Chinese listed firms that visibly disclose stronger responsible workplace data governance achieve higher returns on assets: a one‑SD increase in an employee‑weighted disclosure index is associated with a 0.49 percentage‑point ROA gain (≈13% of the mean). The effect is amplified by firms' learning capability, weakened by technical complexity, and partially mediated by lower administrative compliance costs, though disclosure-based measurement and observational design limit causal interpretation.

Main Finding

An employee-informed, disclosure-based index of responsible workplace data governance for Chinese listed firms is positively associated with firm performance: a one–standard-deviation increase in the governance index predicts a 0.49 percentage-point higher ROA (≈13.2% of the sample mean). Evidence indicates part of this association operates through lower administrative compliance costs (≈10.8% mediation). Organizational learning amplifies the association; technical complexity attenuates it. Tobin’s Q shows a positive but less stable relationship.

Key Points

• Constructed a firm-year index of visible responsible workplace data governance using:
  • 519 employee questionnaires to weight five governance dimensions and 19 indicators.
  • Annual-report scoring for Chinese A‑share non‑financial listed firms (2011–2023).
• Primary outcome: return on assets (ROA). Secondary outcome: Tobin’s Q.
• Main estimate approach: two-way fixed effects panel models, supplemented by timing tests, dynamic-panel checks, strengthened fixed effects, mediation exploration, and moderation analyses.
• Quantitative result: +1 SD in the index → +0.49 pp ROA (13.2% of the sample mean).
• Mechanism: ~10.8% of the governance–ROA association is attributed to reduced administrative compliance costs.
• Moderation:
  • Organizational learning capability strengthens the positive governance–performance link.
  • Technical complexity of data/AI systems weakens the link (higher verification/audit burden).
• Ownership heterogeneity: stronger ROA association in non–state-owned enterprises than in state-owned enterprises.
• Limits: disclosure-based proxy can reflect “machinewashing” or regulatory signaling; observational design prevents definitive causal or individual-level psychological claims.

Data & Methods

• Measurement:
  • Employee stage: 519 valid questionnaires from employees of listed firms used to weight five governance dimensions (reported dimensions include organizational governance, technical trustworthiness, rights protection, social impact, operational mechanisms) and 19 specific indicators.
  • Firm-year stage: manual/algorithmic scoring of annual reports to produce the visible governance index for firm‑years (Chinese A-share non-financial firms, 2011–2023).
• Sample: panel of Chinese A‑share non‑financial listed firms (2011–2023). (Paper reports firm-year panel; exact N of firm-years in full paper.)
• Outcomes:
  • Primary: ROA (accounting performance).
  • Secondary: Tobin’s Q (market recognition).
• Estimation strategy:
  • Two-way fixed effects (firm and year) as the baseline.
  • Robustness: timing/lag specifications, dynamic panel checks, strengthened fixed effects.
  • Mediation: exploratory check linking governance → lower administrative compliance costs → higher ROA (estimated share ≈10.8%).
  • Moderation: interaction tests with organizational learning capability and technical complexity.
• Robustness and caveats explicitly acknowledged: possibility of strategic disclosure, regulatory pressure, and limits of disclosure to capture lived employee experience.

Implications for AI Economics

• Measurable private returns to visible data-governance investments: The effect size (0.49 pp ROA per SD) implies meaningful accounting returns to firms that make responsible governance visible and systematized, suggesting governance is an actionable strategic input in AI/data investments.
• Complementarity between governance and organizational capital: Organizational learning amplifies governance returns — firms should pair governance policies with investments in training, cross-functional coordination, and absorptive capacity to capture benefits.
• Complexity imposes frictions: Highly complex AI/data architectures reduce the payoff to governance visibility (auditability and verification costs rise). Economists modeling AI adoption should include technical complexity as a moderator of governance value and a source of implementation cost.
• Compliance-cost channel and cost-benefit accounting: A nontrivial share of governance value operates by lowering administrative compliance costs. Firms and analysts should incorporate these recurring cost savings (not only reputational or risk mitigation) when valuing governance investments.
• Market recognition is weaker/less stable: Tobin’s Q evidence is positive but less robust than ROA, implying markets sometimes recognize governance value but not consistently; this may reflect information frictions, signaling noise, or heterogeneous investor priors about disclosure credibility.
• Policy and regulation design: Regulatory encouragement of transparent governance disclosure could help internalize coordination and fairness externalities, but policymakers should be wary of disclosure-only regimes that could encourage symbolic compliance (machinewashing). Complementary auditing/assurance and incentives for organizational learning will increase social welfare.
• Empirical research guidance: The two-stage, employee-weighted disclosure index is a scalable approach for large-sample work on AI governance, but researchers should guard against overinterpreting disclosure as lived experience and use complementary microdata where possible.
• Heterogeneity matters for comparative statics: Ownership and institutional context alter returns to governance; cross-country or cross-sector AI economics models should allow for such heterogeneity when predicting welfare or firm behavior.

Limitations to keep in mind for application: index is disclosure-based (visible practices), evidence is associational, and results are from Chinese listed non‑financial firms—generalization to other institutional settings requires caution.