Methodology behind the Global Slavery Index

The Global Slavery Index was designed to shed light on the extent of modern slavery and level of vulnerability to modern slavery for 160 countries, as well as the actions taken by 176 governments to address these crimes and human rights violations. The methodology that enables these assessments is described in detail in three parts:

1. Part A – Estimating prevalence.

2. Part B – Measuring vulnerability.

3. Part C – Assessing government action.

Appendix Part A: Estimating prevalence

The regional estimates presented in the 2021 Global Estimates of Modern Slavery (Global Estimates) produced by the International Labour Organization (ILO), Walk Free, and the International Organization for Migration (IOM), are used by Walk Free as the starting point for our independently produced national level estimates, which are presented in the 2023 Global Slavery Index (GSI).

This section summarises the methods used to produce the global and regional estimates presented in the 2021 Global Estimates and details the process to get from the regional estimates to national prevalence estimates. A detailed account of the methodology of the 2021 Global Estimates can be found in the Global Estimates of Modern Slavery report on the Walk Free website.

Global Estimates of Modern Slavery

The 2021 Global Estimates were comprised of two sub-estimates: an estimate of forced labour and an estimate of forced marriage. The sub-estimate of forced labour was then further broken down into three categories: forced labour in the private economy, forced commercial sexual exploitation, and state-imposed forced labour.

Figure 1: Typology of modern slavery

As no single source provides data that is suitable for the measurement of all forms of modern slavery, a combined methodological approach was adopted for the 2021 GEMS, drawing on three sources of data to calculate the sub-estimates:

Nationally representative surveys

The estimates of forced labour in the private economy (excluding the sex industry) and forced marriage are derived from 68 nationally representative surveys on forced labour and forced marriage jointly conducted by ILO and Walk Free, and implemented through the Gallup World Poll.1 During the 2017-2021 period, a total of 77,914 respondents aged 15 years and over were interviewed either face-to-face or by telephone across the 68 survey countries. Surveys also collected information on forced labour and forced marriage among immediate family members (spouse, biological parents, children, and biological siblings) of respondents who were alive at the time of interview. As a result, the full network sample (respondents and their family network) was 628,598 persons and included children below 15 years of age as well as individuals 15 years old and over. Only cases of modern slavery that occurred between 2017 and 2021 were included in these estimates, and all situations of forced labour were counted in the country where the exploitation took place. In the five-year reference period for the estimates, while surveys were conducted in 68 countries, men, women, and children were reported to have been exploited in 129 countries (see Figure 2).

Figure 2: Countries of exploitation identified from the countries surveyed

The estimates of forced marriage also draw on national telephone surveys on forced marriage conducted in four countries in the Arab States region: Kuwait, Qatar, Saudi Arabia, and United Arab Emirates.2 During 2021, Arabic speaking residents aged 18 and over were recruited via random digit dialling with quotas based on the most recent national census used to achieve an approximately representative sample of 2,000 respondents per country. Respondents were asked about their own experiences of forced marriage and those of their immediate family members. The inclusion of these surveys brought the total network sample for forced marriage to 109,798 persons.

Estimates for countries in which national surveys were not conducted were produced through an imputation model.3 A weighted linear model on the prevalence rate of forced labour included geographic variables and a variable on the number of international migrant workers in the country as covariates. For forced marriage, the weighted linear model consisted of geographic variables only. Because no national surveys were available for the North America region, that region was assimilated with the Northern, Southern, and Western Europe region.

Counter Trafficking Data Collaborative (CTDC) dataset

Administrative data from IOM’s CTDC dataset4 was used in combination with the 68 survey datasets to estimate forced commercial sexual exploitation of adults and commercial sexual exploitation of children. The CTDC dataset comprises cases of trafficking for both sexual and forced labour exploitation and includes information on the profile of the survivors of human trafficking (e.g., age, gender, citizenship, country of birth) and on the trafficking situation (e.g., country of exploitation, type of exploitation, industry of exploitation, means of control). Statistical models were used to estimate the odds ratios of forced commercial sexual exploitation relative to forced labour exploitation separately for adults and children by gender using the CTDC database. These odds ratios were applied to the corresponding global estimates of forced labour exploitation of adults and children derived from the national surveys.

Comments from the ILO Committee of Experts on the Application of Conventions and Recommendations relating to state-imposed forced labour, and other secondary sources

Because the surveys focused on the non-institutionalised population, meaning that people in prisons, labour camps or military facilities and other institutional settings are not sampled, they are not suitable for estimating state-imposed forced labour. Instead, the estimate of state-imposed forced labour was derived from validated secondary sources and a systematic review of comments from the ILO Committee of Experts on the Application of Conventions and Recommendations relating to state-imposed forced labour.

The estimates are calculated as stock estimates; that is, the average number of persons in modern slavery at a given point in time during the 2017 to 2021 reference period. The stock estimate is calculated by multiplying the total flow by the average duration (the amount of time in which people were trapped in forced labour) of a spell of modern slavery.

From global and regional to national estimates

The national estimates presented in this GSI were calculated using individual and country-level risk factors of modern slavery. A risk model was used to generate average predicted probabilities of modern slavery by country. The regional totals in the 2021 Global Estimates were then apportioned based on each country’s average predicted probability of modern slavery. This process involved the following key steps:

Validating individual-level risk factors of modern slavery and predicting modern slavery at the individual-level

During the development of the 2018 GSI risk model to estimate the risk of modern slavery, a set of individual-level risk factors for forced labour and forced marriage were identified using national surveys that included questions on experiences of forced labour and forced marriage. This included using a series of statistical tests to identify relationships between instances of victimisation and other variables collected in the Gallup World Poll.5 In developing the 2023 GSI vulnerability model, the relationship between these individual-level risk factors and modern slavery was assessed. The individual-level risk factors are presented in Table 1. Tests confirmed the variables identified in 2018 remained significantly associated with forced labour and/or forced marriage and were therefore retained as variables in the 2023 risk model. Individual-level risk factors included age, gender, marital status, education level, urban/rural, employment, life evaluation, health, and ability to live on current income.

Table 1: Individual-level predictors of modern slavery

Multi-level models 

Multilevel models (MLM) were fitted to the data to enhance the predictions of the individual-level models and account for the hierarchical nature of these data. MLMs allow for the extrapolation of model results beyond the sample of 68 countries. Model coefficients were estimated using Bayesian hierarchical linear models with random intercepts, with weakly informative normal priors (mean = 0, SD = 2.5). Model coefficients were calculated separately for forced labour and forced marriage which were set as outcome variables. Model coefficients were then applied to survey data in each country to calculate individual-level risk. Individual risks were then aggregated into a country average, which was calculated using post-stratification weights. This followed the same approach that was used in 2018 GSI estimate calculations. This model produced unrealistically high risk of forced labour for several countries in the Latin America and the Caribbean and Eastern Europe subregions. As a result, the sample of surveys on which calculations of individual-level risk were based was limited to those where there was greater confidence based on alignment with the vulnerability model. As a result, 12 countries were removed and the models re-estimated. The final model is represented by equation 1:

Equation 1.

where

is the logit of the probability  of forced labour or forced marriage for each individual  in a country.

₀ is a constant term (intercept).

_1ij is a vector of individual-level demographic control variables with values varying for each individual within a country, and with unknown coefficients ₁.

_2ij is a vector of individual-level predictor variables , with values varying for each individual within a country, and with unknown coefficients ₂.

_3j  is the vulnerability score , with values varying for each country_, and with an unknown coefficient ₃.

_j is a random coefficient that is allowed to vary by country.

_ij is an individual error term.

Model Performance 

The overall accuracy of a model was measured by the area under the ROC curve (AUC), with an AUC of 1 representing a perfect model, and an AUC of .50 representing a model with no discrimination, as good as a random guess. The assessed AUC value of 0.75 indicates the model has acceptable discrimination, according to the Hosmer & Lemeshow (2013) heuristic.

Actual versus fitted prevalence plots for the 55 countries with modern slavery survey data showed a similar or better fit (Pearson’s r = 0.71). The random intercepts model showed a poorer fit with the actual values than the other two models (a perfect fit is exemplified by the red dotted line). Alternate models were examined for improved fit; for example, a model removing random intercepts but leaving country as a fixed effect. However, a random intercepts model with country level predictors provides the most comprehensive framework to undertake these inferences and was the model on which estimates were based.

Estimating prevalence and aligning with Global Estimates of Modern Slavery regional estimates

Individual predictions were aggregated into risk scores at the country level. For the 29 countries that were missing Gallup Word Poll data, risk factors were imputed as an average over several multiple imputation approaches (glm, amelia, multiple imputation by chained equations). Country risk scores were used to estimate country prevalence by apportioning the regional counts of modern slavery from the 2021 Global Estimates based on the risk of modern slavery in a country relative to the risk of other countries in the region. This was undertaken as follows:

First, country risk was adjusted by country of exploitation. The basic premise is to apply an adjustment factor equal to the ratio of victims identified in the national surveys in a subregion, to total exploited victims in the same subregion. If no national surveys were conducted in a given subregion, we estimate that the prevalence is equal to modelled risk multiplied by population. This is calculated using the following steps:

1. Calculate number of victims identified by the country surveys who are exploited in a different country, by country of exploitation.

2. Code countries as either “net sending” or “net receiving” based on income level of country and international migrant stock as a percentage of the total population (2020).6 Countries with an international migrant stock of less than 5 per cent of their population and grouped as either “low income,” “lower-middle income,” or “upper-middle income” were assigned as sending countries, as were “high income” countries with an international migrant stock of less than 10 per cent of their population.

3. Calculate aggregate number of victims by place of exploitation in sending and receiving areas.

4. Adjust down the risk score of regions that have a lower number of victims being exploited in country.

5. Adjust up the risk score of regions that have a higher number of victims being exploited in country.

Second, taking adjusted country risks, estimate prevalence in a country based on the regional prevalence and the distance between the adjusted country risk and the weighted average regional risk score, following these steps:

1. Normalise the adjusted and imputed country risk scores to a 1-100 range, with 1=min risk, 100=max risk.

2. Multiply the normalised risk score by the country population.

3. Calculate regional prevalence by dividing aggregates for total modern slavery (excluding state-imposed forced labour) over total population.

4. Calculate average normalised regional score by dividing the sum of normalised risk scores by the country population.

5. Calculate country prevalence by multiplying the regional average by the ratio of normalised country risk score over the average normalised regional score.

To simplify, since normalised modern slavery risk in Afghanistan (60.1) is 2.28 times higher than the average risk in Asia and the Pacific region (26.3), we estimate that prevalence in Afghanistan is 2.28 times greater than the regional average.

Third, use the survey estimate for Mauritania (3.2 per cent) rather than the modelled risk score due to the distinct nature of slavery in the country. The practice is entrenched in Mauritanian society with slave status being inherited and deeply rooted in social castes and the wider social system. Those owned by masters had no freedom to own land and cannot claim dowries from their marriages nor inherit property or possessions from their families.7 When it abolished slavery in 1981, it was the last country to do so. Hereditary slavery continues to impact the Haratine and Afro-Mauritanian communities, with many survivors and their descendants dependent upon former “masters” because of limited skills and alternate economic opportunities. Given the evidence available that supports the higher survey estimate, that estimate is taken from Mauritania, and other countries in Sub-Saharan Africa are adjusted down to ensure totals are aligned with the Global Estimates of Modern Slavery.

To account for heightened risk experienced by migrants in the Arab States, we made an additional adjustment within the region based on national surveys of returned migrant workers in six origin countries across Asia and Africa.

Final calculation of estimated prevalence

The process outlined in steps 1 and 2 produces prevalence estimates for all forms of modern slavery except state-imposed forced labour. Given the nationally specific manifestations of state-imposed forced labour where it does occur, it was excluded from the steps outlined above and a final adjustment based on publicly available data sources was made to account for this. A final estimate of the prevalence of all forms of modern slavery is then calculated. Additionally, estimates of the number of people in modern slavery for each country are calculated with reference to UN population estimates.8 The resulting estimates are presented in Table 2.

Data limitations

Limitations of the source data

As with all empirical research, there are some limitations of the data used to produce the 2021 Global Estimates, within which the findings of this Index should be interpreted.

First, the use of imputations models introduce some error and, as such, the national estimates should not be interpreted as hard findings.

Second, while the sample of countries on which the estimates were based in the 2021 estimates is larger than in previous editions, there remain some regions where the coverage is limited or lacking — this specifically concerns North America and the Arab States (however, regional forced marriage data gaps have been somewhat addressed). The sample of countries also omits some of the most populous countries, namely China, India, Pakistan. While surveys were conducted in India and Pakistan, fragility of the underlying data led to their exclusion. Fielding of these surveys during COVID-19 restrictions is likely to have had an impact on data quality. Similarly, it is usually not possible to survey in countries that are experiencing profound and current conflict, such as Syria, Iraq, Yemen, Libya, South Sudan, and parts of Nigeria and Pakistan. Yet it is known that conflict is a significant risk factor — the breakdown of the rule of law, the loss of social supports, and the disruption that occurs with conflict all increase risk of both forced labour and forced marriage. The lack of data from countries experiencing conflict means that modern slavery estimates in regions in which conflict countries are situated will understate the problem.

Third, COVID-19 affected the data collected from countries surveyed during 2020 and 2021, during which time data was collected via telephone rather than face-to-face, as had been done exclusively in previous rounds of data collection.

Fourth, the estimates of forced commercial sexual exploitation and forced labour of children were built on models of profiles of assisted cases of human trafficking in the CTDC dataset compiled by IOM and its partners. While the dataset provided rich data for global estimation, the regional distribution must be taken with caution.

Lastly, due to changes in some areas of the methodology and the expansion of the data coverage, the 2021 global and regional estimates on forced labour are not truly comparable with the estimates of the previous edition. The high variability of the estimates, especially, at the regional level also warrants caution when comparing between editions. Forced labour and forced marriage are not only difficult to capture in sample surveys and administrative sources, but also hard to measure through survey questionnaires and administrative reporting systems. The result is that the estimates have relatively high sampling errors and a low degree of replicability. Even without changes in methodology and data coverage, the estimates are likely to exhibit high variability making comparison over time somewhat hazardous.

Limitations of the risk modelling

This analysis is not without the limitations inherent to any cross-sectional research endeavour. Our selection of variables is driven by both theoretical and statistical considerations, but unfortunately the field of modern slavery lacks a unifying causal theory that can be used to inform variable selection. Finally, we have a limited sample size of confirmed individual cases, which limits the extent to which we can expand our predictive models and enhance the accuracy of our predictions. Further surveys will lead to an increase in our sample, thereby enabling the study of more complex effects and refinement of the modelling.

Table 2: Estimated prevalence and number of people in modern slavery, by country

Appendix Part B: Measuring vulnerability

Walk Free measures the extent to which a population is vulnerable to modern slavery across 160 countries. The vulnerability model has three main aims:

• Inform prevalence estimation, contributing to the risk model which allows national estimates to be made at the country-level, including for countries where there is no national survey.

• Identify and quantify individual, systemic, and structural factors that make people vulnerable to modern slavery, informing the allocation and direction of anti-modern slavery efforts.

• Help to identify potential data “blind spots” where future research should be targeted.

The vulnerability model maps 23 risk variables across five major dimensions:

1. Governance Issues

2. Lack of Basic Needs

3. Inequality

4. Disenfranchised Groups

5. Effects of Conflict

The following section provides an overview of the methodology of the vulnerability model.

Development of the vulnerability model

The vulnerability model methodology has evolved since the 2018 edition of the Global Slavery Index with the input from an independent Expert Working Group. During 2016 and 2017, the Expert Working Group provided feedback on addressing theoretical and empirical gaps, normalisation, and standardisation of the data, dealing with missing data, and weighting of the data. This advice has carried through to the methodology used to calculate the vulnerability model. Further detail on the feedback of the Expert Working Group can be found in the 2018 GSI methodology.

Theoretical framework

The vulnerability model is guided by human security and crime prevention theories. The human security theory was developed by the UN Development Programme to capture seven major areas of insecurity: economic, political, food, community, personal, health, and environment. The most basic shared characteristic of human security as a concept involves a focus on the safety and well-being of individuals regardless of their citizenship status or relationship to a nation state. Importantly, the field of human security allows us to situate our understanding of modern slavery — a complex crime that is both a cause and a symptom of many other global problems such as pandemics, environmental disasters, conflict, and financial crises — within the larger discourse on vulnerability and to ensure that we were not missing significant dimensions of vulnerability to modern slavery. The use of human security theory also emphasises the global importance of the Sustainable Development Goals and links our vulnerability theory and modelling exercises to the developing global discussion on common metrics and goals for international development. Finally, this approach allows for the inclusion and exclusion of variables to be grounded in theory.

Review of the 2018 vulnerability model

Developing the 2018 vulnerability model involved selection of variables based on the human security theory that were published and updated regularly, were transparent about their methodology and source of the data and were the product of a methodology that did not suffer significant limitations that would impact the reliability of the data. A total of 35 variables were collated and, where necessary, normalised to a 1-100 scale and inverted so that higher numbers represented greater vulnerability. The 35 variables were then tested for collinearity, with any variables with a variance inflation factor above 10 and tolerance score below 0.1 dropped from the model. Principal factor analysis was then performed on the 24 variables which were retained. This reduced the number of variables to 23 and grouped them into five factors. The next step involved conceptualising the factors as distinct dimensions based on the final factor loadings and focused on risk of modern slavery. Imputation was used where threshold levels of missing data were met, and subregional averages imputed. Vulnerability scores were finally weighted by eigenvalues to give more weight to dimensions that have the most explanatory power in our overall vulnerability score. Quality assurance checks were performed on data transcription and calculations. A detailed description of the 2018 methodology can be found in Appendix 1 of the 2018 Global Slavery Index.

Data collation

The 23 variables used in the 2018 vulnerability model were assessed for updated data and changes to methodology. Where updated data was available, the most recent version was used, as of 31 December 2021. Six variables were discontinued or covered too few countries. Substitute variables were identified through a literature search and selected based on theoretical and empirical similarity to maintain comparability to the 2018 model. The discontinued variables and their substitutes are listed in Table 3. Table 4 presents the placement of variables within dimensions in the vulnerability model.

Table 3: Variables replaced with substitutes in the 2023 vulnerability model

Table 4: 2023 vulnerability model with new variables indicated in italics

Impact of changes to variables on total scores and ranks

The substitution of discontinued variables with theoretically and empirically similar variables had the effect of increasing data availability for some countries which, in addition to methodological differences between the 2018 variable and its 2023 substitute, led to indicator and dimension level changes for some countries that do not necessarily reflect actual changes in vulnerability. Similarly, changes to methodology for three of the variables used in the 2018 vulnerability model saw a reduction in data availability for those variables and thus for some countries. This meant that there were some countries for which imputation was required whereas for others, increased data meant that averages were no longer imputed. This led to significant shifts in scores and ranks for some countries. For example, Papua New Guinea previously benefitted from imputed regional scores based on Australia and New Zealand, however more complete data in 2023 meant imputation was no longer required and led to an increase in overall vulnerability relative to 2018.

Data preparation and analysis

Updated data was cleaned and analysed using Microsoft Excel. The variable Internally Displaced Persons was calculated by summing four different datasets sourced from UNHCR: Internally Displaced Persons (IDPs), New Asylum Applications, Returned IDPs, and Returned Refugees. Internally Displaced Persons was then logarithmically transformed. Where necessary, variables were normalised to a 1-100 scale and inverted so that higher numbers represented greater vulnerability. For more information on normalisation and inversion of variables, please download the 2023 vulnerability data and codebook from the Global Slavery Index 2023 downloads page.

To address issues relating to missing data, the proportion of missing data was calculated at the dimension level for each country. For dimensions Inequality, Disenfranchised Groups, and Effects of Conflict, the threshold was 50 per cent. For dimensions Governance Issues and Lack of Basic Needs, the threshold for missing data was 51 per cent due to the larger number of total vulnerability variables included in the first two dimensions of vulnerability. Subregional averages were imputed for variables within a dimension where the proportion of missing data met the threshold for that dimension. Exceptions to this approach were made in order to maintain variability within the regions where some data may have been more limited and are described in Table 5. Dimension averages were then calculated for each dimension and weighted by the factor eigenvalue determined in development of the 2018 vulnerability model. Weighting by eigenvalue is performed to give more weight to factors that have the most explanatory power in our overall vulnerability score. That is, the factors are not equal, and eigenvalues indicate the amount of variance explained by a certain factor. Factors with greater eigenvalues explain more of the overall model and have thus been weighted accordingly in the overall score which was then calculated by averaging the eigenvalue-weighted dimension scores and was normalised so to scale from 1-100. As a final step, quality assurance checks were performed by external consultants to ensure that no errors were made in the transcription from original sources, nor in calculations made in Excel.

Table 5: Exceptions to the general approach to missing data

Limitations

The vulnerability model should be interpreted with the following limitations in mind:

1. How well the vulnerability variables measure the real-world phenomena they are approximating in our model is limited by the need to select variables that cover most of our 160 countries, are published regularly, and explain clearly how these measures were developed.
2. The lag in administrative data reflecting real world situations on the ground affected the quality of the vulnerability model, as even the most recent information may still not reflect current situations on the ground at this moment.
3. Collinearity checks on our variables resulted in dropping several empirically redundant but conceptually important variables such as corruption, gender inequality, and environmental performance.
4. Data imputation ensured that missing data points were supplemented with regionally specific trends and information on affected vulnerability variables. However, imputed values are unlikely to be the true values for those countries.
5. Comparability between the 2023 vulnerability model and the 2018 vulnerability model is reduced by the substitution of discontinued variables with variables that are conceptually and empirically similar, which nonetheless differ in how they measure the real-life phenomena they approximate or measure slightly different phenomena.

Appendix Part C: Government responses

Governments play a critical role in the developing and implementing the laws, policies, and programs that are needed to prevent and respond to modern slavery. To complement the prevalence estimates and assessment of vulnerability, as with previous editions, the GSI includes an assessment of the actions governments are taking to respond to modern slavery.

This assessment is based on tracking government progress towards the achievement of five milestones:

1. Survivors of slavery are identified and supported to exit and remain out of modern slavery.

2. Criminal justice mechanisms function effectively to prevent modern slavery.

3. Coordination occurs at the national and regional level and across borders, and governments are held to account for their response.

4. Risk factors, such as attitudes, social systems, and institutions that enable modern slavery are addressed.

5. Government and business stop sourcing goods and services produced by forced labour.

Theoretical framework: crime prevention theory

Our assessment of government responses is underpinned by situational crime prevention theory (Figure 3).9 This is based on the understanding that in order for the crime of modern slavery to occur, there needs to be a vulnerable victim, a motivated offender, and the absence of a capable guardian. It also recognises that crime does not happen in a vacuum and that broad contextual factors like state instability, discrimination, and disregard of human rights are critical to any government response.

Figure 3: Situational crime prevention theory

Therefore, to reduce the prevalence of modern slavery crimes, governments need to:

• Reduce the opportunity for offenders to commit the crime.

• Increase the risks of offending.

• Decrease the vulnerability of potential victims.

• Increase the capacity of law enforcement and other guardians.

• Address the people or factors that stimulate or facilitate slavery.

Development of the conceptual framework

Using this theoretical framework as a starting point and drawing on the UN Trafficking Protocol10 and the European Convention on Action against Trafficking Beings,11 as well as literature on effective responses to modern slavery,12 we first devised (in 2014) a conceptual framework of what constitutes a strong response to modern slavery. We further refined this conceptual framework in consultation with our independent Expert Working Group and scholars in fields related to modern slavery, such as harmful traditional practices, health, social welfare, and migration. It was organised around the five milestones outlined above, which, if achieved, would ensure that governments are taking sufficient steps to address modern slavery, and underpinned our assessment of government responses in the 2014, 2016, and 2018 Global Slavery Indices.

Updating the conceptual framework

After the publication of the 2018 Global Slavery Index and following consultations with several stakeholders from the end of 2018 until data collection began in late 2020, further refinements were made to the conceptual framework in advance of the 2023 Global Slavery Index. After three editions of the GSI, we needed to keep apace of legislative and policy developments and consult with those with lived experience to review and strengthen our framework.

To this end, we held two workshops with our independent Expert Working Group to identify gaps, enhance the ability to measure outputs, and refine the ways issues such as harmful traditional practices were incorporated in the conceptual framework. The Expert Working Group also provided guidance on the processes of data collection and the weighting of milestones. Broadly, the rounds of review with the Expert Working Group culminated in the inclusion of indicators that assess concrete outputs as opposed to processes and of more indicators that focused on labour rights and child rights, as well as aligning our understanding of underlying risk factors and how to address these with international conventions and grey literature.

The need to include survivors in developing solutions to modern slavery has long been identified but rarely actioned.13 To systematically embed survivor perspectives in the assessment of government responses, Walk Free partnered with Survivor Alliance and NGO partners to convene more than 50 survivors in four Lived Experience Expert Groups in Ghana, India, Kenya, and the United Kingdom.

Survivors were asked to share their perspectives on the actions governments should take to end modern slavery, to rank the milestones from most to least important, and any other feedback. The vulnerability of women and children to modern slavery was emphasized across the Lived Experience Expert Groups, as well as the need for sensitisation campaigns to raise awareness of the risks of modern slavery, and for the provision of proactive social safety nets as part of prevention activities. Depending on location, the ranking of which milestone was most important differed: Milestone 1 (UK and Ghana), Milestone 4 (Kenya), and Milestone 5 (India) (see Table 6). Interestingly, despite several groups noting the need for laws to criminalise modern slavery and corruption, no group rated Milestone 2 as the most important part of a government’s response.

Table 6: Lived Experience Expert Groups ranking of government response milestones (ranked from most important (1) to least important (5))

Based on the consultations with the Lived Experience Expert Groups, indicators within the conceptual framework were edited to take into account survivor’s assessments of essential services. This included ensuring indicators better measure government action on registering recruitment agencies, ensuring that survivor compensation is delivered in practice, and ensuring that National Referral Mechanisms include providing survivors with the right to work. All survivor groups emphasised the importance of involving survivors in the development of policy and advocacy; in the UK and India, survivors suggested they should be incorporated in the data collection process to overcome data gaps for output indicators. Although this did not form part of the data collection process for the purposes of this report (discussed in greater detail below), Walk Free will consider how to further involve survivors in future rounds of data collection analysing government responses.

In cases where feedback received from the Expert Working Group or the Lived Experience Expert Groups was not incorporated into the conceptual framework, the proposed change either was already captured under another indicator within the framework, there was limited data available, or the suggested indicator was not conceptually consistent. For example, the suggestion to include a specific indicator to capture whether governments had publicly committed to using a survivor-centred approach was not included as it was considered preferable from a conceptual perspective to examine whether a government had streamlined a survivor-centred approach throughout their entire response, rather than simply making a public declaration. To test the robustness and availability of data for suggested indicators, we worked with Regenesys BPO, an offshore ethical sourcing business which employs survivors of modern slavery in the technology sector. From the review of indicators and this round of testing, we excluded 54 proposed indicators where suggestions were not conceptually consistent with the current framework or where data was not available consistently and at the level required for data collection. This included some of the indicators that measured “outputs” as opposed to existence of policies and processes. We will continue to review these suggested indicators to potentially incorporate in the next round of data collection. For examples of indicators excluded in this round, refer to Table 7.

Table 7: Example of suggested indicators which cannot be included in the current conceptual framework due to conceptual or data gaps

The Expert Working Group and the Lived Experience Expert Groups identified a greater need to reflect the rights of the child and our understanding of how governments should address underlying risk factors in the conceptual framework. As such, an internal review was conducted to map the conceptual framework to international rights instruments such as the United Nations Convention on the Rights of the Child and Guidelines on International Protection No. 8.14 The purpose of this mapping exercise was to ensure all relevant standards are reflected in the conceptual framework. The conceptual framework was also compared to areas of vulnerability identified by the Alliance 8.7 Migration Action Group15 and joint research by Walk Free and the International Organization for Migration16 to ensure sites of vulnerability, victim characteristics, and guardian responses are all reflected within the framework. Finally, the conceptual framework was mapped against the vulnerability model developed by Walk Free, as discussed in Part B, to ensure vulnerability was comprehensively represented in the conceptual framework.

The full updated conceptual framework can be found at the end of this section.

Process of assessing government responses to modern slavery

Throughout 2020 and 2021, data was collected for 176 countries for the government response component of the GSI. As in previous editions of the Global Slavery Index, this included data on Commonwealth countries17 in our government response database. Where data is limited18 in smaller Commonwealth nations, we have excluded these countries from our analysis. Due to the ongoing conflict and extreme disruption to government, we have excluded ratings for Afghanistan, South Sudan, Syria, and Yemen this year.19

The conceptual framework underpinning our assessment is arranged around the five milestones, which are then broken down into activities, which are further disaggregated into indicators. There are a total of 141 indicators in the conceptual framework and 42 activities. This included five additional indicators that were developed by the Commonwealth Human Rights Initiative (CHRI) in 2018 and are referenced in a report released by Walk Free and CHRI in 2020 that assessed the modern slavery responses of Commonwealth governments. These additional indicators are:

1. ILO Forced Labour Convention, 1930 (No. 29) is ratified.

2. Protocol Against the Smuggling of Migrants by Land, Sea and Air (2000) is ratified.

3. NEGATIVE: Certain groups, such as migrant workers or domestic workers, are not able to unionise.

4. NEGATIVE: There are lower primary school enrolment rates for girls.

5. NEGATIVE: Homosexuality is criminalised.

The breakdown by milestone is described in Table 8 below.

Table 8: Breakdown of milestones into activities and indicators

Taiwan and Kosovo have 41 activities, not 42, as they are unable to ratify international conventions.

Data collection

Data is collected at the indicator level, where each indicator describes an element of an activity. Take Activity 1.2 under Milestone 1, “Comprehensive reporting mechanisms operate effectively” as an example, set out in Table 9.

Table 9: Activity 1.2, Milestone 1

* These indicators were added to the conceptual framework following consultation and review with the Expert Working Group and the Lived Experience Expert Groups. However, data was not collected against them due to limited available sources. These indicators are “aspirational” and will be retained in the conceptual framework for future rounds of data collection and analysis.

There are seven indicators under this activity, each of which determine the existence of the reporting mechanism, and how well it is operating. Desk research was conducted for five of these indicators and others in the conceptual framework by a team of 17 researchers and research assistants following a strict protocol. Through AnnieCannons, a non-profit that provides technology and software focused vocational training to survivors of modern slavery, and Regenesys BPO, 12 researchers were survivors of modern slavery. Specific protocols were developed on an indicator-level for these researchers to provide further technical support and to ensure consistency in data collection across the global research team. All protocols described both the types of reports and sources to be reviewed and what constitutes “relevant” information. The multilingual global team20 conducted research either by country or by indicator and saved these references in the government response database.21

These data points were then verified, as far as possible, by NGO contacts within each country. NGOs were given the opportunity to either respond via email, hold Zoom interviews, or complete a survey. Over 25 survey responses were received, and a further 51 NGOs responded to individual requests for information via email or Zoom calls.

Data is current as of 31^st August 2022.

Rating

The 2023 scoring system, which is based on activities, has not been used in previous editions of the GSI. Activities are used to clearly communicate the actions a government is taking, or not taking, to eradicate modern slavery in its simplest form.

Of the 141 indicators in the conceptual framework against which data was collected, 125 are what we have called “positive indicators.” Put simply, these cover the actions the government is taking to implement each activity and ultimately each milestone. The indicators described under Table 8 above are all positive indicators. In some instances, they go beyond implementation to measure effectiveness of an activity (There has been an increase in number of victims being identified through the hotline in the above example).

These indicators are supplemented by 16 standardised “negative indicators,” which attempt to measure implementation of a particular activity (listed in Table 10). For example, if shelters exist for modern slavery victims, the negative indicator victims are held in shelters against their will and do not have a choice about whether or not to remain in a shelter would capture whether victims are detained and experience secondary victimisation despite the existence of these shelters. The negative indicators also cover broader factors, which, if conducted by governments, would increase the risk of human trafficking and child exploitation. These include state-sanctioned forced labour, high levels of government complicity, criminalisation of victims, deportation of potential victims, and policies that tie migrant workers to their employers.

Table 10: Example of negative implementation indicators, Activity 2.1, Milestone 1

Once data had been collected and verified, each indicator was scored on a 0 to 1 scale. On this scale, 0 meant no information was identified or available, or information explicitly demonstrated that the government did not meet any indicators, and 1 meant that the indicator had been met. Negative indicators were scored on a 0 to -1 scale, where 0 meant no information was identified or available, or information explicitly demonstrated that the government did not meet any indicators, and -1 meant that the indicator had been met. As an advocacy tool, we have retained our rating where no information found is rated as “0.” However, we have identified indicators within the government response database that have consistently had no information found since 2014. In future rounds of data collection, we plan to prioritise testing these indicators to ascertain if no information is available, whereupon we will ultimately remove the indicators for future rounds of data collection. As part of the data collection that informed this assessment of government responses, an indicator that assessed whether evaluations of anti-slavery projects were provided to government officials to inform their future programming was removed from data collection due to consistent gaps in available information (Milestone 1, Indicator 3.3.3).

The data and ratings then underwent a series of quality assurance — first by country, where each country was reviewed against the rating descriptions to determine if ratings were sound. Secondly, following data collection being completed, each indicator was reviewed across all countries to check for consistency in the applied logic. Any final edits were then reviewed, and final edits made in the database.

The data was then imported into STATA, where indicators were grouped into activities and scored. Indicators are organised into 42 activities, which are given a score from -2 to 2, dependent on the number of indicators met and the nature of those indicators. Activity scores are outlined in Table 11.

Table 11: Scoring activities

The three negative rated activities in Table 12 are subtracted from the overall score as they typically involve systemic issues that significantly undermine the veracity of the entire government response to modern slavery, rather than increasing vulnerability. For example, these systemic issues are represented by indicators relating to corruption and complicity, the use of state-imposed forced labour and when a government has identified no victims of modern slavery, which indicates that any systems in place to support survivors in the country are ineffective.

Table 12: 16 Negative indicators, and three negative activities in which the government undermine their own response to modern slavery

Activity scores are summed to give a total score for each milestone. Milestone scores are then summed to give a total government response score out of 78.

Each activity is weighted equally so that a country can only obtain a total of 78 points, noting that the final score is presented as a percentage. This does lead to an implicit weighting of milestones, where the more activities in a milestone, the more weight it is given. Table 13 describes the milestone weightings below.

Table 13: Implicit weighting of each milestone

Limitations

Collecting data for 141 indicators across 176 countries is a complicated undertaking. Access to data is limited for all indicators where information is not available publicly or available in languages spoken by the research team. The continued absence of Arabic speakers prevented verification with NGOs in countries where these are the primary languages spoken. Limits also remain in measuring the implementation of a response — while the negative indicators and NGO verification are the first steps in measuring this, more remains to be done in getting at the reality of what is occurring on the ground as opposed to what is reported publicly.

Comparability to previous assessments

The government response assessment is broadly comparable with previous iterations of the GSI (Table 14). However, due to the significant changes in both the method of analysis and the conceptual framework, this is limited to comparisons at the indicator rating level rather than comparisons against milestone or total scores.

In 2023, the method of calculating total scores for milestones was updated to better reflect a government’s commitment to completely achieving an activity. Percentage scores for milestones were calculated based on the achievement of activities rather than the number of met indicators as has been done in the past. While the data remains comparable at the indicator level, comparing milestone percentage scores between 2018 and 2023 will be indicative of both changes made to the scoring system and actual government improvement. When seeking government improvement, it is most accurate to compare indicator ratings between 2018 and 2023.

The comprehensive consultations and review of the 2018 conceptual framework led to the addition of 43 indicators and removal of 49 indicators across all five milestones in the conceptual framework. Additionally, some indicators have been moved to different activities in different milestones after a final review of the framework and the shift to activity-based scoring.

Table 14: 2018 rankings when considered under the 2023 framework

Other changes at the indicator level involved edits to the existing indicators rather than removing or introducing new indicators: however, across milestones 1, 2, and 3 only 11 indicators were substantially changed. No substantial changes were made to indicators in either milestone 4 and 5. As a result of the additions, removals, and edits there was a related change in the categorisation of indicators at the activity level. Most changes were concentrated in Milestone 2 and were largely the result of regrouping the criminal justice indicators into activities responding to specific types of modern slavery or to support particular vulnerable groups. In 2018, all conventions were grouped into one activity and all criminal justice provisions into a separate activity. In 2023, we organised these by type of modern slavery, so all conventions and domestic legislation related to trafficking were grouped into one activity. We also split out the implementation in legislation versus implementation in practice related to access to justice mechanisms.

Table 15: Conceptual framework for measuring government responses

Methodology for identifying the highest value products at risk of forced labour imported by the G20

The world’s most developed countries are connected to modern slavery not only through exploitation occurring within their own borders but also through the goods they import. In the GSI, we identify the highest value products at risk of being produced by forced labour which are currently being imported into G20 countries. The G20 are the largest importers (and exporters) in the world, accounting for 75 per cent of global trade.22

As a first step we developed a list of products at risk of being produced by modern slavery. This was informed by high-risk countries and industries as well as recent suspected cases of forced labour identified in the production of these goods.23 We then compiled import data for all G20 countries for these products.

Identifying a list of imports at risk of modern slavery

Our starting point was the 2022 US Department of Labor list of goods produced by forced labour and child labour.24 The list was first filtered by “forced labour” to ensure that products suspected of being produced only by child labour were excluded. A simple country count of products was performed to determine a ranking: the product with the highest number of countries listed against it was ranked first, the product with the second highest numbers of countries against it was ranked second, and so on. This produced an initial list of 22 product/source country combinations at risk of modern slavery.

Table 16: Initial list of goods produced by forced labour as reported by the US Department of Labor

As a next step, we conducted a literature review of the product/source country combinations to independently validate the list, using the following parameters: 

• Reference period: 1 January 2017 to 31 July 2022.

• Mix of media and non-media sources (peer-reviewed journal articles, research reports, government documents, international organisation reports, NGO reports, etc.), whenever possible.

The following hierarchy of sources was used in conducting this research:

1. Peer reviewed publications, e.g., articles from journals identified through database searches and, if required, through Google Scholar. 

2. Reports of international organisations, e.g., ILO, IOM, UN.

3. Reports of international NGOs, e.g., Human Rights Watch, Amnesty International.

4. Government websites, e.g., Ministry of Foreign Affairs.

5. National NGOs.

6. Media, through Google searches.

It should be noted that this list is not exhaustive, and we performed additional searches where suggested sources did not provide the information required.  

Once the literature review was completed, a product/source country combination was included if it was on the 2022 US Department of Labor list of goods produced by forced labour that are listed in Table 16. In addition, the product/source country combination had to be independently verified by credible secondary sources, such as journal articles, primary research reports, reports from an international organisation or an NGO, or media reports. If no relevant references were found or the information was more than five years old, the product/source country combination was excluded.

The literature review resulted in the final list of at-risk products seen in Table 17. Source countries marked in red were deleted from the list as we could not find recent evidence to verify the occurrence of forced labour. The countries marked in green were added to the final list based on Walk Free primary research into modern slavery in the cocoa sector in Ghana25 and well-known exploitation occurring in the products of solar panels in China.26

Table 17: Final list of products at risk of modern slavery by source country.

Identifying the most valuable imported products at risk of modern slavery

Trade data was obtained for the 19 G20 member countries. South Africa was included in this analysis for the first time: in 2018 it was excluded as it reported trade data via the Southern African Customs Union. The European Union was excluded to avoid double counting trade data from France, Germany, and Italy.

The final list of G20 countries are:

1. Argentina

2. Australia

3. Brazil

4. Canada

5. China

6. France

7. Germany

8. India

9. Indonesia

10. Italy

11. Japan

12. Mexico

13. Russia

14. Saudi Arabia

15. South Africa

16. South Korea

17. Türkiye

18. United Kingdom

19. United States

Data source and definitions

BACI dataset 

The import data used for this analysis was taken from the BACI dataset.27 BACI is the world trade database developed by the French research centre Centre d’Études Prospectives et d’Informations Internationales (CEPII) at a high level of product disaggregation.

Original trade data is provided by the United Nations Statistical Division (COMTRADE database). BACI is constructed using a procedure that reconciles the declarations of the exporter and the importer. This harmonisation procedure enables the extension of the number of countries for which trade data is available. The dataset gives information about the value of trade (in thousands of US dollars) and the quantity (in tonnes).

For this project, we used the 2021 BACI trade dataset with the 2017 Harmonized System (HS) nomenclature, which was the most recent available at the time of writing.

Harmonized Commodity Description and Coding System 

The Harmonized System (HS) is an international nomenclature for the classification of products. It allows participating countries to classify traded goods on a common basis for customs purposes. At the international level, the HS for classifying goods is a six-digit code system.

The HS comprises approximately 5,300 product descriptions that appear as headings and subheadings, arranged in 99 chapters, and grouped into 21 sections. The six digits can be broken down into three parts. The first two digits (HS-2) identify the chapter the goods are classified in, e.g. 09 = Coffee, Tea, Maté and Spices. The next two digits (HS-4) identify groupings within that chapter, e.g. 09.02 = Tea, whether or not flavoured. The next two digits (HS-6) are even more specific, e.g. 09.02.10 = Green tea (not fermented). Up to the HS-6 digit level, most countries classify products in the same way (a few exceptions exist where some countries apply old versions of the HS).

The HS was introduced in 1988 and has been adopted by most countries worldwide. It has undergone several revisions in the classification of products, which entered into force in 1996, 2002, 2007, 2012, and 2017.

Data compilation

Each of the products from the final list in Table 17 is represented by multiple HS 2017 product categories within the BACI trade dataset. The relevant categories were identified using the Foreign Trade Online directory.28 Using STATA, import data for all relevant product categories and source countries was extracted from the 2021 BACI dataset for 19 G20 countries.

The products were then ranked from highest to lowest according to import value in US$. The resulting list of top five products at risk of modern slavery (according to US$ value) imported by each of the G20 countries is presented in Table 18. This product list has changed between 2018 and 2023. Textiles, palm oil, and coffee were not included in the top five most valuable products at risk of modern slavery of any G20 country in 2018. Cotton and carpets remain products at risk of modern slavery, however they have dropped off the highest value list since 2018.

Table 18: Top five products at risk of modern slavery according to US$ value imported by G20 countries29