How does algorithmic management affect gig workers?
A two-wave study of 389 Chinese delivery workers links platform management with both opportunity and threat appraisals. The self-reported, non-random design cannot prove cause, but it points to tracking transparency and appeals—not resilience training alone.
The short version
- What happened
- The final matched sample contained 389 delivery workers from Shanghai, Nanjing and Suzhou, retained from 571 valid first-wave respondents after a three-week follow-up.
- Who it affects
- Delivery workers may experience platform management as useful coordination and as threatening control at the same time.
- What remains unproven
- Observational self-report design cannot establish causal direction
Keep in mind
Observational self-report design cannot establish causal direction
Only 389 of 571 valid first-wave respondents were retained in the matched analysis
Three-city convenience sample recruited through delivery stations
No audited platform rules, earnings, safety incidents or device logs
Secondary dimension analyses were not experimental tests of design changes
Source analysis · 1 linked source. Primary analysed source date: 15 October 2026.
Read the full report · 6 min The evidence, methods, implications and limitations
The answer: workers can experience the same system as help and control
In this study, algorithmic management was linked with two opposing interpretations among food-delivery workers. It was associated with seeing work demands as a challenge that could support achievement, and with seeing them as a threat that could bring loss or pressure. Challenge appraisal related positively to workplace wellbeing; threat appraisal related negatively. The finding is more useful than a simple verdict that algorithms are either good or bad for work.
It is not causal evidence. Workers rated the management system, their own appraisals, resilience and wellbeing in surveys. Although the researchers separated measurements by three weeks and tested several statistical models, an observational self-report design cannot show that a platform feature changed wellbeing. Worker circumstances, station culture, selection into the job and earlier experiences can influence both how management is perceived and how work feels.[1]
Who participated and how attrition shaped the sample
Researchers recruited adult food-delivery workers through delivery stations in Shanghai, Nanjing and Suzhou from August to November 2025. At the first wave, 657 people entered the survey and 571 valid responses remained after screening. Three weeks later, 496 completed a second questionnaire. After matching and excluding responses completed in under five minutes, the analysis contained 389 workers—a retention rate of 68.1% from the valid first-wave group.
The sample was 86.9% male, 43.2% aged 18–25, 82.0% full-time and 62.2% unmarried. Workers received CNY 20 for each survey wave. Platform regional managers helped researchers gain access to stations, but the paper states that responses were confidential and unavailable to managers. Attrition matters: people who left the job or declined the second survey may have had different experiences, and the final sample is not a probability sample of all delivery workers in China.[1]
What the path model found
The structural model estimated a standardised path of 0.500 from algorithmic management to challenge appraisal and 0.324 to threat appraisal. Challenge appraisal was associated with workplace wellbeing at 0.451, while threat appraisal had a negative path of minus 0.202. Bootstrap estimates suggested a positive indirect path through challenge of 0.226 and a negative indirect path through threat of minus 0.065; both confidence intervals excluded zero.
These values describe fitted associations among questionnaire constructs, not the size of a real-world intervention. The researchers controlled for demographic factors and reported good measurement-model fit. Statistical tests found no clear evidence that one general response tendency explained the results, yet every core variable still came from the same person. Such tests cannot eliminate shared-method bias, reverse direction or omitted causes.[1]
Tracking stood out as a possible threat signal
A post-hoc analysis separated four dimensions of algorithmic management. Assessment and feedback, plus allocation and pricing, were positively associated with challenge appraisal. Tracking and monitoring was the only dimension significantly associated with threat appraisal in that model, with a standardised coefficient of 0.144. Training and guidance was not significantly related to either appraisal after the other dimensions and controls were included.
Because this was a secondary analysis, it should guide questions rather than platform policy on its own. Still, it gives a concrete hypothesis: support functions may feel enabling when rules are understandable, while opaque monitoring can feel punitive or uncontestable. A better study would test specific design changes—clearer tracking rules, explanations, realistic deadlines and appeal channels—and measure stress, earnings, safety and retention before and after implementation.[1]
Resilience is not a substitute for fair system design
Psychological resilience strengthened the positive association between algorithmic management and challenge appraisal, but it did not significantly weaken the association with threat or the negative indirect path to wellbeing. The authors therefore caution against treating resilience as a personal resource that can compensate for platform pressure. A worker may be skilled at coping and still face opaque penalties, unstable pricing or monitoring that cannot be challenged.
For platform operators, the practical response is procedural: explain how allocation, ratings and monitoring affect work; give workers usable routes to correct data and appeal decisions; and monitor whether rules create unsafe rushing or unpredictable income. For regulators and worker representatives, the result supports examining concrete management functions rather than accepting a generic claim that an algorithm optimises efficiency. Individual wellbeing training cannot resolve a system-level lack of transparency or voice.[1]
Evidence limits and what would change the assessment
The sample covered three prosperous eastern Chinese cities and one kind of gig work. It cannot represent drivers, care workers, freelancers or delivery workers elsewhere. Recruitment through stations may miss the least connected workers. Attrition between waves can bias results, and the three-week design does not establish temporal order for all constructs because algorithmic-management perceptions and resilience were measured together before appraisal and wellbeing. The study also measures perceived management, not audited platform rules or device logs.
Chinese education grants and Jiaxing Nanhu University funded the research; the authors declared no commercial conflict. Stronger evidence would combine independent platform data with repeated worker measures, recruit representative panels and evaluate rule changes prospectively. Randomised or phased introduction of explanations and appeals could test effects on wellbeing, earnings, task acceptance, safety and discrimination. Until then, the study supports transparency and worker voice as prudent safeguards, not a numerical estimate of how much any platform causes harm or benefit.[1]