Experimental Design Details
The experiment involved five stages that are presented in Table~\ref{tab:Recruitment-Process} and described in more detail below.
\paragraph{Stage 1: Announcement.}
In the first stage we announced the positions through newspapers, university employment boards and social media In this announcement we provided general information about the position. Table~\ref{tab:Information} presents a summary of the information that participants received in this and subsequent stages. In this stage we tried to get a large pool of subjects interested in the positions over which we randomized the treatments. We invited recently graduated applicants from all areas of study. Note that at this stage the exact nature and dangers of the job have \textit{not} been communicated yet.
\paragraph{Stage 2: Statement of Interest.}
Job-seekers were allowed to state their interest in the position by filling out a low hurdle \emph{statement of interest} form. %\autoref{app:Statement_Interest} presents the complete list of socioeconomic information collected in this stage.
The announcement evoked substantial initial interest and around 2,200 people expressed interest in the positions. We randomly assigned 1,105~job-seekers to the three treatments that we explain below.
In this stage we elicited basic information of the job-seekers by asking them to fill out an ``Expression of interest'' form. We obtained information about gender, age, level of studies (undergraduate, master), area of studies and year of graduation.\footnote{Asking for gender is common in Colombia and there are no legal implications in doing so. }
The measures that we obtained in this stage can be considered predetermined as we observe them before participants are exposed to the treatment. These measures are used to evaluate the qualification of job-seekers as explain in more detail below. Using self-reported measures, truthful reporting is of some concern as people have a tendency to misreport in order to increase their chances to be hired. We therefore provided incentives for truthful reporting by announcing that supporting documents will be required in order to be eligible for employment if selected. This policy was communicated whenever they had to enter verifiable information during the application process.
\paragraph{Stage 3: Invitation to apply.}
In this stage, about half of the job-seekers who submitted a statement of interest form were randomly assigned to be invited to continue the application process. This group to which we refer to as invited job-seekers, received an email explaining the particular conditions of employment regarding job responsibilities, salary and duration of the employment and an invitation to apply for the job (see~Table \ref{tab:Information}). At this stage we introduced the treatments and varied: the location of the job (low or high conflict area) and the wage compensation paid in the risky area. %A job offer to work in low risk regions at a specified salary of 1.5~million Colombian Pesos (COP) per month, a job offer to work in the high risk region at a specified salary of 1.5~million COP or to work in the high risk regions at a salary of 1.8~million COP.
% say how we determined the risk premium (was sth liek the additional premium one has to pay for insurances in this region)
Finally, we asked invited job-seekers to complete a lengthy application questionnaire in order to apply for the position. Filling out the application questionnaire took between 40 and 60~minutes and required to search supporting information on several questions which was announced to job-seekers. By using a demanding and time-consuming application questionnaire that would increase the cost of the application (time required), we wanted to capture the effect of opportunity cost of applying to the position. As recruitment processes are usually very comprehensive, we expected that invited job-seekers would not be surprised about the demanding application process.\footnote{We stratified job-seekers into treatment and control group based over gender and the main residence in Bogotá.}
\paragraph{Stage 4: Application.}
Invited job-seekers had access to a personalized page and could complete the application form over different sessions saving the information and continuing the application over several days. Yet, a strict deadline date was set after which no application was accepted. The measures of qualifications obtained in this stage are endogenous as performance in the tests and responses to the questionnaire could have been affected by the treatment. A summary of the information job-seekers obtained is presented in Table~\ref{tab:Information}.
Invited job-seekers who were not interested in the job also had the chance to actively drop out of the application by clicking a button. In the questionnaire we asked the reasons why they left the application process. However, the turnout was low and only 6.4~percent of subjects who did not start the application process actively dropped out.
Field experiments that last for a longer period and communication between subjects or potential subjects could cause information spillovers. We tried to minimize this by opening the position at the same time and by recording the starting time of the applications, in order to control for potential timing effects.
\paragraph{Stage 5: Hiring.}
Invited job-seekers who completed the application processes, to which we will refer to as applicants, were ranked upon qualifications. The top 10~applicants were invited for an interview. The best applicants received a job offer. We hired 3~applicants. We do not consider measures of on-the-job performance, as the limited number of positions offered does not allow us to conduct a statistical analysis of the impact on job performance.
We use a between subject design that varied the the location of the job (low or high conflict area) and the wage compensation paid in the risky area.\footnote{As "no risk" jobs do not exist, we are not interested in looking at a certainty effect, a discontinuity in risk perception from no-risk to very small risks. Although theoretically interesting and documented in laboratory studies, "no-risk" environments do not exist and even in laboratory studies a "no risk" condition entails some not specified background risk (i.e. the experimenter has run out of cash or acts deceptively).}