A cross-sectional correlational design was utilized in this study. The survey was designed and administered using the Wenjuanxing Platform (https://www.wjx.cn/app/survey.aspx) and was conducted between August and December 2022 in Shenzhen and Guangzhou, two major metropolitan cities in southern China. Shenzhen is a special economic zone with a population of 17.66 million and a GDP of ¥3.24 trillion, while Guangzhou is the capital city of Guangdong province with a population of 18.73 million and a GDP of ¥2.88 trillion.

A combined street-intercept (convenience) and snowball sampling strategy was used. To assist with data collection, ten research assistants were selected from psychology students and received a half-day survey training. They conducted street-intercept recruitment in five main urban districts of Shenzhen and Guangzhou, targeting public locations where platform delivery riders aggregate (e.g., commercial centers, residential complexes, restaurant clusters, and rider waiting points outside merchants). Data were collected at varied time slots throughout the day and across weekdays and weekends to capture temporal variation in rider availability. At each site, research assistants approached riders during non-peak waiting periods, briefly explained the study purpose and content, and showed the rider a printed card displaying the survey QR code. Riders who agreed to participate scanned the QR code with their smartphones and completed the questionnaire on-site (approximately 15 min); upon completion, they received a small gift (worth approximately 15 RMB—riders could choose two items from a raincoat, a smartphone waterproof case, and a sports towel) or an equivalent monetary compensation of 15 RMB delivered via WeChat transfer. To extend reach to riders who were difficult to access on-site, snowball sampling was additionally used. Participating riders were invited to forward the survey QR code to their colleagues and riders’ work WeChat groups. Riders who shared the QR code via WeChat and completed the questionnaire online received the equivalent monetary compensation through WeChat transfer. The first page of the online questionnaire presented the research objectives, the voluntary nature of participation, confidentiality and anonymity assurances, and the principal investigator’s contact information. Participants were required to actively check an “I agree to participate” option before being permitted to proceed to the survey items. This study received ethical approval from the Ethics Committee of Shenzhen University (Approval No. 202200044). All the participants provided informed consent. All procedures were performed in accordance with relevant guidelines.

To ensure data quality and guard against fraudulent responses, multi-layered quality control measures were implemented. At the platform level, the Wenjuanxing system enforced mandatory responses on every item, built-in range checks on numerical inputs, and IP address tracking to detect and block duplicate submissions from the same device or network. Three attention/lie-detection items (e.g., “Please select ‘strongly disagree’ for this item”) were embedded throughout the questionnaire, together with cross-item logical consistency checks. For identity screening, on-site research assistants approached only individuals who could be reliably identified as active platform delivery riders based on contextual cues (e.g., presence at rider waiting points, branded delivery equipment), and confirmed rider status through a screening question before sharing the QR code. For online snowball-recruited respondents specifically, an additional manual verification step was implemented before disbursing the monetary compensation: the research team manually reviewed each submitted response and verified the participant’s identity as an active platform delivery rider by requesting a screenshot of their most recent delivery record from the platform app (with customer and order details redacted to protect privacy). Monetary compensation was disbursed only to respondents who passed both the identity verification and the response-quality screening.

A total of 1976 participants consented to participate in the survey, and 1949 of them completed the survey (959 in Shenzhen and 990 in Guangzhou, completion rate among consenters = 98.6%). Given that on-site riders who declined participation due to time constraints could not be reliably tracked, a precise response rate based on initial contacts could not be computed. It was applied the following predefined criteria to identify and exclude invalid responses: (a) completion time below 250 s (the lower 5th percentile), considered unrealistically short for thorough reading; (b) straight-line or patterned responding (e.g., selecting the same option for all items in a scale); (c) failed embedded attention/lie-detection items; and (d) cross-item logical inconsistencies. After data cleaning, 1879 valid questionnaires were retained for analysis (valid retention rate = 96.4%). Participants’ platform affiliations were assessed by a multi-select question covering 22 platforms; some riders worked across multiple platforms simultaneously. The sample was predominantly composed of food-delivery riders—73.0% were affiliated with Meituan and 31.2% with Ele.me (multiple selections allowed)—with the remainder coming from fresh-grocery and on-demand errand platforms, including Hema (4.5%), Dingdong Maicai (3.8%), Shansong (3.8%), Pupu Maicai (3.2%), JD Daojia (2.7%), and Paotui errand services (2.2%), among others. Conventional parcel-courier workers were not included, as their employment structure, dispatch logic, and labor-relations status differ fundamentally from platform-mediated on-demand delivery.

Independent-samples t-tests and one-way analysis of variance (ANOVA) were used to examine variations in uncertainty stress across basic demographic groups, personal lifestyles, and job-related characteristics. Pearson correlation analyses were conducted to explore bivariate associations between uncertainty stress and the continuous psychosocial job-demand variables (work conflict, work pace, job insecurity). Subsequently, hierarchical multiple linear regression was then employed to identify independent predictors of uncertainty stress.

The model-building followed a three-stage variable selection process. First, guided by the Job Demands-Resources (JD-R) framework, three substantive predictor blocks were theoretically specified. Second, within each theoretical block, only variables that showed a statistically significant association with uncertainty stress in univariate analyses (t-tests/ANOVA) or bivariate correlations were retained for entry into the multivariate model to ensure model parsimony. Third, demographic variables (e.g., marital status) that reached univariate significance were additionally entered as covariates to adjust for potential confounding. Accordingly, the retained variables were entered in four hierarchical blocks: Model 1 (base model) included the demographic covariate (the marital status dummy variables); Model 2 added structural job characteristics (monthly delivery income and working years); Model 3 added the personal lifestyle block (smoking and regular exercise); Model 4 (full model) further added the psychosocial job-demands block (work conflict, work pace, and job insecurity). Marital status was dummy-coded with “unmarried” as the reference category (Married_d:1 = Married, 0 = otherwise; Divorced/widowed_d:1 = Divorced or widowed, 0 = otherwise). Monthly delivery income and working years were entered as ordinal variables coded in ascending order; this specification assumes an approximately linear trend in uncertainty stress across ordered categories. Smoking and regular exercise were entered as dichotomous variables (0 = No, 1 = Yes).

The standardized coefficients (β) were used to assess the relative strength of these predictors. Regression assumptions were evaluated for the full model: normality of residuals was assessed via histogram and normal P-P plot inspection; homoscedasticity was evaluated by plotting standardized residuals against standardized predicted values; and multicollinearity was examined using Variance Inflation Factor (VIF) values. No serious violations were detected. All analyses were conducted in SPSS Statistics 26.0 (IBM Corp., Armonk, NY, USA).

Divorced or widowed riders reported higher levels of uncertainty stress than married or unmarried riders. The loss of a long-term partner typically removes a primary source of emotional, financial, and social support [33], and the resulting loneliness and insecurity may contribute to heightened uncertainty stress.

Lower monthly delivery income was associated with higher uncertainty stress, consistent with prior findings linking lower family income to severe uncertainty stress in college students [34]. Higher income may buffer against uncertainty by enhancing economic security and access to social resources, whereas lower-income workers are more exposed to market fluctuations and employment instability [35], both of which can amplify perceived uncertainty and stress.

Working years was negatively associated with uncertainty stress in the full model, suggesting that experience may buffer against perceived uncertainty. As riders accumulate experience, they develop familiarity with platform algorithms, delivery routes, and customer interactions, which can reduce perceived unpredictability [36]. The non-significant descriptive fluctuation at >4 years warrants further longitudinal investigation, since prolonged exposure to precarious work may eventually erode this protective effect [37]. Overall, Hypothesis 1 was only partially supported: among structural job characteristics, only lower monthly income and shorter working years were significantly associated with higher uncertainty stress, whereas platform type, employment type, social insurance, and daily working hours were not.

Smoking and regular exercise were both associated with uncertainty stress in univariate analyses, but only regular exercise remained significant in multivariate models—riders who exercised regularly reported lower uncertainty stress. This finding aligns with prior evidence that physical activity buffers stress [38,39] through both physiological mechanisms (e.g., reduced cortisol reactivity [40]) and psychological mechanisms (e.g., enhanced self-efficacy and perceived control [41]), which are particularly relevant to coping with unpredictable work environments. Exercise has also been shown to reduce anxiety sensitivity [42,43], a construct closely linked to intolerance of uncertainty [44,45]. Hypothesis 3 was therefore partially supported: regular exercise was independently protective, whereas smoking and alcohol use were not independently associated with uncertainty stress in the multivariate model.

Work conflict, fast work pace, and job insecurity were each independently associated with uncertainty stress. Job insecurity is widely conceptualized as a subjectively experienced stressor [46], and platform delivery riders—often working under temporary or contract-based arrangements [2]—face acute insecurity about job continuity, valued job features, and career prospects, which lies at the core of uncertainty perceptions. A fast work pace likely compounds these effects, as constant time pressure and worry about meeting delivery deadlines fuel anxiety and a sense of loss of control.

Work conflict emerged as the strongest psychosocial correlate of uncertainty stress. Customer incivility—through negative feedback, verbal aggression, or unfair complaints—can trigger immediate algorithmic penalties and damage to platform ratings, threatening riders’ social-evaluative standing and economic stability in unpredictable ways [47]. By contrast, a fast work pace, although demanding, may become routinized over time as riders develop time-management strategies. The unpredictable and interpersonal nature of work conflict likely poses a greater challenge to one’s sense of control, a core element of uncertainty stress [48].

The structure of platform work itself may amplify these stressors. Opaque algorithmic management—dynamic order assignment, punitive systems for delays or low ratings—creates an environment of low controllability and high unpredictability [47]. Capricious customer demands compounded by inflexible platform rules may trap riders in a cycle of uncertainty, where performance and livelihood depend on factors beyond their control. Taken together, Hypothesis 2 was fully supported: work conflict, work pace, and job insecurity were all independently and positively associated with uncertainty stress, with work conflict emerging as the strongest correlate.

Several potential implications for mitigating uncertainty stress among platform delivery riders are discussed below. Because the present study used a cross-sectional correlational design, the associations identified above do not support causal inference. The suggestions that follow should therefore be interpreted as hypothesis-generating directions rather than evidence-based policy prescriptions, and they warrant rigorous testing in longitudinal or intervention research before adoption. First, at the policy and regulatory level, our finding that lower income and higher job insecurity were each associated with greater uncertainty stress points to possible directions for strengthening labor protections for gig workers, such as establishing frameworks for minimum earnings guarantees and mandating access to social insurance schemes (e.g., for injury or unemployment) to reduce financial precarity [49]. Clarifying riders’ employment status may be a critical step in this direction.

Secondly, at the platform and organizational level, our finding that work conflict was the strongest correlate of uncertainty stress points to a potential role for platforms to reform their management systems to enhance transparency and fairness. This may include creating transparent and appealing rating systems to protect riders from unfair customer feedback, and designing algorithms that incorporate realistic delivery timelines and adequate rest periods [50]. Platforms could also consider instituting training programs for customers to promote respectful interactions and provide riders with de-escalation training for handling conflicts.

Thirdly, given the observed protective association between regular exercise and lower uncertainty stress, it may be beneficial for platforms and rider associations to actively promote physical well-being. This could involve subsidized gym memberships, organized sports events, or digital health challenges. Furthermore, mental well-being programs, such as stress management workshops and access to confidential counseling services, tailored to the flexible schedules of gig workers, may also warrant consideration.

Several limitations should be noted. First, this study was a cross-sectional survey, which inherently precludes the establishment of causal relationships. All reported associations should be interpreted as correlational, and the directionality of relationships cannot be determined from the present data. In the future study, a longitudinal study for exploring how work conflict, work pace, and job insecurity are associated with sustained uncertainty stress over time should be conducted. Second, it is important to acknowledge the potential presence of selection bias. The street-intercept and snowball sampling strategy may overrepresent riders active in high-traffic urban locations and those with stronger social-network connections (e.g., active in rider WeChat groups), while underrepresenting riders in suburban or peri-urban areas or those less digitally connected. Additionally, riders who declined participation due to time pressures during peak delivery hours may systematically differ in workload and stress levels from those who participated. However, our sampling strategy covered diverse locations across five major urban districts in each city, including streets and communities in commercial, residential, and mixed-use areas where delivery riders are typically active. Data were collected at various time slots throughout the day to account for temporal variations. While conducting research on delivery riders presents significant logistical challenges, the total sample size of this study is substantial compared to prior research in this field. We believe these efforts enhance the representativeness of our findings, despite potential limitations in sampling bias. Third, it is important to acknowledge that our study sample was drawn exclusively from two megacities. Consequently, the findings regarding uncertainty stress may not be fully generalizable to platform delivery riders operating in smaller cities, rural areas, or other geographical regions, where distinct socioeconomic contexts and working conditions could lead to different experiences of uncertainty stress. Therefore, caution should be exercised when attempting to generalize the study’s conclusions. Fourth, regular physical exercise was assessed using a single dichotomous item. Future studies should employ validated instruments such as the International Physical Activity Questionnaire (IPAQ) to obtain more granular estimates of physical activity. Fifth, data collection coincided with a period of intermittent COVID-19 control measures in southern China; pandemic-driven reliance on platform delivery substantially elevated demand for riders during this period, while occasional local restrictions affected on-site recruitment in some districts and time slots. Elevated workload during outbreak waves, sporadic mobility restrictions, and broader macroeconomic volatility likely contributed to the elevated uncertainty stress levels reported here, limiting direct comparison with non-pandemic periods. Caution is therefore warranted when generalizing the present findings to non-pandemic contexts.