Tighter algorithmic control on food-delivery platforms fuels covert worker resistance: riders who feel their skills are underused are more likely to game or defy platform algorithms, with perceived overqualification explaining much of the effect.

Main Finding

Perceived algorithmic control (normative guidance, real‑time surveillance, behavioral constraints) increases gig workers’ anti‑algorithm behaviors, and this effect is significantly mediated by perceived overqualification: algorithmic management can make workers feel their skills are underused, which in turn catalyzes covert resistance (e.g., rule‑bending, exploiting loopholes).

Key Points

• Core constructs
  • Perceived algorithmic control: workers’ sense of being governed by algorithmic norms, monitoring, and enforced behavioral constraints.
  • Perceived overqualification: subjective belief that one’s skills, experience, or education exceed job demands.
  • Anti‑algorithm behaviors: informal/covert resistance such as delaying order acceptance, circumventing rules, exploiting system loopholes.
• Theoretical framing: Conservation of Resources (COR) theory. Excessive algorithmic control depletes autonomy and other resources (loss spiral), fostering perceived overqualification and motivating resource‑protective/resistance behaviors.
• Mechanism: Algorithmic standardization reduces workers’ opportunities for autonomous decision‑making and skill use → workers interpret this as underutilization of capabilities (overqualification) → increased propensity to engage in anti‑algorithm behavior to restore autonomy/resource balance.
• Practical observation: While algorithmic control may raise short‑term operational efficiency, it can undermine service quality and platform outcomes by provoking covert resistance and stress.
• Boundary/context: Results are drawn from food delivery riders in three eastern Chinese cities; phenomena likely generalize to other algorithmically managed, low‑autonomy gig roles, but cultural/institutional variation may moderate effects.

Data & Methods

• Design: Three‑wave longitudinal survey.
• Sample: 483 food delivery riders across three cities in eastern China.
• Measurement/operationalization:
  • Perceived algorithmic control measured across dimensions of normative guidance, real‑time surveillance, and behavioral constraints.
  • Perceived overqualification as subjective person–job mismatch.
  • Anti‑algorithm behavior captured as self‑reported covert resistance behaviors.
• Analysis: Partial least squares structural equation modeling (PLS‑SEM) to test direct and mediated paths.
• Strengths/limits of method:
  • Strengths: longitudinal design helps temporal ordering; PLS‑SEM suitable for complex mediation with latent constructs.
  • Limits: self‑report measures, single industry/geography, potential unobserved confounders; PLS‑SEM does not by itself establish definitive causality.

Implications for AI Economics

• Platform efficiency vs. behavioral externalities: Algorithmic management improves matching and throughput but can create negative behavioral externalities (covert resistance) that reduce realized productivity and service quality—models of platform performance should account for endogenous worker responses to control intensity.
• Labor supply and turnover: Perceived overqualification and resultant resistance may affect retention, hours supplied, and gig workers’ reservation utilities. Empirical models estimating supply elasticities or lifetime earnings on platforms should control for perceived autonomy and skill utilization.
• Market competition and platform design: Platforms that balance algorithmic control with mechanisms for skill utilization, feedback, and explainability may gain a competitive advantage by lowering resistance costs. Investments in transparent algorithms, human channels for exceptions, and skill‑based task allocation could be economically justified.
• Welfare and regulation: Regulators should consider algorithmic intensity as a labor market friction that can harm worker welfare and service reliability. Policies encouraging algorithmic transparency, appeals processes, and minimum participation of human managers may reduce social costs.
• Measurement suggestions for researchers/economists:
  • Include survey modules for perceived algorithmic control and perceived overqualification when studying platform labor outcomes.
  • Use administrative logs to correlate control intensity (e.g., frequency of reassignments, tightness of time windows) with behavioral anomalies (late arrivals, cancellations) as objective proxies for anti‑algorithm behavior.
  • Explore heterogeneity: skill level, prior occupation, local labor market tightness, and regulatory environment as moderators.
• Research agenda: Experimental or quasi‑experimental work (A/B tests, policy changes) to estimate causal impact of reducing algorithmic control or increasing channels for skill use on productivity, earnings, churn, and welfare; structural models integrating worker psychology into platform equilibrium analyses.