Aberrant Riding Behavior of Food Delivery Riders – Model

1. Abstract

The Aberrant Riding Behavior of Food Delivery Riders – Model is an empirical psychometric instrument and structural diagnostic framework developed by Qian Qian, Junyi He, and Jing Shi (2024) to evaluate the cognitive, attitudinal, perceptual, and behavioral mechanisms underlying dangerous electric bicycle (e-bike) operation among app-based food couriers. Motivated by the surge in urban micromobility logistics and escalating traffic casualty rates within the on-demand gig economy, the instrument synthesizes socio-cognitive safety models with occupational traffic psychology. The assessment battery comprises 31 primary operational items structured across five overarching latent dimensions: Safety Knowledge (SK), Safety Attitudes divided into Safety Attitudes toward Responsibility (SA-r) and Safety Attitudes toward Laws (SA-l), Risk Perception bifurcated into Perceived Risk Severity (RP-s) and Perceived Collision Probability (RP-p), and self-reported Aberrant Riding Behaviors encompassing general violations, distraction, and aggressive riding maneuvers. Administered via an electronic survey platform using 5-point Likert scales, the measurement model was validated using a sample of professional food delivery riders in China.

Psychometric evaluation demonstrated strong construct validity, internal consistency, and structural stability. Exploratory factor analysis (EFA) demonstrated high sampling adequacy with Kaiser-Meyer-Olkin (KMO) metrics of 0.821 and 0.912 across structural blocks, accounting for over 60% of the total cumulative variance. Confirmatory factor analysis (CFA) affirmed an acceptable model fit: χ²/df = 1.929, Root Mean Square Error of Approximation (RMSEA) = 0.064, Comparative Fit Index (CFI) = 0.916, Incremental Fit Index (IFI) = 0.917, and Tucker-Lewis Index (TLI) = 0.904. Scale reliability yielded Cronbach’s alpha (α) coefficients ranging between 0.706 and 0.944 and Composite Reliability (CR) values spanning 0.836 to 0.934. Convergent validity was established with Average Variance Extracted (AVE) values exceeding the 0.50 threshold, and discriminant validity met the Fornell-Larcker criterion. This tool provides fleet managers, urban transportation authorities, and traffic safety researchers with an empirically substantiated diagnostic protocol to quantify occupational risk factors, design algorithmic accountability safeguards, and implement evidence-based training interventions.

2. Keywords

Aberrant Riding Behavior, Food Delivery Riders, Electric Bikes, Safety Attitudes, Risk Perception, Safety Knowledge, Gig Economy, Traffic Psychology, Occupational Safety, Driver Behavior Questionnaire.

3. Authors

The scale and its foundational empirical model were formulated by traffic engineering and occupational safety specialists:

• Qian Qian — School of Transportation, Southeast University, Nanjing, Jiangsu, China.
• Junyi He — School of Transportation, Southeast University, Nanjing, Jiangsu, China.
• Jing Shi (Corresponding Author) — School of Transportation, Southeast University, Nanjing, Jiangsu, China. Contact: [email protected].

4. Purpose

The accelerated expansion of commercial on-demand delivery platforms has transformed urban mobility paradigms globally. However, this growth has precipitated an alarming surge in micromobility traffic accidents, injuries, and fatalities. Food delivery couriers navigate dense urban street networks predominantly via lightweight electric bicycles (e-bikes) and motorized two-wheelers. Their operating reality is characterized by dynamic navigational challenges, adverse weather conditions, and severe institutional pressures. These pressures stem from platform dispatch algorithms that impose rigid delivery windows, dynamic wage penalties for delay, and instant customer rating systems. In response, delivery couriers frequently resort to non-compliant, aberrant riding practices—such as running red lights, riding against the flow of traffic, riding on designated pedestrian sidewalks, operating smartphones while moving, and executing aggressive overtaking maneuvers. The purpose of the Aberrant Riding Behavior of Food Delivery Riders – Model is to establish a psychometrically robust, diagnostically sensitive instrument capable of capturing the cognitive, normative, and behavioral antecedents that drive these high-risk occupational maneuvers.

From an applied perspective, the tool was developed to address critical theoretical and practical gaps. Prior assessments of two-wheeled traffic behavior predominantly adapted generic instruments such as the Driver Behavior Questionnaire (DBQ) or focused on non-commercial commuter bicyclists. Conventional commuter-focused instruments fail to capture the unique occupational hazards, algorithmic time pressures, platform insurance dynamics, and specialized vehicle-handling demands inherent to professional food courier operations. The Qian et al. (2024) instrument directly isolates these occupational facets by integrating formal safety training indicators, operational legal knowledge, specific perceptions of subjective liability, and deliberate compliance-violation trade-offs (e.g., weighing traffic violations against punctual product handover).

In clinical, research, and industrial settings, this measurement model serves multiple critical functions:

• Occupational Risk Profiling: It facilitates the screening and baseline profiling of commercial fleet couriers, identifying specific cohorts characterized by low perceived collision susceptibility or high attitudinal tolerance for law violations under time constraints.
• Algorithmic and Regulatory Auditing: Urban governance bodies and municipal police departments can utilize the tool to evaluate how algorithmic dispatches, delivery radius thresholds, and platform penalty mechanisms warp occupational compliance and safety margins.
• Evaluation of Educational Interventions: Transportation agencies and digital platforms (e.g., Meituan, Ele.me, Uber Eats, Deliveroo) can administer the scale longitudinally as a pre- and post-intervention metric to measure the efficacy of digital safety training modules, simulator-based hazard awareness courses, and structural safety incentives.
• Predictive Traffic Modeling: Psychometric profiles gathered through this instrument can be directly incorporated into macro-level urban traffic microsimulations, casualty prediction frameworks, and active collision avoidance infrastructure designs.

5. Psychological Construct

The scale measures a multi-tiered psychological architecture comprising cognitive assets, attitudinal orientations, risk appraisals, and overt operational actions. Specifically, the framework operationalizes five interrelated constructs:

Safety Knowledge (SK)

Safety Knowledge represents the rider’s internal comprehension of operational rules, institutional safety protocols, and statutory regulations governing electric bicycle transport. It is not merely an index of abstract traffic sign identification, but an operationalized measure of procedural awareness. The construct assesses whether couriers have undergone structured occupational training, their practical grasp of national and municipal e-bike road codes (such as legal speed thresholds and lane designations), and their working comprehension of accident liability insurance policies. Riders with deficient procedural safety knowledge often operate under operational misconceptions, underestimating legal penalties or lacking defensive riding competencies required in emergency urban scenarios.

Safety Attitudes (SA)

Safety Attitudes denote the stable evaluative orientations couriers hold toward safety compliance, moral responsibility, and occupational risk mitigation. The Qian et al. (2024) model segments this psychological construct into two distinct, latent sub-dimensions:

• Safety Attitudes toward Responsibility (SA-r): Measures the internal moral compass, professional duty, and perceived social accountability of the rider. It assesses feelings of personal accountability for pedestrian welfare, intrinsic motivation to actively avert collisions, anticipated post-crash guilt, and an understanding of the devastating socioeconomic ramifications a severe collision inflicts upon the rider’s household. High scores reflect an internal locus of control and elevated social-altruistic responsibility.
• Safety Attitudes toward Laws (SA-l): Captures utilitarian compliance trade-offs and pragmatic road opportunism. This dimension specifically probes the rider’s willingness to condone traffic violations if it guarantees rapid product delivery or protects against platform punctuality deductions. A high score on the original operational scale indicates an instrumental rationalization of rule-breaking, where legal statutes are perceived as negotiable friction against economic incentives.

Risk Perception (RP)

Grounded in dynamic cognitive appraisal paradigms, Risk Perception measures how delivery riders interpret environmental hazards and subjective vulnerability while navigating road networks. It comprises two complementary cognitive components:

• Perceived Risk Severity / Situational Risk (RP-s): Assesses the cognitive gravity and perceived hazard intensity attributed to specific dangerous driving actions, including running red lights, riding under the influence of alcohol, wrong-way riding (contraflow), abruptly diverting into fast-moving motorized lanes to bypass congestion, and maintaining speeds exceeding surrounding traffic. Riders possessing low situational risk sensitivity normalize severe road hazards as standard operating maneuvers.
• Perceived Collision Probability (RP-p): Measures subjective vulnerability or optimism bias. It evaluates the self-estimated likelihood of being directly involved in a vehicular collision or suffering acute physical injury over the subsequent 12-month period. Disproportionately low probability estimates reflect cognitive invulnerability biases common among experienced commercial riders.

Aberrant Riding Behaviors

The behavioral endpoint of the model reflects self-reported frequencies of dangerous, non-compliant, or suboptimal riding behaviors executed in urban traffic. Drawing structural parallels to the foundational taxonomy established by Reason et al. (1990), this construct captures:

• General Violations: Deliberate contraventions of formal traffic codes to optimize operational velocity (e.g., speeding beyond legal e-bike limits of 25 km/h, ignoring signal phases, riding counter to one-way systems, and unauthorized occupation of high-speed motor vehicle lanes).
• Distraction and Inattention Mistakes: Cognitive and sensory failures arising from dual-task interference, specifically the handling of telecommunication devices, answering calls, monitoring dispatch screens while moving, or losing situational awareness resulting in near-misses with pedestrians and physical obstacles.
• Aggressive Riding: High-risk maneuvers executed around other road users, including abrupt turning without signaling, failure to yield when traversing pedestrian zones (such as school precincts and transit hubs), aggressive cut-throughs in front of turning vehicles, and forceful merging maneuvers.

6. Theoretical Framework

The theoretical architecture of the Aberrant Riding Behavior of Food Delivery Riders – Model synthesizes principles from cognitive safety psychology, occupational health frameworks, and behavioral economics:

Central to its structure is the Theory of Planned Behavior (TPB) formulated by Icek Ajzen. TPB asserts that overt behavior is governed directly by behavioral intentions, which are structured by attitudes toward the behavior, subjective norms, and perceived behavioral control. Within Qian et al.’s framework, Safety Attitudes toward Laws (SA-l) and Safety Attitudes toward Responsibility (SA-r) serve as foundational attitudinal anchors. When couriers develop permissive attitudes toward rule contravention—evaluating traffic violations as instrumental necessities for economic survival—their subjective intentions to abide by traffic laws deteriorate, directly precipitating aberrant riding actions.

The model concurrently integrates the cognitive threat processing mechanisms of the Health Belief Model (HBM) and Rogers’ Protection Motivation Theory (PMT). Under PMT, threat appraisal hinges upon the cognitive integration of perceived severity and perceived vulnerability. Qian et al. explicitly decouple this appraisal into Perceived Risk Severity (RP-s) and Perceived Collision Probability (RP-p). The model posits that adequate Safety Knowledge (SK) acts as an upstream cognitive precursor. Procedural knowledge enhances cognitive awareness of environmental hazards, which calibrates risk appraisals: riders who comprehend kinetic impact forces, stopping distances, and traffic laws evaluate infractions as significantly more hazardous (elevated RP-s) and perceive themselves as susceptible to harm (elevated RP-p). This heightened risk perception subsequently tempers reckless riding impulses.

Furthermore, the behavioral classification maps directly onto the taxonomy of human error popularized by James Reason in the Driver Behavior Questionnaire (DBQ) framework. Reason differentiated between violations (deliberate deviations from practices deemed necessary to maintain safety) and errors/lapses (failures of planned actions to achieve desired outcomes due to attentional failures or cognitive overload). In the food delivery context, this boundary is exacerbated by the Job Demands-Resources (JD-R) Model. Platform dispatch algorithms impose extreme time-related job demands, driving couriers into cognitive overload where mobile phone interactions (checking new orders, navigating digital maps) generate cognitive-sensory distraction, transforming occupational pressure into both deliberate violations and attentional mistakes.

7. Validity

The validity of the Aberrant Riding Behavior Model was rigorously assessed using modern multivariate statistical techniques within structural equation modeling (SEM):

Content and Construct Validity

Content validity was established through systematic literature adaptation (incorporating validated items from Wang et al., 2018; Nguyen-Phuoc et al., 2020; and Jamil et al., 2021) followed by focus-group screening with active couriers and traffic management authorities. The final measurement structure was refined through item purification procedures, which eliminated four aberrant riding items (F15 through F18) that exhibited high cross-loadings or insufficient communalities during pilot testing. Construct validity was validated via both exploratory and confirmatory structural equation workflows.

Convergent Validity

Convergent validity captures the extent to which indicator items of a specific construct share a high proportion of common variance. In Qian et al.’s (2024) investigation, convergent validity was confirmed across all latent dimensions:

• The Average Variance Extracted (AVE) for every latent factor uniformly exceeded the recommended benchmark of 0.50, indicating that the latent constructs explained greater than 50% of the variance observed among their respective manifest indicators.
• Composite Reliability (CR) estimates across the latent variables ranged between 0.836 and 0.934, substantially exceeding the conventional psychometric threshold of 0.70.
• Standardized factor loadings of individual items onto their target latent factors were statistically significant (p < 0.001), with the vast majority exceeding the 0.60 to 0.70 threshold.

Discriminant Validity

Discriminant validity evaluates whether constructs that theoretically should be distinct are empirically differentiated. This was verified through the standard Fornell-Larcker criterion:

• The square root of the AVE for each latent variable was systematically compared against the inter-construct correlation coefficients between that variable and all other constructs in the model.
• The diagonal values (square roots of AVEs) were demonstrably greater than any off-diagonal inter-construct correlation coefficients across the corresponding rows and columns.
• This statistical confirmation established that dimensions such as Safety Attitudes toward Responsibility (SA-r) and Safety Attitudes toward Laws (SA-l) represent distinct psychological concepts rather than redundant expressions of a single general attitude factor.

Predictive and Nomological Validity

Path analysis and structural equation regressions confirmed robust nomological networks. Safety Knowledge exhibited significant positive indirect effects on compliance behaviors, mediated through the cultivation of risk perception and pro-social safety attitudes. Furthermore, self-reported aberrant riding behaviors significantly and positively correlated with historical near-miss frequency and documented traffic collision records among surveyed delivery personnel, demonstrating predictive ecological validity.

8. Reliability

Reliability analyses evaluated the internal consistency and stability of the scale across its independent latent dimensions:

Internal Consistency Reliability

Internal consistency was assessed using Cronbach’s alpha (α). Across all subscales within the battery, alpha coefficients fell comfortably within the desirable range of 0.706 to 0.944, indicating high internal consistency without excessive item redundancy:

• Safety Knowledge (SK): Demonstrates strong internal coherence (α > 0.75), confirming that regulatory awareness, training exposure, and insurance familiarity align as a single knowledge construct.
• Safety Attitudes (SA-r and SA-l): Achieved alpha coefficients exceeding 0.80, indicating consistent moral and statutory attitudinal baselines.
• Risk Perception (RP-s and RP-p): Displayed alpha values above 0.82, affirming reliable measurement of cognitive hazard evaluation.
• Aberrant Riding Behaviors: Displayed high internal consistency (α > 0.90 across retained behavioral indicators), highlighting unified reporting across general violations, mobile phone distractions, and aggressive riding behaviors.

Composite Reliability and Split-Half Stability

To overcome known limitations of Cronbach’s alpha under violations of tau-equivalence, Composite Reliability (CR) was examined. The computed CR parameters ranged between 0.836 and 0.934, verifying measurement precision. In addition, split-half reliability tests were conducted on targeted attitudinal and perceptual factors using the Spearman-Brown prophecy formula:

• The Safety Attitudes toward Laws (SA-l) subscale achieved an unequal-length Spearman-Brown split-half reliability coefficient of 0.804.
• The Perceived Collision Probability (RP-p) subscale exhibited a Spearman-Brown coefficient of 0.821.

These values demonstrate that the instrument yields stable measurement across halves of the item pools, minimizing susceptibility to random measurement error.

9. Factor Analysis

The factorial validity of the tool was examined using a two-stage structural equation modeling workflow comprising Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA):

Exploratory Factor Analysis (EFA)

Initial factor structure identification was conducted on calibration subsamples. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy yielded scores of 0.821 for the attitudinal and knowledge constructs and 0.912 for the behavioral inventory, well above the 0.60 ceiling recommended for stable factor extraction. Bartlett’s Test of Sphericity attained statistical significance (χ² p < 0.001), rejecting the identity matrix hypothesis and establishing inter-item correlation suitability.

Principal Axis Factoring with orthogonal (Varimax) and oblique rotations extracted factors accounting for over 60% of the cumulative variance. During factor extraction, four behavioral candidate items (F15: checking delivery routes on mobile phones; F16: drinking or eating while riding; F17: failing to decelerate when passing vehicles; F18: failing to decelerate on rainy road surfaces) demonstrated low communalities (< 0.40) or complex cross-loadings across violation and error dimensions. In accordance with psychometric purification criteria, these four items were excluded, resulting in an optimized 14-item behavioral matrix (F1 through F14).

Confirmatory Factor Analysis (CFA)

Confirmatory factor analysis using maximum likelihood estimation was executed on validation data to confirm the five-construct structural model. The empirical model demonstrated an acceptable fit to the observed covariance matrices:

• Chi-Square to Degrees of Freedom Ratio (χ²/df): 1.929. Values falling below 3.0 (and particularly under 2.0) represent good model parsimony and fit.
• Root Mean Square Error of Approximation (RMSEA): 0.064 (with 90% confidence interval within accepted bounds ≤ 0.08), denoting low residual discrepancy.
• Comparative Fit Index (CFI): 0.916, surpassing the standard 0.90 threshold of acceptable comparative fit.
• Incremental Fit Index (IFI): 0.917, reflecting robust fit relative to the null baseline model.
• Tucker-Lewis Index (TLI / NNFI): 0.904, exceeding the 0.90 benchmark and confirming acceptable parameter adjustment for model complexity.

10. Instrument / Measurement Tool

The Aberrant Riding Behavior of Food Delivery Riders – Model is a standardized, self-report psychometric questionnaire administered electronically via mobile or web interfaces. Below are the administrative specifications of the instrument:

• Test Type: Multi-construct psychological and behavioral inventory (Self-report questionnaire).
• Target Population: Professional commercial food delivery riders, courier gig-workers, and commercial electric two-wheeler operators (Ages 18 and older).
• Item Count: 31 active measurement items retained in the empirical model (3 Knowledge items, 7 Safety Attitude items, 7 Risk Perception items, and 14 Aberrant Riding Behavior items). Four supplemental behavioral items (F15–F18) from the original 35-item drafting pool are documented for historical-comparative applications.
• Response Formats:
  • Safety Knowledge (Section A): 5-point Likert-type scale: 1 = Totally Uninformed, 2 = Uninformed, 3 = Average / Neutral, 4 = Informed, 5 = Very Well-Informed.
  • Safety Attitudes (Section B – Attitudes): 5-point Likert-type agreement scale: 1 = Strongly Disagree, 2 = Disagree, 3 = Neutral, 4 = Agree, 5 = Strongly Agree.
  • Risk Perception – Severity (Section B – RP-s): 5-point Likert-type hazard scale: 1 = Not Risky at All, 2 = Slightly Risky, 3 = Moderately Risky, 4 = Very Risky, 5 = Extremely Risky.
  • Risk Perception – Probability (Section B – RP-p): 5-point subjective probability scale: 1 = Highly Improbable / Zero Possibility, 2 = Low Possibility, 3 = Moderate Possibility, 4 = High Possibility, 5 = Extremely High Possibility / Certainty.
  • Aberrant Riding Behaviors (Section C): 5-point Likert-type frequency scale: 1 = Never, 2 = Rarely, 3 = Occasionally, 4 = Sometimes, 5 = Always.
• Scoring and Transformation Procedures:
  • Construct scores are generated by calculating the unweighted arithmetic mean or summative composite across the constituent items of each designated subscale.
  • Reverse Scoring: Item A8 (“If a crash occurs, it will seriously affect the family’s life”) is reverse-scored during specific composite modeling routines if utilized in aggregate risk-tolerance indices, ensuring uniform directional alignment with risk-permissive constructs.
  • In Safety Attitudes toward Laws (SA-l: items A9, A10), elevated scores indicate high willingness to violate traffic regulations to maintain delivery speed.
  • In Aberrant Riding Behaviors (items F1–F14), higher aggregate scores reflect elevated engagement in dangerous, illegal, or inattentive driving actions.

11. Permissions & Fee and Test Year

The scale was developed and published in 2024 by Qian Qian, Junyi He, and Jing Shi. It appeared in the academic journal Transportation Research Part F: Traffic Psychology and Behaviour, published by Elsevier Ltd. The instrument is intended for academic research, non-commercial institutional assessments, and municipal traffic safety evaluations under standard fair use academic conventions.

Commercial deployment, integration into proprietary logistics fleet-management platforms, or algorithmic evaluation systems generally requires explicit formal permission or licensing from the copyright holder (Elsevier Ltd.) and consultation with the corresponding author, Prof. Jing Shi (Southeast University). Researchers seeking authorized institutional access, translation adaptations into alternate languages, or commercial deployments must contact the corresponding author directly.

12. References

• Jamil, H. M., Roslan, N. F., Noor, M. M., & Prasetijo, J. (2021). The relationship between safety attitude and risky riding behavior among food delivery riders in Malaysia. International Journal of Road Safety, 2(2), 45–53.
• Nguyen-Phuoc, D. Q., De Gruyter, C., Nguyen, H. A., & Mark, P. (2020). Risky behaviors among motorcycle taxi drivers in Vietnam: The role of work-related factors, safety attitudes, and risk perception. Accident Analysis & Prevention, 146, 105739. https://doi.org/10.1016/j.aap.2020.105739
• Qian, Q., He, J., & Shi, J. (2024). Analysis of factors influencing aberrant riding behavior of food delivery riders: A perspective on safety attitude and risk perception. Transportation Research Part F: Traffic Psychology and Behaviour, 100, 273–288. https://doi.org/10.1016/j.trf.2023.12.007
• Reason, J., Manstead, A., Stradling, S., Baxter, J., & Campbell, K. (1990). Errors and violations on the roads: A real distinction? Ergonomics, 33(10-11), 1315–1332. https://doi.org/10.1080/00140139008925335
• Shi, J., Bai, Y., Ying, X., & Atchley, P. (2010). Aberrant driving behaviors: A study of drivers in Beijing. Accident Analysis & Prevention, 42(4), 1031–1040. https://doi.org/10.1016/j.aap.2009.12.010
• Wang, C., Xu, C., Xia, J., & Qian, Z. (2018). The effects of safety knowledge and risk perception on aberrant riding behaviors of electric bike riders. Accident Analysis & Prevention, 117, 265–275. https://doi.org/10.1016/j.aap.2018.04.022
• Wang, Y., Hao, W., & Xiang, W. (2021). Delivery riders’ red-light running behavior: An integrated perspective of theory of planned behavior and protection motivation theory. Travel Behaviour and Society, 24, 263–271. https://doi.org/10.1016/j.tbs.2021.04.008
• Zhao, X., Ding, H., & Ji, T. (2021). Mobile phone use while cycling among delivery workers: An empirical study using the extended theory of planned behavior. Transportation Research Part F: Traffic Psychology and Behaviour, 81, 102–114. https://doi.org/10.1016/j.trf.2021.05.020

13. Items of the Scale