9th Biennial ACSPRI Social Science Methodology Conference

Background
Estimates of forced labour in the Arab States region are conservative due to difficulties in conducting population-based surveys of this target group in this region. Forced labour in the Arab States is expected to be more prevalent in sectors which rely heavily on migrant workers, many of whom do not speak Arabic and reside in locations that are not accessible such as dormitories or employers’ homes. In 2019, the International Labour Organisation estimated the Arab States hosted 24.1 million migrant workers, 14% of the global figure. This study sought to overcome barriers to conducting surveys on forced labour in the Arab States using a novel approach to sample those who had recently worked in the Arab States and since returned to one of six major sending countries.
Aim
The primary aim of this study was to estimate prevalence of forced labour among migrant workers in the Arab States.
Methodology
To achieve the target sample size of 2,000 respondents in each survey country, a hybrid design was used, combining a multi-stage probability design (nationally representative household surveys) with respondent driven sampling (RDS). The target population was persons aged 15 years and older who had worked in an Arab State in the preceding five years, regardless of their type of work (returned migrant workers). The surveys were conducted face-to-face in six countries known to be major source countries for migrants to the Arab States, namely; India, Bangladesh, Pakistan, Indonesia, Philippines, and Ethiopia.
The sampling design aimed to recruit 50 percent of the sample via multi-stage household sampling and 50 percent via RDS to balance the increased efficiency from the RDS sampling with the well-established probability design. RDS “seeds” who referred other members of the target population were identified through the household sampling and by relevant local organisations.
Probability weights were calculated for the household survey sample and calibrated to population benchmarks from the most recent national census. Statistical matching via a nearest neighbours’ method was used to determine appropriate weights for each record in the nonprobability sample by imputing weights from the matched records in the probability sample.
Results
A total of 10,302 returned migrant workers were surveyed, with 48.5 percent recruited via RDS and 51.5 percent recruited via household sampling. Of the total returned migrant workers, 18.5 percent had been in forced labour in an Arab States country in the preceding five years. There was no significant difference in occurrence of forced labour between those recruited by RDS and household sampling. Issues encountered during the collection of the RDS combined with the experimental nature of the approach to weighting the RDS sample limited our ability to produce reliable estimates of the prevalence of forced labour among returned migrant workers from the Arab States.
Conclusion
A hybrid design that combines probability and nonprobability sampling approaches offers the potential to recruit a sample from which reliable estimates can be drawn. In this study, the use of random household sampling and respondent-driven sampling provided insights into the strengths and challenges of this approach. We present a summary of these findings to inform future research with similar objectives.