The effect of destination safety information on job search and migration: Evidence from South Africa

We recruit jobseekers aged 18–35 living in Amathole District (Eastern Cape, South Africa) to participate in an information experiment. All respondents first complete two sets of discrete choice experiments (DCEs), designed to identify both preferences for and subjective beliefs about migration destination attributes (such as destination safety, earnings opportunities, relocation cost), which are difficult to distinguish using observational or revealed preference data.

All respondents receive urban wage information for a set of feasible migration destinations identified during the DCE.

Treatment-group participants then receive an information package about the relative safety of destinations among a set of feasible migration destinations identified during the DCE. The package summarizes ward-level safety in urban migration destinations, with a focus on risk of crime exposure, using the local CBD of Amathole District as a reference point. Safety information is aggregated from diverse data sources. Primary data sources were collected by the researchers in advance of the intervention in high-traffic migration locations between March-July 2026. These data sources include:
- Safety audits of the physical urban environment using the SafetiPin methodology and a South Africa-specific questionnaire developed by the researchers
- Perceptions of relative ward-level safety elicited from enumerators and urban residents.
- Experiences of crime data collected from urban residents
- Secondary data sources include:
- South African Police Service (SAPS) statistics
- the Victims of Crime Surveys (VCS)
- Meta population data
- nightlight data
- open street maps data

The intervention is intended to reduce uncertainty and correct inaccurate beliefs about destination safety. The impact of information about a given destination will be respondent-specific. We will analyse the treatment as both a respondent-specific information shock in absolute magnitude, and relative to their prior beliefs. Depending on the respondent's prior beliefs, the information may shift beliefs toward greater or lesser perceived safety. Participants assigned to the control group receive no destination-safety information.

Both groups receive the same wage information, compensation and complete the same survey up to the information intervention. Recruitment, contact effort, and follow-up procedures are otherwise identical across arms. In the follow-up survey, both groups’ ward-level safety beliefs are re-elicited to observe responses to the information treatment.

2026-09-10

2026-12-10

We will collect data on realized migration outcomes, migration intentions, willingness to work or live in our set of urban migration destinations, and job search intensity and employment by geographic area. The primary outcome will be a measure of job search re-allocation between wards. We hypothesize that there will be shifts toward destinations that are safer than their prior. We will pre-specify a final primary outcome in a pre-analysis plan.

Migration is a rare event in this setting, so we are unlikely to be powered to observe effects on realized migration outcomes. We are therefore likely to rely on a measure of job search intensity and migration intentions.

To better understand the mechanisms underlying the impact of information treatment, we will both directly measure beliefs about and preferences for destination safety and elicit them through the DCEs to observe how salient destination safety is to individual migration decisions, how safety beliefs change (both the mean and the variance), and whether these effects are persistent. We will explore heterogeneity with respect to gender

The study plans to enroll and randomize 2,700 eligible individuals. Eligibility requires that respondents: (i) are aged 18–35; (ii) currently reside in Amathole District (iii) are actively seeking work via the SAYouth platform, or receiving the Social Relief of Distress (SRD) unemployment grant, and (iv) can provide informed consent. Recruitment is planned through the SA Youth jobseeker platform and via sampling from queues for the Social Relief of Distress (SRD) unemployment grant in towns in the Amathole district.

We will first construct a sampling frame from the set of individuals who consent to participating in the trial. Sampling is then stratified by recruitment channel, gender, and migration intentions. We will up-sample respondents who intend to migrate within the next year to improve power to observe realized migration outcomes and to allow heterogeneity analysis by gender. The study targets approximately equal numbers of women and men.

After consent, eligibility confirmation, enrollment, and baseline measurement, named respondents are assigned 1:1 to:
- Safety-information treatment: approximately 1,350 respondents
- Control: approximately 1,350 respondents

The baseline survey collects basic information on demographics, socio-economic status and risk aversion.

Respondents are then administered two discrete choice experiments to identify preferences for and beliefs about destination attributes – most importantly, destination safety but also classic economic variables including wages, unemployment risk, and relocation cost. In the first experiment, participants make migration choices between urban destinations with fully known attributes. These hypothetical discrete choices are constructed from real ward examples (with names removed) where both safety information and wages are known, or linear interpolation between real examples. This identifies preferences over destination attributes, including safety. In the second discrete choice experiment, respondents make choices between wards with which they are familiar. In these choices, some destination attributes — notably safety — are uncertain, allowing identification of ex-ante beliefs about these attributes, through participants’ choices under uncertainty. The discrete choice experiment occurs before treatment to prevent the information intervention from changing the beliefs and preference parameters elicited at baseline.

Finally all respondents (in both treatment and control groups) are provided with urban wage information, while only respondents in the treatment group are provided with the safety information intervention detailed above. Treatment and control groups will be interviewed at different times of day, limiting contamination via spread of information.


Hypotheses
Using our randomized treatment, we will first test whether the provision of information shifts safety beliefs. This is essentially a first stage. We are unsure ex ante whether safety beliefs are unbiased compared to the information (which is near ground truth), and how precise they are. At the individual level, we expect that beliefs will shift in the direction of information, but the global average effect will depend on the distribution of ex ante beliefs.
Second, we test whether information changes the geographic allocation of job search, migration and employment destinations. Our core hypothesis is that there will be re-allocation toward relatively safe destinations and away from less safe destinations. Uncertainty about destination safety may also reduce, increasing job search among risk-averse respondents

Not available

Randomization is done in the office by one of the PIs following the construction of the final sampling frame, and prior to the commencement of the baseline activity. Randomization is blocked within recruitment channel, gender and baseline migration intention.

Randomization is at the individual level.

No

2700 respondents

2700 respondents

1,350 respondents – Treatment Group: receive safety and wage information
1,350 respondents – Control Group: receive wage information only