Call Me When You Get Home”: New Technologies, and Safety in Public Spaces

Safety in public spaces is a global concern that can negatively impact individuals’ well-being, cause them to avoid certain activities, and lead to additional expenditures to feel secure. In this study, we document the prevalence of safety concerns and experiences of harassment among university students in the UK. The experiment has three layers of randomization.
First, in the baseline survey, a random subset of students (20%, Round 2 only) receive information about the prevalence of harassment among university students drawn from prior survey evidence. Second, all students are invited to download the safety app with randomized monetary incentives (£0 or £25). Third, among students who download the app, half are randomly assigned to receive a real-time video-call feature connecting them to a trained operator available 24/7, while the other half receive the same app, without the video-call feature.
The study is conducted in two rounds: Round 1 (Spring 2025) and Round 2 (Spring 2026). A supplementary Prolific study (N=600 UK university students) collects additional demand and travel data. Endline surveys are conducted approximately 2-3 months after baseline. We study effects on crime experiences, perceived safety, beliefs about crime risk, anxiety, and mobility. We characterize selection into adoption, and study how effects vary across individuals. We additionally estimate willingness to pay for safety and develop a structural model of nighttime travel choices.

2025-03-24

2026-09-30

Primary outcomes are organized into four families:
Family 1: Transport-related perceived safety (safety walking in public spaces, using public transport, and in a cab/Uber; 1-5 Likert).
Family 2: Anxiety (GAD-7 score and minimal anxiety binary; nighttime walking anxiety intensity 0-10).
Family 3: Beliefs and perceived risk (perceived probability of own harassment in 100 outings; perceived probability on usual route; confidence in own estimate).
Family 4: Activities and mobility. Extensive margin: nights returning home after 10pm, stayed home when wanting to go out. Intensive margin: nights alone, taxi use when alone, share of solo returns by taxi.

For each primary family, we construct a summary index following Kling et al. (2007): components are standardized using the control group mean and standard deviation, then averaged. We correct for multiple hypothesis testing within primary families when examining individual components.
Family 1 index: average of standardized safety scores across three situational items. We report both continuous and binary (safe = 4 or 5) versions. (survey data)
Family 2 index: average of standardized GAD-7 score and nighttime anxiety intensity. (survey data)
Family 3 index: average of standardized own harassment probability, route-specific probability, and confidence. Safety salience frequency is measured in Round 2 only and reported as an individual component, not included in the index. (survey data)
Family 4: we construct separate extensive margin (nights out, binary any night out) and intensive margin (taxi share) sub-indices rather than a single index, as these capture different behavioral responses. (survey + admin)

Secondary outcomes are organized into four families:
Family 5: Situational safety perceptions (safety along usual route from campus to home, near home, near campus). - survey data
Family 6: Experienced harassment and safety incidents (past two months) - survey data
Family 7: Beliefs on prevalence: outcome of information treatment - survey data
Family 8: . Academic engagement, performance, and choices - admin data - analyzed descriptively and treated as exploratory due to limited coverage.

1. Family 5 (Situational Safety) index: average of three standardized items — safety along campus-to-home route (1-5), safety near home (1-5), safety near campus (1-5).

2. Family 6 (Crime Experiences): we construct three levels of aggregation. (a) Overall crime index (Kling): average of standardized binary indicators for all incident types. (b) Two sub-indices: a sexual harassment sub-index (unwanted sexual comments, unwelcome staring, someone physically following, sexual exposure, unwanted sexual contact, overt sexual touching, attempt to rape/rape) and a non-sexual crime sub-index (theft/attempted theft, physical harassment, verbal harassment). (c) Individual binary indicators for each specific incident type. We additionally report: a binary for any incident in the past 2 months, as well as separate dummies for each specific type of incident. Harassment experiences are measured by asking respondents whether they experienced specific events in the past two months, in order to avoid relying on potentially varying interpretations of the term “sexual harassment.” At baseline, we also include a randomized subsample to which we apply hard garbling techniques, allowing us to estimate average misreporting. For the garbled subsample, we report garbling-corrected prevalence estimates.

3 Family 7 (Beliefs on prevalence): perceived percentage of female university students in London who experienced sexual harassment in public spaces in the past two months (0-100), and the same for male students (0-100). These beliefs are the direct target of the information treatment, which provides the actual statistics from Round 1 survey. This family is the primary outcome for the effect of information treatment.

4.Family 8: Educational outcomes are only for students for which we have admin data (analyzed descriptively and treated as exploratory due to limited coverage):
Class attendance (% of scheduled classes)
Exam performance (0-100 scale)
Engagement (on a 1-4 scale, as reported by class teachers)

The study is a multi-stage randomized experiment among university students in the UK. To maximize participation, most of the students are invited to complete a baseline survey during class time. In Stage 1 (Round 2 only), a random subset of students receive information about harassment prevalence. In Stage 2, all students are invited to download the app with randomized monetary incentives. In Stage 3, among students who download a mobile safety app, half are randomly assigned to receive an additional protection safety feature while the other half receive the basic version. The study is conducted across two rounds with an endline survey approximately 2-3 months after baseline.

Not available

Randomization done by the computer directly on Qualtrics

Individual

No

N/A. Randomization at the individual level.

Across two rounds, we have surveyed approximately 3,600 students (Round 1: 1,100; Round 2: approximately 2,500 across multiple UK universities). A supplementary Prolific study collects additional data from 600 UK university students. Eligible participants are students who (i) are at least 18 years old and (ii) provide informed consent. We may conduct additional rounds subject to funding availability.

Info treatment (Round 2 only): 80% control group, 20% treatment group.
Incentives (Round 2): 40% low incentives, 60% high incentives.
Incentives (Round 1): 50% low incentives, 50% high incentives.
Video-call (among adopters): 50% control group, 50% treatment group.