From “transitions” to “trajectories”: towards a holistic interactionistic analysis of educational inequality in contemporary China

China’s post-1990s education expansion sharply increased higher education enrollment, approaching mass higher education by 2018. Prior sociological research often focuses on terminal outcomes (e.g., college entry) or models stagewise “transitions,” which can obscure the holistic, selective, and heterogeneous nature of educational careers. Methodological issues in transition models (e.g., comparability of logit coefficients, unobserved heterogeneity, dynamic selection) complicate interpretation and may produce the well-known “waning coefficients.” This study proposes a person-centered, holistic approach using sequence analysis (SA) combined with tree-based modeling to typologize full educational trajectories for the 1976–1988 cohort and relate them to background characteristics. The goal is to reveal cumulative advantage/disadvantage across entire trajectories and provide a complementary perspective to variable-centered models of educational inequality.

The paper reviews the evolution from linear and stagewise models (Duncan; Blau & Duncan; Mare) toward frameworks recognizing persistent background effects, including MMI (Raftery & Hout) and EMI (Lucas). It details methodological critiques: scale/comparability problems in logits, heteroskedasticity, selection bias, and the strong assumptions of dynamic discrete choice models. Extensions (heterogeneous choice, sequential logit, multinomial/latent class, IV-based multinomial) mitigate but do not eliminate issues such as IIA violations and dynamic selectivity. In China, most studies using logit/transition models emphasize college entry and often find “waning coefficients,” though findings can be mixed across periods and specifications; EHA variants face challenges with repeated events, frailty, and complexity. The paper argues for SA as a person-centered method aligning with holistic interactionism, capturing cumulative (dis)advantage and trajectory heterogeneity without strong distributional assumptions. Research hypotheses: (1) With compulsory education universalization and higher education expansion, early termination declines and higher education trajectories increase. (2) Higher parental education is associated with more stable and higher-quality trajectories. (3) Rural origin is associated with more terminations and fewer cumulative-advantage pathways relative to urban origin.

Data: CGSS 2008 (n≈6000). Focus on 1976–1988 birth cohort to capture trajectories spanning the 1994 unified enrollment reform through 2008. After excluding 21 logically inconsistent sequences, the analytic sample is 1305 respondents (3915 person-stage records). A life history calendar of education is converted to spell format and state sequences, distinguishing institution types and levels (e.g., key vs non-key junior/senior high, vocational/technical tracks, junior college, university tiers, postgraduate), and including an empty-phase state between stages. Variables: Outcomes are trajectory clusters from SA. Predictors include residence at age 14 (village/town–county/prefecture city/provincial capital–municipality), parental highest education (years), learning situation at age 14 (10-item, 5-point scale; composite score; Cronbach’s alpha≈0.65), gender, and age (proxy for cohort birth year within 1976–1988). Sequence analysis: Distances computed via OMloc (optimal matching with location costs), substitution cost matrix derived from transition rates; exponential cost parameter set to 0. Hierarchical clustering with Ward’s method on the distance matrix; 16 clusters selected for interpretability and fit, capturing key/non-key tracking and academic vs vocational bifurcation. Tree-based modeling: To avoid reversion to variable-centered multinomial regression and high-order interactions, the study uses a conditional inference tree (CIT) to relate background factors to trajectory clusters. CIT employs permutation tests for unbiased variable selection and determines stopping without post-pruning. Data split 70/30 for train/test; accuracy≈31.5% (train) and 33.9% (test), suggesting no overfitting; random forest baseline≈30.1%. Feature importance in RF (mean decrease in accuracy): urban/rural highest, followed by parental education and learning situation; gender minimal. Statistical association between leaf nodes and clusters is strong (e.g., χ2 tests p<0.001).

- The 1976–1988 cohort’s trajectories show clear path dependency, branching, and selectivity, with expansion policies increasing senior high and tertiary participation. - Sixteen trajectory clusters group into three broad categories: (1) general education pathways (Clusters 3, 4, 5, 8, 10, 11, 12, 13) totaling 33.0%; (2) termination pathways (Clusters 1, 2, 9, 15, 16) totaling 51.1%; (3) alternative vocational pathways (Clusters 6, 7, 14) totaling 15.9%. - Cumulative advantage and disadvantage are evident: elite pathways (e.g., Cluster 8 super-elite; Cluster 4 general elite) show strong accumulation; termination types reflect cumulative disadvantage. “Reversal” or antitype patterns (Clusters 9, 15) exist but are minority cases. - The CIT’s primary split is urban–rural: rural youth are markedly more likely to follow Cluster 2 (termination at junior high), evidencing lower prevalence of compulsory education completion and fewer higher-education trajectories (supports Hypothesis 3). - Parental education exerts strong, persistent effects in both urban and rural contexts, shaping entry into both high- and low-quality trajectories (supports Hypothesis 2). Thresholds around 6 years (primary completion) and 9 years (junior high) are consequential in splits. - Age (period/birth-year proxy) captures cohort effects of expansion: younger rural individuals (>6 years parental schooling) are less likely to terminate and more likely to follow stable or improved pathways; among urban youth, those ≤27 show more movement into higher-quality academic clusters (supports Hypothesis 1). - Learning situation at age 14 differentiates urban youth with higher parental education: better scores (e.g., >35) increase probabilities of elite or higher-quality pathways (e.g., Clusters 4, 8, 5). - Even the worst urban node shows higher chances of advantageous pathways and lower chances of inferior ones than rural nodes. Urban groups benefit distinctly from junior colleges and prefectural-level institutions as stepping stones. - Identification of a “well-educated echelon” (largely urban, with higher parental education) and a “small-town swot” subgroup benefiting from expansion yet experiencing status dissonance. - Model diagnostics: CIT yields eight terminal nodes with significant distributional differences in cluster memberships (omnibus χ2 p<0.001). Predictive accuracy is modest by design (interpretability prioritized), indicating substantial unobserved heterogeneity.

The findings demonstrate that educational inequality in China emerges early and accumulates throughout the schooling career, contradicting interpretations of diminishing background effects during successive transitions. A holistic, person-centered trajectory view reveals consistent cumulative advantage/disadvantage shaped by urban–rural structures, parental education, and learning context. The prominence of the urban–rural split underscores institutional and structural barriers limiting rural students’ access to higher-quality trajectories, while parental education persistently channels children toward more advantageous paths. Apparent local equalities (e.g., similar shares of high-school termination among some nodes) do not translate into systemic equality because many rural students exit before reaching high school. The approach reframes selectivity from a bias source to an analyzable feature: clustering entire sequences and using tree-based splits enables identification of heterogeneous, conditional effects and subgroups (including rare reversals) that variable-centered transition models can miss. Overall, results support Hypotheses 1–3 and suggest that expansion policies benefitted younger cohorts and urban/small-town youth more, with compensatory effects narrowing some gaps but not eliminating structural inequalities.

This study advances the analysis of educational stratification by shifting from segmented transitions to holistic trajectories via sequence clustering combined with conditional inference trees. It documents persistent cumulative advantage/disadvantage across the Chinese 1976–1988 cohort, with urban–rural origin and parental education strongly shaping trajectory type, and with expansion policies benefiting younger cohorts and urban/small-town youth. Methodologically, the person-centered approach complements econometric transition models by addressing heterogeneity and selection at the trajectory level, improving subgroup comparability and interpretability. Future research should integrate richer time-varying covariates and mechanisms with trajectory methods (e.g., multistate SA, mixed methods), leverage larger datasets to refine typologies and subgroup identification, and explore advanced algorithms (graph-based models, simulation, deep learning) to capture dynamic processes underlying trajectory shifts.

Sequence analysis faces challenges incorporating time-varying variables and relies on data-driven clustering that may produce complex or unstable patterns in high-dimensional settings. The approach is computationally intensive and depends on detailed life-history data. The study’s analytic sample (n=1305) for the focal cohort is relatively small, potentially limiting robustness of some subgroup results. Missing information on parental occupation/income constrains background measurement (parental education used as proxy). While the method mitigates selection issues through holistic clustering and inclusive sampling, unobserved heterogeneity remains. Generalization requires validation with larger or additional cohorts; future work should integrate multistate SA and complementary typological or causal models.