China's educational expansion since the 1990s, driven by factors like education commercialization and employment pressures, led to a dramatic increase in higher education enrollment. While many studies suggest this exacerbated inequality, particularly in higher education, the methodologies used often focus solely on terminal outcomes (university admission) or analyze trajectories in segmented stages. This study addresses this gap by adopting a holistic perspective on educational trajectories using sequence analysis, cluster analysis, and decision tree analysis to create a typology of educational pathways and examine the relationship between background characteristics and these pathways. This approach avoids the limitations of traditional models like the Mare model, which often show "waning coefficients," suggesting a diminishing effect of family background as students progress. The researchers hypothesize that educational expansion increases overall opportunities but that inequalities persist, especially concerning qualitative differences in education. They further hypothesize that highly educated parents enhance their children's educational trajectories and that rural individuals face significant disadvantages compared to urban counterparts.

Existing research on educational attainment frequently employs linear regression models (Duncan, 1967; Blau and Duncan, 1967) or the Mare model (Mare, 1979, 1980, 1981), which uses a series of logit models for each educational stage. The Mare model often reveals "waning coefficients," indicating a diminishing influence of family background as students progress. This has been interpreted to support the "maximally maintained inequality" (MMI) hypothesis (Raftery and Hout, 1993), suggesting that inequality diminishes with educational expansion. However, the Mare model has been criticized for potential biases due to heteroskedasticity, selectivity-based endogeneity, and unrealistic assumptions about rational agents (Cameron and Heckman, 1998, 2001; Holm and Jæger, 2011). Alternative models, like dynamic discrete choice models (DDCM) and those addressing effectively maintained inequality (EMI) (Lucas, 2001), suggest that family background effects persist even with universal education. Studies in China have predominantly used logit models focusing on college enrollment, often observing "waning coefficients." Event history analysis (EHA) offers an alternative but has limitations in handling the complexities of educational pathways. The authors argue that sequence analysis, a person-centered approach, offers a more suitable methodology for capturing the heterogeneity and cumulative effects within educational trajectories.

The study utilizes data from the 2008 Chinese General Social Survey (CGSS 2008), focusing on the 1976-1988 birth cohort (N=1305). Sequence analysis, using the TraMineR and TraMineRextras R packages, is employed to analyze educational trajectories, which are categorized into 21 distinct educational states. The optimal matching distance method (OMloc) with hierarchical clustering (Ward method) identifies 16 distinct clusters representing different educational trajectory types. A conditional inference tree (CIT) model is then used to investigate the relationship between these trajectory clusters and background characteristics, including gender, age, type of residence at age 14, parents' highest education level, and learning situation at age 14. The CIT model avoids the limitations of traditional regression models by allowing the data to shape the model structure and handling complex, non-linear relationships without distributional assumptions. The researchers justify their causal inference approach by arguing that the clustering process addresses unobserved heterogeneity and mitigates selectivity-based endogeneity.

A Sankey diagram visualizes the dynamic and heterogeneous nature of educational trajectories within the cohort. Sequence analysis reveals a general progression from elementary to higher education, with a rise in informal education after formal schooling. Cluster analysis identifies 16 distinct trajectory types, categorized into general education, education termination, and alternative pathways (primarily vocational). These clusters reveal cumulative advantages and disadvantages. For example, Cluster 8 (super-educated elite) shows a higher proportion of students from elite high schools, while Cluster 2 (junior high school termination) represents cumulative disadvantage. The CIT model reveals that the urban-rural divide is the most significant predictor of educational trajectory, with rural individuals more likely to experience early termination (Cluster 2). Parental education significantly impacts educational outcomes in both urban and rural areas, supporting Hypothesis 2. The model also identifies a period effect related to the expansion of education, with younger individuals benefiting more from increased access to higher education, particularly in rural areas. The study identifies distinct subgroups, including a highly-educated elite and a "small-town swot" group, highlighting the complex interplay of background factors and individual choices. The analysis does not support the "waning coefficients" effect; instead, it suggests persistent inequality throughout educational careers. The learning situation at age 14 also plays a significant role, particularly for urban youth with highly educated parents.

The findings challenge the prevailing notion of diminishing inequality with educational expansion. The study demonstrates that inequalities manifest early and accumulate over time, contradicting the "waning coefficients" often observed in traditional models. The holistic interactionistic approach, using sequence analysis and decision trees, provides a more nuanced understanding of the complex interplay between social background and educational trajectories. The urban-rural divide and parental education emerge as key determinants of educational success, with rural students facing significant disadvantages. The identification of distinct subgroups like the "small-town swots" highlights the complexity of educational stratification in contemporary China. The results underscore the importance of adopting a person-centered approach that considers the entire educational pathway rather than focusing solely on specific transitions.

This study offers a novel approach to analyzing educational inequality, demonstrating the limitations of traditional models and highlighting the importance of a holistic perspective. The findings reveal persistent inequality across educational trajectories, emphasizing the cumulative effects of social background. Future research could explore the micro-mechanisms driving these trajectories, integrating qualitative and quantitative methods to gain a deeper understanding of the complexities involved. Larger datasets and more sophisticated models, such as those incorporating graph theory or deep learning, could further enhance the analysis of these intricate educational pathways.

The study's reliance on cross-sectional data from a single survey limits the ability to fully capture dynamic changes in educational trajectories. The data-driven nature of the cluster analysis may also influence the results. The sample size, although substantial, may not be fully representative of the entire population of China. The lack of detailed information on certain variables, such as parental occupation, could also affect the interpretation of the results.