Reducing the Prevalence of Human Trafficking through Increasing the Financial Literacy and Capability of Survivors and Those At-Risk in SSA

2025-09-01

2028-06-30

Prevalence of domestic and cross-border labor trafficking among YYA aged 18-37, based on direct survey questions in the labor trafficking module; forced labor prevalence estimates implied by list experiments; results of conjoint analysis on the factors that matter most for YYA employment decisions

Labor trafficking will be defined using existing PRIF indicators and thresholds, consistent with the UN definition of trafficking in the Palermo Protocol

Baseline information on socio-economic factors, health, substance use and emotional well-being and correlation with labor trafficking experiences; creation of vulnerability indices for labor trafficking

What makes youth and young adults vulnerable to labor trafficking is important for informing and designing policy initiatives to prevent human trafficking and to support victims and survivors. But there is no accepted vulnerability index in the existing social science literature. We will explore the overlap and differences in the underlying variables among multiple a priori reasonable ways of constructing such an index. We will also analyze how well these indices predict actual labor trafficking experiences in our baseline data. Vulnerability indices will be created using the following approaches: 1) based on risk factors identified in the existing literature; 2) based on PRIF indicators, both individually and in combinatory groups; 3) integrative analysis of vulnerability based on the PRIF indicators and the potential correlates as mentioned in the above paragraph; 4) using unsupervised learning approaches such as principal component analysis and latent class analysis; 5) created by machine learning approaches such as Least Absolute Shrinkage and Selection Operator (LASSO), classification (such as random forest), clustering (such as k-means and hierarchical clustering).

Baseline data collection on labor trafficking prevalence, to be followed by a financial capability RCT

Not available

Randomization done in office by a computer

Randomization is at the individual level

No

Note: clusters are not used in experimental design, randomization is at the individual level
Geographic clusters:
2 districts in Malawi, 4 districts in Zambia
A two-stage stratified sampling design will be employed for the household survey. The primary sampling unit will be the enumeration areas (EAs). The secondary sampling unit will be the households. We define a household as a person or a group of persons who make common provisions of food, shelter and other essentials for living. A sample of census EAs will be selected with probability proportional-to-size (PPS) based on the most recent census. Households will be selected via systematic random sampling from a randomly selected starting point in each selected EA, where a predetermined number will be selected from each EA. Sampling weights will be accounted for using the Horvitz-Thompson method.

Traditional household surveys: 500 households per district, so 1,000 households in Malawi and 2,000 households in Zambia. All YYAs in a household will be interviewed, with an expected 1.7 YYAs per household on average. For each household, we will also ask some household-level questions to the most knowledgeable person in the household. This implies about 2,700 respondents in Malawi and 5,400 respondents in Zambia. RDS: 500 YYAs who have returned from working abroad in the past 2 years per district, so 1,000 YYAs in Malawi and 2,000 YYAs in Zambia. There are a planned 100 seeds per district who will be a combination of at-risk YYAs and survivors, recruited through the household survey and qualitative interviews with key informants, participants from focus group discussions, and in-depth interviews with YYA survivors and those at risk.

Given randomization at the individual level, about half the sample will be in the treatment group and about half the sample in the control group for the list experiment

List experiments: List experiments are less efficient than direct survey questions since the indirect question technique introduces noise. Blair and Imani (2012) suggest power is not an issue in samples with around 1500 respondents, although that will also depend on the number and variance of the control statements. This may mean that the Malawian RDS sample with a target size of 1000 is slightly underpowered. Chuang et al. (2021) find that choosing less innocuous control statements that are at least somewhat sensitive improves the precision of the estimates in their study. We follow a similar approach, with control statements that are also expected to minimize floor or ceiling effects. Blair, Graeme and Kosuke Imai. 2012. Statistical Analysis of List Experiments, Political Analysis, 20: 47-77. Chuang, Erica, Pascaline Dupas, Elise Huillery and Juliette Seban. 2021. Sex, Lies, and Measurement: Consistency Tests for Indirect Response Survey Methods, Journal of Development Economics, 148, 102582. Conjoint analysis: Power calculations for the conjoint analysis are based on two recent working papers: Schuessler and Freitag (2020) and Stefanelli and Lukac (2020), which both yield similar results. Using the Schuessler and Freitag approach, we expect the following given a sample of 1,000 respondents in Malawi and 2,000 respondents in Zambia, with 4 rounds of 2 hypothetical jobs and up to 4 values per attribute (5 values in Zambia for location): - In Malawi sample: 89% power to detect an effect of 0.05 at alpha = 0.05 (and 81% power for an effect of 0.04 at alpha = 0.1) - In Zambia sample: 99% power to detect an effect of 0.05 at alpha = 0.05 , 95% power for an effect of 0.04 at alpha = 0.05 (and 85% power for an effect of 0.03 at alpha = 0.1) References: Schuessler, Julian and Markus Freitag. 2020. Power Analysis for Conjoint Experiments. Working paper. Stefanelli, Alberto and Martin Lukac. 2020. Subjects, Trials, and Levels: Statistical Power in Conjoint Experiments. Working paper.