Safety of female ride-hailing passengers: Perception and prevention

The rapid growth of app-based ride-hailing services (e.g., Uber, Lyft, Didi, Ola) has significantly impacted urban mobility worldwide, including large user bases and notable mode share in major cities. Despite benefits such as addressing information asymmetries and complementing public transport, safety concerns related to driver behavior, sexual assault, and victimization persist and vary by region. Reports highlight incidents involving theft, assault, and harassment, with women disproportionately experiencing fear, anxiety, and victimization in transport settings, particularly at night. This study focuses on female passengers’ safety perception and prevention actions in the ride-hailing context, aiming to understand how drivers, vehicles, and travel environments shape perceived safety and to identify individual heterogeneity in perceptions and behaviors. The purpose is to develop evidence to inform safety evaluation, crime prevention, and management mechanisms tailored to vulnerable users, especially women.

The literature identifies ride-hailing safety as encompassing vehicle, driver, and platform factors, including traffic/personal, property, and information safety. Proposed measures in prior work include technological and procedural enhancements (e.g., dash cams, alarms, identity verification, privacy protections), yet quantitative passenger-focused validation is limited. Gendered mobility research shows women undertake more complex trip chains, rely more on public transport, and report lower perceived safety and greater fear, especially in poorly surveilled environments and at night. Individual attributes (age, ethnicity, income, disability) and contextual conditions (neighborhood lighting/maintenance) influence fear and victimization risk. Preventative actions taken by women commonly include avoidance (e.g., avoiding night travel, high-crime areas, certain routes/times) and protective behaviors (e.g., heightened situational awareness, seating choices, traveling with companions, using phones to communicate/verify information). Concerns about privacy and sharing rides with strangers affect willingness to carpool. Gaps remain in segmenting proactive versus reactive prevention and in understanding expectations specific to ride-hailing. The reviewed studies (summarized in tables within the paper) highlight factors affecting perception and prevention: demographics, physical and perceived security, risk attitudes and knowledge, and travel habits. Effective protective behaviors frequently cited include mobile phone use for information and communication, surveillance, and traveling in pairs, though effectiveness varies by person and situation.

The study integrates objective and subjective approaches. Objective data: Using China Judgments Online for the dominant Chinese ride-hailing platform Didi Chuxing, the authors retrieved criminal judgment documents related to ride-hailing. The text reports both a search yielding 789 criminal judgment documents and references to 7898 Didi-related criminal cases; approximately 674 verdicts involved drivers as perpetrators and passengers as victims. Incidents included traffic accidents, illegal gains, intentional offenses, sexual assault, indecent behavior, and other violations; motives categorized as monetary gain, sexual gratification, disputes (minor/fare/route), and loading violations. Guangdong and Shandong provinces had the highest counts, with incidents rising from 2016 to 2018 and decreasing around 2020 during COVID-19. Platforms have responded with measures (e.g., Didi’s same-sex late-night pickups; women-only services in some countries). Subjective data: An exploratory survey targeted Chinese female ride-hailing users. Instrument development drew on prior literature and platform-incident keywords. The questionnaire comprised individual characteristics, travel habits, and situational items. Data collection involved a face-to-face on-site survey (October 18, 2020) with 5 Yuan incentives and an online survey (October 20, 2020–March 15, 2021) distributed to women with 2 Yuan incentives post-screening. Random availability sampling was used, with strategies to mitigate selection/sampling biases (larger sample, diverse regions/city types, socio-demographics, mixed online/offline, controlled survey period). Of 1055 responses, 753 were complete and, after screening (trap questions, plausibility, completion time), 596 valid responses remained. Sample profile: majority aged 18–35 (62.6%); students (47.3%) and practitioners (43.6%); regional concentration in southeastern China (e.g., Chongqing 26.8%). Measurement and analysis: Items used 5-point Likert scales. Analyses in SPSS 26 included descriptive statistics, reliability/validity tests, Harman’s single-factor test for common method bias, Pearson correlations, independent-samples t-tests, one-way ANOVA with post-hoc comparisons (LSD/Tamhane/Welch corrections), and box-whisker plots. Perceived safety dimensions (traffic, information, property, personal) were rated on 1–5 (extremely unsafe to extremely safe). Safety Perception Scale (SPS): items tapping driver image (DI), driver behavior (DB), and traveling together (TT). Exploratory factor analysis (EFA) indicated KMO=0.859, Bartlett p<0.001, three factors retained, total variance explained 60.673%; loadings 0.593–0.874; Cronbach’s alpha: DI α=0.896, DB α=0.793, TT α=0.817. Prevention Action Tendency (PAT): constructs included mobile phone dependence (PD), information attention (IA), and risk avoidance (RA). EFA indicated KMO=0.890, p<0.001, total variance explained 67.06%; Cronbach’s alpha: IA α=0.894, RA α=0.917; PD items had acceptable loadings (0.571–0.805). Common method bias: total variance of first factor 26.296% (<50% threshold), inter-construct correlations r<0.90. Reliability/validity: CR 0.82–0.92; AVE mostly >0.5 (TT slightly below but acceptable with CR>0.6); discriminant validity supported as sqrt(AVE) exceeded inter-construct correlations. Difference testing: ANOVA and t-tests assessed heterogeneity in DI, DB, TT, PD, IA, RA, and overall perceived safety (PS) across age, education, income, ride-hailing frequency, student status, victimization, residence (municipal or above vs county-level or below), and business-trip experience.

• Overall perceived safety: Mean ratings for traffic safety (2.84), information safety (2.57), property safety (2.73), and personal safety (2.42) were all below 3, indicating ride-hailing was perceived as marginally unsafe among surveyed women.
• Safety perception specifics: Driver behaviors strongly influenced perceived safety; items like detours without notice (mean ~1.95), intrusive privacy questions (~1.96), and staring/ogling (~1.97) were among the least safe scenarios. Traveling with companions increased perceived safety; TT items had high means (e.g., carpooling with an acquaintance 4.15; with a friend 4.02; with a child 4.09).
• Prevention actions: High endorsement of mobile phone–based precautions and information attention. High-mean PAT items included: IA5 (driver has no criminal record) 4.18; IA4 (strict identity audit) 4.15; IA6 (complaints handling affects trust) 4.14; PD4 (verify route on map) 4.02; PD6 (uneasy if phone out of power/signal) 4.02; PD3 (share driver/car/route info) 4.00; RA2 (avoid remote suburbs) 4.08; RA1 (avoid riding alone at night) 3.99; RA3 (avoid riding alone in unfamiliar places) 3.90.
• Heterogeneity (ANOVA): Age significantly affected driver image perception (DI) (F=6.790, p<0.001) and traveling together (TT) (F=4.638, p<0.01). Education, income, and ride-hailing frequency also significantly affected DI (p<0.05). Education and frequency influenced overall perceived safety (PS) (p<0.05). Risk avoidance (RA) varied by education (F=4.899, p<0.01).
• Heterogeneity (t-tests): Students differed from non-students on DI (t=-4.350, p<0.001) and RA (t=-2.126, p=0.034). Victims differed on DI (t=4.078, p<0.001). Residence differences were significant for RA (t=-5.489, p<0.001) and PS (t=2.668, p=0.008). Business-trip experience affected DI (t=-2.648, p=0.009) and PS (t=-2.007, p=0.045).
• Broad patterns: Older women (>40) were more sensitive to driver image and preferred traveling with companions. Low-income women and those in less-developed residential areas reported lower perceived safety and stronger protective/avoidance tendencies. Greater ride-hailing dependence/frequency was associated with higher perceived safety. Victimization experience heightened attention to drivers’ appearance/behavior but did not necessarily increase all preventative actions.
• Platform/industry context: Crime incidents involving ride-hailing drivers as perpetrators are documented; platforms introduced measures such as same-sex rides at late hours and women-only services in some countries.

The study set out to understand how female passengers perceive safety in ride-hailing and what preventative behaviors they adopt, and how these vary by individual and contextual factors. Findings show that drivers’ words and actions are the most salient cues shaping safety judgments, and companionship during rides increases perceived safety. Preventative behaviors are predominantly information- and communication-based via mobile phones, coupled with risk-avoidance strategies (avoiding night rides alone, remote areas, unfamiliar places). These behaviors align with international evidence and confirm the relevance of targeted interventions in ride-hailing. Heterogeneity analyses demonstrate that age, education, income, residential context, ride-hailing frequency, student status, victimization history, and business-trip needs significantly shape both perceptions and behaviors. This highlights the need for differentiated policy and platform responses tailored to user subgroups. The results bridge gaps in the literature by quantitatively validating passenger-focused safety measures and by distinguishing proactive and reactive prevention tendencies among women in a ride-hailing context. The practical implications emphasize improving driver vetting and conduct, vehicle monitoring and visibility, in-app safety tools (identity verification, route checking, emergency contacts, recording, itinerary sharing), and transparent handling of complaints, alongside environmental improvements (lighting, surveillance) and gender-sensitive transport planning to elevate perceived and actual safety.

The study reveals a persistent gap between actual safety measures in ride-hailing and women’s perceived safety. Female passengers’ safety judgments are particularly shaped by driver image and behavior, while traveling with companions enhances perceived safety. Preventative practices—especially mobile phone dependence for verification and communication, attention to safety-related information, and risk-avoidance behaviors—are widely adopted. Perceptions and prevention vary across age, education, residence, frequency of use, and victimization experience, with lower education, poorer residential environments, infrequent use, and prior victimization linked to lower perceived safety. Building on these insights, ride-hailing platforms and policymakers should strengthen driver conduct regulation, identity audits, vehicle monitoring, and in-app safety features, and develop clear emergency protocols and transparency in complaint handling. Reactive prevention guidance for passengers includes avoiding risky contexts (night, remote areas), protecting privacy and belongings, maintaining connectivity, cooperating with platforms (e.g., location sharing), and traveling with companions. These measures aim to narrow the gap between perceived and actual safety, rebuild trust in ride-hailing, and support safer, more inclusive, and sustainable shared mobility.

The survey targeted all female ride-hailing passengers rather than focusing exclusively on women with victimization experiences, resulting in relatively few victimized respondents for subgroup analysis. Potential selection and sampling biases inherent in the random-availability sampling approach are acknowledged, despite mitigation steps (larger and more diverse sample, mixed online/offline modes, controlled survey period). The analysis is based on data from China, and contextual factors (e.g., regional crime prevalence, platform policies, urban form) may limit generalizability. Some inconsistencies in reported counts of criminal cases in secondary data sources are noted.