Why measure vulnerability?

Understanding a problem is fundamental to being able to respond effectively and efficiently to it. This is particularly true for crimes that are as complex as modern slavery. Data that enable us to understand the systemic, individual, and environmental factors that enable modern slavery to occur are critical to being able to design effective preventative measures, and also to being able to better determine where modern slavery may be occurring completely out of sight, within “blind spots.”

Complementing the prevalence estimates, the Vulnerability Model is designed to enable us to identify and better understand the potential drivers of this crime. The existing literature and expert input suggests a connection between modern slavery and related systemic factors such as corruption,1 conflict,2 and adverse environmental change3 among many other vulnerability factors. While this evidence provides an important starting point, without measurement to better understand relationships and interactions between these factors, we cannot understand their significance. To provide a reliable evidence base upon which governments, civil society groups, and businesses can build more effective responses, a statistical approach to identifying the factors that are correlated with increased risk of enslavement has been applied.4 In other words, the Vulnerability Model uses statistical testing and processes to identify the factors that explain or predict the prevalence of modern slavery. Reflecting the limits of existing data (particularly on prevalence but also on key variables), the Vulnerability Model is necessarily in the early stages of development and, as such, it should be viewed as iterative. Nonetheless, the Vulnerability Model provides an important resource to better understand and predict where modern slavery is most likely to occur based on our present best available data.

The 2018 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 methodology that was used to develop the Vulnerability Model is explained in this section. It includes, initially, steps taken in the development of the Vulnerability Model. As this drew upon recent Expert Working Group consultations and a review process, we briefly describe this process and note the decisions and changes that have been undertaken as a result of that review in implementing the methodology. Finally, this section provides a summary of the factors and variables that comprise the final 2018 Vulnerability Model.

Development of the Vulnerability Model

Theoretical framework

The Vulnerability Model is guided by human security and crime prevention theories. Human security as a developing security sub-field has many overlapping and diverging definitions without any clear “consensual definition”5 among scholars. 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 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 (SDGs) 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, while remaining an empirically exploratory approach.6

The current Vulnerability Model

The 2018 GSI includes an assessment of vulnerability that is used to measure the factors linked to the risk of modern slavery in each country. The importance of this work is twofold:

  1. To improve our understanding of the drivers of modern slavery through quantification such that we can assess changes in these drivers, and therefore in rates of prevalence, over time; 
  2. It provides important data that are used to arrive at estimates in countries for which no reliable, national-level data exist.

The major refinements made since the 2016 Global Slavery Index and the process by which these decisions were arrived at, are set out below.

Overview of 2018 Vulnerability Model development

The Vulnerability Model development process included the following phases:

  1. Review of 2016 Vulnerability Model
  2. Data Collation
  3. Data Preparation (Normalisation, Inversion, and Logarithmic Transformation of certain variables including Refugees, Internally Displaced Persons, and GDP (PPP))
  4. Collinearity Checks (dropped if Variance Inflation Factor (VIF) above 10 and Tolerance below 0.1)
  5. Principal Factor Analysis
  6. Final Factor Loadings and Placements
  7. Missing Data Solutions
  8. Eigenvalue Weighting by Factor. Throughout this process, the major decision points and a summary of the statistical team’s determinations are captured for transparency
  9. Quality Assurance Checks

Phase 1. Review of 2016 Vulnerability Model

After an internal review of the 2016 Vulnerability Model, our Expert Working Group members were consulted between August and December 2016 regarding the areas for improvement that had been identified. That feedback was then summarised and a second round of consultations took place in October and November 2017.

We sought feedback on the following areas, and below each topic is a summary of key feedback received:

Theoretical and Empirical Gaps

Generally, our experts maintained the importance for continuity and did not identify significant gaps in our model that we had not already attempted to address through sufficient alternative data sources.

Generally, experts were supportive of the use of human security theory, but desired further elaboration on how crime prevention theory was formally utilised. This highlighted the need for articulation and finalisation of a generalisable theory related to determinants of slavery, which will be dealt with in a forthcoming publication by Joudo Larsen and Durgana. 

Normalisation and Standardisation

Experts recommended that we consider different approaches to determine the overall final data transformation method for the 2018 Vulnerability Model. Some suggested that we employ statistical standardisation with a mean of 0 and a standard deviation of 1. Others cautioned us to consider this question both philosophically (do we believe that these variables are normally distributed?) and empirically (how significant to our work are the outlier figures?). As it stands, our current normalisation process features outliers prominently in our calculations, while statistical standardisation of our variables would collapse/lose these elements.

Missing Data

Experts recommended that we consider alternative forms of imputation that would allow us to avoid dimension level imputation by employing either country vulnerability averages (as we have done) or using regional averages at the variable level per affected country.

Weighting by Eigenvalue

Experts considered and supported the issue of weighting the factors by eigenvalues. Even though weighting by eigenvalue presents a slight change to our traditional vulnerability range beyond a 100-point scale, weighted values can be (and ultimately were) normalised on a 1-100 scale.

The actions taken as a result of these recommendations and the final decisions made are summarised in the relevant sections of the process, as set out below.

Phase 2. Data collation

Data requirements for model inclusion

In 2016 and 2017, we reviewed the Vulnerability Model, taking account of human security theory, and considering issues related to data quality, availability, and limitations. Key reasons for adding or removing variables from the model include:

  • To ensure continual availability of data – data that were irregularly published and updated, or lacked transparency about original data source, were removed.
  • To ensure we get as close to the source of the data as possible: for example, using original source data rather than composite scores from other indices.
  • To replace weaker measures with potentially stronger variables.
  • To address conceptual gaps in our framework and model.

We collated all tested vulnerability data (35 variables listed below) for the reference period ending on 15 April 2017. This list of variables includes some that were added following the expert review; data on Environmental Performance Index were added and data from the Gender Inequality Index (which had been in the original 2014 Vulnerability Model but dropped in the 2016 Vulnerability Model for reasons of collinearity with other variables) were added for re-testing. A further change from 2016 was the exclusion of “Internet usage” due to cessation of data collection on that variable by the World Bank.

The final list of tested vulnerability variables is as follows:

  1. Political Rights
  2. Civil Rights
  3. Financial Inclusion – Received Wages
  4. Literacy
  5. Child Mortality
  6. Corruption
  7. Alternative Social Safety Net measure
  8. GDP (PPP)
  9. Government Effectiveness
  10. Gender Inequality Index
  11. Environmental Performance Index (EPI)
  12. Financial Inclusion – Ability to Borrow Money
  13. Financial Inclusion – Ability to Obtain Emergency Funds
  14. Cell Phone Users
  15. Social Safety Net
  16. Undernourishment
  17. Access to Clean Water
  18. Tuberculosis
  19. Confidence in Judicial Systems
  20. Political Instability
  21. Impact of Terrorism
  22. Internal Conflicts Fought
  23. Violent Crime
  24. Women’s Physical Security
  25. Weapons Access
  26. GINI Coefficient
  27. Same Sex Rights
  28. Disabled Rights
  29. Acceptance of Immigrants
  30. Acceptance of Minorities7
  31. Global Slavery Index Government Response
  32. Alternative Political Rights measure
  33. Regulatory Quality
  34. Internally Displaced Persons
  35. Refugees 

Phase 3. Data preparation - highlights

As recommended by the Expert Working Group, both methods of standardising and normalising the data were tested and evaluated. We determined that normalisation would be retained for its ease of use and 1-100 scale, particularly in aggregation with the other components of the Index. There were also conceptual concerns about forcing standardisation on these variables, given many of them could not be assumed to have a normal distribution. The standardised range of values was much closer and also resulted in negative values. This would have presented a challenge in terms of our prior approaches to vulnerability values and scores and would not have been as intuitive to our policy audience as our existing normalisation scales.

Normalisation

The following variables were normalised using the normalisation formula below: Political Rights, Civil Rights, Cell Phone Users, Social Safety Net, Child Mortality, Tuberculosis, Political Instability, Impact of Terrorism, Internal Conflicts Fought, Violent Crime, Women's Physical Security, Weapons Access, GSI Government Response, Alternative Social Safety Net measure, Alternative Political Rights Variable, Government Effectiveness, Regulatory Quality, and Gender Inequality Index (GII).

Normalisation formula:

Normalisation: Normalised Value = 1+(Reported Value – minimum value)*(100-1)/(maximum value – minimum value)

Inversion formula for normalised variables:

101-normalised value = Inverted Value

Certain selected variables were inverted to ensure that a high value indicates higher vulnerability on every variable. The variables affected are: Cell Phone Users, Literacy, Social Safety Net, Access to Clean Water, Corruption, Global Slavery Index Government Response, Alternative Political Rights, Government Effectiveness, Regulatory Quality, and Environmental Performance Index.

Phase 4. Collinearity testing and results

Collinearity among the vulnerability variables was assessed to identify where variables are already highly correlated. The collinearity results for any pairs of variables with values above 0.80 are reported in Table 1.

Table 1Collinearity results for pairs of variables with values above 0.80
Variable 1
Variable 2
Collinearity Result
Child MortalityAccess to Water0.8008
Child MortalityAlt. Social Safety net0.8143
Child MortalityLiteracy0.8040
CorruptionGovernment Effectiveness0.9371
CorruptionRegulatory Quality0.9048
Government EffectivenessPolitical Instability0.8089
Alt. Social Safety NetGender Inequality Index0.8277
Alt. Social Safety NetEnvironmental Performance Index0.8794
CorruptionPolitical Instability0.8063
Civil LibertiesPolitical Rights0.9435
Political RightsAlt. Political Rights Measure0.8536
Alt. Political Rights MeasureCivil Liberties0.8261
Civil LibertiesPolitical Instability0.8063
Alt. Social Safety NetFinancial Inclusion – Received Wages0.8156
Financial Inclusion – Received WagesGovernment Effectiveness0.8333
Financial Inclusion – Received WagesGender Inequality Index0.8506
Financial Inclusion – Received WagesEnvironmental Performance Index0.8097
Child MortalityGender Inequality Index0.8048
Government EffectivenessRegulatory Quality0.9377
Government EffectivenessGender Inequality Index0.8072
Gender Inequality IndexEnvironmental Performance Index0.8226
Government EffectivenessGDP (PPP)0.8236
Gender Inequality IndexGDP (PPP)0.8074
Environmental Performance IndexGDP (PPP)0.8253

Our Experts had previously recommended that any variables with VIF scores above 10 and Tolerance scores below 0.1 would be dropped from the model, and we followed this approach. Despite the conceptual gaps that were potentially addressed by their inclusion in the model, the Gender Inequality Index and Environmental Performance Index variables suggested by our Experts were ultimately dropped due to high collinearity with other vulnerability measures, suggesting a degree of redundancy in their explanatory power within the model given existing variables. A full list of variables dropped from the model is presented in Table 2.

Table 2Variables Dropped from Model following collinearity check on normalised data (with VIF and Tolerance scores)
Variable
VIF
Tolerance
Political Rights817.890.0559
Civil Rights922.910.0436
Financial Inclusion – Received Wages1010.360.0966
Literacy1113.360.0749
Child Mortality1213.030.0768
Corruption1310.880.0919
Alt. Social Safety Net1414.620.0684
GDP (PPP)1515.990.0625
Government Effectiveness1622.010.0454
Gender Inequality Index1720.050.0499
Environmental Performance Index1818.120.0552

Following edits to the list reflecting the collinearity checks described above, our final list of variables retained for factor analysis was as follows:

  1. Financial Inclusion – Ability to Borrow Money
  2. Financial Inclusion – Ability to Obtain Emergency Funds
  3. Cell Phone Users
  4. Social Safety Net
  5. Undernourishment
  6. Access to Clean Water
  7. Tuberculosis
  8. Confidence in Judicial Systems
  9. Political Instability
  10. Impact of Terrorism
  11. Internal Conflicts Fought
  12. Violent Crime
  13. Women’s Physical Security
  14. Weapons Access
  15. GINI Coefficient
  16. Same Sex Rights
  17. Disabled Rights
  18. Acceptance of Immigrants
  19. Acceptance of Minorities
  20. Global Slavery Index Government Response
  21. Alternative Political Rights Measure19
  22. Regulatory Quality
  23. Internally Displaced Persons
  24. Refugees

Phase 5: Principal factor analysis

Principal Factor Analysis or Factor Analysis is a statistical technique used to reduce the number of variables so that relationships between variables can be easily understood. It does so by regrouping variables into a limited set of clusters, with each cluster representing a latent construct that has not been directly measured (such as governance issues, inequality, etc.). Hence, it helps to isolate constructs and concepts from an array of many variables. Principal Factor Analysis typically retains all factors with eigenvalues scores over 1.0. A six-factor solution is naturally occurring with the following eigenvalues expressed (Table 3):

Table 3Initial Factor Analysis solution table
Factor
Variance
Difference
Proportion
Cumulative
Factor One5.680673.205180.34220.3422
Factor Two2.475490.537180.14910.4913
Factor Three1.938310.032900.11670.6080
Factor Four1.905410.072820.11480.7228
Factor Five1.832590.158460.11040.8332
Factor Six1.674140.169790.10080.9340

Cumulative

When a six-factor solution is forced in the factor analysis, the values change slightly to the following:

Table 4Six-factor solution table
Factor
Variance
Difference
Proportion
Cumulative
Factor One5.826753.472220.35100.3510
Factor Two2.354530.160670.14180.4928
Factor Three2.193850.117510.13210.6249
Factor Four2.076340.183380.12510.7500
Factor Five1.892970.008610.11400.8640
Factor Six1.88435
0.11350.9975

When a four-factor solution is forced in the factor analysis, the values are as follows:

Table 5Four-factor solution table
Factor
Variance
Difference
Proportion
Cumulative
Factor One5.524631.831180.33280.3328
Factor Two3.693451.065530.22250.5552
Factor Three2.627920.184740.15830.7135
Factor Four2.44318
0.14720.8607

A forced five-factor solution yields the following values:

Table 6Five-factor solution table
Factor
Variance
Difference
Proportion
Cumulative
Factor One5.761302.338480.34700.3470
Factor Two3.422821.189200.20620.5532
Factor Three2.233620.141570.13450.6877
Factor Four2.092050.153770.12600.8137
Factor Five1.93828
0.11670.9305

The five-factor model (Table 6) resulted in a consolidated second factor that closely matches the 2016 model’s factor loadings. On this basis, we decided to proceed with a five-factor solution.

Phase 6: Final factor loadings and placement

In the five-factor solution, the Factor Analysis variable loadings are as set out in Table 7.

Table 7Final factor loadings and placement table
Variable
Factor One
Factor Two
Factor Three
Factor Four
Factor Five
Uniqueness
Ability to Borrow Money
0.6194


0.5226
Ability to Obtain Emergency Funds

0.5324

0.5590
Cell Phone Users
0.5010


0.5959
Social Safety Net
0.7023


0.3238
Undernourishment
0.7377


0.2985
Access to Clean Water0.56250.6366


0.2753
Tuberculosis
0.6174


0.4603
Confidence in Judicial Systems

0.4174

0.5452
Political Instability0.8902



0.1806
Impact of Terrorism



0.81370.2805
Internal Conflicts Fought



0.71290.4397
Violent Crime0.5462
0.5980

0.2135
Women’s Physical Security0.6270



0.3277
Weapons Access0.7040
0.4533

0.2334
GINI Coefficient

0.7416

0.3165
Same Sex Rights0.6218

0.4467
0.2636
Disabled Rights0.5396



0.3219
Acceptance of Immigrants


0.8332
0.2960
Acceptance of Minorities


0.7414
0.4080
GSI Government Response0.6805



0.3622
Political Rights0.7576



0.3431
Regulatory Quality0.8436



0.1485
Internally Displaced Persons0.6976


0.45000.2368
Refugees




0.5995

We then started to conceptualise the factors as distinct dimensions based on the final factor loadings from Table 7. In consultation with our Expert Working Group, employed analytical frameworks focused on concept-variable consistency to help determine how closely empirical data or "measured concepts" match the phenomena they are meant to capture. This framework is employed not only in the selection of the vulnerability variables themselves, but then also their resulting role in the overall dimension and, consequently, its label. Further, the recommendation that latent factor construction be re-focused on risk to slavery, and not expressed as resilience, was also implemented when naming the dimensions. The results of this process are set out in Table 8, where the dimension headings are presented. Please note that the refugees variable has also been dropped as it does not load on any of the retained factors. 

Table 8Initial factor groupings by variables (final factor loading in bold, multiple loadings in italics)
Factor One (5.76 Eigen)
Governance Issues
Factor Two (3.422 Eigen)Lack of Basic Needs
Factor Three (2.233 Eigen)
Inequality
Factor Four (2.092 Eigen)
Disenfranchised Groups
Factor Five (1.938 Eigen)
Effects of Conflict
Political InstabilityCell Phone UsersAbility to Obtain Emergency FundsAcceptance of ImmigrantsImpact of Terrorism
GSI Government ResponseUndernourishment
Acceptance of MinoritiesInternal Conflicts Fought

Social Safety Net (0.7023)GINI Coefficient

Political RightsAbility to Borrow MoneyConfidence in Judicial Systems

Regulatory QualityTuberculosis


Access to Clean Water (0.5625)Access to Clean Water (0.6366)


Violent Crime (0.5462)
Violent Crime (0.5980)

Weapons Access (0.7040)
Weapons Access (0.4533)

Same Sex Rights (0.6218)

Same Sex Rights (0.4467)
Disabled Rights



Internally Displaced Persons (0.6976)


Internally Displaced Persons (0.4500)
Women’s Physical Security



With reference to the initial dimension headings presented in Table 8, decisions were then made regarding placement of variables, which loaded onto multiple dimensions (variables indicate the final placement and italicised variables indicate multiple loadings), and the dimension headings were refined. The final dimension headings and final placement of variables are set out in Table 9.

Table 9Final dimension headings and final variable placement
Factor One (5.76 Eigen)
Governance Issues
Factor Two (3.422 Eigen)
Lack of Basic Needs
Factor Three (2.233 Eigen)
Inequality
Factor Four (2.092 Eigen)
Disenfranchised Groups
Factor Five (1.938 Eigen)
Effects of Conflict
Political InstabilityCell Phone UsersAbility to Obtain Emergency FundsAcceptance of ImmigrantsImpact of Terrorism
GSI Government ResponseUndernourishmentViolent Crime (0.5980)Acceptance of MinoritiesInternal Conflicts Fought
Women’s Physical SecuritySocial Safety Net (0.7023)GINI CoefficientSame Sex Rights (0.4467)Internally Displaced Persons
(0.4500)
Political RightsAbility to Borrow MoneyConfidence in Judicial

Regulatory QualityTuberculosis


Disabled RightsAccess to Clean Water (0.6366)


Weapons Access (0.7040)



The following decisions were made on final dimension placements for variables that had multiple loadings (Table 10). As recommended by the Expert Working Group, these decisions were taken to ensure a level of conceptual clarity across the set of variables within each overall dimension. Table 10 also includes a brief explanation of the rationale behind the subsequent conceptualisation of each dimension.

Table 10Final Dimension Placement Rationales
VariablesDimension placement and rationale
WaterWater was placed in Factor Two (Lack of Basic Needs) due to conceptual consistency with other variables within the dimension as conceptualised (covering issues such as access to food and health) despite its slightly higher factor loading on Factor One.
Violent CrimeViolent Crime remains in Factor Three (Inequality) due to its higher factor loadings and greater conceptual clarity with other variables in that dimension as conceptualised. That is, this variable represents a qualitative assessment of the problems posed by violent crime for government and business, reflecting a government’s capacity to address crime. Violent crime often disproportionately affects individuals in a society, often consistent with other sociological markers of inequality.20
Weapons AccessWeapons Access remains in Factor One (Governance) due to its higher factor loadings and greater conceptual clarity within that dimension. That is, this variable represents a qualitative assessment of the ease of access to weapons, essentially reflecting legislation and regulatory requirements.
Same Sex RightsDespite the slightly higher factor loadings for Factor One, Same Sex Rights remains placed in Factor Four (Disenfranchised Groups) due to conceptual consistency with the other variables on Immigrants and Minorities in that dimension.
DisplacedDespite the slightly higher factor loadings for Factor One, Displaced remains in Factor Five (Effects of Conflict) alongside variables on refugees and impact of terrorism, for greater conceptual clarity within that dimension as conceptualised.

The Governance Issues dimension was constructed to represent elements of vulnerability strongly linked to government intervention and regulation. Both Weapons Access and Women's Physical Security fit within Governance Issues because they essentially measure a government's ability to provide for the safety of its population. The Women's Physical Security scale takes into account the presence and enforcement of laws against domestic violence, rape and marital rape, the existence of taboos or norms against reporting these crimes, and the occurrence of honour killings and femicide. The presence and enforcement of laws against domestic violence, rape, marital rape, and the comfort of the public in reporting these crimes and whether honour killings/femicide occur (basically if they can occur without penalty), all fit within Governance Issues as consistently defined with Government Response measures. Weapons Access is also a qualitative assessment of the ease of access to weapons, both small and light weapons, essentially reflecting government legislation and regulation requirements. Regulatory Quality evaluates the ability of governments to foster private sector development, and Political Instability measures how well a country’s political institutions can support the needs of its citizens, businesses, and overseas investors. Additionally, there is a strong rationale for including Disabled Rights in this dimension because some of the criteria for people to find an area a good place to live for those with intellectual abilities also has to do with government intervention on their behalf and overall legal protections for these populations.

The label “Lack of basic needs” was applied to Dimension Two upon consideration of the variables that loaded on this dimension and commonalities between them. The variables Undernourishment, Access to Clean Water, and Tuberculosis all reflect basic needs around healthcare and nourishment. The variables Social Safety Net, Ability to Borrow Money, and Cell Phone Users (a proxy measure of socio-economic capacity) are effectively measures of the ability to obtain necessary goods/services.

The Inequality dimension reflects developments from sociology that suggest that inequality is often a driving force behind populations that are disproportionately affected by violent crime and ability to access funds/emergency funds.21 The GINI Coefficient measure is a direct measure of financial inequality in a nation. Confidence in Judicial Systems can also be impacted by ability to access or pay for legal representation.

The Disenfranchised Groups dimension measures general acceptance of different racial and ethnic minority groups, immigrants, and same sex groups in a population.

The Effects of Conflict dimension measures impact of terrorism, internal conflicts fought, and internally displaced persons as manifestations of the effects of conflict globally.

Phase 7: Missing data solutions

In reviewing the approach we took to missing data in previous iterations of the Vulnerability Model, experts recommended that we consider alternative forms of imputation to avoid dimension level imputation. This led to two changes: (1) imputation of regional averages for missing variable data points when needed and (2) the setting of a threshold for missing data to determine when imputation would be performed.

Regional average values for vulnerability variables allowed us to impute missing vulnerability scores on a given dimension by using data from similar countries in a given geographic area.

Further, a threshold was set for missing data, such that imputation was undertaken for all dimensions/factors where data were missing on 50 percent of the total number of variables in Dimensions Three, Four, and Five, and a 51 percent missing data threshold was applied on Dimensions One and Two. In Dimensions One and Two, this rule was applied for of Libya, Qatar, Somalia, and South Sudan in Dimension Two: Lack of Basic Needs, due to the larger number of total vulnerability variables included in the first two dimensions of vulnerability. Dimension One: Governance Issues also had a 51 percent or above missing data threshold applied, but no countries in this dimension required imputation for missing data due to this stricter requirement. There were several cases where imputation was deliberately not employed and missing data percentages of 66 percent or 25 percent were retained for certain countries on specific dimensions in order to maintain variability within the regions where some data may have been more limited.

Each instance of missing data at the dimension level is catalogued in Table 11 and Table 12.

Table 11Countries with 100 percent missing data on a dimension
Factor
Country
Factor Three: InequalityBarbados

Brunei

Suriname
Factor Four: Disenfranchised GroupsAlgeria

Angola

Barbados

Brunei

Burundi

Cape Verde

China

Cuba

Djibouti

Equatorial Guinea

Eritrea

Gambia

Guinea-Bissau

Guyana

Jamaica

Laos

Namibia

Democratic People’s Republic of Korea (North Korea)

Oman

Papua New Guinea

Qatar

Sudan

Suriname

Swaziland

Timor-Leste

Trinidad and Tobago

Syria

Mozambique

Malaysia

Sri Lanka
Factor Five: Effects of ConflictLuxembourg
Table 12Countries with 50 percent to 99 percent missing data on a dimension
Factor
Country
Factor One: Governance IssuesBarbados
Brunei
Hong Kong, China
Factor Two: Malnourishment and Lack of Basic NeedsKosovo

Taiwan

Libya*not imputed at 50 percent due to greater number of variables on Factor Two

Qatar*not imputed at 50 percent due to greater number of variables on Factor Two

Somalia*not imputed at 50 percent due to greater number of variables on Factor Two

South Sudan*not imputed at 50 percent due to greater number of variables on Factor Two
Factor Three: InequalityAlgeria

Bahrain*reduced to 25 percent missing data

Cape Verde

Cuba

Djibouti

Equatorial Guinea

Eritrea

Gambia

Guinea-Bissau

Guyana

Hong Kong

Jamaica

Jordan*reduced to 25 percent missing data

Kuwait*reduced to 25 percent missing data

Laos

Libya

Morocco

Mozambique

Myanmar

North Korea

Oman*reduced to 25 percent missing data

Papua New Guinea

Qatar*reduced to 25 percent missing data

Saudi Arabia*reduced to 25 percent missing data

Suriname

Swaziland

Syria*reduced to 25 percent missing data

Timor-Leste

Trinidad and Tobago

Turkmenistan

United Arab Emirates*reduced to 25 percent missing data

Uzbekistan
Factor Four: Disenfranchised GroupsBahrain*maintained at 66 percent missing data

Algeria*maintained at 66 percent missing data

Angola

Oman*maintained at 66 percent missing data

Qatar*maintained at 66 percent missing data

Syria*maintained at 66 percent missing data

Egypt*maintained at 66 percent missing data

Barbados

Trinidad and Tobago

Cuba

Jamaica

Equatorial Guinea

Angola

Burundi

Eritrea

Mozambique

Djibouti

North Korea

China

Sudan*maintained at 66 percent missing data

Papua New Guinea

Guyana

Suriname

Brunei

Laos

Timor-Leste

Malaysia

Swaziland

Namibia

Sri Lanka

Gambia

Guinea-Bissau

Cape Verde

Iran

Iraq*maintained at 66 percent missing data

Jordan*maintained at 66 percent missing data

Kuwait*maintained at 66 percent missing data

Lebanon*maintained at 66 percent missing data

Libya*maintained at 66 percent missing data

Morocco*maintained at 66 percent missing data

Saudi Arabia*maintained at 66 percent missing data

Turkmenistan

United Arab Emirates*maintained at 66 percent missing data

Yemen*maintained at 66 percent missing data
Factor Five: Effects of ConflictBarbados

Brunei

Cape Verde

Hong Kong

Suriname

Luxembourg

Phase 8: Eigenvalue weighting by factor

We calculated the unweighted and eigenvalue weighted vulnerability scores after consultation with our Expert Working Group and strong recommendations 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 or dimensions with greater eigenvalues explain more of the overall model and can be weighted accordingly in the overall vulnerability score.

After calculating the unweighted averages across factors (simple Average calculation function) and the Unweighted Overall Vulnerability Score (Factor 1 Average + Factor 2 Average + Factor 3 Average + Factor 4 Average + Factor 5 Average) divided by five factors, the following formula was then employed to determine the eigenvalue weighted vulnerability scores (Figure 1):

Eigenvalue Weighting Formula:

(((Factor 1 Average*5.76)+(Factor 2 Average*3.422)+(Factor 3 Average*2.233)+(Factor 4 Average*2.092)+(Factor 5 Average*1.938))/(5*5.76*3.422*2.233*2.092*1.938))*100 = Eigenvalue Weighted Value

Normalisation of Eigenvalue Weighted Variable:

Normalisation: Normalised Value = 1+(Reported Value – minimum value)*(100-1)/(maximum value – minimum value)

Ultimately, we decided to proceed with the eigenvalue weighted values because it provided appropriate context for the relative importance and strength of factors rather than treating them all as equally important.

Phase 9: Quality assurance checks

A final step prior to finalising the 2018 Vulnerability Model involved turning over all data to Ernst and Young,22 which conducted quality assurance checks on the transcription of vulnerability data from the original sources, the exported data files from Stata, and the final Excel files in order to confirm the data underpinning our 2018 Vulnerability Model is error-free.

Description of variables in final model by dimension

The final dimensions and variables are presented in Table 13. The resulting vulnerability scores are listed in Table 14 for 167 countries. Detailed descriptions of all retained variables and relevant sources are listed in Table 15.

Table 132018 Vulnerability Model
Governance issues
Lack of Basic Needs
Inequality
Disenfranchised groups
Effects of conflict
Political instabilityCell phone usersAbility to obtain fundsAcceptance of immigrantsImpact of terrorism
GSI government responseUndernourishmentViolent crimeAcceptance of minoritiesInternal conflicts fought
Women's physical securitySocial safety net GINI coefficientSame sex rightsInternally displaced persons
Political rightsAbility to borrow moneyConfidence in judicial systems

Regulatory qualityTuberculosis


Disabled rightsAccess to clean water


Weapons access




Table 14Vulnerability to modern slavery by dimension for 167 countries
Country
Governance issuesLack of basic needsInequalityDisenfranchised groupsEffects of conflictOverall weighted average
Central African Republic85.450.262.758.081.6100.0
South Sudan75.751.162.956.185.794.7
Afghanistan81.041.364.746.092.693.9
Syrian Arab Republic85.636.962.533.495.492.3
Congo, Democratic Republic of77.250.855.646.586.791.7
Somalia80.656.849.622.788.489.5
Sudan80.746.642.437.087.487.1
Yemen79.243.149.253.069.986.4
Iraq72.634.965.246.689.485.7
Chad71.843.248.546.546.174.9
Pakistan56.836.245.955.392.874.1
Nigeria54.141.350.247.195.574.1
Korea, Democratic People's Republic of (North Korea)87.652.030.332.412.373.3
Libya81.423.049.628.163.173.1
Burundi72.442.642.148.141.772.9
Kenya55.148.749.644.566.870.6
Guinea-Bissau77.840.147.644.117.170.5
Cameroon65.936.546.246.353.969.6
Haiti62.449.754.156.820.169.6
Eritrea71.050.633.748.125.969.6
Congo75.137.648.546.119.669.2
Zimbabwe66.345.536.653.025.366.4
Guinea68.332.454.746.428.666.3
Myanmar58.143.826.146.070.265.9
Niger61.941.237.045.050.465.6
Swaziland69.950.039.438.811.764.8
Ethiopia62.447.527.334.655.364.5
Cambodia66.338.541.656.714.863.5
Malawi55.451.540.961.519.163.4
Iran, Islamic Republic of74.625.535.837.339.563.3
Angola60.243.448.248.519.862.3
Mauritania67.333.739.350.522.362.0
Madagascar54.446.851.056.817.362.0
Papua New Guinea64.863.346.29.513.361.9
Rwanda56.640.840.055.734.061.7
Equatorial Guinea68.440.836.748.510.161.7
Togo70.031.545.342.317.161.3
Djibouti66.838.033.948.121.361.2
Uganda52.848.338.250.335.360.8
Tanzania, United Republic of55.547.334.952.729.160.5
Egypt61.618.444.252.851.160.4
Philippines50.535.345.736.469.360.2
Liberia55.044.044.154.918.259.3
Lebanon59.122.648.144.847.858.9
Gambia66.828.141.844.120.858.4
Lesotho53.850.744.641.918.658.3
Turkmenistan80.221.531.432.615.958.1
Venezuela, Bolivarian Republic of65.119.760.434.327.857.9
Lao People's Democratic Republic70.735.126.441.213.957.5
Mexico47.323.759.037.868.857.3
Côte d'Ivoire59.530.141.737.540.957.2
Mozambique48.648.340.548.130.057.0
Mali55.324.435.535.966.355.9
Tajikistan67.430.932.827.830.155.8
Honduras55.526.558.936.532.755.5
India46.229.832.441.180.055.5
Zambia45.854.444.949.113.155.2
Sierra Leone50.946.141.248.118.155.2
Ukraine54.015.946.439.062.254.4
South Africa46.738.361.036.926.953.8
Burkina Faso58.431.640.335.226.253.1
Timor-Leste58.441.937.241.23.952.8
Cuba60.225.937.647.817.352.4
Ghana52.629.142.053.721.652.2
Guatemala51.025.858.140.927.452.1
Algeria63.217.927.837.043.652.0
Colombia45.719.256.432.663.551.6
Russia59.313.538.634.151.951.6
Turkey47.022.247.048.647.951.6
Thailand50.921.835.345.151.951.1
El Salvador50.523.059.843.622.750.7
China61.420.526.932.444.250.6
Indonesia43.738.035.853.332.250.5
Oman68.720.537.833.46.450.1
Bangladesh54.138.425.720.945.350.0
Jordan57.915.741.847.426.249.9
Bahrain63.025.834.524.025.449.6
Gabon56.527.136.647.512.449.1
Morocco60.718.838.135.722.048.3
Namibia44.638.455.938.810.448.1
Azerbaijan60.321.223.935.732.547.8
Uzbekistan71.720.332.69.018.047.5
Belarus64.916.723.939.420.847.3
Brunei Darussalam53.530.931.741.218.247.2
Bosnia and Herzegovina52.016.431.750.734.146.4
Saudi Arabia63.221.930.114.232.246.3
Senegal43.934.835.641.030.946.2
Kuwait59.720.129.329.328.545.9
Macedonia, the former Yugoslav Republic of 48.417.442.550.627.345.6
Guyana49.525.660.428.112.445.4
Albania46.020.744.348.427.045.2
Benin51.128.839.935.315.845.0
Cape Verde48.719.744.144.122.144.5
Peru44.324.748.038.227.544.3
Jamaica39.524.262.247.815.544.2
Nepal52.035.632.28.734.744.1
Bolivia, Plurinational State of50.925.846.332.113.444.1
Nicaragua48.224.543.335.322.843.9
Kosovo53.116.039.349.712.043.8
Armenia51.118.933.846.322.143.6
Mongolia40.936.835.147.118.143.5
Kazakhstan60.414.525.138.219.543.3
Dominican Republic42.528.746.138.821.843.1
Kyrgyzstan49.619.735.442.623.242.8
Sri Lanka44.127.033.534.935.942.5
Botswana43.337.937.337.69.742.1
Suriname55.510.750.828.116.342.1
Barbados47.614.352.547.89.241.9
Moldova, Republic of42.022.935.358.318.141.6
Vietnam53.623.228.132.518.541.5
Ecuador46.023.046.429.123.041.3
Paraguay38.321.064.732.722.740.9
Malaysia36.228.439.641.227.839.2
Tunisia47.215.434.831.933.739.2
Georgia41.519.333.943.931.439.2
Trinidad and Tobago38.613.062.447.813.739.1
Qatar56.313.829.533.47.037.7
Greece38.514.436.456.023.637.1
Israel35.819.127.548.538.636.4
Panama44.221.042.633.19.436.4
Brazil43.113.656.219.824.036.4
Montenegro39.415.037.450.918.335.8
Serbia39.115.231.640.927.533.9
Romania35.819.532.652.016.133.9
Croatia35.720.234.148.312.232.7
Bulgaria33.014.743.344.117.431.3
Korea, Republic of (South Korea)33.929.425.733.813.429.8
Estonia35.213.727.452.212.429.2
Argentina39.311.445.023.613.428.9
Costa Rica35.216.740.729.412.228.4
Italy31.714.445.431.019.328.3
Slovakia29.915.129.951.214.227.2
United Arab Emirates47.915.124.77.811.926.8
Lithuania29.215.435.646.39.726.2
Chile28.513.850.023.520.325.6
Hong Kong, China39.39.624.728.415.024.7
Latvia31.715.923.844.010.324.6
Poland24.513.727.559.613.624.4
Hungary23.914.832.948.315.523.6
Mauritius25.517.733.631.112.221.2
Taiwan, China24.524.740.621.11.420.3
Slovenia22.416.630.645.66.420.1
Uruguay31.913.534.315.49.519.7
Cyprus24.516.732.629.710.119.1
Czech Republic25.113.921.037.118.219.1
United States18.318.230.315.628.615.9
France17.315.429.421.228.515.3
Japan21.513.115.531.917.813.8
Singapore30.816.35.018.79.013.4
Belgium20.015.029.919.312.313.1
Spain17.218.333.515.114.212.8
United Kingdom15.915.625.112.427.811.1
Germany15.915.022.815.724.710.4
Ireland17.217.024.310.920.110.4
Canada16.620.720.19.221.510.2
Portugal12.215.631.720.79.78.5
Luxembourg17.713.724.512.114.38.4
Finland18.616.015.017.811.28.2
Netherlands12.813.626.016.012.26.1
Norway15.717.813.19.410.84.5
Australia11.915.720.712.013.04.3
Sweden10.217.017.413.018.34.3
Iceland20.611.721.14.11.84.2
Austria12.612.218.223.53.13.4
New Zealand12.218.416.27.07.01.9
Switzerland11.612.215.220.14.91.5
Denmark8.715.313.815.212.51.0
Table 15Variable descriptions and sources23
Factor
Variable Name
Data Description and Source
Variable Description Min-max normalisation has been performed on all variables to ensure comparability via a linear transformation (scale of 1 to 100).
Factor One: Governance IssuesPolitical InstabilityData are from The Global Peace Index, which measures the level of peace in 162 countries according to 22 qualitative and quantitative indicators aligned with the absences of violence and fear of violence.24This variable represents an assessment of political instability ranked from 1 to 5 (very low to very high instability). This measure is assessed by the Economist Intelligence Unit (EIU) based on five questions relating to the degree to which the country's political institutions are sufficiently stable to support the needs of its citizens, businesses, and overseas investors. This indicator aggregates five other questions on social unrest, orderly transfers, opposition stance, excessive executive authority, and an international tension sub-index. These data are analysed and updated on a quarterly basis. The score provided for March 2014–March 2015 is the average of the scores given for each quarter.
Weapons AccessData are from The Global Peace Index, which measures the level of peace on 162 countries according to 22 qualitative and quantitative indicators aligned with the absences of violence and fear of violence.25The variable Weapons Access is a qualitative assessment of the accessibility of small arms and light weapons. This measure ranges from a value of 1, which indicates very low access to weapons, to a value of 5, which indicates very high access to weapons. Data are from 2016.
Women's Physical SecurityData are from Women Stats.26This variable reflects women’s physical security on a range of indicators. It is a multivariate scale derived from WomenStats Database that provides the ordinal ranking of the physical security of women. Data are from variable "MULTIVAR-SCALE-1." The scale takes into account the presence and enforcement of laws against domestic violence, rape and marital rape, the existence of taboos or norms against reporting these crimes, and the occurrence of honour killings and femicide. This scale ranges from 1 "high security" to 4 "low security." Data are from 2014.
Disabled RightsData are from Gallup Analytics, which presents a detailed assessment of global attitudes for more than 100 countries.27The variable Disabled Rights measures the proportionally representative percentage of respondents who responded that their region is "not a good place" in direct response to the question: "Is the city or area where you live a good place or not a good place to live for people with intellectual disabilities?" 0 represents lower vulnerability, 100 represents higher vulnerability. Select World Poll as Data Source, Good Place for Intellectually Disabled as your Metric, Aggregate Demographics, and Time set to Year 2015. Data are from 2015.
Government Response GSIData are from Global Slavery Index.The variable Government Response from the Global Slavery Index represents the average evaluation score for each country across five main indicators including: 1) Survivors are Supported, 2) Criminal Justice Responses, 3) Coordination and Accountability, 4) Attitudes, Social Systems, and Institutions, and 5) Business and Government Responses to Modern Slavery in Supply Chains. Each section sub-score is aggregated to provide a response rating. The quantitative evaluation scores are used for each country. Data are from 2016. These values range from 0 representing poor government response to 100 representing strong government response.
Political RightsData are from the Center for Systemic Peace, Polity IV Dataset hosted by the Integrated Network for Societal Conflict Research.28This variable measures governments on a scale from strongly democratic to strongly autocratic. Alt. Political Rights Variable "Polity IV" score is the POLITY2 score reported in the POLITY IV dataset on "Regime Authority Characteristics and Transitions Database" under the Polity IV: Annual Time-Series, 1800-2015 Excel file. This variable modifies the original Polity score that measures the "Combined Polity Score: The POLITY score is computed by subtracting the AUTOC score from the DEMOC score; the resulting unified polity scale ranges from +10 (strongly democratic) to -10 (strongly autocratic). This variable was then inverted to represent increasing risk to slavery through less democratic governance. Data are from 2015.
Regulatory QualityData are from the Worldwide Governance Indicators (WGI). Regulatory Quality is one of six broad dimensions of governance covered by the WGI (others include Voice and Accountability, Political Stability and Absence of Violence/Terrorism, Regulatory Quality, Rule of Law, Control of Corruption).29The variable Regulatory Quality measures perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development. The variable ranges from approximately -2.5, which indicates low regulatory quality, to a score of 2.5, which indicates high regulatory quality. Data are from 2014.
Factor Two: Lack of Basic NeedsCell Phone Users Data are from the World Bank's World Development Indicators.30The variable Cell Phone Users measures the number of cellular subscriptions per 100 people for each country. The indicator includes (and is split into) the number of postpaid subscriptions, and the number of active prepaid accounts (i.e. that have been used during the last three months). The indicator applies to all mobile cellular subscriptions that offer voice communications. It excludes subscriptions via data cards or USB modems, subscriptions to public mobile data services, private trunked mobile radio, telepoint, radio paging, and telemetry services. Data are from 2015. The potential data values range from 0 out of 100 people in a country to 100 out of 100 people in a country.
Social Safety NetData are from the International Labour Organization's (ILO) World Social Protection Report.31The variable Social Safety Net evaluates countries on their national social security system based on the number of policy areas covering at least one program including total social security provisions available to citizens including protections for sickness, maternity, old age, invalidity, survivors, family allowances, employment injury, and unemployment. Scores range from 0 to 8, where a value of 0 indicates that none of these provisions are provided, and a value of 8 indicates that all of these provisions are provided. Data are from the World Social Protection Report 2014/2015.
UndernourishmentThese data are from the Statistics Division of the Food and Agricultural Organization (FAO) of the United Nations.32This variable measures prevalence of undernourishment. The prevalence of undernourishment expresses the probability that a randomly selected individual from the population consumes an amount of calories that is insufficient to cover her/his energy requirement for an active and healthy life. When a value is listed as <5.0, the value is reported as 2.5. Data provided are for 2014-2016.The indicator is measured as a percentage and computed by comparing a probability distribution of habitual daily dietary energy consumption with a threshold level called the minimum dietary energy requirement. Both are based on the notion of an average individual in the reference population. This is the traditional FAO hunger indicator, adopted as official Millennium Development Goal indicator for Goal 1, Target 1.9. It is listed as Dataset V.2 Note: Many of the developed countries that are scored as <5.0, which will translate to a 2.5 value in our dataset, are actually listed beneath the section "Developed Countries" at the end and you must apply this proportion for all of the countries in that category.The indicator is calculated in three year averages, from 1990-92 to 2014-16, to reduce the impact of possible errors in estimated DES, due to the difficulties in properly accounting of stock variations in major food. Data are from 2014-2016. Data in Excel are also available here http://bit.ly/14FRxGV.
Access to Clean WaterData are from the World Bank, World Development Indicators.33This variable measures improved water source access as a percentage of the total population. Access to an improved water source refers to the percentage of the population using an improved drinking water source. The improved drinking water source includes piped water on premises (piped household water connection located inside the user’s dwelling, plot or yard), and other improved drinking water sources (public taps or standpipes, tube wells or boreholes, protected dug wells, protected springs, and rainwater collection). This variable is measured in percentage points of the total population with improved access to water with potential values ranging from 0 to 100 percent of the population.
TuberculosisData are from the World Bank, World Development Indicators.34This variable measures the incidence of tuberculosis (per 100,000 people). Incidence of tuberculosis is the estimated number of new pulmonary, smear positive, and extra-pulmonary tuberculosis cases. Incidence includes patients with HIV. Data are from 2015.
Ability to Borrow MoneyData are from World Bank Global FINDEX which provides data on how individuals save, borrow, make payments, and manage risks. The indicators in the 2014 Global Financial Inclusion (Global Findex) database are drawn from survey data covering almost 150,000 people in 143 economies—representing more than 97 percent of the world’s population.35This variable represents the percentage of those age 15+ who borrowed any money in the past year. The percentage of respondents who report borrowing any money for any reason and from any source in the past 12 months. The potential data values range from 100 percent of respondents age 15 and over to 0 percent of respondents who reported borrowing any money for any reason and any source in the past 12 months. Data are from 2014.
Factor Three: InequalityConfidence in Judicial SystemsData are from Gallup Analytics, which presents a detailed assessment of global attitudes for more than 100 countries.36The variable Confidence in the Judicial System measures the proportionally representative percentage of respondents who responded "no" in direct response to the question: "In this country, do you have confidence in [the] judicial system and courts?" Data are from 2016.
Violent CrimeData are from The Global Peace Index, which measures the level of peace in 162 countries according to 22 qualitative and quantitative indicators aligned with the absences of violence and fear of violence.37The variable Violent Crime measures the peacefulness of a society based on the likelihood of violent crime. These data are calculated on a five-point scale, where a value of 1 indicates a country that is "most peaceful," and a value of 5 indicates a country that is "least peaceful." This assessment by the Economist Intelligence Unit (EIU) based on the question, “Is violent crime likely to pose a significant problem for government and/or business over the next two years?” Country analysts assess this question on a quarterly basis. Data are from 2016
GINI CoefficientData are from the World Bank's World Development Indicators.38The variable GINI Coefficient is a measure of income inequality. This measure ranges from a value of 0, which indicates absolute income equality, to 100, which indicates absolute income inequality. Data are from the most recent year in the period 2005 to 2013.
Ability to Obtain Emergency FundsData are from World Bank Global FINDEX which provides data on how individuals save, borrow, make payments, and manage risks. The indicators in the 2014 Global Financial Inclusion (Global FINDEX) database are drawn from survey data covering almost 150,000 people in 143 economies – representing more than 97 percent of the world’s population.39The percentage of respondents age 15+ who report that in case of an emergency it is not at all possible for them to come up with 1/20 of gross national income (GNI) per capita in local currency within the next month. The potential data values range from 100 percent of respondents to 0 percent of respondents reporting that they are capable of coming up with 1/20 of gross national income per capita in their local country within the next month. Data are from 2014.
Factor Four: Disenfranchised GroupsSame Sex RightsData are from Gallup Analytics, which presents a detailed assessment of global attitudes for more than 100 countries.40The variable Same Sex Rights measures the proportionally representative percentage of respondents who responded that their region is "not a good place" in direct response to the question: "Is the city or area where you live a good place or not a good place to live for gay or lesbian people?" 0 represents lower vulnerability, 100 represents higher vulnerability. Data are from 2016.
Acceptance of ImmigrantsData are from Gallup Analytics, which presents a detailed assessment of global attitudes for more than 100 countries.41The variable Acceptance of Immigrants measures the proportionally representative percentage of respondents who responded that their region is "not a good place" in direct response to the question: "Is the city or area you live a good place or not a good place to live for immigrants from other countries?" 0 represents lower vulnerability, 100 represents higher vulnerability. Data are from 2016.
Acceptance of MinoritiesData are from Gallup Analytics, which presents a detailed assessment of global attitudes for more than 100 countries.42The variable Acceptance of Minorities measures the proportionally representative percentage of respondents who responded that their region is "not a good place" in direct response to the question: "Is the city or area where you live a good place or not a good place for racial and ethnic minorities?" 0 represents lower vulnerability, 100 represents higher vulnerability. Data are from 2016.
Factor Five: Effects of ConflictImpact of TerrorismData are from the Global Terrorism Index, Institute for Economics and Peace (IEP).43The variable Impact of Terrorism is drawn from the scores calculated for the Global Terrorism Index, which measures the number of deaths, injuries, property damage and psychological impact resulting from terrorism in 162 countries. Data are from the 2016 Global Terrorism Index and is scored on a 1 to 10 scale, where 1 is low impact and 10 represents high impact of terrorism.
Internal Conflicts FoughtData are from The Global Peace Index, which measures the level of peace in 162 countries according to 22 qualitative and quantitative indicators aligned with the absences of violence and fear of violence.44This indicator measures the number and duration of conflicts that occur within a specific country's legal boundaries. Information for this indicator is sourced from three datasets from Uppsala Conflict Data Program (UCDP): the Battle-Related Deaths Dataset, Non-State Conflict Dataset and One-sided Violence Dataset. The score for a country is determined by adding the scores for all individual conflicts that have occurred within that country's legal boundaries over the last five years, based on the following factors: 1. Number: a) the number of interstate armed conflicts, internal armed conflict (civil conflict), internationalised internal armed conflicts, one-sided conflict and non-state conflict located within a country's legal boundaries, b) If a conflict is a war (1,000+ battle-related deaths) it receives a score of one; if it is an armed conflict (25-999 battle-related deaths) it receives a score of 0.25. 2. Duration: a) a score is assigned based on the number of years out of the last five that conflict has occurred. For example, if a conflict last occurred five years ago that conflict will receive a score of one out of five. The cumulative conflict scores are then added and banded to establish a country's score. Scores range from 1 out of 5 to 5 out of 5. A score of 1 indicates no internal conflicts and a score of 5 indicates very high levels of internal conflict. Data are from the 2016 Global Peace Index Indicators.
Internally Displaced PersonsData are from UNHCR.45This variable measures the number of displaced persons. These data represent the total number of displaced (internally displaced, returned internally displaced, returned refugees, asylum seekers, stateless persons) by territory of origin. For 2015 data where an * is used to indicate between 1-4 individuals to protect their identities, we retained the value of 1 person. 0 = low displacement 100 = high displacement of our normalised data, but as reported the data values range from 1 person, which is small displacement, to values over 7,800,000 for the highest displacement in 2014. Data are from 2015.

Data limitations

There are several areas of data limitations relevant to our vulnerability model that should be kept in mind as these results are interpreted. These major limitations include: 1) concept-variable consistency or the fit of the vulnerability variables to the real world phenomena they are approximating in our model, 2) data availability, transparency, and publication regularity, 3) lag in administrative data reflecting real world situations on the ground, 4) collinearity checks on our variables that resulted in dropping several empirically redundant but conceptually important variables such as corruption, gender inequality and environmental performance, and 5) data correction methods for missing data, such as imputation.

In developing a theoretically based model of vulnerability to modern slavery, there are several common challenges that must be overcome. Global models of vulnerability will face data limitations in terms of available data covering a majority of our 167 countries for prevalence and vulnerability. All variables included in the Vulnerability Model must cover most of our 167 countries, be published regularly, and explain clearly how these measures were developed. Then there is the conceptual exercise of ensuring that these measured variables match the phenomena we seek to capture in our model. This exercise in ensuring concept-variable consistency is often limited by data availability but requires the intentional selection of variables that represent the potential risks that individuals vulnerable to modern slavery may face across a broad array of potential factors consistent with the areas of insecurity reflected by human security theory.

Lags in administrative data also affect our Vulnerability Model, as even the most recent information may still not reflect current situations on the ground at this moment. Finally, as a result of standard statistical methods to refine our model, we perform collinearity checks on our variables to ensure that we do not retain redundant variables. However, as a result of this process, we were required to drop empirically redundant but conceptually important measures such as Corruption, Gender Inequality, and Environmental Performance. We must also note that we have employed imputation to resolve missing data issues for Dimensions 1 and 2 for above 51 percent missing data and for Dimensions 3, 4, and 5 for above 50 percent missing data by using regional averages. Where these missing data thresholds were met, we replaced missing data points with subregional averages for the affected variable. These efforts ensured that missing data points were supplemented with regionally specific trends and information on affected vulnerability variables.

Footnotes

1Liberty Asia & The Freedom Fund 2015, Modern Slavery and Corruption. Available from: http://un-act.org/publication/modern-slavery-and-corruption/. [14 March 2018].
2The Freedom Fund 2016, Modern Slavery and Trafficking in Conflict: The UN’s Response. Available from: freedomfund.org/wp-content/uploads/UN-trafficking-in-conflict-WEB.pdf. [14 March 2018].
3Bales, K 2016, Blood and Earth: Modern Slavery, Ecocide, and the Secret to Saving the World, Spiegel and Grau, New York.
4Within the Global Slavery Index itself, the vulnerability modelling has served an additional purpose. For many of the countries that we do not or cannot obtain a national level estimate either through Multiple Systems Estimation (MSE) or nationally representative surveys, the vulnerability model is the key to our understanding of the risks of modern slavery. In the 2014 and 2016 GSI, we used the vulnerability model as the basis of clustering countries into risk groupings. This was the first step of a process, from which we extrapolated prevalence estimates. In 2018, the country-level vulnerability model was one layer of risk data that contributed to our prevalence model for GSI 2018.
5Tadjbakhsh, S & Chenoy, A M 2007, Human Security: Concepts and Implications, London: Routledge.
6See Joudo Larsen, J & Durgana, D P 2017, ‘Measuring vulnerability and estimating prevalence of modern slavery,’ Chance: Special issue on modern slavery, vol. 30, no. 3, pp. 21-29, American Statistical Association. 
7The variable “Minorities” refers to ethnic and racial minorities.
8Freedom House 2017, Freedom in the World 2017 - Political Rights. Available from: https://freedomhouse.org/report/fiw-2017-table-country-scores. [14 March 2018].
9Freedom House, 2017, Freedom in the World 2017 – Civil Liberties. Available from: https://freedomhouse.org/report/fiw-2017-table-country-scores. [14 March 2018].
10World Bank 2016, Global FINDEX. Available from: http://databank.worldbank.org/data/reports.aspx?source=1228#. [14 March 2018].
11UNESCO Institute for Statistics 2015, Literacy. Available from: http://data.uis.unesco.org/Index.aspx?queryid=166. [14 March 2018].
12World Bank Development Indicators 2015, Child Mortality. Available from: http://data.worldbank.org/indicator/SH.DYN.MORT. [14 March 2018].
13Transparency International 2016, Corruption Perceptions Index. Available from: https://www.transparency.org/news/feature/corruption_perceptions_index_2016#resources. [14 March 2018].
14International Labour Organization 2015, World Security Report 2014/2015. Available from: http://www.ilo.org/global/research/global-reports/world-social-security-report/2014/WCMS_245201/lang--en/index.htm. [14 March 2018].
15World Bank 2015, Gross Domestic Product Per Capita in terms of Purchasing Power Parity. Available from: http://www.ilo.org/global/research/global-reports/world-social-security-report/2014/WCMS_245201/lang--en/index.htm. [14 March 2018].
16World Governance Indicators Project 2015, Government Effectiveness. Available from: http://info.worldbank.org/governance/wgi/index.aspx#home. [14 March 2018].
17United Nations Human Development Programme 2015, Gender Inequality Index. Available from: http://hdr.undp.org/en/composite/GII. [14 March 2018].
18Yale University 2016, Environmental Protection Index (EPI). Available from: https://epi.envirocenter.yale.edu/. [14 March 2018].
19As only one Political Rights measure was retained, this Alternative Political Rights Measure will now be referenced as “Political Rights” for simplicity.
20Fajnzylber, P, Lederman, D & Loayza, N 2002, ‘Inequality and Violent Crime,’ Journal of Law and Economics, vol. XLV, pp. 1-39. Available from: www.jstor.org/stable/10.1086/338347. [14 March 2018]
21As above.
22Ernst & Young (EY) is a professional services firm. Further information on the services provided is available from: http://www.ey.com/au/en/home.
23All data are the most recent publicly available data for each variable.
24The Institute for Economics and Peace 2016, Global Peace Index: Global Rankings. Available from: http://www.visionofhumanity.org/#page/indexes/global-peace-index/2016. [12 January 2017].
25As above.
26WomanSTATS Project n.d. Available from: http://www.womanstats.org/data.html. [12 January 2017].
27Gallup, Inc. n.d., Gallup Analytics. Available from: http://www.gallup.com/products/170987/gallup-analytics.aspx. [10 February 2017].
28Center for Systemic Peace 2016, The Polity Project. Available from: http://www.systemicpeace.org/polityproject.html. [10 February 2017].
29World Bank n.d., World Governance Indicators Project. Available from: http://info.worldbank.org/governance/wgi/index.aspx#home. [5 March 2017].
30World Bank n.d., Mobile Cellular Subscriptions (per 100 people). Available from: http://data.worldbank.org/indicator/IT.CEL.SETS.P2. [12 January 2017].
31International Labour Organization 2015, World Social Protection Report 2014/15, ILO. Available from: http://www.ilo.org/global/research/global-reports/world-social-security-report/2014/WCMS_245201/lang--en/index.htm. [12 January 2017].
32Food and Agriculture Organization of the United Nations n.d., Food Security. Available from: http://faostat3.fao.org/download/D/*/E. [12 January 2017].
33WHO & UNICEF n.d., Data & estimates from WHO/UNICEF Joint Monitoring Programme for Water Supply and Sanitation. Available from: http://www.wssinfo.org/data-estimates/tables/. [12 January 2017].
34World Health Organization 2016, Global Tuberculosis Report, The World Bank. Available from: http://data.worldbank.org/indicator/SH.TBS.INCD/countries. [12 January 2017].
35The World Bank n.d., Global Financial Inclusion Database. Available from: http://databank.worldbank.org/data/reports.aspx?source=1228. [12 January 2017].
36Gallup, Inc. n.d., Gallup Analytics. Available from: http://www.gallup.com/products/170987/gallup-analytics.aspx. [10 February 2017].
37The Institute for Economics and Peace 2016, Global Peace Index: Global Rankings. Available from: http://www.visionofhumanity.org/#page/indexes/global-peace-index/2016. [12 January 2017].
38United Nations Development Programme n.d., Table 3: Inequality-Adjusted Human Development Index, United Nations Development Programme Reports. Available from: http://hdr.undp.org/en/composite/IHDI. [12 January 2017].
39The World Bank n.d., Global Financial Inclusion Database. Available from: http://databank.worldbank.org/data/reports.aspx?source=1228. [12 January 2017].
40Gallup, Inc. n.d., Gallup Analytics. Available from: http://www.gallup.com/products/170987/gallup-analytics.aspx. [10 February 2017].
41As above.
42As above.
43The Institute for Economics and Peace 2016, Global Peace Index: Global Rankings. Available from: http://www.visionofhumanity.org/#page/indexes/global-peace-index/2016. [12 January 2017].
44As above.
45United Nations High Commissioner for Refugees n.d., Population Statistics. Available from: http://popstats.unhcr.org/en/time_series. [10 February 2017].