Preventing Sexual Harassment in Urban Transport: Measurement Pilot and Cluster Randomized Evaluation in Lima, Peru

The study contains a pilot measurement and implementation-learning phase and a full RCT that includes a census/baseline measurement round, urban-transport safety intervention rollout, and endline measurement round.
In the ongoing pilot, respondents are individually assigned within participating stops to one of three survey-administration modes: guided interviewing, private ACASI, or private ACASI with a hard-garbling randomized-response feature. The pilot is designed to assess measurement feasibility, respondent comprehension, survey privacy, and reporting patterns across modes.
The pilot also includes two stop-level implementation variants that will inform the full-RCT intervention package: a volunteer-based intervention and a signage-only campaign without volunteer presence. These pilot components are intended to assess implementation feasibility, visibility, message salience, field logistics, and operational procedures rather than to produce definitive intervention-effect estimates.
In the full-RCT census/baseline, respondents at the 217 study stops will be assigned within stop to alternative survey-administration modes, including direct questions, private ACASI, and private ACASI with hard garbling. The current minimum target is approximately 10 respondents per stop, with the possibility of increasing the target if field capacity permits. This round will estimate harassment prevalence, test the measurement modes at scale, generate baseline stop-level measures for stratification and balance checks, and refine the full-RCT measurement strategy; it is not intended to estimate intervention effects.
In the full RCT, urban-transport stops will be assigned to status quo control, a volunteer-based intervention, or an awareness campaign without volunteers. The full RCT will estimate effects on harassment, perceived safety, reporting and help-seeking, guardianship, mobility disruption, and related mechanisms. Structured observations of harassment and bystander behavior will be collected where feasible.

2026-07-22

2027-06-30

1. Confirmatory survey-mode contrasts: private ACASI versus guided interview and hard-garbling ACASI versus guided interview.
2. Any of 12 sexual-harassment behaviors reported during the past 30 days at the focal public-transport stop or inside a vehicle boarded there.
3. Any reported severe past-30-day sexual harassment
4. Perceived safety at the focal stop and inside vehicles boarded there.
5. Reporting and help-seeking knowledge or orientation.
6. Observed harassment

The primary outcomes are reporting measures based on the 12 prespecified sexual-harassment behaviors. The main confirmatory survey-mode contrasts are private ACASI versus guided interview/direct questioning and private ACASI with hard garbling versus guided interview/direct questioning.
Any reported past-30-day sexual harassment: an indicator equal to one if the respondent reports any of the 12 prespecified sexual-harassment behaviors during the past 30 days at the focal urban-transport stop or inside a vehicle boarded there.

Any reported severe past-30-day sexual harassment: an indicator equal to one if the respondent reports any prespecified severe harassment behavior during the past 30 days. Severe behaviors include following or pursuit, blocking passage or cornering, threats or intimidation, unwanted sexual touching or bodily pressure, forced kissing, attempted or forced sexual acts, or use of physical or sexual force during an unwanted sexual conduct.

For the pilot and full RCT, the primary outcome families are:
Reported past-30-day sexual harassment: any of the 12 prespecified sexual-harassment behaviors reported during the past 30 days at the focal urban-transport stop or inside a vehicle boarded there.

Reported severe past-30-day sexual harassment: any of the prespecified severe harassment behaviors reported during the past 30 days at the focal stop or inside a vehicle boarded there.

Perceived safety: perceived safety at the focal stop and inside vehicles boarded there, measured using prespecified survey items or a standardized perceived-safety index.

Reporting and help-seeking orientation: knowledge of, confidence in, or willingness to use reporting and support channels, measured using prespecified survey items or a standardized reporting/help-seeking orientation index.

Observed harassment: structured-observation measures of harassment-related incidents, third-party presence, and bystander or volunteer response during prespecified observation windows at study stops.

1. Maximum severity of harassment reported
2. Location of harassment incidents
3. Perpetrator and bystander context
4. Observed harassment of other women
5. Perceived frequency and timing of harassment risk
6. Mobility and activity disruption
7. Additional transport spending
8. Reporting behavior, institutional response, and barriers
9. Safety and environmental conditions at stops
10. Intervention exposure and visibility
11. Intervention comprehension and immediate perceived effects
12. Institutional trust and willingness to seek help
13. Gender norms and normalization of harassment
14. Recent psychological distress
15. Social support and conversations about harassment

Maximum severity of harassment reported: an ordinal measure classifying the most severe harassment behavior reported by the respondent, based on the prespecified hierarchy of the 12 behavior-specific harassment items.

Incident location and context: indicators for whether harassment occurred at the focal stop, near the stop, or inside vehicles boarded there, and measures of perpetrator type, third-party presence, and bystander reactions.

Observed harassment of other women: an indicator for whether the respondent observed another woman experiencing harassment at the focal stop or in transport, with descriptive measures of incident type and bystander response.

Perceived frequency and timing of harassment risk: ordinal or categorical measures of how frequently harassment is perceived to occur and when risk is perceived to be highest.

Mobility and activity disruption: indicators for whether harassment or fear of harassment led the respondent to change routes, travel times, behavior, companionship, clothing, punctuality, attendance, or work/study/activity choices. A composite disruption indicator may be coded as 1 if any such consequence is reported.

Additional transport spending: an indicator for whether the respondent spent more on transport to avoid harassment or travel more safely, and the reported additional amount spent.

Reporting behavior, institutional response, and barriers: indicators and categorical measures for whether harassment was reported, to whom it was reported, whether action was taken, and reasons for not reporting.

Environmental safety conditions: measures of perceived lighting, crowding, ability to ask for help quickly, visible security presence, and visible anti-harassment messages at the focal stop.

Intervention exposure and visibility: indicators for exposure to campaign materials, signage, volunteers, and observed volunteer actions.

Intervention comprehension and immediate perceived effects: indicators for whether campaign materials or volunteers improved understanding of where to seek help, knowledge of where to report, or perceived safety.

Institutional trust and willingness to seek help: measures of trust in police, MIMP/CEM, municipal or security actors, and willingness to seek institutional help after harassment. An institutional-trust index may be constructed by recoding component items to a 0-1 scale and averaging.

Gender norms and normalization of harassment: indices measuring agreement with inequitable gender-norm statements and minimization or acceptance of harassment, coded so higher values indicate more inequitable or more accepting attitudes.

Recent psychological distress: the Kessler K6 score, constructed as the sum of six symptom-frequency items ranging from 0 to 24, with higher values indicating greater distress. A normalized version may be used in index analyses.

Social support and conversations about harassment: the Oslo Social Support Scale, constructed from three items with a total score ranging from 3 to 14, plus indicators for whether the respondent discussed harassment with others in the past 30 days and with whom.

The registration covers two linked study phases: a pilot measurement and implementation-learning phase and a full randomized controlled trial. The full RCT includes a pre-randomization census/baseline measurement round, stop-level intervention assignment and rollout, and an endline measurement round.
In the pilot, eligible women recruited at 20 urban-transport stops are assigned to one of three survey-administration modes: guided interviewing, private ACASI, or private ACASI with hard garbling. The pilot also tests two implementation variants that will inform the full-RCT intervention package: a volunteer-based intervention and a signage-only awareness campaign without volunteer presence. The pilot is designed to assess measurement feasibility, respondent comprehension, implementation feasibility, visibility, message salience, field logistics, and operational procedures, rather than to produce definitive intervention-effect estimates.
After the pilot and before full-RCT intervention rollout, the study will conduct a census/baseline measurement round across the 217 study stops. This round will estimate harassment prevalence, generate baseline stop-level measures for stratification and balance checks, test alternative survey-administration modes at scale, and inform updated power calculations and measurement decisions for the full RCT. The current minimum target is approximately 10 respondents per stop, with the possibility of increasing the target if field capacity permits.
In the full RCT, 217 urban-transport stops will be assigned approximately evenly to status quo control, a volunteer-based intervention, or an awareness campaign without volunteers. Randomization will be stratified using pretreatment stop characteristics where available. Endline outcomes will be measured through commuter surveys, structured observations, and implementation records. The primary analysis will follow intention to treat and use inference clustered at the stop level.

Not available

Study stops are selected from a prespecified frame of eligible urban-transport stops using operational and research criteria, including commuter volume, feasibility of survey and observation fieldwork, suitability for volunteer/signage implementation, geographic coverage, and sufficient separation across stops to limit spillovers where possible.
For the pilot, the survey-mode assignment list was generated by computer before fieldwork using a reproducible random-number seed. Within each of the 20 pilot stops, interview-sequence positions are allocated approximately equally across the three survey-mode arms: guided interviewing, private ACASI, and private ACASI with hard garbling. Field staff do not choose the assigned mode. The pilot stop-level implementation variants are also assigned across the 14 pilot stops, with 7 stops assigned to the volunteer-based intervention and 7 stops assigned to the signage-only campaign without volunteers.
For the full-RCT census/baseline survey-mode experiment, respondents will be randomized within stop to direct questioning, private ACASI, or private ACASI with hard garbling using a computer-generated assignment sequence. The final within-stop allocation will be prespecified before census/baseline fieldwork and may depend on the final respondent target per stop.
For the full RCT, the research team will use statistical software to randomize 217 stops across three intervention arms: status quo control, volunteer-based intervention, and awareness campaign without volunteers. Randomization will be stratified or restricted using prespecified pretreatment stop characteristics, such as geographic area, commuter volume or traffic intensity, and baseline harassment-risk measures where available, to improve balance across arms. The final randomization code, seed, allocation, and balance checks will be archived before intervention rollout.

For the survey-mode experiments, the unit of randomization is the individual consenting respondent, with assignment blocked or sequenced within urban-transport stop. For the volunteer/signage intervention, the unit of randomization is the urban-transport stop.

Yes

For the survey-mode experiments, the unit of randomization is the individual respondent, with assignment blocked or sequenced within urban-transport stop. For the volunteer/signage intervention, the unit of randomization is the urban-transport stop. The full RCT includes 217 urban-transport-stop clusters. The pilot includes 20 recruitment stops used for measurement and implementation learning.

Pilot: maximum 200 respondents across 20 urban-transport stops. Full-RCT: current minimum target of approximately 2,170 respondents across the 217 study stops, based on 10 surveys per stop, with the possibility of increasing the target up to approximately 4,340 respondents if field capacity permits.

Pilot survey-mode experiment: up to 200 respondents, assigned approximately equally across guided interviewing, private ACASI, and private ACASI with hard garbling.
Pilot stop-level implementation test: 20 urban-transport stops: 7 volunteer-based intervention stops, 7 signage-only campaign stops without volunteers, and 6 status quo control stops.
Full-RCT census/baseline survey-mode experiment: respondents at 217 study stops assigned within stop to direct questioning, private ACASI, or private ACASI with hard garbling. The current minimum target is approximately 10 respondents per stop, with the possibility of increasing the target if field capacity permits. The final within-stop allocation across survey modes will be prespecified before census/baseline fieldwork.
Full-RCT stop-level intervention: 217 urban-transport stops assigned approximately evenly across three arms: about 72 status quo control stops, 72 volunteer-based intervention stops, and 73 awareness-campaign-only stops without volunteers.

The pilot sample is up to 200 respondents across 20 urban-transport stops. The pilot is designed primarily for feasibility, measurement, and implementation learning rather than definitive intervention-effect estimation. Pilot survey-mode contrasts and pilot stop-level implementation comparisons will be reported descriptively, with confidence intervals where appropriate. The full-RCT census/baseline measurement experiment is designed to estimate harassment prevalence, test measurement-mode differences, and provide inputs for final full-RCT power calculations. The current minimum census/baseline target is approximately 10 respondents per stop across 217 stops, or about 2,170 respondents total, with the possibility of increasing the target up to approximately 20 respondents per stop, or about 4,340 respondents total, if field capacity permits. Under two-sided alpha 0.025 and 80 percent power, approximate minimum detectable differences for survey-mode contrasts are expected to be about 7-9 percentage points under the 10-respondent-per-stop target and about 5-6 percentage points under the 20-respondent-per-stop target, depending on baseline prevalence, within-stop allocation across direct questioning, private ACASI, and private ACASI with hard garbling, and the additional variance induced by the hard-garbling estimator. The full-RCT intervention-effect power calculations use 217 randomized stops assigned approximately evenly across three arms. With approximately 10 respondents per stop at endline, two-sided alpha 0.025, 80 percent power, binary outcome prevalence in the 20-50 percent range, and intracluster correlations between 0.01 and 0.10, indicative minimum detectable effects for pairwise intervention-arm comparisons are approximately 7-11 percentage points. For lower-prevalence severe-harassment outcomes, indicative detectable absolute differences are smaller, approximately 5-7 percentage points, but represent larger proportional effects. Final minimum detectable effects will be updated before full-RCT randomization using pilot and census/baseline information on outcome prevalence, stop-level intracluster correlation, treatment-arm allocation, covariate adjustment, and the structured-observation fieldwork plan. Confirmatory full-RCT inference will use stop-clustered standard errors and a prespecified familywise multiple-testing adjustment.