Street harassment and the potential of technological tools

Harassment in public spaces is a pervasive global issue that severely undermines individuals’ sense of safety. This persistent threat reduces subjective well-being, deters people from seizing valuable opportunities, and forces them to bear additional financial and emotional costs to protect themselves. In response, safety-focused mobile applications have become increasingly widespread, developed and managed both by governments and by independent organizations.
This study examines the determinants of take-up of such apps, with particular attention to individuals’ knowledge of crime incidence. It also evaluates their effectiveness across several outcomes, including perceived and actual safety, and explores users’ willingness to pay for these tools. Our sample includes more than 2,000 students at Bocconi University.

Gender-based violence remains a pervasive global issue, with nearly one in three women experiencing physical and/or sexual violence in their lifetime (WHO, 2021). Much of the existing research has focused on Intimate Partner Violence (IPV) (e.g., Adams-Prassl et al., 2024; Banerjee et al., 2019; Bindler and Ketel, 2022) and, more recently, workplace harassment (e.g., Adams-Prassl et al., 2024; Boudreau et al., 2022; Folke and Rickne, 2022). Yet harmful acts also occur in public spaces, committed not only by intimate partners but also by strangers. Alarmingly, 80% of women worldwide report having experienced harassment in public spaces, and 90% fear it (IPSOS, 2021). Concerns about safety have been shown to significantly shape women’s behavior and well-being (Cornell Project, 2015).

Despite the salience of the issue, there is a striking lack of causal evidence on how safety concerns affect mental health, economic outcomes, and behavioral choices. This study addresses this gap by examining the determinants of mobile safety app take-up, with particular attention to knowledge about crime incidence. It also evaluates the effectiveness of these tools across multiple outcomes—including perceived and actual safety—and investigates individuals’ willingness to pay for them.

Population of interest. We focus on students at Bocconi University in Milan. Students represent an ideal population, as young people (under 35) are more likely to experience or witness street harassment, perceive heightened risks, and adopt mobile apps more frequently than older groups (IPSOS, 2021; Statista, 2021). Milan also ranks as the second Italian city with the highest crime index (Numbeo, 2025), making it a particularly relevant setting.

Experimental design. The study involves more than 2,000 students (both women and men). To maximize participation, baseline surveys are administered during class time in the first or last 15 minutes.

Step 1 – Informational treatment. Students are randomized to receive (or not) information about the incidence of sexual harassment, to test whether knowledge affects adoption of safety tools.

Step 2 – App encouragement. Students are then assigned to one of three groups:

Control group: no encouragement.

Treatment 1: encouragement to download the government app 112 Where Are U.

Treatment 2: encouragement to download the VIOLA Premium app, using a unique tester ID provided by the research team.

Two months later, an endline survey is again administered in class. Finally, we plan to link survey responses with administrative data from the academic institution to broaden the scope of our analysis.

2025-10-06

2025-12-15

1. Take-up of the mobile safety apps
2. Safety perceptions (at night alone) - survey data
3. Crime experiences - survey data
4. Anxiety- survey data
5. WTP for safety (conjoint experiment)

1. Take-up of the mobile safety apps: This is a dummy (1/0) whether the individual actually downloaded the app. For Treatment 1 we need to rely on self-reported outcome asked in endline, while for Treatment 2 we can check whether the assigned tester-id was actually used.

For the other primary outcomes, we will correct for multiple hypothesis testing when necessary. For each family of outcomes, we will construct summary indices and also analyze individual outcomes separately.

2. Safety perceptions (at night while alone) are measured using survey data, specifically six situational questions rated on a 1–5 Likert scale, as well as the question, “When is the last time you were willing to walk alone at night?” We will construct a general index of safety perception, along with separate indices for: (i) safety while walking or using public transport, and (ii) safety across specific types of situations. We will also divide the indexes in 2 categories.

3. Harassment experiences are measured by asking respondents whether they experienced specific events in the past month, in order to avoid relying on potentially varying interpretations of the term “sexual harassment.” In addition to indexes, we construct a dummy variable indicating whether any incident occurred in the past month, as well as separate dummies for each specific type of incident. We also report the frequency of sexual harassment.

4. We have two different measures of anxiety: 1. GAD-7: we will construct a dummy for "minimal anxiety" 2. Anxiety while walking at night alone (1-10): we will report the unit difference and construct a dummy.

1. Trust in different types of interventions/institutions - survey data
2. Beliefs about harassment — We measure participants’ beliefs regarding their probability of harassment and incidence.
3. Mobility

3. Mobility: The data here comes from survey data. Here as well we will construct first a general index of mobility at night, and then analyze each group and outcome separately.

The project follows a randomized experimental design involving a sample of university students (both men and women) in Italy. To maximize participation, most of the students are invited to complete a baseline survey during class time. At the end of the baseline, students are first randomized to receive information about the incidence of sexual harassment, and then they are (cross) randomized in either not receiving anyting or trying the VIOLA PREMIUM app —a new mobile safety app that became viral in Italy, or a government app 112 Where Are U. An endline survey is then conducted two months later, either via email or during class time.

The project follows a randomized experimental design involving a sample of university students (both men and women) in Italy. To maximize participation, most of the students are invited to complete a baseline survey during class time. At the end of the baseline, students are first randomized to receive information about the incidence of sexual harassment, and then they are (cross) randomized in either not receiving anyting or trying the VIOLA PREMIUM app —a new mobile safety app that became viral in Italy, or a government app 112 Where Are U. An endline survey is then conducted two months later, either via email or during class time.

Individual randomization inside Qualtrics

Individual

No

N/A

We are targeting 2,490 students. However, the final sample will inevitably be smaller, though the exact size cannot be determined ex-ante. This is because some students may be absent from class, and incomplete surveys will not be included in the analysis.

1/3; 1/3; 1/3