Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
In 2017, Walk Free Foundation and the International Labour Organization estimated that there were 40 million victims of modern slavery on any given day in 2016, including 25 million people in forced labor and 15 million in forced marriage. Some regions of the world suffer this problem disproportionately, with modern slavery most prevalent in Africa (7.6 per 1,000 people) and least prevalent in the Americas (1.9 per 1,000 people).
To provide a targeted, objective assessment of the problem at the country level, Walk Free and Gallup have collaborated since 2014 to collect data on modern slavery using nationally representative household surveys implemented through the Gallup World Poll. To date, the module on modern slavery has been fielded in 54 national surveys covering 48 countries (conducted multiple times in some countries), with a total sample of 71,158 individual interviews.
In preparation for the release of Walk Free’s 2018 Global Slavery Index, Gallup and Walk Free have developed an extrapolation methodology using hierarchical Bayes models to inform prevalence estimates. This type of model is of particular interest because we want to extrapolate the results of the model beyond the sample of 48 countries that have included the module on modern slavery.
Hierarchical models can integrate individual-level and country-level predictors to estimate variation in individual and country-level risks, allowing predictors to follow different functions in different parts of the world, as needed. A Bayesian approach is computationally useful because our data deal with rare events that in a frequentist approach may lead to computational problems.
How do we define ‘modern slavery’?
Modern slavery is an umbrella term that includes several crimes against freedom — human trafficking, slavery and slavery-like practices such as servitude, forced labor, forced or servile marriage, the sale and exploitation of children, and debt bondage.
We operationally define modern slavery following the methodology described by Jacqueline Joudo Larsen and Pablo Diego-Rosell (2017) and the International Labour Organization (2017). We set up our models by first defining modern slavery, including cases of forced labor and forced marriage.
Forced labor victims are identified according to the following criteria:
Forced marriage victims are identified according to the following criteria:
What factors predict modern slavery?
We identified suitable predictors of modern slavery from the intersection of the Gallup World Poll core questionnaire and the five dimensions of the Walk Free modern slavery vulnerability model (Joudo Larsen & Davina Durgana, 2018, to be published). The vulnerability model, guided by human security and crime prevention theories, assesses vulnerability at the country level to improve our understanding of the drivers of modern slavery, as well as the quantitative changes over time in these drivers.
The risk scores for 167 countries are based on an analysis of data covering 23 risk variables across five major dimensions:
Using these five dimensions as an organizing framework, we identified a total of 157 variables that could potentially be used to predict forced labor or forced marriage status, including 122 individual-level variables from the World Poll and 35 country-level variables from the Walk Free vulnerability model. From this larger set, we identified a subset of 10 independent variables that were available across most countries and most optimally predicted either forced labor or forced marriage, including:
We then proceed to build our predictive models using a multilevel modeling approach that enables us to simultaneously model individual-level predictors of modern slavery and country-level predictors of the average prevalence of modern slavery in different countries. Multilevel models also allow us to let regression coefficients vary in different parts of the world. For example, we may want to fit a model that allows a predictor such as gender to have different effects in different regions to better fit the risk ecosystem of each region. This is an innovative approach in the field of modern slavery (Gelman & Hill, 2007).
The following charts summarize the key regression coefficients for the final forced labor and forced marriage models. In the case of forced labor, and adjusting for all other factors in the model, the risk of having experienced an instance of forced labor in the past five years varies according to the protective and risk factors shown in the table below.
In the case of forced marriage, and adjusting for all other factors in the model, the risk of having been in a situation of forced marriage varies according to the protective and risk factors shown in the table below.
What countries are at highest risk of modern slavery?
We calculated the average weighted predicted probabilities for each outcome (forced labor/forced marriage). The resulting heat map (below) shows the sum of the predicted probabilities of forced marriage and forced labor, for a global total risk of modern slavery. Based on the sum of the average predicted probabilities of forced labor and forced marriage, Syria, Central African Republic, South Sudan, Somalia and Iraq are the five countries with the highest overall risk of modern slavery.
This study represents the first global effort to model and predict the risk of modern slavery.
While the primary purpose of our analysis is to help estimate the global prevalence of modern slavery, our risk models also have the potential to assist in policy design and to better target programs aimed at eliminating modern slavery. Future modeling efforts will benefit from using these results as priors within a Bayesian framework, which should help further reduce uncertainty regarding the prevalence of modern slavery.
More generally, the hierarchical Bayes methodology used in this project can be used wherever we have individual data nested hierarchically and we are interested in extrapolating to other units for which we don’t have data for the outcome of interest. For example, we might have outcome data on public opinion, health or voting intentions for a subset of countries or regions and want to predict the same outcomes in countries or regions where we don’t have outcome data. Strong predictors, robust statistical models and appropriate model validation are key elements of success.
Read the full paper.
Jacqueline Joudo Larsen is Senior Research Manager at Walk Free Foundation.
Pablo Diego-Rosell is a Senior Researcher at Gallup.
Read more from Gallup’s Methodology Blog and visit our Methodology Center.