This project introduces a novel methodology for treatment assignment, which aims to balance the goals of maximising the precision of treatment effect estimates and maximising the welfare of experimental participants. The assignment algorithm is applied to three interventions designed to improve formal employment outcomes of Syrian refugees and local jobseekers in Jordan.
In a conventional RCT, the randomisation of treatment assignment to experimental subjects is performed in order to precisely estimate the effects of all treatments. However, there are cases where the experimenter may want to not only learn whether policies work, but also to maximise the welfare of program participants, by learning what treatments work best. Information derived from observing treatment outcomes over time can be used in order to adaptively optimise treatment assignment for future participants.
In recent years, there has been a radical shift in how assistance is provided for refugees. The international community has moved away from a model in which refugees are housed in camps to a model focused on identifying solutions that integrate refugees into local communities and labour markets. Thus, aid has shifted from the delivery of basic commodities and food items to the support of individuals to gain access to employment.
Of the 660,000 Syrian refugees registered with the UNHCR as being displaced in Jordan (UNHCR, 2020), it is estimated that 93% live below the five USD per day poverty line. Employment among Syrian refugees in the formal Jordan labour market is likely to be constrained by both demand (e.g. difficulties in processing work permits, consequences of sanctions applied to informal work) and supply side factors (e.g. credit constraints, lack of experience and information, self-control problems, job quality in the formal sector). At the same time, low-skilled Jordanians continue to suffer from pre-existing labour market challenges (IRC, 2017; Government of Jordan, 2019; UNHCR, 2020).
It is within this context that researchers took advantage of the International Rescue Committee’s “Project Match”, which was active in three cities (Amman, Irbid and Mafraq), to recruit study participants. Eligibility criteria included (i) being Syrian refugees or Jordan nationals; (ii) between 18 and 45 years old, and (iii) willing to take up low-skilled formal wage work that pays approximately minimum wage. Participants were selected using a variety of passive and active recruitment methods. For both nationalities, 60 percent of the sample is composed of women and the average age is about 29 years; however, Syrians tend to be much less educated on average. Many individuals in the sample - including refugees - are actively looking for work.
All respondents received four Jordanian dinars (JOD, about 5.60 USD at the time of the intervention) to cover possible costs of transport to a job interview, and an informational flyer covering steps for interview preparation. In addition, three separate job search interventions, each of them designed to represent a distinct form of job search assistance, were designed and administered to three separate treatment groups.
First, one group received a cash transfer, of a value of 65 JOD (about 92 USD at the time of the intervention). This was intended to support the recipient to pay for the cost of job search, and recommendations were provided to participants on how they should use this cash, making it a labelled transfer.
A second group received informational support on (i) how to prepare and interview for a formal job in Jordan, and (ii) the legal rights of employees in formal jobs. Information was provided through face-to-face interaction, two videos and two take-home paper tools.
Lastly, a third group received psychological support, through (i) a four-week job-search planning calendar, (ii) an instructional video on how to use the calendar, (iii) a face-to-face demonstration, and finally (iv) reminder SMSs.
Treatment assignment was delivered through a Tempered Thompson algorithm, implemented within a hierarchical Bayesian model, which represents one of the main novelties of this project. The purpose of this assignment mechanism is to balance the goals of maximising the precision of treatment effect estimates and maximising the welfare of experimental participants, by exploiting information derived in previous rounds of treatment about which treatment works best for which strata of the participants, and using it to update a participant’s probability of receiving treatment.
Six weeks after the intervention took place, none of the interventions increased employment for the average programme participant.
However, all interventions raise the proportion of Syrians who look for work two months after baseline by 5.6 percentage points (a 13 percent increase over a control job-search rate of 43 percent) and lead to 0.5 more job applications placed by Syrians (40 percent increase over a control mean of 1.2 applications). Among Jordanians, only the information intervention has a large and positive impact on job search.
Both the cash and information intervention improve Syrian refugees’ labour market outcomes, up to four months after baseline. Offering cash leads to a significant increase in the employment rate of more than 50 percent, and a significant boost in earnings of about 40 percent after two months and of 65 percent after four months. The information intervention increased employment and earnings by almost the same amount as the cash grant two months after baseline. Four months after baseline, the cash transfer generates a 40 percent increase in employment and a significant 55 percent increase in earnings. The information intervention has weaker and short-lived effects on labour market outcomes of refugees; four months after baseline, no significant impacts are documented.
The magnitude of the effects documented is large relative to estimates reported in the recent literature for other active labour market policies in developing countries (McKenzie, 2017).
These effects, and further analysis included in the paper, suggest that liquidity constraints may be a key labour market barrier for refugees.
Government of Jordan (2019). Jordan Response Plan for the Syrian Crisis: Draft manuscript.
IRC (2017, February). In Search of Work: Creating Jobs for Syrian Refugees: A Case Study of the Jordan Compact. https://www.rescue.org/sites/default/files/document/1343/insearchofworkweb.pdf.
McKenzie, D. (2017). How Effective are Active Labor Market Policies in Developing Countries? A Critical Review of Recent Evidence. The World Bank Research Observer 32(2), 127–154.
UNHCR (2020, March). Syria emergency. https://www.unhcr.org/syria-emergency.html.