Upper secondary education policy adjustment, educational disengagement and juvenile justice involvement: an interrupted time-series analysis in Shandong province, China, 2014–2025

1 Introduction

Juvenile delinquency prevention has become a persistent concern in global public policy and criminological research. Adolescence represents a critical developmental stage characterized by socialization, identity formation, and the transition toward adulthood. During this period, young people's developmental trajectories are shaped by multiple factors, including family environments, school experiences, peer relationships, and access to educational opportunities. Existing criminological research suggests that juvenile delinquency is not solely the consequence of individual characteristics but is embedded in broader social, institutional, educational, and developmental contexts (Farrington, 2007; Hirschi, 1969; Sampson and Laub, 1995).

In this study, educational disengagement (ED) is operationalized as an aggregate educational transition-gap indicator derived from annual provincial statistics. It reflects the difference between the reported number of junior-high graduates and the reported number of students entering the upper-secondary education pathways included in this study. In recent years, juvenile justice involvement in Shandong Province has attracted increasing attention due to emerging concerns regarding younger ages of offending, technology-related delinquency, and group-based offending patterns. Consequently, governance strategies have gradually shifted from punitive responses toward prevention-oriented approaches emphasizing education, early intervention, and social integration (Supreme People’s Procuratorate of China, 2026). International frameworks, including the United Nations Guidelines for the Prevention of Juvenile Delinquency, emphasize that improving developmental environments, expanding educational opportunities, and strengthening social support systems are essential mechanisms for reducing youth offending risks (Cappelaere and Verhellen, 1996). From this perspective, maintaining educational continuity and reducing ED may represent important institutional pathways for juvenile delinquency prevention.

Within the Chinese educational system, transitions during upper-secondary education have particular significance for youth development and social governance. In 2014, China further formalized upper-secondary education admission arrangements, reinforcing the structural differentiation between general high schools and vocational secondary education pathways (Ministry of Education of China, 2014). Under this dual-track system, educational opportunities have been substantially shaped by examination performance and institutional allocation mechanisms, potentially creating unequal developmental pathways for adolescents. Students who face academic difficulties, lack identification with vocational education, or experience insufficient educational adaptation may be at increased risk of ED or premature exit from formal education.

The 2020 reform-period reference point should be understood within a broader trajectory of vocational education reform in China and Shandong Province. At the national level, the State Council issued the Implementation Plan for the National Vocational Education Reform in January 2019 (Ministry of Education of China, 2019), establishing a broader policy framework for vocational education reform. To address these structural challenges, Shandong Province initiated the construction of a provincial vocational education innovation development pilot zone in 2020. The reform package focused on strengthening integration between general and vocational education, improving vocational education quality, and expanding alternative educational pathways (Ministry of Education of China and Shandong Provincial Government, 2020). Furthermore, the revised Vocational Education Law of the People's Republic of China in 2022 legally recognized vocational education as equally important to general education, providing institutional support for improving the inclusiveness of upper-secondary education opportunities (Ministry of Education of China, 2022).

The present study focuses on the 2020 upper-secondary education reform period in Shandong Province as a policy reference point within a broader trajectory of educational reform rather than as a completely isolated policy shock. National vocational education reforms initiated before 2020, together with subsequent provincial policy implementation, may have influenced educational participation before the formal designation of the 2020 reform period. Accordingly, the 2020 intervention point is used to identify a temporal discontinuity associated with the reform period, while recognizing that policy preparation, implementation, and behavioral responses may have preceded the formal intervention date.

Existing studies have extensively examined the relationship between individual educational experiences and criminal behavior, but less attention has been paid to whether changes in educational opportunity structures can reshape patterns of ED and subsequently influence juvenile justice trends at the institutional level. Against this background, the present study addresses this level-of-analysis gap by examining the relationship between upper-secondary education policy adjustment, aggregate educational disengagement (ED), and juvenile justice involvement at the provincial level. Specifically, it first evaluates whether the 2020 upper-secondary education reform period was associated with a change in the trajectory of ED in Shandong Province and then conducts exploratory analyses of whether ED was statistically associated with aggregate juvenile justice involvement. The study therefore does not test individual-level pathways from educational disengagement to delinquency. Rather, it examines whether changes observed at the level of an educational system and its aggregate participation patterns correspond temporally with changes in an aggregate juvenile justice indicator. This study makes three contributions:

First, methodologically, it conceptualizes upper-secondary education reform in Shandong Province as a quasi-experimental policy intervention and applies interrupted time series analysis to evaluate the potential policy effects of educational reform.

Second, empirically, based on a regional education reform case in Shandong Province, this study provides evidence regarding whether institutional changes in educational opportunity structures influence ED patterns and explores the potential relationship between educational continuity and juvenile justice involvement.

Third, theoretically, by demonstrating the absence of a statistically significant macro-level association between ED and juvenile justice involvement, this study identifies boundary conditions for applying individual-level criminological theories to institutional-level policy evaluation. The findings highlight that educational reforms may improve social integration and developmental opportunities without producing immediate measurable reductions in juvenile justice involvement.

2 Literature review

2.1 Educational opportunity, educational continuity, and juvenile justice involvement

The relationship between education and juvenile delinquency has been examined from several complementary criminological perspectives. Social control theory emphasizes the role of social bonds in constraining delinquent behavior, with schools serving as important institutional settings in which adolescents develop attachments to conventional norms and social relationships (Hirschi, 1969). Life-course criminology likewise emphasizes that educational experiences form part of broader developmental trajectories and that disruptions in education may interact with family environments, social relationships, and other life-course conditions in shaping subsequent behavioral outcomes (Sampson and Laub, 1995). These perspectives suggest that educational continuity may matter for adolescent behavioral outcomes, but they primarily explain processes occurring at the individual level.

Educational disengagement (ED) is thus better understood as a process of weakening participation in formal education rather than as a single event. Finn (1989) conceptualized withdrawal from education as a process involving declining engagement, weakening institutional attachment, and eventual separation from school, while Tinto (2012) emphasized the progressive weakening of academic and social integration within educational institutions. Balfanz et al. (2007) made a comparable argument, contending that disengagement develops over time rather than occurring suddenly. From a developmental perspective, adolescence is also a period in which important transitions may shape subsequent behavioral trajectories (Farrell et al., 2014; Oberwittler and Svensson, 2025). Farrington (2007) identified adolescence as a period of heightened vulnerability to delinquency and discussed educational difficulties and social exclusion as relevant factors, while Moffitt (2018) distinguished between adolescence-limited and life-course-persistent patterns of antisocial behavior, highlighting the possibility that educational disruption may have heterogeneous implications across developmental trajectories.

The central theoretical premise of this study is narrower. Educational opportunity structures refer here to the availability, accessibility, and institutional organization of upper-secondary educational pathways. Shifts in these structures may influence the extent to which adolescents continue into formal upper-secondary education. Educational continuity is accordingly treated as the principal conceptual pathway linking educational opportunity structures with broader youth outcomes (Manno et al., 2015; Schoon and Bynner 2019). If more accessible and institutionally inclusive upper-secondary pathways reduce the proportion of adolescents who fail to enter available educational pathways, aggregate ED may decline. Subsequent shifts in educational continuity may have broader social implications, although the individual-level processes through which these implications arise are not directly measured here.

Several complementary perspectives help explain why educational continuity may be relevant to later social and justice-related outcomes. From a rational-choice perspective, education may increase the expected returns to legitimate activities and thereby raise the opportunity costs of offending (Becker, 1968). Consistent with this argument, Lochner and Moretti (2004) found an association between educational expansion and reduced criminal involvement in the United States, while Machin et al. (2011) reported evidence from educational reforms in the United Kingdom that expanded educational opportunities were associated with reductions in youth offending. Cross-national evidence has likewise linked educational expansion with differences in criminal involvement (Buonanno and Leonda, 2006). Social capital perspectives further emphasize that schools are institutional settings in which social relationships, norms, and forms of social integration may develop (Coleman, 1988). These perspectives are complementary rather than competing explanations here.

Importantly, the mechanisms proposed by these theories should not be interpreted as directly tested mechanisms in this analysis. Processes such as school attachment, social integration, peer influence, opportunity costs, and employment expectations are not directly measured. Previous research has proposed such pathways to explain why educational participation may be associated with later delinquency or justice-related outcomes (Anderson, 2014; Lochner, 2020). This study instead examines whether changes in educational opportunity structures are associated with aggregate changes in ED and subsequently explores whether ED is statistically associated with an aggregate indicator of juvenile justice involvement.

2.2 Empirical evidence on education, disengagement, and delinquency

Empirical research provides substantial evidence that educational experiences are associated with delinquency and other behavioral outcomes at the individual level. Using linked educational and criminal justice records, Hjalmarsson et al. (2015) found that higher educational attainment was associated with fewer criminal convictions and incarceration, providing evidence consistent with the possibility that extended educational participation may have longer-term preventive implications. Oreopoulos (2007), examining compulsory schooling reforms in Canada, likewise found persistent effects of extended education on employment and health outcomes. These studies suggest that the consequences of educational participation may extend beyond educational attainment itself and may emerge over longer periods rather than through immediate changes in annual outcomes.

Research focusing more specifically on school disengagement and dropout has provided additional evidence. Sweeten et al. (2009) examined the relationship between school dropout and delinquency and found that dropping out was associated with subsequent delinquent behavior. Henry et al. (2012) further showed that school disengagement predicted dropout, delinquency, and problem substance use during adolescence and early adulthood. Research on school belonging and engagement has also emphasized the importance of students' relationships with educational institutions, although the strength and nature of these associations vary across contexts (Allen et al., 2018; Bernburg et al., 2006). Bae (2020), using longitudinal evidence among adolescents who had dropped out of school, found that school disengagement and adverse childhood experiences were associated with accumulated delinquency risks, further illustrating the importance of considering educational disengagement as part of a broader developmental process.

Recent empirical research continues to support the relevance of educational transitions and school engagement while also illustrating the complexity of these relationships. Freelin et al. (2023), using a quasi-experimental design and longitudinal data on more than 14,000 students, examined the transition from middle school to high school and found differences in subsequent delinquency between students experiencing different school-transition structures. The study also identified changes in peer and social environments as possible pathways. Such findings strengthen the evidence that educational transitions and institutional contexts may be related to adolescent behavioral outcomes, while simultaneously demonstrating that the mechanisms involved operate through multiple individual and contextual processes.

Recent research involving juvenile justice populations likewise continues to examine school engagement as an important correlate of justice-related outcomes. Lang et al. (2024), using data from the Florida Department of Juvenile Justice, examined school engagement in relation to recidivism among adolescents receiving community-based juvenile justice services and found evidence for a mediating role of school engagement in the relationship between temperament and recidivism. This evidence further supports the relevance of educational engagement in juvenile justice research, but it remains an individual-level analysis of justice-involved adolescents rather than a population-level analysis of education policy.

Taken together, these studies provide a relatively substantial individual-level empirical foundation: educational attainment, school engagement, educational transitions, and discontinuity in educational participation have been associated with delinquency, substance use, criminal convictions, or recidivism in different populations and research designs. At the same time, these findings should not be interpreted as establishing that an aggregate reduction in ED will necessarily produce an immediate reduction in aggregate juvenile justice involvement. The timing, magnitude, and direction of such a relationship may depend on developmental processes, institutional contexts, and broader social conditions.

2.3 Upper-secondary and vocational education in the Chinese context

The relationship between educational opportunity and educational continuity is particularly relevant in the Chinese upper-secondary education system, where students may enter different academic and vocational pathways after completing compulsory education. Research on Chinese upper-secondary vocational education suggests that participation is shaped not only by individual preferences but also by institutional and structural conditions. Huang and Ye (2021), for example, examined factors associated with participation rates in upper-secondary vocational schools in China and identified relationships involving school staffing, student composition, youth unemployment, and the social recognition of vocational education. These findings indicate that participation in vocational education is embedded within broader institutional and socioeconomic conditions rather than being determined solely by individual choice.

Recent research has also examined the consequences of the supply of vocational education in China. Dai and Martins (2024) used variation in education provision across cities and years to investigate the returns to secondary vocational education and found substantial labor-market returns, particularly among individuals with lower earnings potential. Although their study focuses on labor-market outcomes rather than educational disengagement or juvenile justice, it provides further evidence that the availability and organization of vocational educational opportunities can have consequences beyond immediate school participation.

This literature is relevant to this study for two reasons. First, it indicates that upper-secondary educational participation in China is influenced by the organization and supply of educational pathways. Second, it suggests that vocational education should not simply be treated as an alternative educational category but as part of the institutional opportunity structure through which adolescents continue their education after compulsory schooling. However, existing Chinese research has generally focused on participation determinants, educational quality, labor-market outcomes, or individual educational experiences. Comparatively less attention has been given to whether large-scale changes in upper-secondary educational opportunity structures are associated with changes in aggregate educational disengagement over time.

2.4 Level-of-analysis gap and remaining uncertainty

A substantial level-of-analysis gap remains. Much of the empirical literature linking education and delinquency is based on individual-level longitudinal, survey, administrative, or quasi-experimental data. These studies are valuable for identifying associations between particular educational experiences and subsequent behavioral outcomes and for examining potential individual-level mechanisms. However, individual-level evidence does not by itself establish whether changes in an educational system or education policy are accompanied by changes in population-level educational disengagement.

This distinction is particularly important for this study. At the individual level, educational disengagement may be associated with school attachment, peer relationships, social integration, employment expectations, family conditions, and other factors. At the aggregate level, however, annual shifts in ED may simultaneously reflect demographic change, educational supply, economic conditions, institutional adjustments, migration, and other population-level processes. Similarly, an administrative indicator of juvenile justice involvement may be influenced not only by underlying offending behavior but also by prosecutorial practices, justice policies, reporting patterns, and institutional responses. Several theoretical perspectives offer complementary rather than competing explanations for why educational continuity may matter for youth outcomes. Importantly, the mechanisms they describe—including social integration, school attachment, peer influence, opportunity costs, and community context—are not directly measured in the present aggregate analysis and therefore should not be interpreted as empirically demonstrated mechanisms of the observed relationships.

The ecological perspective further supports this distinction by emphasizing interactions among individuals, families, schools, communities, and institutional environments (Bronfenbrenner, 1979; Evans, 2014; Sampson et al., 1997). From this perspective, relationships observed at one analytical level may be weakened, modified, or obscured at another level. This study accordingly treats individual-level criminological theories as theoretical context rather than as directly tested causal mechanisms. Their value lies in explaining why educational continuity may plausibly matter, while the empirical analysis is restricted to the aggregate relationships that can actually be examined using annual provincial data.

Several uncertainties consequently remain. First, although individual-level research has established important associations between educational experiences and delinquency-related outcomes, it remains unclear whether large-scale changes in educational opportunity structures are accompanied by comparable changes in aggregate ED. Second, even if aggregate ED changes following an education-policy reform, it is uncertain whether such changes will correspond immediately with changes in juvenile justice involvement, because the relevant social and developmental processes may unfold over longer periods. Third, the Chinese context provides relatively limited longitudinal evidence connecting upper-secondary education reform, aggregate educational disengagement, and juvenile justice involvement. These uncertainties define the specific empirical gap addressed by this study.

2.5 Why an aggregate interrupted time-series approach?

The methodological implication of this literature gap is that a population-level policy evaluation design is required to examine a question that individual-level studies cannot directly answer. Interrupted time-series (ITS) analysis is particularly useful when randomized experiments are unavailable but repeated observations are available before and after a policy implementation point. ITS allows researchers to examine whether a policy period is associated with an immediate change in the level of an outcome and/or a subsequent change in its trajectory (Bernal et al., 2017; Wagner et al., 2002). It has thus been widely used in policy evaluation where interventions affect broad populations and randomization is not feasible (Fei et al., 2020; Linden, 2015).

This study applies this logic to the upper-secondary education reform period in Shandong Province. Rather than attempting to reproduce the individual-level mechanisms identified in previous criminological studies, the ITS analysis asks a more specific population-level question: whether the 2020 upper-secondary education reform period was associated with a change in the level or trajectory of aggregate ED. A subsequent exploratory analysis examines whether annual variation in aggregate ED is statistically associated with aggregate juvenile justice involvement.

This design does not replace individual-level longitudinal research, nor can it determine whether individual adolescents who experience educational disengagement subsequently become involved in the juvenile justice system. Its contribution is narrower. By examining annual provincial-level data, it provides an exploratory test of whether a major change in educational opportunity structures coincides with changes in an aggregate indicator of educational disengagement and whether those changes exhibit a detectable statistical correspondence with juvenile justice involvement.

Based on this theoretical and empirical framework, the study develops hypothesis and exploratory question. H1 represents the primary policy-related expectation, whereas RQ1 is explicitly exploratory:

H₁: The 2020 upper-secondary education reform period was associated with a change in the level and/or subsequent trend of aggregate educational disengagement (ED) in Shandong Province.

RQ₁: Is aggregate educational disengagement associated with juvenile justice involvement?.

Accordingly, the empirical framework of this study is intentionally narrower than the broader theoretical framework: the ITS analysis examines the association between the reform period and aggregate ED, while the subsequent exploratory models examine whether aggregate ED is statistically associated with aggregate juvenile justice involvement.

3 Empirical strategy

3.1 Model specification

3.1.1 ITS model of education policy reform and ED

Amid the ongoing development of juvenile protection and prevention systems, understanding how educational institutions contribute to reducing developmental risks among adolescents has become an important issue in public governance. A more specific and empirically testable question is whether reforms in upper-secondary education policy have altered the trajectory of ED.

To examine whether changes in upper-secondary educational opportunity structures affected ED, this study first employs a segmented ITS regression model to estimate changes in the level and trend of ED before and after the implementation of the 2020 vocational education reform in Shandong Province. The model is specified as follows:where ED_t represents the upper-secondary ED rate in year t, calculated according to Equation (2). Time_t is a continuous time variable coded from 1 to 12, corresponding to the years 2014–2025. Policy_t is a binary intervention indicator, coded as 1 for years from 2020 onward and 0 otherwise. PostTime_t represents the post-intervention time trend, coded as 0 before 2020 and increasing from 1 to 6 after policy implementation. ε_t represents the random error term.

EDt=β0+β1Timet+β2Policyt+β3PostTimet+ϵt(1)

The 2020 reform-period reference point coincided with the COVID-19 pandemic and other contemporaneous changes in the educational, social, and administrative environment. Because the present study uses a single-group interrupted time-series design without a comparison series, the model cannot independently separate changes associated with the reform period from changes associated with the pandemic or other contemporaneous factors. The ITS specification is therefore used to characterize whether a change in the level and subsequent trend of aggregate educational disengagement coincided with the reform-period reference point, rather than to provide a uniquely identified causal estimate of the reform effect.

This specification follows the conventional ITS framework, which estimates both immediate level changes and subsequent trend changes following policy interventions (Bernal et al., 2017; Linden, 2015). The construction of ED is defined as follows:Where N_g,t represents the reported number of junior-high graduates in Shandong Province in year t, and N_h,t represents the reported number of students entering the upper-secondary education pathways included in this study in the corresponding annual reporting period, comprising general senior secondary schools and secondary vocational schools. However, because the available data are annual aggregate statistics, the indicator should therefore be interpreted as an aggregate educational transition gap rather than as a direct measure of confirmed discontinuity in educational participation or individual psychological disengagement.

EDt=Ng,t−Nh,tNg,t×100%(2)

More specifically, the indicator represents the proportionate difference between the reported number of junior-high graduates and the reported number of entrants into the upper-secondary pathways included in the study. The observed gap may reflect educational non-transition, but it may also incorporate delayed enrollment, alternative educational or training pathways not captured by the included categories, migration or cross-regional enrollment, and differences in statistical reporting conventions.

Worked example. To illustrate the calculation, in 2019, Shandong Province reported 1.053 million junior-high graduates and 0.855 million students entering general senior secondary schools and secondary vocational schools. The aggregate upper-secondary transition gap was therefore calculated as: ED₂₀₁₉ = (1.0530−0.8551)/1.0530 × 100% = 18.79%. In this calculation, students entering secondary vocational education are included in N_h,t, together with students entering general senior secondary education. Thus, entry into vocational education is not treated as educational disengagement in the present study. The resulting 18.79% represents the proportionate gap between reported junior-high graduates and reported entrants into the two included upper-secondary pathways during the corresponding annual reporting period. It should not be interpreted as indicating that 18.79% of students definitively dropped out of education or experienced psychological disengagement, because other post-junior-secondary pathways, delayed transitions, population mobility, and differences in statistical reporting may also contribute to the observed gap.

Higher ED values indicate a larger aggregate gap between reported junior-high completion and reported entry into the upper-secondary education pathways included in the study. The measure is therefore interpreted as an indicator of educational transition discontinuity at the provincial level, rather than as a direct measure of individual discontinuity in educational participation or psychological disengagement.

In Equation (1), β₁ captures the pre-intervention annual trend in ED rates. β₂ estimates the immediate level change associated with the 2020 policy intervention, while β₃ captures the change in the post-intervention trend. A statistically significant β₂ or β₃ would indicate that upper-secondary education reform was associated with changes in ED patterns and would provide evidence consistent with Hypothesis 1.

The descriptive statistics of ED rates from 2014 to 2025 are presented in Table 1. The subsequent ITS analysis evaluates whether the observed changes are statistically associated with the 2020 policy intervention. The juvenile justice outcome was retained as an annual aggregate count rather than converted into a population-adjusted rate. Although a denominator based on the relevant adolescent population would theoretically improve comparability across years, the publicly available population statistics did not provide a sufficiently consistent age-specific denominator that could be matched to the population represented by the prosecutorial administrative data throughout the 2014–2025 study period. Constructing a rate using an approximate denominator could therefore introduce additional measurement uncertainty. The annual count is consequently used as the primary outcome, with the limitation that temporal variation may partly reflect changes in the size or composition of the relevant population.

Table 1. Construction of the upper-secondary ED indicator in Shandong province, China, 2014–2025, unit: ten thousand persons.

3.1.2 Exploratory model examining the association between ED and juvenile justice involvement

To examine whether variations in ED were statistically associated with temporal changes in juvenile justice involvement, this study constructed an exploratory regression model:where Y_t represents the number of juveniles involved in prosecution procedures in Shandong Province in year t, which is used as an aggregate proxy indicator of juvenile justice involvement; ED_t represents the upper-secondary ED rate in the corresponding year; Time_t and Policy_t are defined consistently with Equation (1); and μ_t denotes the random error term.

Yt=α0+α1EDt+α2Timet+α3Policyt+μt(3)

It should be emphasized that Equation (3) is designed as an exploratory analysis rather than a causal model. The purpose of this analysis is not to establish a causal relationship between ED and juvenile justice involvement, but rather to examine whether a statistically detectable association exists between the two variables at the annual aggregate level. Therefore, rather than confirming or rejecting individual-level criminological mechanisms, this analysis aims to identify the boundary conditions under which theoretically expected relationships may or may not emerge in macro-level policy contexts. By examining the absence or presence of aggregate-level associations, this study further discusses possible explanations and directions for future research. The standardized values of the core variables are presented in Table 2.

Table 2. Standardized values of core variables (2014–2025).

3.2 Data sources and descriptive statistics

This study uses annual provincial-level time-series data from Shandong Province, China, covering the period from 2014 to 2025. The data were obtained from multiple official sources. Educational indicators, including the number of junior high school graduates and the scale of upper-secondary enrollment, were collected from annual educational statistics released by the Shandong Provincial Department of Education. Macroeconomic indicators, including regional economic development and urbanization levels, were obtained from statistical publications issued by the Shandong Provincial Bureau of Statistics. Data on juvenile justice involvement were obtained from publicly available reports and statistical materials released by procuratorial authorities in Shandong Province, including the number of juveniles involved in prosecution procedures. The descriptive comparison of key variables before and after the 2020 policy reform is presented in Table 3.

Table 3. Descriptive comparison of key variables before and after the 2020 policy reform variable.

It should be noted that juvenile justice system follows the principle of balancing leniency and strictness in handling juvenile cases and has implemented policies emphasizing reduced use of arrest, cautious prosecution, and limited detention for minors where appropriate. Therefore, changes in officially recorded juvenile justice involvement may reflect not only underlying offending behaviors but also variations in judicial practices and institutional responses.

As shown in Tables 1, 3, ED and juvenile justice involvement in Shandong Province exhibited divergent patterns before and after the 2020 reform. While the ED rate declined after the policy intervention (approximately 2.74 percentage points), the number of juveniles involved in prosecution procedures continued to increase during the same period (approximately 40.81%). This divergence suggests that improvements in educational opportunity structures may not immediately correspond to observable reductions in juvenile justice involvement. The potential influence of educational continuity on juvenile justice outcomes may involve temporal lags and may be further shaped by demographic changes, judicial practices, and broader social conditions.

This descriptive pattern further supports the rationale for treating RQ₁ as an exploratory expectation rather than a confirmatory hypothesis.

3.3 Robustness checks for the association between education policy reform and ED

This study employed an ITS segmented regression model to examine whether the upper-secondary education policy reform implemented around 2020 was associated with changes in ED patterns. Given the relatively small number of annual observations, ordinary least squares (OLS) estimation was applied. The analysis primarily focused on the coefficient changes of the policy intervention variable and the post-intervention trend variable to identify temporal association patterns rather than to establish individual-level causal effects.

To evaluate the reliability and stability of the estimated results, three robustness checks were conducted. First, the Durbin–Watson statistic was applied to examine whether the residuals exhibited substantial serial correlation, which may affect the validity of OLS estimates in time-series models (Durbin and Watson, 1950). Second, placebo tests were conducted by artificially assigning the intervention points to 2018 and 2019 to examine whether the estimated policy effect depended on a specific intervention year rather than reflecting a general temporal trend (Bernal et al., 2017). Third, a leave-one-year-out sensitivity analysis was performed to assess whether the main findings were driven by observations from any single year.

The results presented in Table 4 indicate that the pre-intervention time trend coefficient was positive but statistically insignificant (β = 0.231, p > 0.05), suggesting that ED did not exhibit a statistically stable linear trend before the policy intervention between 2014 and 2019.

Table 4. ITS regression results for education policy reform and ED.

Standard errors are reported in parentheses.

p < 0.05.

Following the implementation of the 2020 vocational education innovation development reform, the coefficient of the policy intervention variable was negative and statistically significant (β = −2.397, p < 0.05). This result indicates that the policy reform was associated with an immediate reduction of approximately 2.4 percentage points in the ED rate in Shandong Province, providing support for Hypothesis H1.

However, the post-intervention trend coefficient remained statistically insignificant (β = −0.053, p > 0.05). This finding suggests that although ED showed a slight declining direction after the reform, the available data do not provide sufficient statistical evidence to confirm a sustained long-term downward trend. Instead, the observed change may represent an immediate level adjustment following the policy intervention rather than a persistent structural decline.

3.3.1 Serial correlation test

Because ITS analysis relies on consecutive annual observations, residual terms may exhibit temporal dependence, whereby unobserved factors influencing adjacent years are correlated. Such serial correlation may reduce the efficiency of ordinary least squares estimates. Therefore, this study first examined first-order autocorrelation using the Durbin–Watson statistic.

The Durbin–Watson statistic for the baseline ITS model was 2.22, which is close to the theoretical value of 2. This result indicates that the model residuals did not exhibit substantial first-order positive or negative serial correlation.

The serial correlation test suggests that residuals in the annual variation of ED in Shandong Province from 2014 to 2025 did not demonstrate significant temporal dependence, indicating that the OLS estimates were unlikely to be substantially biased by autocorrelation.

The analysis included 12 annual observations from 2014 to 2025, with six observations preceding and six observations following the 2020 reform-period reference point. Given the small number of observations, the ITS analysis was treated as exploratory. The model was used to characterize changes in the aggregate level and post-reform trend of ED, rather than to provide definitive causal estimates of the reform effect.

3.3.2 Placebo tests

To further examine whether the year 2020 represented a meaningful intervention point rather than an arbitrary temporal breakpoint, this study conducted placebo tests by artificially shifting the intervention year to 2018 and 2019 and re-estimating the ITS models. If the estimated policy effect was uniquely attributable to the 2020 reform, the artificially assigned placebo intervention points should not generate statistically significant effects.

Table 5 presents the results of the placebo tests. When the intervention point was artificially assigned to 2018, the placebo policy coefficient was negative but statistically insignificant (β = −0.876, p > 0.10), suggesting that no comparable structural change occurred around 2018.

Table 5. Placebo tests using alternative policy intervention years.

Standard errors are reported in parentheses.

p < 0.10,

p < 0.05,

p < 0.01.

However, when the intervention point was shifted to 2019, the placebo policy coefficient became statistically significant (β = −2.817, p < 0.01), and the post-intervention trend variable also reached statistical significance. This finding suggests that the decline in ED may have already emerged before the formal implementation of the 2020 policy.

These findings indicate that the decline in ED was not uniquely initiated at the 2020 intervention point. Rather, part of the observed change appears to have emerged before 2020, suggesting that the 2020 reform period should be interpreted as a point within a broader policy and implementation trajectory rather than as a clean, isolated policy shock

3.3.3 Leave-one-year-out sensitivity analysis

To further assess whether the baseline findings were driven by specific years with unusual fluctuations, this study conducted a leave-one-year-out sensitivity analysis. Specifically, three potentially influential years were excluded separately: 2018, when ED reached its observed peak; 2020, the year of policy implementation; and 2021, a year strongly affected by the COVID-19 pandemic context. The ITS model was then re-estimated after removing each observation. The results are presented in Table 6.

Table 6. Leave-one-year-out sensitivity analysis.

Additional analyses were conducted to assess the temporal specificity and stability of the primary ITS estimate. The placebo specification using 2018 as the intervention year did not produce a statistically significant policy coefficient (p > 0.10), whereas the 2019 placebo specification produced a statistically significant negative policy coefficient (p < 0.01), indicating that an intervention-like change may have occurred before the formal 2020 reform-period reference point.

Leave-one-year-out analyses further indicated that the direction of the estimated reform-period association remained negative across the tested specifications, although its statistical significance varied with the observation period. The estimated coefficient was −1.853 (p = 0.091) when 2018 was excluded, −1.331 (p = 0.249) when 2020 was excluded, and −2.941 (p = 0.017) when 2021 was excluded, compared with −2.397 (p = 0.050) in the primary specification.

Taken together, these analyses provide some evidence regarding the stability and timing of the observed association but do not resolve the identification problem. In particular, the significant 2019 placebo result and the sensitivity of statistical significance to individual observations indicate that the 2020 estimate should be interpreted cautiously as exploratory temporal evidence rather than as a uniquely identified causal effect. The result is more consistent with the possibility that changes in aggregate ED were already underway before 2020. The available annual data, however, do not allow us to determine whether this earlier change reflected policy preparation, preceding educational reforms, changes in institutional implementation, or other contemporaneous factors. Accordingly, the 2020 estimate should be understood as marking a period of observed temporal change within a broader reform trajectory rather than as identifying the independent causal effect of a discrete policy intervention.

Compared with the baseline model, the coefficient of the Policy variable remained negative after excluding different years, indicating that the estimated direction of association was not driven by a single observation. Specifically, after excluding 2018, the policy coefficient remained negative (β = −1.853). After excluding 2020, the coefficient remained negative but became statistically insignificant (β = −1.331, p > 0.10). After excluding 2021, the estimated policy effect increased in magnitude and remained statistically significant (β = −2.941, p < 0.05).

Together with the placebo test results, these findings suggest that the 2020 reform period was likely embedded within a broader process of vocational education restructuring rather than representing a completely isolated policy shock. Therefore, the estimated effect should be interpreted as reflecting the association between a period of institutional reform intensification and changes in ED, rather than a fully exogenous treatment effect.

3.4 Exploratory analysis of the temporal association between ED and juvenile justice involvement

To further examine whether changes in ED were statistically associated with variations in juvenile justice involvement, this study estimated annual time-series regression models. In addition to the contemporaneous model, a one-year lagged model was estimated to explore whether changes in ED predicted subsequent changes in juvenile justice involvement. The results are presented in Table 7.

Table 7. Exploratory associations between ED and juvenile justice involvement: concurrent and lagged models.

Standard errors are reported in parentheses. The concurrent model includes ED, Time, and Policy as predictors, whereas the 1-year lagged model replaces ED with its one-year lag (ED_Lag). Both models are exploratory aggregate-level analyses.

Table 7 reports the exploratory associations between ED and juvenile justice involvement. In the concurrent model, the coefficient of ED was positive (β = 0.227), indicating that higher levels of ED were associated with increases in juvenile justice involvement at the aggregate level. However, this association did not reach statistical significance (p = 0.436).

The lagged model further examined whether ED predicted juvenile justice involvement in the subsequent year. The coefficient of the lagged ED variable remained statistically insignificant (β = 0.139, p = 0.667), suggesting that changes in ED did not demonstrate a stable predictive relationship with subsequent juvenile justice involvement within the observed annual time frame.

Accordingly, this study does not interpret ED as a direct causal determinant of juvenile justice involvement. Instead, the findings suggest that ED may represent one structural factor embedded within broader developmental risk processes. The absence of a statistically detectable macro-level association highlights that relationships identified at the individual level may not automatically translate into observable aggregate policy effects. Future research should further examine potential mechanisms through longer observation periods, individual-level longitudinal data, and multilevel analytical approaches.

4 Results

4.1 Immediate effects of educational institutional reform on ED

This section reports the estimation results of Equation (1) and addresses whether changes in upper-secondary educational opportunity structures significantly altered the level of ED.

As shown in the descriptive trends in Table 1, the ED rate in Shandong Province generally increased between 2014 and 2018, reaching a peak of 23.74% in 2018. Following the implementation of the 2020 vocational education reform, the ED rate declined to 14.95%, the lowest level observed during the study period, and subsequently increased and stabilized at approximately 18%. Compared with the average disengagement rate before the reform (20.55% during 2014–2018), the post-reform average (17.81% during 2020–2025) remained lower, although the improvement did not continue to expand in subsequent years.

The ITS regression results reported in Table 4 indicate that ED experienced a statistically significant immediate decline following the implementation of the 2020 policy reform. The estimated coefficient of the policy intervention variable was −2.397 (p = 0.050), suggesting that ED decreased by approximately 2.4 percentage points after the intervention, after controlling for the underlying time trend.

The pre-intervention trend coefficient was not statistically significant (β = 0.231, p = 0.305), indicating that ED did not exhibit a statistically detectable linear trend between 2014 and 2019. Similarly, the post-intervention trend coefficient (PostTime) was statistically insignificant (β = −0.053, p = 0.864), suggesting that the observed decline primarily represented an immediate level shift rather than a sustained long-term downward trend.

Taken together, the ITS estimates and descriptive evidence provide support for H₁. The 2020 vocational education innovation development reform was significantly associated with an immediate reduction in ED in Shandong Province, suggesting that institutional restructuring of educational opportunities improved access to upper-secondary education in the short term. However, this effect did not translate into a statistically detectable long-term declining trajectory.

4.2 Descriptive patterns and statistical tests of the relationship between ED and juvenile justice involvement

This section reports the estimation results of Equation (3) and examines whether changes in ED were statistically associated with temporal variations in juvenile justice involvement.

The descriptive statistics presented in Tables 1, 2 indicate that ED and juvenile justice involvement did not change synchronously between 2014 and 2025. ED reached its lowest point in 2020 (14.95%), whereas juvenile justice involvement peaked in 2022 before declining in 2023. The timing difference between these turning points was approximately two to three years.

The regression results reported in Table 7 show that, in the concurrent model, the coefficient of ED was positive but statistically insignificant (β = 0.227, p = 0.436). In the lagged model, the coefficient of the one-year lagged ED variable (ED_t₋₁) was also statistically insignificant (β = 0.139, p = 0.667).

Regarding model explanatory power, the concurrent model produced an R² of 0.682 (adjusted R² = 0.562), while the lagged model produced an R² of 0.663 (adjusted R² = 0.537). Overall, at the provincial annual aggregate level, ED did not demonstrate a stable predictive relationship with either contemporaneous or subsequent juvenile justice involvement.

Therefore, the results do not identify a statistically significant direct association between ED and juvenile justice involvement at the macro annual level. RQ₁ was not statistically supported within the context of provincial-level longitudinal data.

4.3 ED as a potential mechanism linking education policy and juvenile justice involvement

Although RQ₁ was not statistically supported, this finding has important theoretical and methodological implications.

First, descriptive temporal patterns were broadly consistent with theoretical expectations, although statistical evidence remained insufficient. The approximately two- to three-year temporal gap between changes in ED and juvenile justice involvement corresponds, in direction, with the theoretical pathway of ED → weakened social integration → increased juvenile justice involvement risk. However, due to the limited number of annual observations, the statistical power of the models may have been insufficient to reliably detect moderate effects. Therefore, the absence of statistical significance should not be interpreted as evidence that ED is unrelated to juvenile justice involvement.

Second, macro-level aggregated data may not fully capture mechanisms proposed by individual-level criminological theories. As discussed in the literature review, previous studies identifying associations between ED and delinquency have primarily relied on longitudinal individual-level data, allowing researchers to control for individual characteristics and observe behavioral trajectories over time. In contrast, when such relationships are aggregated to the provincial annual level, ED reflects population-level structural changes, while juvenile justice involvement may simultaneously be influenced by demographic composition, judicial policies, social governance practices, and broader socioeconomic conditions (Deming, 2011; Hirschfield, 2018). Under these structural influences, the independent contribution of ED may be attenuated or obscured in aggregate data.

Third, the findings should not be interpreted as establishing or rejecting any individual-level mechanism linking educational disengagement with juvenile justice outcomes. Rather, they indicate that provincial-level annual data alone may be insufficient to examine the mechanisms proposed in previous individual-level research. Future research should employ longer observation periods, individual-level longitudinal datasets, and multilevel analytical approaches to examine whether the relationships proposed in previous research among educational continuity, social integration, and juvenile justice involvement can be observed across different analytical levels.

5 Discussion

5.1 Research findings

Drawing on annual provincial data from Shandong Province between 2014 and 2025, this study employed an ITS design to examine the relationship between upper-secondary education policy reform, ED, and juvenile justice involvement. The main findings are summarized as follows.

First, upper-secondary education policy reform was significantly associated with a reduction in ED; however, the observed effect primarily reflected an immediate level change rather than a sustained long-term trend. The primary ITS specification estimated an immediate reduction of approximately 2.4 percentage points in aggregate ED around the 2020 reform-period reference point (β = −2.397, p = 0.050). Given the single-group design, the small number of annual observations, the significant 2019 placebo result, and the temporal overlap with the COVID-19 pandemic, this estimate should be interpreted as exploratory evidence of a temporal association rather than as a uniquely identified causal effect of the reform. Accordingly, the 2020 estimate should be understood as marking a period of observed temporal change within a broader reform trajectory rather than as identifying the independent causal effect of a discrete policy intervention.

However, the post-intervention trend coefficient remained statistically insignificant (β = −0.053, p = 0.864), indicating that the reform did not generate a statistically detectable continuous downward trajectory. Therefore, H₁ was supported, but the primary ITS specification was consistent with the hypothesized association between the reform period and a reduction in aggregate ED. However, the placebo and sensitivity analyses indicate that this association should not be interpreted as a uniquely identified causal effect of the 2020 reform.

Second, this study did not identify a statistically significant direct association between ED and juvenile justice involvement (p = 0.436), and the expected statistical relationship proposed in RQ₁ was not supported in the present analysis. This null finding has important theoretical and methodological implications.

The descriptive temporal patterns showed that ED reached its lowest level in 2020, whereas juvenile justice involvement peaked in 2022 before declining, indicating a temporal gap of approximately two to three years. Although this temporal pattern cannot substitute for statistical evidence, the difference in timing indicates that changes in aggregate ED and juvenile justice involvement did not occur synchronously. This temporal pattern may be relevant to future research on possible lagged relationships, but the present data are insufficient to establish a specific lag structure or delayed effect. Within the annual time-series models and one-year lag specification employed in this study, however, no stable statistical association between ED and juvenile justice involvement was detected.

Third, this study provides a methodological insight into the relationship between education and juvenile justice involvement. Existing research has demonstrated associations between ED and delinquent behavior primarily through individual-level longitudinal analyses. However, when individual-level relationships identified in previous research are examined using provincial annual indicators, their aggregate-level associations may not be directly observable or may be obscured by structural factors, including population composition, judicial policy changes, and socioeconomic conditions. Therefore, the mechanism linking ED to juvenile justice involvement may be more appropriately examined through individual-level longitudinal data rather than expected to appear directly in provincial aggregate statistics.

Taken together, the findings demonstrate that theoretical expectations derived from individual-level research cannot be directly translated into observable macro-level policy effects. Upper-secondary education reform generated identifiable short-term improvements in educational participation; however, the pathway from reduced ED to reduced juvenile justice involvement was not statistically confirmed at the macro level. This finding does not undermine the preventive value of educational institutions, nor does it challenge individual-level evidence regarding education and delinquency. Instead, it identifies an important boundary condition: the relationship between educational opportunity structures and juvenile justice involvement may be conditional, multilevel, and time-dependent, requiring more refined data and longer observation periods for further investigation.

5.2 Policy implications

The findings have several implications for education policy and policy evaluation, although they should be interpreted in proportion to the exploratory nature of the study. The primary ITS specification was consistent with a reduction in the aggregate upper-secondary transition gap during the reform period, but the single-group design, limited number of annual observations, significant 2019 placebo result, and overlap with other contemporaneous changes prevent the observed change from being uniquely attributed to the 2020 reform. In addition, the exploratory analyses did not provide statistically significant evidence of a concurrent or one-year lagged association between aggregate educational disengagement and juvenile justice involvement. The findings suggest that educational continuity and aggregate transition patterns warrant continued policy attention, while the specific mechanisms linking educational transitions to juvenile justice involvement require further empirical investigation.

First, the findings suggest that educational continuity and transition patterns warrant continued policy attention. Because the ED indicator captures the aggregate gap between reported junior-high completion and entry into the upper-secondary pathways included in the study, monitoring changes in this indicator may provide a useful descriptive basis for identifying periods or groups requiring closer policy attention. This does not imply that the observed transition gap represents confirmed individual discontinuity in educational participation or that reducing the gap would necessarily reduce juvenile justice involvement.

Second, longitudinal monitoring and evaluation of education reforms should be strengthened. The present study illustrates the limitations that arise when policy evaluation relies on a short annual time series and aggregate administrative indicators. More consistent longitudinal data, including more frequent observations where available and comparable measures across regions and years, would allow future research to examine educational transitions and policy changes with greater temporal resolution. Where appropriate data are available, comparison groups or alternative quasi-experimental designs could also strengthen the identification of policy effects.

Third, educational policy evaluation should consider outcomes beyond enrollment or transition rates. Future evaluations could examine whether changes in educational participation are accompanied by improvements in educational persistence, progression, completion, and subsequent social outcomes. Such analyses would help distinguish changes in aggregate transition patterns from broader changes in educational opportunity and continuity.

Finally, the relationship between educational trajectories and juvenile justice involvement requires further empirical investigation rather than direct policy inference from the present findings. Future studies using longer time series, individual- or school-level data, and appropriately specified lag structures could examine whether and under what conditions educational disengagement is related to subsequent juvenile justice involvement. Until such evidence is available, the policy implications of the present study are best understood as priorities for monitoring, data development, and further evaluation rather than as evidence supporting a specific intervention as an established solution to juvenile justice involvement.

5.3 Limitations and future research

Before discussing specific limitations, it should be emphasized that the findings of this study should be interpreted as temporal association evidence within a policy reform context rather than as strict causal estimates.

First, the ITS analysis was based on only 12 annual observations. This small sample substantially limits statistical power and makes it difficult to reliably distinguish baseline trends, immediate level changes, and post-intervention slope changes. It also increases the sensitivity of coefficient estimates to individual observations. Consequently, the statistical significance and magnitude of the estimated intervention-period association should be interpreted cautiously. Longer time series with more pre- and post-reform observations would provide a stronger basis for estimating segmented trends and evaluating policy-associated changes.

Second, an important consideration concerns the timing of the observed reduction in ED. Although the 2020 reform period was associated with an immediate reduction in ED in the ITS model, the placebo analysis identified a statistically significant change when 2019 was used as the intervention point. This suggests that the decline in ED may have begun before the formal 2020 policy designation. One possible interpretation is that national policy preparation, provincial implementation measures, institutional adjustments, or behavioral responses preceding 2020 may have contributed to the observed transition. However, the present study cannot distinguish among these possibilities because the available annual aggregate data do not provide sufficient temporal resolution or a comparison group for identifying their separate effects. Therefore, the 2020 intervention coefficient should be interpreted as an association with the reform period rather than as an independently identified causal effect of a discrete policy introduced precisely in 2020. The 2020 reform period may be better understood as an institutional consolidation or intensification stage within a broader trajectory of upper-secondary education reform.

Third, the macro-level nature of the data limits the ability to capture individual-level mechanisms. Previous research has proposed several possible pathways through which educational disengagement may be related to juvenile justice outcomes, including peer influence, social support, employment expectations, and social integration. However, these constructs were not directly measured in the present study and therefore cannot be evaluated as explanatory mechanisms using the available provincial annual data. Similarly, aggregate ED rates cannot distinguish different forms of disengagement, such as failure to enter upper-secondary education, high school discontinuity in educational participation, or withdrawal from vocational education. Moreover, previous research suggests that the relationship between educational disengagement and behavioral or justice-related outcomes may be bidirectional; this possibility cannot be examined directly using the present aggregate data. Prior research has shown that juvenile justice involvement itself may disrupt educational trajectories and increase the likelihood of school withdrawal (Hirschfield, 2009). Therefore, the absence of a significant macro-level association in the present study may partly reflect reciprocal processes that cannot be adequately captured using aggregated annual data.

Fourth, external validity is limited. As a pilot province for vocational education reform, Shandong Province has unique policy conditions and socioeconomic characteristics, which may limit the generalizability of the findings to other regions. Future research should extend the analysis to multiple provinces and compare heterogeneous effects across different reform contexts.

Despite these limitations, this study provides exploratory evidence from a Chinese local reform context regarding how educational opportunity structures relate to juvenile development risks. By examining the gap between educational institutional change and observable juvenile justice outcomes, this study contributes to understanding the broader social governance function of education systems.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The authors acknowledge the use of ChatGPT (OpenAI) as a tool for text editing and refinement during the manuscript preparation. Although ChatGPT played a role in enhancing language and sentence structure, all ideas, analyses, and perspectives presented in this article are the original contributions of the authors.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/feduc.2026.1955861/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Babar Nawaz Abbasi, Zhejiang University, China

Reviewed by: Abdo Hasan AL-Qadri, Xi'an Eurasia University, China

Mohammed Suliman, Shaanxi Normal University, China