Abstract
Background
Delivery gig workers face elevated risks of occupational injury, yet little is known about how working time exposures contribute to injury risk and the pathways underlying this association. This study examined the associations between multiple dimensions of working time exposures and occupational injury and whether sleep quality represented an indirect pathway co-occurring with these associations.
Methods
Using cross-sectional survey data from 1,537 delivery gig workers in Shanghai, China, we examined the associations between three dimensions of working time exposures (daily working hours, weekly working days, and night-shift frequency) and occupational injury. Sleep quality was examined as a potential indirect pathway based on the Effort–Recovery Model. We estimated a series of regression models and used bootstrap methods to decompose direct and indirect associations while adjusting for sociodemographic and job-related covariates.
Results
Longer daily working hours and more frequent night shifts were directly associated with higher occupational injury counts. Weekly working days and night-shift frequency showed significant indirect associations with occupational injury through poorer sleep quality. Robustness checks using a binary injury indicator yielded consistent results.
Conclusion
Working time exposures are associated with occupational injury risk among delivery gig workers through both direct exposure–related associations and indirect pathways involving sleep quality. Interventions and policies targeting excessive working hours, night work, and sleep quality may help reduce injury risks in this vulnerable transport workforce.
Keywords: Delivery gig workers, Occupational injury, Sleep quality, Working time exposures
1. Introduction
Gig work is short-term contract work mediated by virtual platforms, and laborers engaged in this work are commonly referred to as gig workers [[1], [2], [3]]. The global gig workforce has grown rapidly, with recent World Bank estimates ranging from 154 to 435 million workers and higher growth in developing countries such as China [4]. Within this expanding workforce, delivery gig work has scaled dramatically: the global online food delivery market was estimated at 288.84 billion USD in 2024 and projected to reach 323.29 billion USD in 2025 [5]. Delivery gig workers constitute a precarious segment of the urban transport workforce [6], performing time-sensitive tasks embedded in continuous road use under insecure and contingent working arrangements.
Occupational injury has emerged as a major health concern in this workforce. In a large survey of food delivery gig workers in New York City, 21.9% reported having been injured while working, which is twice the rate among car drivers [7]. Among motorcycle food delivery riders in Chiang Mai, Thailand, 18.9% reported experiencing traffic accidents within six months [8]. Existing evidence further suggests that delivery gig workers face elevated risks of traffic-related crashes, falls, collisions, and musculoskeletal injuries during delivery work [[7], [8], [9], [10]]. However, less attention has been paid to how working time exposures in delivery gig work may shape injury risk.
Working time arrangements are a central feature of delivery gig work, shaped by algorithmic management, performance-based pay, and fluctuating consumer demand. Drawing on international frameworks for analyzing working time and worker health, working time exposures encompass multiple dimensions, including the duration of work within a day, the number of working days across a week, and the timing of work such as night work [11,12]. Because most gig workers are classified as independent contractors rather than formal employees, they are not protected by standard labor laws on maximum working hours or overtime pay [13] and often work long hours under unstable schedules [2]. According to the Effort–Recovery Model [14], sustained work demands without adequate recovery lead to cumulative physiological and psychological strain. From this perspective, working time exposures may elevate occupational injury risk by disrupting the balance between effort expenditure and recovery in traffic-intensive work that demands sustained attention.
Sleep quality represents one of the most critical recovery processes through which working time exposures may be associated with occupational injury [15]. Sleep restores cognitive resources such as attention, reaction speed, and executive functioning that are depleted during intensive work. Prior studies have shown that poorer sleep quality is associated with higher occupational injury risk among public sector employees [16] and electronic industry workers [17]. Malik et al. further found that night-shift workers experienced poorer sleep quality, higher fatigue, and more injuries than day-shift workers [17]. Among delivery gig workers, opportunities for sufficient sleep may be constrained by unpredictable schedules and financial precarity [18]. To date, few studies have examined sleep quality as an indirect pathway linking working time exposures to occupational injury in this population [19,20].
Building on these gaps, the present study aims to examine the associations between three dimensions of working time exposures and occupational injury and to explore whether sleep quality represents an indirect pathway co-occurring with these associations (Fig. 1). Given the cross-sectional design, this analysis is framed as a decomposition of associational pathways rather than a test of causal mediation. We hypothesize that (H1) greater working time exposures are significantly associated with more occupational injuries; (H2) greater working time exposures are significantly associated with poorer sleep quality; and (H3) poorer sleep quality is significantly associated with more occupational injuries.
Fig. 1.
Conceptual model.
Note: ∗p < .05, ∗∗p < .01, ∗∗∗p < .001.
Dashed lines indicate insignificant paths.
2. Materials and methods
2.1. Participants and procedures
This cross-sectional study recruited 1,537 delivery gig workers in Shanghai, China, between October and December 2025. Participants were eligible if they were currently engaged in delivery gig work and were at least 18 years of age. We randomly selected 23 “couriers’ hubs” in Putuo and Minhang districts. Within each hub, participants were recruited through voluntary self-selection, whereby workers chose to scan a study QR code posted at the hub. Couriers’ hubs are shared rest facilities established for delivery gig workers to take breaks during their working hours, typically providing seating, water, and phone charging, and are commonly supported jointly by local governments, labor unions, and platform-related entities in urban China. The survey was administered via the “Wenjuanxing” mini-program accessible through WeChat (Changsha Ranxing Information Technology Co., Ltd., Changsha, Hunan Province, China). Each participant received compensation of 10 RMB (approximately 1.43 USD) upon completion. The survey required approximately 15 minutes to complete. All participants provided informed consent prior to participation and were informed of the study purpose, voluntary nature, and confidentiality protections. The 23 hubs were randomly selected from couriers’ hubs across the two districts, and within each hub, participants were recruited by posting the study QR code, which workers voluntarily scanned to complete the survey. Because recruitment relied on voluntary scanning at the hubs, a response or refusal rate could not be computed. The survey used required-response items, so there were no missing data, and all regression models were estimated on the full analytic sample (N = 1,537).
2.2. Measures
2.2.1. Occupational injury
Occupational injury was measured using a self-reported item asking participants in Chinese: “您在送外卖期间发生过几次工伤?” (“How many occupational injuries have you experienced during your delivery gig work?”). The Chinese the term “工伤” refers to any injury event sustained in the course of work and corresponds to the standard concept of occupational injury in the occupational health literature. Response options included five ordered categories: no injuries, 1–3 injuries, 4–6 injuries, 7–9 injuries, and 10 or more injuries. These categorical responses were converted to approximate count values using midpoint coding (0, 2, 5, 8, and 11, respectively), a common approach in occupational health research when exact counts are unavailable [21,22]. The resulting injury variable was treated as a count outcome. Although the item did not assess injury type, prior research with delivery gig workers indicates that the most commonly reported occupational injuries include traffic-related crashes, falls, collisions, and musculoskeletal strains [[7], [8], [9], [10]]. Because the item asked about injuries experienced across the entire period of delivery work, it captures cumulative self-reported injury experience rather than incidence within a fixed reference window and refers to any work-related injury without invoking formal recognition or compensation. As a robustness check, we re-estimated all models using the binary and original five-category ordinal injury variable with ordered logistic regression (Supplementary Table S1 and S2).
2.2.2. Working time exposures
Working time exposures were operationalized using three indicators reflecting distinct dimensions of work organization over time, drawing on international frameworks for analyzing working time and worker health [11,12]. Daily working hours were measured as a continuous variable (1–24 hours) based on the typical number of hours worked per day. Weekly working days were measured as a continuous variable (1–7 days). Night-shift frequency was measured using ordinal categories (never, 1–2 days, 3–4 days, or 5+ days per week) because the fluid and irregular nature of night work among delivery couriers makes exact daily counts unreliable. In regression models, night-shift frequency was treated as a continuous exposure (0–4) to capture dose–response relationships with progressively higher circadian disruption [23]. The three indicators were included simultaneously in the regression models because they capture distinct but inter-related dimensions of work organization: daily working hours reflect within-day effort intensity, weekly working days capture cumulative workload across days, and night-shift frequency represents circadian disruption. Including all three indicators in a single model also reduces the risk of omitted-variable bias, given that delivery gig workers who work long hours per day are often the same individuals who work many days per week or engage in night work. Respondents reported the total hours they typically spend working as delivery couriers. The measures did not separately distinguish active delivery time from waiting time between orders and referred specifically to delivery gig work rather than any other employment.
2.2.3. Sleep quality
Sleep quality was assessed using the Jenkins Sleep Scale (JSS-4), a widely used self-report measure of sleep problems in population-based and occupational health research [24]. The scale consists of four items assessing the frequency of common sleep difficulties during the past month, rated on a 6-point frequency scale (0 = “Never” to 5 = “Every day”). Item scores were summed to create a continuous index ranging from 0 to 20. All items were reverse-coded such that higher scores indicate better sleep quality. The scale demonstrated good internal consistency in the present study (Cronbach α = 0.88).
2.2.4. Covariates
Sociodemographic covariates included age, sex (0 = female, 1 = male), educational attainment, marital status, the number of children, household registration type, monthly income, and self-rated financial status (1 = very difficult to 10 = very comfortable). Job-related covariates included years of delivery gig work experience, whether having other income sources, type of gig worker (platform employed, crowdsourced, or other), contract type, and whether having social insurance. Platform-employed (dedicated) couriers are affiliated with a platform or its outsourcing partner and operate within assigned zones under stronger algorithmic scheduling, whereas crowdsourced couriers freely accept orders with greater flexibility. Very few Chinese riders hold a direct labor contract with the platform itself, so platform-employed here denotes the dedicated-delivery arrangement rather than direct employment. Type of gig worker and contract type therefore capture distinct constructs, work arrangement versus legal contractual form.
2.3. Statistical analysis
All analyses were conducted in Stata 18. Sleep quality was modeled using ordinary least squares regression, and occupational injury was modeled using negative binomial regression because of the overdispersed distribution of injury counts. All regression models adjusted for sociodemographic and job-related covariates. To examine the proposed indirect pathway framework, we used a regression-based decomposition approach with four models: (1) Model 1 regressed sleep quality on the three working time exposure indicators and covariates (a path); (2) Model 2 regressed injury counts on working time exposure indicators and covariates (total associations); (3) Model 3 regressed injury counts on sleep quality and covariates (b path); and (4) Model 4 regressed injury counts on both working time exposure indicators and sleep quality, with covariates (direct associations). Indirect associations through sleep quality were estimated using 1,000 bootstrap resamples, with 95% bias-corrected confidence intervals (CIs). Indirect associations were considered significant if the CI did not include zero. Because the data were cross-sectional, all pathways are interpreted as covariation rather than causal mediation. All statistical tests were two tailed at α = 0.05. As a robustness check, we re-estimated all models using a binary injury indicator (0 = no injuries, 1 = at least one injury) with logistic regression. Because injury was an over-dispersed count modeled with negative binomial regression, we used this bootstrap regression-based decomposition rather than a linear structural equation model. We further re-estimated the models treating injury as a rate per year of delivery experience using a tenure-based exposure offset (Supplementary Table S3). All robustness analyses are reported in Supplementary Tables S1–S3.
3. Results
3.1. Descriptive statistics
Table 1 presents the descriptive statistics of the study sample (N = 1,537). Participants had a mean age of 29.54 years (standard deviation [SD] = 6.81) and were predominantly male (90.94%). More than half were single (53.81%) and reported having no children (53.09%). Nonlocal rural household registration was most prevalent (56.34%), indicating that many participants were migrant workers from rural areas outside Shanghai. The mean monthly income was 8,137.14 RMB (SD = 2,596.50), and the mean self-rated financial status was 4.62 (SD = 2.47). Most participants (74.24%) had no other income sources, and about half (51.79%) were platform-employed couriers. A notable proportion (21.21%) reported not having signed a formal labor contract, and fewer than half (44.11%) reported social insurance coverage.
Table 1.
Descriptive statistics (N = 1,537)
| Variables | n | Mean or percentage | SD | Min | Max |
|---|---|---|---|---|---|
| Sociodemographic variables | |||||
| Age | 1,537 | 29.54 | 6.81 | 18 | 63 |
| Sex | |||||
| Female | 139 | 9.04% | |||
| Male | 1,398 | 90.94% | |||
| Education attainment | |||||
| Primary school or below | 27 | 1.76% | |||
| Junior high school | 336 | 21.86% | |||
| Vocational/technical school | 401 | 26.09% | |||
| Senior high school | 228 | 14.83% | |||
| Associate degree | 355 | 23.10% | |||
| Bachelor’s degree | 184 | 11.97% | |||
| Master’s degree or above | 6 | 0.39% | |||
| Marital status | |||||
| Single | 827 | 53.81% | |||
| Married | 643 | 41.83% | |||
| Other | 67 | 4.36% | |||
| Number of kids | |||||
| No kids | 816 | 53.09% | |||
| One | 410 | 26.68% | |||
| Two | 273 | 17.76% | |||
| Three or more | 38 | 2.47% | |||
| Household type | 1 | ||||
| Local urban | 197 | 12.82% | |||
| Local rural | 182 | 11.84% | |||
| Nonlocal urban | 292 | 19.00% | |||
| Nonlocal rural | 866 | 56.34% | |||
| Monthly income∗ | 1,537 | 8137.14 | 2,596.50 | 1 | 20,000 |
| Financial status | 1,537 | 4.62 | 2.47 | 1 | 10 |
| Job-related variables | |||||
| Number of years working as a gig driver | |||||
| Six months or below | 207 | 13.47% | |||
| Six months to one year | 260 | 16.92% | |||
| One to three years | 538 | 35.00% | |||
| Three to five years | 298 | 19.39% | |||
| Five years or above | 234 | 15.22% | |||
| Whether having other income sources | |||||
| No | 1,141 | 74.24% | |||
| Yes | 396 | 25.76% | |||
| Type of gig worker | |||||
| Platform-employed couriers | 796 | 51.79% | |||
| Crowdsourced couriers | 385 | 25.05% | |||
| Other | 356 | 23.16% | |||
| Contract type | |||||
| No contract signed | 326 | 21.21% | |||
| Standard labor contract | 509 | 33.12% | |||
| Labor service contract | 435 | 28.30% | |||
| Independent contractor agreement | 267 | 17.37% | |||
| Social insurance† | |||||
| No | 678 | 44.11% | |||
| Yes | 859 | 55.89% | |||
| Average daily working hours | 1,537 | 10.63 | 2.14 | 1 | 24 |
| Average weekly working days | 1,537 | 6.53 | 0.74 | 1 | 7 |
| Average weekly night-shift days‡ | |||||
| None | 402 | 26.15% | |||
| One to two days | 515 | 33.51% | |||
| Three to four days | 290 | 18.87% | |||
| Five days or more | 330 | 21.47% | |||
| Number of occupational injuries experienced§ | |||||
| None | 781 | 50.81% | |||
| One to three times | 541 | 35.20% | |||
| Four to six times | 170 | 11.06% | |||
| Seven to nine times | 36 | 2.34% | |||
| Ten times or above | 9 | 0.59% | |||
| Sleep quality | 1,537 | 14.25 | 4.44 | 0 | 20 |
SD, standard deviation.
Note: ∗ Monthly income is measured in RMB.
Insurance measures whether the participant is paying for the social insurance, whether by himself or by the employer.
The average night-shift frequency per week is measured as an ordinal variable reflecting the number of nights worked per week and was treated as a continuous exposure in data analyses.
For analytical purposes, occupational injury counts were approximated using midpoint values (0, 2, 5, 8, and 11) corresponding to the response categories no injuries, 1–3, 4–6, 7–9, and ≥10 injuries.
Regarding working time exposures, participants worked an average of 10.63 hours per day (SD = 2.14) and 6.53 days per week (SD = 0.74). Night-shift work was common, with only 26.15% reporting no night shifts. About half (49.19%) reported having experienced occupational injuries during their delivery gig work. Sleep quality scores ranged from 0 to 20, with a mean of 14.25 (SD = 4.44). These intensities are high but broadly consistent with evidence that full-time Chinese delivery riders commonly work well beyond standard statutory hours [34]. Because recruitment occurred at couriers’ hubs used during working hours, the sample likely over-represents active, full-time riders, and the findings may generalize best to this more intensive subgroup.
3.2. Cross-model regression analyses
Results from the regression-based indirect pathway analyses are shown in Table 2. In Model 1 (ordinary least squares), more weekly working days (β = −0.672, Standard Error = 0.155, p < 0.001) and higher night-shift frequency (β = −0.558, Standard Error = 0.104, p < 0.001) were both significantly associated with poorer sleep quality, while daily working hours were not. In Model 2 (negative binomial; total associations), daily working hours (β = 0.058, SE = 0.019, p < 0.01) and night-shift frequency (β = 0.230, SE = 0.036, p < 0.001) were positively associated with injury counts, whereas weekly working days were not significantly associated. In Model 3, better sleep quality was significantly associated with fewer injuries (β = −0.106, SE = 0.008, p < 0.001); Fig. 2 visualizes this relationship. In Model 4, which included both working time exposure indicators and sleep quality, sleep quality remained significantly associated with lower injury counts (β = −0.096, SE = 0.009, p < 0.001); the associations of daily working hours (β = 0.054, SE = 0.018, p < 0.01) and night-shift frequency (β = 0.157, SE = 0.034, p < 0.001) with injury also remained significant, whereas weekly working days were not directly associated with injury counts after adjusting for sleep quality.
Table 2.
Cross-model regression results
| Model 1 |
Model 2 |
Model 3 |
Model 4 |
|
|---|---|---|---|---|
| β (SE) | β (SE) | β (SE) | β (SE) | |
| Working hours per day | −0.063 (0.058) | 0.058∗∗ (0.019) | 0.054∗∗ (0.018) | |
| Working days per week | −0.672∗∗∗ (0.155) | 0.088 (0.05) | 0.02 (0.049) | |
| Night-shift days per week | −0.558∗∗∗ (0.104) | 0.23∗∗∗ (0.036) | 0.157∗∗∗ (0.034) | |
| Sleep quality | −0.106∗∗∗ (0.008) | −0.096∗∗∗ (0.009) | ||
| Sex | 0.657 (0.396) | 0.017 (0.13) | 0.14 (0.124) | 0.084 (0.123) |
| Age | 0.017 (0.022) | −0.02∗∗ (0.007) | −0.027∗∗∗ (0.007) | −0.024∗∗∗ (0.007) |
| Education attainment | −0.294∗∗∗ (0.089) | 0.082∗∗ (0.029) | 0.042 (0.028) | 0.05 (0.028) |
| Marital status | ||||
| Single | Ref. | Ref. | Ref. | Ref. |
| Married | −0.329 (0.378) | 0.032 (0.122) | −0.007 (0.118) | 0.035 (0.116) |
| Other | −0.756 (0.621) | 0.054 (0.212) | −0.031 (0.204) | 0.012 (0.201) |
| Number of kids | 0.271 (0.21) | 0.057 (0.068) | 0.119 (0.065) | 0.082 (0.064) |
| Monthly income | 0.529∗∗∗ (0.124) | −0.077 (0.04) | 0.04 (0.037) | −0.002 (0.038) |
| Financial status | 0.303∗∗∗ (0.05) | −0.008 (0.017) | 0.002 (0.017) | 0.016 (0.016) |
| Household type (ref.) | ||||
| Local urban | Ref. | Ref. | Ref. | Ref. |
| Local rural | −0.856 (0.447) | 0.043 (0.141) | −0.082 (0.134) | −0.071 (0.132) |
| Nonlocal urban | −0.534 (0.41) | −0.163 (0.131) | −0.224 (0.125) | −0.245∗ (0.123) |
| Nonlocal rural | −0.562 (0.383) | −0.348∗∗ (0.123) | −0.399∗∗∗ (0.117) | −0.416∗∗∗ (0.116) |
| Years working as a gig driver | −0.401∗∗∗ (0.103) | 0.151∗∗∗ (0.036) | 0.151∗∗∗ (0.035) | 0.132∗∗∗ (0.034) |
| Whether having other income sources | ||||
| No | Ref. | Ref. | Ref. | Ref. |
| Yes | −0.433 (0.267) | −0.064 (0.089) | −0.126 (0.085) | −0.121 (0.084) |
| Type of gig worker | ||||
| Platform employed | Ref. | Ref. | Ref. | Ref. |
| Crowdsourced | −0.197 (0.29) | −0.084 (0.098) | −0.247∗∗ (0.094) | −0.148 (0.094) |
| Other | −0.41 (0.283) | 0.183∗ (0.093) | 0.072 (0.089) | 0.108 (0.088) |
| Contract type | ||||
| No contract signed | Ref. | Ref. | Ref. | Ref. |
| Standard labor contract | 0.395 (0.376) | 0.273∗ (0.126) | 0.242∗ (0.12) | 0.308∗ (0.12) |
| Labor service contract | 0.242 (0.328) | 0.196 (0.113) | 0.104 (0.108) | 0.174 (0.109) |
| Independent contractor | 0.14 (0.362) | 0.096 (0.125) | 0.119 (0.12) | 0.129 (0.119) |
| Social insurance | ||||
| No | Ref. | Ref. | Ref. | Ref. |
| Yes | −0.088 (0.28) | 0.295∗∗ (0.095) | 0.233∗ (0.091) | 0.262∗∗ (0.09) |
Note: ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05. SE = Standard Error. Model 1 is the OLS regression on sleep quality with three main predictors and covariates. Model 2 is the negative binomial regression on the number of occupational injuries with three main predictors and covariates. Model 3 is the negative binomial regression on the number of occupational injuries with sleep quality and covariates. Model 4 is the negative binomial regression on the number of occupational injuries with three main predictors, sleep quality, and covariates.
OLS, ordinary least squares.
Fig. 2.
Predicted occupational injury counts across levels of sleep quality.
Note: Predictions are based on negative binomial regression models, holding other variables at their mean values.
3.3. Bootstrap indirect pathway analysis
Table 3 presents the bootstrap estimates of the indirect associations through sleep quality. Weekly working days showed a significant indirect association with occupational injury through sleep quality [indirect estimate = 0.064, 95% CI: (0.035, 0.094)], as did night-shift frequency [indirect estimate = 0.053, 95% CI: (0.031, 0.076)]. The indirect association of daily working hours through sleep quality was not statistically significant [indirect estimate = 0.006, 95% CI: (−0.005, 0.017)]. Sleep quality therefore represented a significant indirect pathway for weekly working days and night-shift frequency but not for daily working hours.
Table 3.
Bootstrap indirect effects of work intensity on occupational injuries via sleep quality
| Indirect effect | Boot SE | Boot LLCI | Boot ULCI | |
|---|---|---|---|---|
| Working hours per day | 0.006 | 0.006 | −0.005 | 0.017 |
| Working days per week | 0.064∗∗∗ | 0.015 | 0.035 | 0.094 |
| Night-shift days per week | 0.053∗∗∗ | 0.012 | 0.031 | 0.076 |
Note: ∗p < .05, ∗∗p < .01, ∗∗∗p < .001. SE = Standard error. LLCI = Lower level confidence interval. ULCL = Upper level confidence interval.
3.4. Robustness check
Re-estimating all models using a binary injury indicator yielded consistent results. In Model 1, weekly working days (β = −0.672, SE = 0.155, p < 0.001) and night-shift frequency (β = −0.558, SE = 0.104, p < 0.001) remained associated with poorer sleep quality. In Model 2, daily working hours (β = 0.129, SE = 0.030, p < 0.001) and night-shift frequency (β = 0.253, SE = 0.052, p < 0.001) remained significantly associated with injury status. Sleep quality remained significantly associated with injury status in Model 3 (β = −0.168, SE = 0.014, p < 0.001) and in Model 4 (β = −0.162, SE = 0.015, p < 0.001), alongside daily working hours (β = 0.132, SE = 0.031, p < 0.001) and night-shift frequency (β = 0.189, SE = 0.055, p < 0.001). Bootstrap analyses also confirmed significant indirect associations through sleep quality for weekly working days [indirect estimate = 0.109, 95% CI: (0.058, 0.160)) and night-shift frequency [indirect estimate = 0.091, 95% CI: (0.050, 0.131)] but not for daily working hours [indirect estimate = 0.010, 95% CI: (−0.009, 0.029)])
4. Discussion
This study examined how multiple dimensions of working time exposures are associated with occupational injury among delivery gig workers in urban China and explored whether sleep quality represents an indirect pathway co-occurring with these associations. Longer daily working hours and more frequent night shifts were directly associated with higher occupational injury risk, while weekly working days and night-shift frequency showed significant indirect associations with injury through poorer sleep quality. The findings suggest that working time exposures form a multidimensional set of conditions with distinct associational pathways to injury risk. Specifically, daily working hours showed a direct association with injury that was not mediated by sleep quality, weekly working days were associated with injury entirely indirectly through poorer sleep quality, and night-shift frequency was associated both directly and indirectly. Although demonstrated among delivery couriers, these fatigue- and recovery-related pathways may extend to other transport and platform-work groups, even as their intensity is heightened by the traffic-intensive, algorithmically managed nature of delivery work.
4.1. Working time exposures and occupational injury
The direct associations between longer daily working hours, higher night-shift frequency, and occupational injury are consistent with the existing literature. Analyses of nationally representative U.S. longitudinal data have shown a dose–response pattern in which working very long hours (e.g., ≥12 hours/day) is associated with significantly higher hazards of occupational injury and illness [25]. Similar patterns have been observed among shift workers in hospital settings [26] and among food delivery drivers in Bangkok [27]. These findings are also consistent with the Job Demands–Resources Model [28], which posits that excessive job demands deplete workers’ cognitive resources and increase strain when recovery is inadequate. In traffic-intensive delivery work, this strain may translate into impaired alertness and reduced safety margins, elevating injury risk. The absence of a significant direct association between weekly working days and injury, after controlling for daily working hours and night-shift frequency, suggests that compressed workweeks operate primarily through fatigue accumulation and impaired recovery rather than through immediate within-day exposure [29].
4.2. Sleep quality as an indirect pathway
Sleep quality showed significant indirect associations linking weekly working days and night-shift frequency to occupational injury. This pattern aligns with prior longitudinal evidence that long working hours predict subsequent sleep problems [30,31] and with meta-analytic findings that sleep problems are associated with a significantly higher risk of occupational injury [32]. From the Effort–Recovery perspective [14], frequent night shifts and dense workweeks constrain opportunities for recovery, leading to sustained sleep impairment and cumulative fatigue. Lack of recovery undermines alertness, reaction speed, and hazard perception in traffic-intensive delivery work, thereby increasing susceptibility to injury. Notably, longer daily working hours were not significantly associated with injury through sleep quality, suggesting that within-day exposure may operate through other channels, such as longer road exposure time or physical fatigue from prolonged driving, rather than through disrupted sleep recovery.
5. Practical and policy implications
These findings highlight the need for interventions that address work scheduling and recovery opportunities rather than focusing solely on individual risk behaviors. Platform companies could consider limiting excessive night-shift frequency, discouraging overly compressed workweeks, and promoting minimum rest periods through algorithmic scheduling. Interventions aimed at improving sleep quality, such as psychoeducation on sleep health, fatigue management programs, and scheduling practices that reduce circadian disruption, may be particularly effective in mitigating injury risk. From a policy perspective, delivery gig workers represent a vulnerable transport workforce often disproportionately composed of young, male, rural-to-urban migrants with limited educational attainment and restricted access to formal labor protections. Policymakers could consider extending occupational health and safety standards to delivery gig work, including regulations on maximum working hours, night-shift load, and access to social protections that support recovery, regardless of employment classification. Comparable policy developments are underway in China: the Ministry of Human Resources and Social Security issued guidance on protecting workers in new forms of employment in 2021 and, since 2022, has piloted occupational-injury insurance for new-employment-form workers such as delivery riders, expanding it toward nationwide coverage, while major platforms have introduced fatigue management systems that suspend order assignment after prolonged continuous work [33,34]. The counterintuitive positive association between social insurance coverage and injury counts may reflect that more intensively exposed full-time riders are both more likely to be insured and more exposed to injury, greater recognition and recall of injuries among insured workers, or adverse selection and is consistent with these schemes prioritizing higher-risk workers.
5.1. Limitations
Several limitations should be acknowledged. First, the cross-sectional design precludes causal inference, and all pathways are interpreted as associational rather than causal. Longitudinal or experimental designs are needed to establish temporal ordering. In addition, the injury outcome is cumulative over the entire period of delivery work, whereas sleep quality was assessed over the past month. This temporal mismatch means a reverse association cannot be excluded, and the path-specific results should be read as a decomposition of cross-sectional covariation rather than directional effects. Second, all measures were self-reported and may be subject to recall bias, particularly for occupational injury and sleep quality. Third, the occupational injury item assessed the overall number of injuries but did not capture injury type or mechanism (e.g., traffic crash, fall, musculoskeletal strain), so we could not examine whether the observed associations varied across injury types. Fourth, the sample was drawn from delivery gig workers in Shanghai, which may limit generalizability to other national or cultural contexts where platform governance, labor regulations, and traffic environments differ. Finally, other unmeasured factors, such as weather conditions or real-time algorithmic pressures, may also contribute to injury risk and warrant further investigation. Relatedly, because the injury measure is cumulative over the period of delivery work, it scales partly with accumulated exposure; although years of delivery work were adjusted for and a rate-based sensitivity analysis was conducted (Supplementary Table S3), accumulated road exposure, delivery volume, vehicle type, and survivor effects could not be fully separated. The survey also did not capture other dimensions of platform work organization, such as delivery volume, distance, vehicle type, platform penalties, customer ratings, dispatch pressure, rest breaks, or protective equipment use, so the analysis reflects working-time exposures rather than the broader organization of platform delivery work. Moreover, Shanghai is a high-income, high-density megacity with a comparatively advanced local social insurance environment and a large rural-to-urban migrant workforce, features that may further shape the magnitude of the observed associations.
CRediT authorship contribution statement
Hao Wen: Writing – review & editing, Writing – original draft, Visualization, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.
Ethics statement
This study was approved by the Institutional Review Board of Columbia University (Study ID: AAAV3154) before the research was conducted.
Data availability statement
Data will be available upon reasonable request to the corresponding author.
Funding
This study is funded by the Summer 2024 Sasakawa Young Leaders Fellowship Fund Grant.
Conflicts of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Footnotes
Supplementary data to this article can be found online at https://doi.org/10.1016/j.shaw.2026.07.001.
Appendix A. Supplementary data
The following are the Supplementary data to this article:
References
- 1.Kuhn K.M. The rise of the “gig economy” and implications for understanding work and workers. Ind Organ Psychol. 2016;9(1):157–162. [Google Scholar]
- 2.Tran M., Sokas R.K. The gig economy and contingent work: an occupational health assessment. J Occup Environ Med. 2017;59(4):e63–e66. doi: 10.1097/JOM.0000000000000977. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Keith M.G., Long A.C., Harms P.D. In: Eby L.T., Allen T.D., editors. vol. 18. Emerald Publishing; 2020. Worker health and well-being in the gig economy: a proposed framework and research agenda; pp. 1–33. (Entrepreneurial and small business stressors, experienced stress, and well-being). [Google Scholar]
- 4.World Bank Group . World Bank; Washington (DC): 2023. The global gig economy: assessing the scope, growth, and economic impact. [Google Scholar]
- 5.Grand View Research . Grand View Research; 2024. Online food delivery market size, share & trends analysis report. [Google Scholar]
- 6.Fu H., editor. Temporary and gig economy workers in China and Japan: the culture of unequal work. Oxford University Press; 2023. [Google Scholar]
- 7.Laskaris Z., Hussein M., Stimpson J.P., Vignola E.F., Shahn Z., Cohen N., Baron S. A price too high: injury and assault among delivery gig workers in New York City. J Urban Health. 2024;101(3):439–450. doi: 10.1007/s11524-024-00873-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Kwangsukstith S., Surawattanasakul V., Mahakkanukrauh C., Panumasvivat J., Sirikul W., Kitro A., Siviroj P. Major injuries and associated factors in traffic accidents among motorcycle food delivery riders during the COVID-19 pandemic in Thailand. Heliyon. 2024 Oct 30;10(20) doi: 10.1016/j.heliyon.2024.e39032. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Abdullah M.S., Azmin A.A., Kassim M.A.M. Risky riding behaviours among p-Hailing riders in Malaysia: implications for business operations and workforce safety in the gig economy. Int J Bus Technopreneur. 2025;15(1):51–68. [Google Scholar]
- 10.Subramaniam S., Borhan M.N., Nguyen-Phuoc D.Q., Ibrahim A.N.H. Exploring determinants of risk-riding behaviour among food delivery riders: a systematic literature review. IEEE Access. 2025;13:1–20. [Google Scholar]
- 11.Eurofound . Publications Office of the European Union; Luxembourg: 2017. Sixth European working conditions survey – overview report (2017 update) [Google Scholar]
- 12.Tucker P., Folkard S. International Labour Office; Geneva: 2012. Working time, health and safety: a research synthesis paper. Conditions of Work and Employment Series No. 31. [Google Scholar]
- 13.Wibowo S.G., Susanto F. Legal protection for gig workers in the digital economy: a critical review of labor relations in Indonesia. Easta J Law Hum Rights. 2025;3(3):231–240. [Google Scholar]
- 14.Meijman T.F., Mulder G. A handbook of work and organizational psychology. Psychology Press; 2013. Psychological aspects of workload; pp. 5–33. [Google Scholar]
- 15.Hyndych A., El-Abassi R., Mader Jr EC. The role of sleep and the effects of sleep loss on cognitive, affective, and behavioral processes. Cureus. 2025;17(5) doi: 10.7759/cureus.84232. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Salminen S., Oksanen T., Vahtera J., Sallinen M., Härmä M., Salo P., Virtanen M., Kivimäki M. Sleep disturbances as a predictor of occupational injuries among public sector workers. J Sleep Res. 2010 Mar;19(1p2):207–213. doi: 10.1111/j.1365-2869.2009.00780.x. [DOI] [PubMed] [Google Scholar]
- 17.Malik A., Ali A., Akram M., Rasheed R. Assessment of sleep quality, fatigue and its association with occupational injuries among shift workers in an electronic industrial environment. Int J Occup Saf Ergon. 2025;31(1):69–76. doi: 10.1080/10803548.2024.2404326. [DOI] [PubMed] [Google Scholar]
- 18.Crain T.L., Brossoit R.M., Robles-Saenz F., Tran M. Fighting fatigue: a conceptual model of driver sleep in the gig economy. Sleep Health. 2020;6(3):358–365. doi: 10.1016/j.sleh.2020.02.004. [DOI] [PubMed] [Google Scholar]
- 19.Hafeez S., Gupta C., Sprajcer M. Stress and the gig economy: it’s not all shifts and giggles. Ind Health. 2022;61(2):140–150. doi: 10.2486/indhealth.2021-0217. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Margerison C.E., Joachim G.E., Michling T.R., Johnson C.Y. Platform-mediated gig work and health: a scoping review. Am J Ind Med. 2025;68(9):729–739. doi: 10.1002/ajim.70007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Heitjan D.F., Rubin D.B. Ignorability and coarse data. Ann Stat. 1991;19(4):2244–2253. [Google Scholar]
- 22.Shardell M., Scharfstein D.O., Vlahov D., Galai N. Inference for cumulative incidence functions with informatively coarsened discrete event-time data. Stat Med. 2008;27(28):5861–5879. doi: 10.1002/sim.3397. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Akerstedt T. Shift work and disturbed sleep/wakefulness. Occup Med (Lond) 2003;53(2):89–94. doi: 10.1093/occmed/kqg046. [DOI] [PubMed] [Google Scholar]
- 24.Jenkins C.D., Stanton B.A., Niemcryk S.J., Rose R.M. A scale for the estimation of sleep problems in clinical research. J Clin Epidemiol. 1988;41(4):313–321. doi: 10.1016/0895-4356(88)90138-2. [DOI] [PubMed] [Google Scholar]
- 25.Dembe A.E., Erickson J.B., Delbos R.G., Banks S.M. The impact of overtime and long work hours on occupational injuries and illnesses: new evidence from the United States. Occup Environ Med. 2005;62(9):588–597. doi: 10.1136/oem.2004.016667. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Horwitz I.B., McCall B.P. The impact of shift work on the risk and severity of injuries for hospital employees: an analysis using Oregon workers’ compensation data. Occup Med (Lond) 2004;54(8):556–563. doi: 10.1093/occmed/kqh093. [DOI] [PubMed] [Google Scholar]
- 27.Prakobkarn P., Luangwilai T., Prempree P., Kunno J. Prevalence of motorcycle accidents among food delivery drivers and its relation to knowledge, attitudes, and practices in urban areas in Bangkok, Thailand. PLoS One. 2024;19(5) doi: 10.1371/journal.pone.0303310. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Bakker A.B., Demerouti E. The job demands–resources model: state of the art. J Manag Psychol. 2007;22(3):309–328. [Google Scholar]
- 29.Folkard S., Lombardi D.A. Modeling the impact of the components of long work hours on injuries and “accidents”. Am J Ind Med. 2006;49(11):953–963. doi: 10.1002/ajim.20307. [DOI] [PubMed] [Google Scholar]
- 30.Virtanen M., Ferrie J.E., Gimeno D., Vahtera J., Elovainio M., Singh-Manoux A., Marmot M.G., Kivimäki M. Long working hours and sleep disturbances: the Whitehall II prospective cohort study. Sleep. 2009 Jun 1;32(6):737–745. doi: 10.1093/sleep/32.6.737. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Qiu D., Li Y., Li R., He J., Ouyang F., Luo D., Xiao S. Long working hours, work-related stressors and sleep disturbances among Chinese government employees: A large population-based follow-up study. Sleep Med. 2022 Aug 1;96:79–86. doi: 10.1016/j.sleep.2022.05.005. [DOI] [PubMed] [Google Scholar]
- 32.Uehli K, Mehta AJ, Miedinger D, Hug K, Schindler C, Holsboer-Trachsler E, Leuppi JD, Künzli N. Sleep problems and work injuries: a systematic review and meta-analysis. Sleep Med Rev. 2014;18(1):61–73. doi: 10.1016/j.smrv.2013.01.004. [DOI] [PubMed] [Google Scholar]
- 33.Ran X., Zhao Y. Behind the flexibility: insufficient occupational injury protection of gig workers in China. Front Public Health. 2023;11 doi: 10.3389/fpubh.2023.1117180. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.International Labour Organization . International Labour Office; Geneva: 2024. New forms of employment and labour protection in China. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data will be available upon reasonable request to the corresponding author.