’ ability development. This study utilizes data from the China Education Panel Survey to examine the effects of education subsidies on junior high school students’ ability development, employing the framework of new human capital theory along with least squares, propensity score matching, and quantile regression models. The findings reveal that, first, education subsidies exert a significant positive effect on students’ noncognitive abilities, although no similar effect is observed for their cognitive abilities. Second, education subsidies exhibit a “compensatory effect” in enhancing noncognitive abilities, with a more pronounced impact observed among students with lower levels of noncognitive ability. Third, education subsidies particularly benefit female students, students from less affluent families, and those residing in rural areas, with these groups experiencing greater improvements in noncognitive abilities. The study suggests refining education subsidy policies and implementation strategies to foster a collaborative system among families, schools, and society, thereby promoting junior high school students’ comprehensive development through multiple avenues.

1. Introduction

Efficient and rational allocation of educational resources is essential for promoting educational equity and improving quality of life (Bird et al., 2022; Bhandari, 2013). In practice, a shortage of educational resources not only limits students’ equal access to education but also significantly impacts the development of individual abilities (Cartwright & Allen, 2002; Moshtari & Safarpour, 2024). As a critical educational resource, educational subsidies have been widely discussed worldwide. The U.S. Elementary and Secondary Education Act (ESEA), for example, outlines types, amounts, and eligibility requirements for federal education subsidies in five areas, aiming to “strengthen and improve the quality of and access to elementary and secondary education throughout the Nation” (Hopkins, 2024). Similarly, as the largest developing country with a significant educational mandate, China’s Compulsory Education Law of 1986 mandates the establishment of scholarships to support economically disadvantaged students in completing their education (Hannum & Park, 2007; Li et al., 2025). In 2001, China implemented the decision on the Reform and Development of Basic Education, formally instituting policies to reduce miscellaneous and textbook fees, subsidizing boarding fees, and establishing a comprehensive compulsory education support system. This initiative, known as “two exemptions and one subsidy”, reflects China’s commitment to educational equity (X. Liu, 2023). According to China’s Ministry of Education, by 2021, tuition and miscellaneous fees had been entirely waived at the compulsory education level, free textbooks were provided, and living subsidies were available to students from low-income families (Fan et al., 2019). These progressive policies from 1986 to the present yielded measurable outcomes. However, the question remains whether enhanced education subsidies can increase students’ abilities (Tsang, 1996; Xiao et al., 2017).

Extensive research has shown that education subsidies impact students’ development mainly through educational access and academic achievement (Jackson et al., 2015). As a vital aspect of anti-poverty efforts, policy interventions at the compulsory education level significantly influence individual short- and long-term human capital accumulation and income potential (Bianchi et al., 2022; Xiao et al., 2017). With the introduction of the “Double Reduction” policy and a reduction in subject-based training time, middle school students now have more significant opportunities to develop comprehensive skills. This shift underscores the need for research on educational subsidies’ effects on cognitive and noncognitive abilities beyond academic performance (Olsson et al., 2022). Although cognitive and noncognitive abilities are essential for students’ holistic development, influencing not only their academic success but also their future careers and position in the labor market (Vittadini et al., 2022; Farkas, 2003), few studies have specifically examined the impact of educational subsidies on the competence development of lower secondary school students.

Traditional human capital theories often utilize educational attainment or cognitive ability as proxies for “ability” (Ng & Feldman, 2010). Kautz et al. defined cognitive ability as intelligence and problem-solving skills encompassing comprehension, memory, reasoning, and critical thinking. These are typically assessed through I.Q. tests, standardized reading scores, math scores, and similar measures (Heckman & Kautz, 2013). Bowles and Gintis introduced the concept of noncognitive skills (NCSs) in their study on returns to education, finding that even after accounting for demographic factors, family socioeconomic status, and cognitive ability, approximately two-thirds of the variance remained unexplained (Bowles & Gintis, 1976; Foster & Rosenzweig, 1993). They argued that human capital theory should encompass both cognitive and noncognitive abilities rather than being confined to cognitive skills alone. Heckman’s research expanded on this perspective, introducing the new human capital theory, which differentiates between cognitive and noncognitive competencies, thereby partially demystifying the “cognitive abilities” concept (Bowles et al., 2001; Heckman & Kautz, 2012). Accordingly, under the new human capital framework, investigating whether education subsidies can foster students’ personal ability development necessitates considering both cognitive and noncognitive growth (C. Wang & Zhong, 2018).

The main contributions of this study are as follows: First, building on previous research that broadly examines the impact of education subsidies on academic performance and other cognitive skills, this study broadens the scope by empirically exploring how education subsidies affect noncognitive abilities, deepening the understanding of their role in enhancing students’ overall quality. Second, it systematically reveals the mechanism of the “compensatory effect” of education subsidies on noncognitive ability development, offering precise empirical evidence supporting the relevance and fairness of these policies. Third, this study investigates the heterogeneity of education subsidies’ effects across different groups on the basis of gender, family income, and urban–rural distinctions, enriching research on how these subsidies influence students from diverse socioeconomic backgrounds. Finally, the study proposes recommendations for optimizing education subsidy allocation and implementation, providing new empirical insights and policy directions to foster equitable resource distribution and capacity development.

The China Education Panel Survey (CEPS) data were selected for this study. The sampling method employed was a multi-stage probability proportional to size (PPS) approach, which was conducted in four stages. In the first stage, the data were drawn from the 2010 Sixth Population Census of China, covering 2870 county-(district-)level administrative units across the 31 provinces, autonomous regions, and municipalities directly under the central government (excluding Hong Kong, Macao, and Taiwan). These units were divided into three sampling frames, with a total of 28 counties (districts) selected. Specifically, Sampling Frame 1 included 15 districts (counties) selected as core samples from the full list of 2870 counties (districts). To account for the unique characteristics of Shanghai, Sampling Frame 2 included 3 districts from the 18 districts under Shanghai’s jurisdiction as supplementary samples. Furthermore, to capture the effects of migrant populations on education and educational inequality, Sampling Frame 3 selected 10 counties (districts) from the 120 counties (districts) with large migrant populations. In the second stage, a PPS sampling method was applied to select four schools from each of the sampled counties (districts). Local partners collected basic information on schools, including types and sizes, from the education departments for the 2013–2014 school year. These data formed the basis for constructing the sampling frame for this stage. In the third stage, four classes were selected from each sampled school, comprising two seventh-grade classes and two additional classes. If a school had only one or two classes in the surveyed grade, all were included; if there were three or more classes, two classes were randomly selected using a random number table. The fourth stage of the sampling process involved selecting all students, parents, subject teachers (Language, Math, and English), and school leaders from the selected classes. These formed the final survey sample1. Using this dataset, the study employed least squares modeling, propensity score matching, and quantile regression techniques to analyze the impact of educational subsidies on the competency development of junior high school students, as well as the heterogeneity of these effects.

2. Literature Review and Research Hypotheses

2.1. Influencing Factors of Cognitive and Noncognitive Abilities

With the increasing recognition of the socioeconomic benefits associated with cognitive and noncognitive abilities, researchers have begun to examine the factors that influence these abilities. Current studies focus on four primary dimensions. First, within the economic and cultural capital dimension, evidence suggests that a financially affluent family environment and higher parental education levels positively influence students’ cognitive and noncognitive abilities (Li et al., 2023; W. Wang et al., 2020; Duncan et al., 2011). Second, in the interpersonal interaction dimension, various types of interactions—such as those between parents and children, between teachers and students, and between peers—have been shown to enhance students’ cognitive and noncognitive abilities (Carrell et al., 2013; Burke & Sass, 2013; Ye & Wang, 2024). Some scholars further note that authoritative parenting styles and collaborative learning environments significantly contribute to students’ noncognitive development (S. Wang & Zheng, 2024; Deng & Tong, 2020). Third, in the human capital investment dimension, interventions such as preschool education and extracurricular tutoring have been identified to enhance cognitive and noncognitive abilities (Zheng et al., 2021; Chen, 2020). Fourth, with respect to personal traits, factors such as good physical appearance and family labor participation positively influence cognitive and noncognitive abilities (Ritts et al., 1992; Bezerra et al., 2009). Additionally, in recent years, increasing attention has been given to social phenomena such as left-behind children, family migration, and boarding. Studies indicate that boarding and staying behind significantly impact academic performance but do not directly influence cognitive ability. However, these experiences often reduce students’ emotional stability and sense of belonging within urban schools (Martin et al., 2014; Zhong et al., 2024). In contrast, migration with family has a more favorable impact, significantly enhancing cognitive and noncognitive abilities (H. Liu et al., 2021; Zhao & Chen, 2022). Despite extensive research into these influencing factors, few studies have focused on the impact of educational subsidies on cognitive and noncognitive abilities. This research gap may be due to the historical emphasis on academic achievement-oriented outcomes (Tao & Hong, 2014) and the inherent challenges in measuring cognitive and noncognitive abilities, compounded by limited data availability (Kautz et al., 2014).

2.2. Impact of Education Subsidies

A literature review on education subsidies indicates that existing research has focused primarily on their impact on educational access and outcomes. In terms of access, subsidies have been shown to reduce poverty, decrease dropout rates, and increase enrollment (Facchini et al., 2021; M. Wang et al., 2024). With respect to outcomes, research has focused predominantly on academic achievement, although the findings are mixed. While some studies suggest a significant positive effect of subsidies on academic performance, others report negative or negligible impacts. In a meta-analysis, Hanushek (2003) reported that the positive effect of subsidies on academic achievement became insignificant when controlling for students’ family backgrounds. Moreover, previous research has overlooked essential aspects of educational outcomes. Noncognitive abilities are essential educational outcomes and are regarded as a core component of literacy and overall development (Abott et al., 2020; Garcia, 2016). However, they have received less attention in existing studies. This study, therefore, focuses on the impact of educational subsidies on students’ cognitive and noncognitive abilities from a microlevel perspective, thereby offering key innovations. The CEPS data utilized in this study measure cognitive ability through assessments of verbal, graphical, computational, and logical skills (Stenning & Oberlander, 1995). This approach aligns closely with American psychologist Cattell’s concept of fluid intelligence, which is neurologically based, develops within a limited “window of time”, and is primarily concentrated in early childhood (Heckman et al., 2006). Previous research has identified several factors influencing cognitive ability, including preschool education, early childhood experiences, and family background (Fujisawa et al., 2019; Burger, 2010; Anger & Schnitzlein, 2017). This study posits that educational subsidies, as an intervention introduced later in life, may have limited effects on middle school students’ cognitive abilities, as they miss the crucial early “window” of cognitive development.

In contrast, educational subsidies are likely to impact noncognitive abilities in multiple ways. First, subsidies directly reduce students’ financial burdens, mitigating the adverse effects of disadvantaged family backgrounds (Dearden et al., 2014; Machin, 2006). Second, subsidies foster the development of interpersonal skills by promoting social interactions, thereby enhancing noncognitive abilities (Bedwell et al., 2014). Third, by increasing family disposable income, subsidies encourage more significant investment in extracurricular educational activities, further supporting students’ noncognitive growth (Morris, 2003; Feng et al., 2023). Based on these considerations, this study proposes the following hypotheses:

Hypothesis 1.

Education subsidies have no significant effect on students’ cognitive ability.

Hypothesis 2.

Education subsidies have a positive effect on students’ noncognitive abilities.

2.3. Heterogeneity in the Impact of Educational Subsidies on Noncognitive Abilities

The heterogeneous impact of educational subsidies, as a form of public financial investment and resource allocation, on educational outcomes has garnered significant scholarly attention. A review of the literature reveals an ongoing debate about whether educational subsidies exhibit a “Matthew effect”—whereby benefits accrue to those already advantaged—or a “compensatory effect” that helps disadvantaged students close the gap. Some studies argue that subsidies reinforce a competitive “arms race” among families, further disadvantaging lower-ability students (Kim, 2010; Rigney, 2010). Conversely, other research suggests that educational subsidies help mitigate the influence of family income on academic performance, thereby reducing educational inequality (Goldrick-Rab et al., 2016). This study suggests that educational subsidies, as a means of allocating educational resources, play a compensatory role, supporting those with fewer advantages. This compensatory effect is evident in three primary aspects. First, educational subsidies increase years of schooling and alleviate poverty, effectively counteracting the negative impact of disadvantaged family backgrounds on students, thereby supplementing the role of family income in students’ educational outcomes (Roberts, 2003; Johnson & Jackson, 2019). Second, drawing from social interaction theory, a cohort effect may arise when preferences and behaviors are influenced through peer interactions (Ivanova et al., 2019). The cohort effect of educational investment tends to be more effective in rural settings than in urban areas, enhancing the implementation of subsidy policies in rural communities (Glewwe & Muralidharan, 2016). Third, in patriarchal family cultures, where women often occupy a subordinate status and gender discrimination is common, Chinese families tend to invest more in boys’ education than in girls’ education (Akabayashi et al., 2020; Hannum et al., 2009; Hannum, 2005). However, with the expansion of educational subsidies and the diversification of subsidy types—such as free textbooks and free school breakfasts—support can now be applied directly to individual students. This shift helps female students benefit independently from family-based resource allocation dynamics, thereby reducing gender disparities (Schultz, 2001). Based on these considerations, this study proposes the following hypotheses:

Hypothesis 3.

Educational subsidies have a “compensatory effect” on the noncognitive abilities of students, with a more pronounced effect for individuals with low noncognitive abilities.

3. Methodology

3.1. Models

This study employs Stata for modeling analysis, which is divided into the following three sections:

3.1.1. Baseline Model

This study estimates the impact of educational subsidies on middle school students’ cognitive and noncognitive abilities via the traditional ordinary least squares (OLS) method. The specific model setup is as follows:

$${\mathrm{Y}}_{\mathrm{i}}={{\mathsf{\beta }}_0+\mathsf{\beta }}_1\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}{\mathrm{y}}_{\mathrm{i}}+\mathsf{\gamma }{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i}}+\mathsf{\mu }$$

(1)

where ${\mathrm{Y}}_{\mathrm{i}}$ is the dependent variable, including cognitive and noncognitive abilities; $\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}{\mathrm{y}}_{\mathrm{i}}$ represents educational subsidies (received = 1, not received = 0); ${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i}}$ represents the control variables; β and γ are the coefficients in the regression; and $\mathsf{\mu }$ represents the residual term.

3.1.2. Propensity Score Matching

This study employs propensity score matching (PSM) to estimate the impact of education subsidies. Access to education subsidies depends on various individual-, family-, and school-level factors and is not a random process; therefore, addressing the issue of sample self-selection is essential for accurately estimating the impact of these subsidies. Although the OLS model controls for individual-, household-, and school-level variables, it does not eliminate selection bias arising from observable factors. In a nonexperimental setting, the average treatment effect (ATT) model based on propensity score matching is particularly effective at addressing endogenous estimation bias due to sample self-selection. Consequently, this study uses PSM to estimate the impact of educational subsidies on middle school students’ cognitive and noncognitive abilities and to perform a robustness check of the OLS regression results. (1) Selection of Covariates: to satisfy the ignorability assumption and build on previous research, this study treats the control variables as covariates that affect both access to education subsidies and the development of students’ competencies. (2) Estimating Propensity Scores: This study uses logit regression to calculate the propensity score for middle school students to receive educational assistance. The specific formula for this calculation is as follows:

$$\mathrm{P}\left({\mathrm{X}}_{\mathrm{i}}\right)=\mathrm{Pr}\left({\mathrm{D}}_{\mathrm{i}}=1|{\mathrm{X}}_{\mathrm{i}}\right)={\displaystyle \frac{\mathrm{e}\mathrm{x}\mathrm{p}\left(\mathsf{\beta }{\mathrm{X}}_{\mathrm{i}}\right)}{1+\mathrm{e}\mathrm{x}\mathrm{p}\left(\mathsf{\beta }{\mathrm{X}}_{\mathrm{i}}\right)}}$$

(2)

where β is the coefficient estimate of the logit model, and $\mathrm{P}\left({\mathrm{X}}_{\mathrm{i}}\right)$ denotes the propensity score value of middle school students. Specifically, the individual characteristics of junior high school students, family characteristics, and school characteristics are used as independent variables ${\mathrm{X}}_{\mathrm{i}}$, and whether they receive subsidies is used as the dependent variable ${\mathrm{D}}_{\mathrm{i}}$ (${\mathrm{D}}_{\mathrm{i}}$ = 1 for access to subsidies).

(3) Propensity score matching: A successful match is achieved when the standardized difference between groups does not exceed 10%. To ensure the robustness of the regression results, this study utilizes three matching methods: within-caliper one-to-four matching, kernel matching, and radius matching. (4) The average treatment effect is calculated on the basis of the matched sample: using the matched sample, the ATT is calculated to assess the impact of educational subsidies as follows:

$$\begin{gathered} \mathrm{A}\mathrm{T}\mathrm{T}=\mathrm{E}\left(\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}{\mathrm{g}}_1^2-\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}{\mathrm{g}}_0^2|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}\mathrm{y}=1\right) \\ =\mathrm{E}\left(\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}{\mathrm{g}}_1^2|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}\mathrm{y}=1\right)-\mathrm{E}\left(\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}{\mathrm{g}}_0^2\right|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}\mathrm{y}=1) \end{gathered}$$

(3)

where $\mathrm{E}\left(\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}{\mathrm{g}}_0^2\right|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}\mathrm{y}=1)$ denotes the cognitive and noncognitive abilities corresponding to the hypothesis that middle school students who receive educational subsidies do not receive educational subsidies, and $\mathrm{E}\left(\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{o}{\mathrm{g}}_1^2|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}\mathrm{y}=1\right)$ denotes the cognitive and noncognitive abilities of middle school students who receive educational subsidies.

3.1.3. Quantile Regression Model

This study employs quantile regression to investigate whether the impact of educational subsidies on middle school students differs across various distributions of students’ abilities. The model is specified as follows:

$$\mathrm{Q}\mathrm{u}\mathrm{a}\mathrm{n}\mathrm{t}\mathrm{i}\mathrm{l}{\mathrm{e}}_{\mathsf{\tau }}\left({\mathrm{Y}}_{\mathrm{i}}|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}{\mathrm{y}}_{\mathrm{i}}\right)={\mathsf{\beta }}_{0\mathsf{\tau }}+{\mathsf{\beta }}_{1\mathsf{\tau }}\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{s}\mathrm{i}\mathrm{d}{\mathrm{y}}_{\mathrm{i}}+{\mathsf{\gamma }}_{1\mathsf{\tau }}{\mathrm{X}}_{\mathrm{i}\mathsf{\tau }}+{\mathsf{\mu }}_{\mathrm{i}\mathsf{\tau }}$$

(4)

where ${\mathsf{\beta }}_{0\mathsf{\tau }}$, ${\mathsf{\beta }}_{1\mathsf{\tau }}$, ${\mathsf{\gamma }}_{1\mathsf{\tau }}$, and ${\mathsf{\mu }}_{\mathrm{i}\mathsf{\tau }}$ represent, respectively, the first $\mathsf{\tau }$ parameter in the first quartile.

3.2. Variables and Data Description

3.2.1. Explained Variables

This study measures junior high school students’ competency development through cognitive and noncognitive competencies. Given that academic performance (i.e., school test scores) in the CEPS data is not standardized, it is unsuitable as an outcome variable (Shen, 2020). Instead, cognitive ability is proxied by a continuous variable based on standardized scores from aptitude tests, measured via the three-parameter logistic (3PL) model. Owing to the diversity and complexity of noncognitive abilities, a unified measurement standard has yet to be established. However, three main scales are widely recognized for noncognitive ability assessment: Rotter’s locus of control scale (Rotter, 1966), Rosenberg’s self-esteem scale (Rosenberg, 1965), and the Big Five Personality Inventory by Costa and McCrae (1992). On the basis of these scales, scholars have developed two primary models. The first type is the two-factor model, developed from Rotter’s Locus of Control Scale and Rosenberg’s Self-Esteem Scale, which includes factors such as control, self-esteem, leisure time, social networks, and interpersonal relationships (Judge et al., 2003). The second type is the Big Five personality model, which has been extended with the advancement of personality psychology to assess traits such as openness, emotional stability, self-discipline, and sociability (Agastya et al., 2020). Some scholars have refined this model into dimensions, such as academic perseverance, open-mindedness, emotional stability, interactional competence, and school adjustment (Martin et al., 2013). Luo and Dai (2015) developed the Chinese Adjectives Big Five Personality Scale. They confirmed its reliability and validity through empirical data. In addition, Yu et al. (2020) incorporated the unique characteristics of junior high school students along with specific items from the CEPS questionnaire. This work resulted in the first comprehensive indicator system tailored to measure noncognitive abilities among junior high school students. Among various measurement approaches, the two-factor model and its extended models somewhat underestimate the impact of noncognitive abilities (Humphries & Kosse, 2017). Given that this study focuses on middle school students, the Big Five personality scale and its extended models were adapted to include a noncognitive ability measurement system developed explicitly by Yu et al. (2020). This system was ultimately chosen to assess noncognitive abilities in this study.

The following steps were implemented regarding variable processing: (1) The evaluation scores for each secondary indicator were standardized to control for differences in measurement scales. (2) The standardized secondary indicators under each primary indicator were then weighted. They are averaged to yield scores for each primary indicator. (3) A correlation analysis of the five primary indicators was conducted. The results (shown in

Table 1. Matrix of correlation coefficients for noncognitive indicators.

	Conscientiousness	Agreeableness	Extraversion	Openness	Neuroticism
conscientiousness	1
agreeableness	0.317 ***	1
extraversion	0.152 ***	0.230 ***	1
openness	0.402 ***	0.357 ***	0.230 ***	1
neuroticism	0.102 ***	0.121 ***	0.333 ***	0.106 ***	1

Notes: *** denotes significance at the 1% level.

) indicate significant correlations among them. Additionally, a principal component analysis yielded a Kaiser–Meyer–Olkin (KMO) value of 0.69, confirming the data’s suitability for this analysis. Two principal components with eigenvalues greater than one were extracted. (4) The two principal components are weighted by their respective variance contribution ratios, producing a composite index that reflects students’ overall noncognitive abilities.

3.2.2. Core Explanatory Variables

Previous studies have commonly used per-pupil expenditure as a proxy for education subsidies, providing a quantitative measure of subsidies. However, in this study, data analysis revealed that 2272 out of 7068 valid student samples did not receive any education subsidies. This finding suggests that prior proxy indicators may have overlooked individual-level differences, potentially homogenizing students within the same school and failing to distinguish between students who received subsidies and those who did not. To address this, the study derived a more individualized proxy indicator from question C6 in the CEPS student questionnaire, which asked whether the student received various forms of educational support, such as free books, free lunches, or bursaries. On the basis of the responses to this question, the binary variable “education subsidy” was created, assigned a value of 1 if the student received at least one type of subsidy and 0 if the student did not receive any.

3.2.3. Control Variables

Following existing research, this study includes individual, family, and school control variables. At the individual level, the control variables include gender (male = 1), ethnicity (Han = 1), only child status (only child = 1), self-assessed health, and preschool education experience. Self-assessed health reflects students’ subjective evaluation of their health on a scale from “very bad” to “very good”. For analysis, the categories “better” and “very good” were defined as healthy and assigned a value of 1. In contrast, all other categories were defined as unhealthy and assigned a value of 0. Preschool education experience was determined from the question “Have you attended preschool or kindergarten since age three?” and was coded as 1 for “yes” and 0 for “no”.

At the family level, the control variables include the father’s and mother’s years of schooling, self-assessed family economic status, primary household language (speaking Mandarin = 1), and parental relationship quality. Parents’ years of schooling were derived from educational attainment levels. Moreover, family economic status was based on parents’ subjective assessment of their financial situation, ranging from “very difficult” to “very rich”. For analysis, families were classified as economically “rich” if they were rated as “very rich” or “relatively rich”, “average” if they were rated as “average”, and “difficult” if they were rated as “very difficult” or “somewhat difficult”, with the latter group assigned a value of 0. Parental relationship quality was assessed through responses to the question, “What is the relationship between your parents?”, with “good” coded as 1 and “bad” as 0.

At the school level, the several variables included were school rank, classroom achievement level, and classroom teacher’s years of experience. School rank was assessed by the item “current ranking of the school in the district”, and classroom achievement level was assessed by the item “current ranking of the class within the grade”. Both are ordinal variables ranging from “worst” to “best”, with higher values indicating a higher ranking of the school or classroom. For this study, the rankings of “best” and “upper middle” were defined as high-level schools/classes and assigned a value of 1. In contrast, all other rankings were classified as low-level and assigned a value of 0. The years of experience of the classroom teacher were measured as a continuous variable.

3.3. Data Sources

The data for this study were derived from the China Education Panel Survey (CEPS), a nationally representative, large-scale longitudinal survey project conducted by the National Survey Research Center (NSRC) at Renmin University of China. This survey, launched in the 2013–2014 academic year as its baseline, employed structured questionnaires as its primary data collection instrument. It focused on two contemporaneous cohorts: students in grade 7 (junior high school year 1) and grade 9 (junior high school year 3). The survey sites were randomly selected from 28 county-level units (counties, districts, and municipalities) nationwide, with stratification based on the average educational attainment of the population and the proportion of migrant residents. The survey adopted a school-based sampling framework, selecting 112 schools and 438 classes from the identified county-level units. All students within the selected classes were included, yielding a baseline sample of approximately 20,000 students. The questionnaire encompassed four primary areas: (1) students’ basic information, developmental background, and social interactions; (2) parents’ demographic data, family education environment, educational investments, and parent–child interactions; (3) teachers’ professional background, teaching philosophies, daily instructional practices, work-related stress, and job satisfaction; and (4) school characteristics, including infrastructure, enrollment figures, and teacher–student ratios, as well as administrative and teaching practices. Additionally, the survey incorporated assessments of students’ cognitive abilities and personality traits, collected their scores from significant exams (e.g., midterm and college entrance exams), and planned to conduct health and physical assessments to gather biomedical indicators. A combination of advanced methods and technologies was employed to ensure the comprehensive and high-quality collection of data. The CEPS dataset has since become a pivotal resource for research across various fields, including education, economics, management, and sociology, offering robust empirical evidence for policymakers and scholars.

This study utilizes the 2014–2015 tracking data, which was chosen for its large sample size; extensive geographic coverage; and comprehensive range of variables, including students’ demographic characteristics, family educational background, household registration information, social behavior, and school learning environment. Compared to the baseline survey, the 2014–2015 questionnaire featured enhanced measurements for variables, such as parental occupation and students’ noncognitive abilities, thereby improving the validity of the data and better aligning it with the study’s objectives. Using this dataset, the study integrated and merged information from students, parents, teachers, and principals through Stata 17 software. After excluding observations with missing data, the final sample comprised 7068 valid cases, which included 4796 middle school students who received educational assistance and 2272 middle school students who did not. Descriptive statistics for these variables are presented in

Table 2. Variable descriptions and statistics.

	Variables	Mean	SD	Min	Max
Explained Variables	Cognitive ability	0.355	0.799	−3.137	2.063
	Noncognitive ability	0.000	0.973	−4.003	2.255
Core Explanatory Variables	Education subsidy	0.679	0.467	0.000	1.000
Individual Control Variables	Onlychild	0.451	0.498	0.000	1.000
	Healthstatus	3.874	0.931	1.000	5.000
	Gender	0.505	0.500	0.000	1.000
	Ethnicity	0.922	0.268	0.000	1.000
	Preschoo	0.825	0.380	0.000	1.000
Household Control Variables	Fatheredu	10.563	3.158	0.000	19.000
	Motheredu	9.967	3.529	0.000	19.000
	Affluentfamily	0.066	0.248	0.000	1.000
	Averagefamily	0.723	0.447	0.000	1.000
	Language	0.344	0.475	0.000	1.000
	Parentarela	0.902	0.298	0.000	1.000
School Control Variables	Schoolrank	0.802	0.399	0.000	1.000
	Classrank	3.265	1.014	1.000	5.000
	Teachexperience	16.834	8.392	0.000	42.000

.

4. Results and Discussion

4.1. Benchmark Regression

Table 3. Results of the benchmark regression.

Variables	(1)	(2)
	Cognitive Ability	Noncognitive Ability
Subsidy	−0.012	0.049 **
	(0.019)	(0.023)
Controls	YES	YES
Constant	−0.888 ***	−2.176 ***
	(0.076)	(0.090)
Observations	7068	7068
R-square	0.125	0.166

Notes: Robust standard errors in parentheses. *** and ** denote significance at the 1% and 5% levels, respectively.

presents the regression results derived from the OLS estimation model. Models 1 and 2 show the full-sample regression results for cognitive and noncognitive abilities. The coefficient for the effect of educational subsidies on noncognitive ability is positive and significant, even after controlling for individual-, family-, and school-level variables. This finding contributes to the literature on educational subsidies and financial aid evaluation, as previous studies have highlighted mainly the impact of educational subsidies on employment rates and academic achievement (Facchini et al., 2021; Hanushek, 2003), with relatively little focus on noncognitive abilities. In both the Chinese and global educational contexts, strong noncognitive abilities contribute to enhanced academic performance and promote career acquisition and upward mobility in the labor market. From this perspective, educational subsidies serve their intended purpose as an educational resource, positively promoting students’ balanced and holistic development. However, this study found that educational subsidies did not significantly impact cognitive abilities. A possible explanation lies in the numerous factors influencing the cognitive abilities of middle school students, with family background and cultural capital being particularly influential. Family background encompasses economic status, parental education levels, and parenting styles. Affluent families can provide abundant learning resources, highly educated parents prioritize early intellectual development, and democratic parenting styles foster the cultivation of critical thinking. Therefore, relying solely on educational subsidies is unlikely to rapidly improve middle school students’ cognitive abilities. While subsidies may alleviate financial pressure, they cannot alter deeper factors, such as family background and cultural capital. Furthermore, this study enriches the literature on factors influencing noncognitive and cognitive abilities. By including educational subsidies within the framework of noncognitive ability influences, this research provides a foundation for selecting relevant control variables in future studies, enabling more precise and insightful findings.

4.2. Robustness Test: PSM

Since junior high school students’ access to educational assistance is not random, sample self-selection may introduce endogeneity bias. To address this, the study employs an average effects model based on propensity score matching (PSM) to mitigate self-selection bias arising from observable factors.

Effective propensity score matching relies on meeting two assumptions: the overlap assumption and the data balancing assumption. The overlap assumption requires that the treatment and control groups have overlapping subsamples, whereas the data balancing assumption ensures that covariate distributions in the matched treatment and control groups are balanced. Figure 1 illustrates the sample matching results obtained using the Caliper 1:4 Matching method, with noncognitive abilities as the dependent variable. The figure shows that most observations fall within the standard support range, representing the overlapping propensity score region between the treatment and control groups. This indicates that the matching process was successful, and only a few samples were excluded. This is crucial for ensuring the validity of propensity score matching, effectively eliminating selection bias, and ensuring the comparability of the results. Figure 2 shows the probability density functions before and after matching; after matching, the distance between the treatment and control groups significantly decreases, confirming that the data meets the overlap assumption.

To further verify the validity of the PSM, this study conducts a balance test on the sample data before and after matching, testing the conditional independence hypothesis. Figure 3 presents the changes in standardized bias for each variable before (Unmatched) and after (Matched) matching. As clearly shown in the figure, compared to the unmatched state, the standardized bias for most variables significantly decreases after matching. Notably, all variables’ standardized bias (%bias) is below 10% after matching. This indicates that the differences between the treatment and control groups have been effectively controlled through the Propensity Score Matching (PSM) method. Before matching, the standardized bias for some variables was notably large, suggesting significant systematic differences between the two groups. After matching, these differences were substantially reduced, meeting the acceptable balance criteria. Therefore, the PSM model in this study successfully and significantly reduced the self-selection bias between the control and experimental groups, thereby improving the endogeneity issue.

To assess the effects of educational subsidies on junior high school students’ cognitive and noncognitive abilities accurately and ensure the robustness of the results, this study treats students receiving educational subsidies as the treatment group and those without subsidies as the control group. Robustness testing was performed via the radius matching method (r = 0.01), within-caliper one-to-four matching (r = 0.01, n = 4), and kernel matching in the PSM model.

Table 4. Results of the robustness test.

Variables	OLS	PSM
		Radius Matching	Kernel Matching	One-to-Four Matching Within a Caliper
	Y = Cognitive ability
subsidy	−0.012	−0.013	−0.018	−0.019
	(0.019)	(0.020)	(0.019)	(0.024)
	Y = Noncognitive ability
subsidy	0.049 **	0.048 **	0.048 **	0.045 *
	(0.023)	(0.024)	(0.024)	(0.028)
Observations	7068

Notes: Robust standard errors in parentheses. ** and * denote significance at the 5% and 10% levels, respectively.

presents the full-sample robustness test results, which show that educational subsidies significantly enhance students’ noncognitive abilities but have no significant effect on their cognitive abilities, confirming H1 and H2. A comparison of the model coefficients reveals that the OLS model overestimates the effect of educational subsidies on composite noncognitive ability scores.

4.3. Quantile Regression

Building on basic regression and robustness tests, this study employed quantile regression modeling to examine the heterogeneous effects of educational subsidies on cognitive and noncognitive abilities.

Table 5. Results of the quantile regression.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	q = 0.1	q = 0.2	q = 0.3	q = 0.4	q = 0.5	q = 0.6	q = 0.7	q = 0.8	q = 0.9
	Y= Cognitive ability
subsidy	0.004	0.007	−0.016	−0.014	−0.006	−0.007	−0.015	−0.004	−0.011
	(0.037)	(0.032)	(0.027)	(0.024)	(0.023)	(0.022)	(0.021)	(0.021)	(0.028)
	Y= Noncognitive ability
subsidy	0.099 **	0.075 **	0.049 *	0.063 **	0.055 **	0.043	0.041	0.041	0.046
	(0.046)	(0.035)	(0.029)	(0.031)	(0.028)	(0.028)	(0.032)	(0.036)	(0.032)
Controls	YES	YES	YES	YES	YES	YES	YES	YES	YES
Observations	7068

Notes: Robust standard errors in parentheses. ** and * denote significance at the 5% and 10% levels, respectively.

displays the quantile regression results, where q = 0.1 to q = 0.9 correspond to different levels of students’ cognitive and noncognitive abilities at the 10th to 90th quantiles, respectively. For cognitive ability, educational subsidies have no significant impact across all quantiles. This may be because the enhancement of cognitive abilities primarily depends on effective teaching methods, learning strategies, as well as individual talent and effort, with educational subsidies having a relatively limited direct impact on students’ cognitive abilities, such as memory, comprehension, and logical thinking (Jamil et al., 2021). However, the estimated coefficients for noncognitive abilities for q = 0.1 to q = 0.5 are significantly positive, indicating that educational subsidies have a notable impact on middle school students with lower noncognitive abilities. According to scarcity theory, students with low noncognitive abilities often face dual resource constraints: they lack psychological resources such as emotional regulation and self-efficacy while also being limited by their family’s economic capital. Educational subsidies, through direct financial support (e.g., tuition reductions, study material subsidies) and indirect resource compensation (e.g., after-school tutoring, psychological counseling services), can significantly alleviate their cognitive load, thereby exerting a significant impact on their noncognitive abilities (De Bruijn & Antonides, 2022). The estimated coefficients for q = 0.6 to q = 0.9 are also insignificant. This gradient-decreasing pattern suggests that the effect of educational subsidies on noncognitive abilities follows a typical “diminishing marginal returns” characteristic. Specifically, when an individual’s noncognitive abilities are at a lower level (q ≤ 0.5), the resource input provided by educational subsidies can significantly improve their areas of weakness. However, as their baseline abilities increase (q ≥ 0.6), the marginal utility of the same subsidy gradually diminishes. This supports the hypothesis (H3) that educational subsidies have a “gap-filling effect” on middle school students’ noncognitive abilities, meaning that these subsidies target and benefit disadvantaged groups, aligning with the compensation theory of educational resource allocation (Sims, 2011).

4.4. Heterogeneity Analysis

This study further investigates the “compensatory effect” of educational subsidies on noncognitive abilities, focusing on the heterogeneity of this effect across gender and household characteristics. The subsamples are divided by gender, household income, and household registration type, and PSM estimation is employed to verify the robustness of the model results while controlling for other variables. The heterogeneity analysis results are presented in

Table 6. Results of heterogeneity analysis.

Variables		(1)	(2)	(3)	(4)	(5)	(6)
		Male	Female	Higher-Income	Lower-Income	Urban	Rural
	Y = Noncognitive ability
subsidy	OLS	0.008	0.092 ***	−0.130	0.060 **	−0.006	0.103 ***
		(0.033)	(0.031)	(0.098)	(0.023)	(0.034)	(0.031)
	Radius Matching	0.013	0.081 **	−0.134	0.062 ***	0.010	0.096 ***
		(0.031)	(0.032)	(0.133)	(0.024)	(0.038)	(0.037)
	Kernel Matching	0.003	0.078 **	−0.116	0.059 **	0.006	0.088 **
		(0.031)	(0.032)	(0.113)	(0.023)	(0.037)	(0.037)
	One-to-Four Matching Within a Caliper	0.009	0.111 ***	−0.133	0.052 *	0.035	0.110 **
		(0.040)	(0.040)	(0.139)	(0.029)	(0.047)	(0.045)
Controls		YES	YES	YES	YES	YES	YES
Observations		3567	3501	466	6602	3353	3715

Notes: Robust standard errors in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

. Models 1 and 2 in Table 6 display the regression results by gender. The estimated coefficient of educational subsidies is significantly positive in the female sample but not significant in the male sample. This finding indicates that educational subsidies have a more pronounced effect on enhancing noncognitive abilities among female students. In China’s patriarchal cultural context, gender discrimination remains prevalent, particularly within families, where boys often receive more investment than girls do (Greenhalgh, 1985). However, direct and varied forms of educational subsidies help female students circumvent these biases, demonstrating that educational resource allocation should prioritize students’ personal development and avoid influence from traditional customs or subjective family preferences (Eloy et al., 2013).

Models 3 and 4 in Table 6 show the regression results for the household income subsamples. The results show that the estimated coefficient for educational subsidies is significantly positive in the regression for samples from lower-income families. At the same time, it is not significant in the sample from wealthier families. This may be because children from lower-income families or rural areas face more challenges during their development, such as limited educational resources and insufficient parental involvement and guidance. The intervention of educational subsidies can, to some extent, compensate for these shortcomings by providing children in rural areas with more opportunities to engage with the outside world and broaden their horizons, thereby promoting the development of their noncognitive abilities (Raymond & Sadoulet, 2001; Sheehan & Hadfield, 2024). While previous studies have noted a potential “Matthew effect” of educational subsidies in reinforcing preexisting advantages (Pisoni, 2018), this study finds that a “compensatory effect” predominates in the context of noncognitive abilities, suggesting that educational subsidies may help bridge disparities in this domain. These results underscore the need to track the impact of educational subsidies over time, particularly across different age groups and competency dimensions.

5. Limitations

First, due to the lack of large-scale, standardized surveys specifically focused on noncognitive abilities in China, the scales used in this study were constructed by previous researchers based on existing survey data. The data upon which these scales are based may have limitations, such as sample bias and incomplete measurement dimensions. Future research should aim to design specialized questionnaires and conduct large-scale surveys targeting student populations from various regions and backgrounds.

Second, from the perspective of sample selection, this study is limited to data from China. While the survey sample used in this study represents a significant portion of middle school students in China, the sample’s heterogeneity could still be improved. Future studies should incorporate more diverse samples, including individuals from different regions, ethnicities, and family economic backgrounds, to better understand the relationship between educational subsidies and student ability development. Additionally, most of the data in this study come from self-reports. Although the test results suggest no serious standard method bias, subjective factors may still influence self-reported data. Future research could incorporate more objective indicators, such as teacher evaluations, academic records, and psychological test results, to provide a multidimensional and objective assessment of students’ cognitive and noncognitive abilities.

Third, regarding research methodology, Propensity Score Matching (PSM) assumes that, given the propensity scores, the potential outcomes for both the treatment and control groups are independent of the treatment assignment. However, in practical research, ensuring that all potential confounding factors have been observed and included in the model is challenging, which may lead to omitted variable bias. Future research could address this issue by collecting more comprehensive data, selecting variables based on theoretical frameworks, conducting sensitivity analyses and model comparisons, and incorporating qualitative research and other causal inference methods to improve PSM by mitigating the estimation bias caused by unobserved and unaccounted for confounding factors.

6. Conclusions

Educational subsidies provide essential financial support to students, alleviating their economic burdens, improving the fairness of learning opportunities, and promoting the widespread dissemination of educational resources and the comprehensive development of talent. Based on the CEPS data, this study systematically analyzed the impact of educational subsidies on the ability development of middle school students, as well as the heterogeneity of these effects, using Ordinary Least Squares (OLS), Propensity Score Matching (PSM), and Quantile Regression models. The following conclusions were drawn: first, educational subsidies have a positive effect on the ability development of middle school students, but their impact on enhancing cognitive abilities is minimal, a finding that was validated through robustness checks; second, educational subsidies exhibit a “gap-filling effect” in enhancing noncognitive abilities, meaning that the effect is more pronounced for students with lower noncognitive abilities; third, the impact of educational subsidies on the improvement of noncognitive abilities is more significant among female students and those from lower-income families and rural areas. Based on these conclusions, the following three recommendations are made:

First, at the policy level, under the global push for educational equity and the realization of the United Nations Sustainable Development Goals, the comprehensive improvement and strengthening of China’s “Two Exemptions and One Subsidy” policy in compulsory education is of profound significance for promoting educational equity and advancing social development. It also holds considerable international value for broader applications. Current efforts should focus on targeting disadvantaged groups, particularly, continuing to direct resources towards regions such as “old, small, remote, and poor” areas. Policy adjustments can help ensure these groups have equal access to quality educational resources, progressively bridging the gaps in educational resource allocation and outcomes between urban and rural areas and among different families. The current funding model, which allocates resources primarily based on school indicators, guarantees the basic distribution of educational resources but also reveals a singular investment model, a common challenge in global educational funding. Therefore, policy design could aim to build a diversified compensation mechanism for compulsory education, breaking the existing limitations. Implementing strategies such as distributing discounted educational vouchers and introducing housing subsidy policies for school districts are practical approaches to reducing the economic burdens of disadvantaged groups in accessing quality educational resources, further promoting educational equity.

Second, regarding the implementation of subsidies, it is crucial to profoundly and accurately implement relevant subsidy policies to ensure educational equity. As the total investment in educational subsidies continues to grow, abandoning outdated ideas that merely pursue extensive expansion is essential. Instead, efforts should focus on building an educational system centered on fairness and scientific principles. Given the significant disparities in economic development and educational resource distribution between urban and rural areas in China and worldwide, implementing educational subsidies in impoverished rural regions must be a priority. Policymakers must consider the specific needs of these regions and increase investment in rural compulsory education. At the same time, it is essential to account for the varied needs of different families to ensure the accuracy and effectiveness of subsidy policies. Thus, it is necessary to establish diversified subsidy channels, providing convenient and efficient application processes and various forms of support for families and students in need.

Third, to create an environment conducive to student ability development, building a collaborative system for fostering abilities involving family, school, and society has become an essential global educational trend. Efforts should be made to eradicate traditional gender biases, such as son preference, in some regions through various channels, such as community outreach and family education seminars. Families should be made aware that every child has equal rights and potential for development, and allocating resources within the family should prioritize promoting the child’s ability development. From the school perspective, school culture is a foundational pillar of education. Schools should strengthen value education and cultivate a positive, diligent, and scholarly atmosphere. Simultaneously, special attention should be given to the care and protection of students from disadvantaged groups, including the establishment of comprehensive anti-bullying prevention and intervention mechanisms, the reinforcement of legal and mental health education, and a zero-tolerance approach to bullying behaviors. This ensures that students develop their abilities in a safe and harmonious environment. From a societal perspective, all sectors of society should increase their support for disadvantaged groups and expand the available forms of assistance. Through diverse forms of support, society can help disadvantaged students overcome difficulties, enabling them to achieve social mobility more quickly and effectively through their efforts. This contributes to advancing educational equity in China and brings Chinese wisdom and strength to the global pursuit of educational equity and social prosperity, fostering a fairer and more harmonious global society.