Analysing how government-provided vocational skills training affects migrant workers' income: A study based on the Livelihood Capital Theory

Following China’s transition to the Lewis turning point, migrant workers’ absolute wages increased, yet their relative income and income satisfaction declined, contributing to relative poverty and low capital stocks among this group, with implications for social justice and equality. To narrow urban–rural gaps, the government has promoted vocational skills training (e.g., Guiding Opinions, Vocational Skills Upgrading Program 2019–2022, Spring Tide Action) across numerous occupations with varying durations. Key policy questions are whether GPVST improves migrant workers’ income level (IL) and income satisfaction (IS), and through what mechanisms. Prior evidence on training returns is mixed: international studies often find positive but modest returns and context-specific heterogeneity, while Chinese studies report both significant positive income effects and low or insignificant returns due to constraints in training content, quality, and self-selection. Methodologically, the self-selection of trainees, unobservable factors (e.g., social relations, motivation), and varying data sources complicate causal inference. Existing research tends to emphasize objective income effects and the role of human capital, overlooking IS and broader livelihood capitals. This study addresses these gaps by using CLDS 2016–2018 panel data and a PSM-DID approach to estimate causal effects of GPVST on IL and IS, and by applying a livelihood capital framework with mediation analysis to uncover multidimensional mechanisms.

International research since the 1980s–1990s documents positive but heterogeneous returns to training (e.g., ~5% on average), with effects varying by gender, duration, and content; however, most focus on unemployed or general workers rather than migrants, limiting external validity to China’s migrant context. Within China, studies are divided: some find vocational training increases migrant wage rates and employment quality, while others report low or insignificant returns due to training design, quality, budget constraints, and labor market segmentation. Theoretical lenses vary (e.g., life-cycle theory, human capital theory), and many studies differentiate by training type or worker age. Key methodological concerns include self-selection into GPVST, reverse causality, and omitted variables. Moreover, the literature pays more attention to objective income than to subjective income satisfaction, and largely centers on human capital, often neglecting the roles of social, financial, physical, and natural capitals emphasized by the sustainable livelihoods framework. The study positions itself to contribute by jointly evaluating IL and IS and by integrating the DFID Sustainable Livelihoods Framework to examine multiple capital-mediated pathways.

Design and identification: The study employs a propensity score matching combined with difference-in-differences (PSM-DID) framework to mitigate self-selection, reverse causality, and time-invariant unobservables. China’s renewed National Migrant Worker Training Program (2015) underpins the timing. The initial period is 2016 and the intervention period is 2018. Treatment comprises migrant workers who had received GPVST by 2016; controls did not. DID common trend assumptions are supported by matching on initial characteristics. Model specification: Y_it = β0 + β1 D_i + β2 T_t + β3 (D_i × T_t) + Σ β_k Controls_it + ζ_i + ε_it, where Y is income status (IL: log annual total income; IS: 1–5 satisfaction), D indicates treatment, T indicates 2018, and D×T is the core effect. ATT is computed within the matched sample. Controls include gender, age (log), marital status, hukou (local vs. foreign), migration experience, education, annual work hours (log), work distance (1–5), commute time (log minutes), industry (tertiary=1), years worked, settlement intention (1–5), perceived social status (1–10), and fairness perception (1–5). Province fixed effects and sample weights are used. Data and sample: CLDS 2016 and 2018 nationally representative panel covers 29 provinces (excluding Hong Kong, Macao, Taiwan, Tibet, Hainan), targeting ages 15–64. The analysis retains migrant workers present in both waves (females 16–55, males 16–60; agricultural hukou; away >180 days/year; currently in non-agricultural jobs), excluding students, zero workdays, and missing key variables. Final balanced panel: N=1,823 (control=1,686; treatment=137; 7.52% treated), consistent with official 2016 training participation rates. Variables: Explained—IL (log annual total income across all sources), IS (1 very dissatisfied to 5 very satisfied). Explanatory—GPVST participation (yes=1). Time—2016=0, 2018=1. Interaction—Tra×time. Controls—listed above. Mediation: Parallel multiple mediation assesses how livelihood capitals mediate GPVST effects. Livelihood capital indices (human, social, physical, natural, financial) are constructed using 15 indicators weighted by the entropy method (e.g., human: health, dialect mastery, qualifications; social: counts of economic/decision/emotional support; physical: housing area/log, facilities, ownership, environment; natural: land transfer, water source, cropland area/log; financial: savings/log, commercial insurance, financial assets). The KHB method decomposes total, direct, and indirect effects for IL and IS, estimating capital-specific mediation and overall mediation. Matching quality: Common support assessed via distributions and kernel densities shows most observations within overlap with minimal loss. Balance tests show standardized differences <10% post-matching; pseudo-R2 drops and LR test becomes insignificant, indicating successful balance. Robustness: Alternative matching ratios (1:1, 1:3), replacing treatment intensity with number of trainings, and Oster (2019) coefficient stability tests confirm robustness. Heterogeneity: Triple interactions examine differences by generation (older: born 1960s–70s vs. new: post-1980s) and hukou (local vs. foreign).

Main effects (PSM-DID): GPVST significantly increases both IL and IS. Estimated IL gains range from approximately 14.6% to 34.7%, and IS increases by about 6.9% to 9.3%, with significance robust to control inclusion and alternative specifications. Representative coefficient magnitudes include Tra×time for IL around 0.146–0.347 and for IS around 0.069–0.093. Mechanism (mediation via livelihood capital): For IL, total GPVST effect is ~0.182–0.183; overall livelihood capital mediation explains 63.93% of the total effect. Human capital mediates 42.08% (indirect ~0.077), financial capital 29.51% (indirect ~0.054), social capital 2.73% (indirect ~0.005); physical and natural capitals’ indirect effects are not statistically significant. For IS, total effect ~0.141–0.147; the direct effect dominates (59.59%), with overall indirect effect 40.41%. Significant mediators: human capital (21.77%; indirect ~0.032) and financial capital (17.02%; indirect ~0.024). Social capital’s indirect effect is smaller (~9.92%) and sometimes marginal; physical and natural capitals are not significant mediators. Heterogeneity: By generation, GPVST’s effect on IL is not significant for the new generation, while IS benefits are larger for the older generation (negative triple interaction for new generation on IS, indicating older > new). By hukou, IL effects are stronger for foreign (non-local) migrant workers, whereas IS effects are stronger for local hukou workers. Robustness: Results are consistent across matching specifications and with alternative treatment intensity; Oster tests indicate limited omitted-variable bias. Descriptive contrasts also show higher pre/post IL and IS in the treated group versus controls.

GPVST meaningfully raises both objective earnings and subjective income evaluations, directly addressing the study’s questions about impacts and mechanisms. The findings align with human capital theory: training enhances skills, productivity, and market signaling, yielding higher returns and satisfaction. Beyond human capital, GPVST operates through social capital (information-interaction effects), expanding networks that improve job search, bargaining power, and alignment between expectations and outcomes, and through financial capital (a trickle-down pathway), via improved financial literacy and access to financial products that enable investment and risk management. The differential dominance of direct versus indirect channels is noteworthy: for IL, indirect effects via capital accumulation dominate, implying GPVST’s contribution is largely through enhancing individuals’ asset bases. For IS, the direct effect is stronger, consistent with training altering expectations, job matching, and perceived returns independent of measured capital accumulation. The absence of significant mediation by natural and physical capital likely reflects short-term horizons (two-year panel), the stickiness and long-run nature of physical capital accumulation, and institutional/behavioral factors (e.g., incomplete land transfer mechanisms, cultural preferences to retain land as security). Policy relevance is high: strengthening GPVST can promote upward mobility, reduce urban–rural disparities, and enhance perceived fairness, but design should consider heterogeneity, particularly focusing on foreign hukou workers for IL gains and addressing the expectations and integration of older workers to improve IS.

Using CLDS 2016–2018 panel data and a PSM-DID design, the study demonstrates that GPVST significantly improves migrant workers’ income level and income satisfaction. Heterogeneity analyses show stronger IL effects for foreign hukou migrants and stronger IS effects for older and local migrants. Mechanism analysis based on the livelihood capital framework reveals that human, financial, and to a lesser extent social capital mediate the effects—indirect channels dominate for IL, while direct effects dominate for IS. Contributions include jointly assessing objective and subjective income outcomes, clarifying multidimensional capital pathways, and providing evidence-based guidance for optimizing public training. Policy implications: expand GPVST’s breadth and depth by engaging enterprises, social organizations, and educational institutions; align curricula with evolving industrial demand and provide staged, hierarchical content (including digital, safety, and legal competencies); strengthen training–employment linkages; build transparent, authoritative vocational qualification systems to enhance labor market signaling and access to credit and housing; and develop employment information and policy service platforms to bolster social and financial capitals. These measures can enhance migrant workers’ assets and capabilities, thereby sustaining improvements in IL and IS and advancing urban–rural integration. Future research should leverage more recent multi-period data and distinguish GPVST types (skills vs. entrepreneurship training) to refine causal mechanisms and indicator systems.

Data limitations include reliance on two waves (2016, 2018); updating to more recent CLDS waves and extending to 2–3 or more periods would allow dynamic assessments and validation. Training categorization is constrained: GPVST types (skills vs. entrepreneurship) are not separately analyzed due to data limits; future work should differentiate modalities and intensities. Further refinement of theoretical models and measurement of livelihood capitals is warranted to capture nuanced pathways and long-run capital accumulation, particularly for physical and natural capitals.