Synthetic Data for an Imaginary Country, Sample, 2023
A synthetic hierarchical dataset for simulation and training purposes
World, 2023
Created on
Jun 26, 2024
Last modified
Jun 26, 2024
Page views
253
Identification
Survey ID Number
WLD_2023_SYNTH-SVY-EN_v01_M
Title
Synthetic Data for an Imaginary Country, Sample, 2023
Subtitle
A synthetic hierarchical dataset for simulation and training purposes
Country
| Name | Country code |
|---|---|
| World | WLD |
Study type
Synthetic data
Series Information
This dataset is part of a collection of fully synthetic data generated, for training and simulation purposes, for an imaginary middle-income country. The dataset is available in English and French. A full population dataset (~10 million individuals) is also available in English and French as a "synthetic census dataset".
Abstract
The dataset is a relational dataset of 8,000 households households, representing a sample of the population of an imaginary middle-income country. The dataset contains two data files: one with variables at the household level, the other one with variables at the individual level. It includes variables that are typically collected in population censuses (demography, education, occupation, dwelling characteristics, fertility, mortality, and migration) and in household surveys (household expenditure, anthropometric data for children, assets ownership). The data only includes ordinary households (no community households). The dataset was created using REaLTabFormer, a model that leverages deep learning methods. The dataset was created for the purpose of training and simulation and is not intended to be representative of any specific country.
The full-population dataset (with about 10 million individuals) is also distributed as open data.
The full-population dataset (with about 10 million individuals) is also distributed as open data.
Kind of Data
ssd
Unit of Analysis
Household, Individual
Version
Version Description
V. 2023-05-01 8K HH EN
Version Date
2023-05-01T04:00:00.000Z
Version Notes
Dataset generated using RealTabFormer (sample of 8,000 households), with post-processing. English version.
Scope
Keywords
| Keyword |
|---|
| synthetic data |
| open data |
| safe data |
| demographics |
| education |
| mortality |
| fertility |
| child malnutrition |
| labor, employment |
| housing |
| dwelling |
| water and sanitation |
| household expenditure |
| migration |
Coverage
Geographic Coverage
The dataset is a synthetic dataset for an imaginary country. It was created to represent the population of this country by province (equivalent to admin1) and by urban/rural areas of residence.
Geographic Unit
province (admin1), district (admin2)
Universe
The dataset is a fully-synthetic dataset representative of the resident population of ordinary households for an imaginary middle-income country.
Producers and sponsors
Primary investigators
| Name | Affiliation |
|---|---|
| Development Data Group, Data Analytics Unit | World Bank |
Funding Agency/Sponsor
| Name | Role |
|---|---|
| UNHCR-World Bank Joint Data Center on Forced Displacement | Sponsored research work for the development of synthetic data for the purpose of assessing statistical disclosure risk measures. |
Sampling
Sampling Procedure
The sample size was set to 8,000 households. The fixed number of households to be selected from each enumeration area was set to 25. In a first stage, the number of enumeration areas to be selected in each stratum was calculated, proportional to the size of each stratum (stratification by geo_1 and urban/rural). Then 25 households were randomly selected within each enumeration area. The R script used to draw the sample is provided as an external resource.
Response Rate
This is a synthetic dataset; the "response rate" is 100%.
Weighting
Sample weights were calculated that take the stratification into account. See the R script provided as an external resource.
Data Collection
Dates of Data Collection
| Start | End |
|---|---|
| 2023 | 2023 |
Time periods
| Start date | End date |
|---|---|
| 2023 | 2023 |
Data Collection Mode
other
Questionnaires
Questionnaires
The dataset is a synthetic dataset. Although the variables it contains are variables typically collected from sample surveys or population censuses, no questionnaire is available for this dataset. A "fake" questionnaire was however created for the sample dataset extracted from this dataset, to be used as training material.
Data Processing
Data Editing
The synthetic data generation process included a set of "validators" (consistency checks, based on which synthetic observation were assessed and rejected/replaced when needed). Also, some post-processing was applied to the data to result in the distributed data files.
Methodology notes
The dataset was generated using REaLTabFormer, a four-level hierarchical generative model. The first-level model is the household composition generator, which generates variables that define each household's composition (household size and basic demographic profile of members, including age and relationship to the head of household). The second-level model is the household-level variables generator, which generates the variables whose values are common to all household members (such as dwelling characteristics) based on the household composition. The third-level model is the household-head generator, which generates observations for the head of the households based on the output of the previous two models. The fourth-level model is the household member generator, which generates data on the household members, excluding the head, for households of size two and above. The household member generator model uses the data generated by the household composition, household-level variables, and household head generator models. This hierarchical model provides relational dependencies within a household that would not be guaranteed if all records were generated independently.
To implement the different models, we adopted a transformer architecture. The household composition generator is a decoder model that generates data from normally distributed noise. The other three models use a sequence-to-sequence model inspired by the application of deep learning to language translation.
More detailed information is available in the Technical Documentation provided as an external PDF document.
To implement the different models, we adopted a transformer architecture. The household composition generator is a decoder model that generates data from normally distributed noise. The other three models use a sequence-to-sequence model inspired by the application of deep learning to language translation.
More detailed information is available in the Technical Documentation provided as an external PDF document.
Access policy
Restrictions
The dataset was generated as a fully-synthetic dataset. The model used to create the synthetic observations includes multiple procedures to avoid overfitting and data-copying. Also, the data used for training the model went through processes of sampling and recoding that make it impossible to link a synthetic observation to an actual observation. The dataset is thus safe for dissemination. It can be used with no restriction and is shared as open data.
Access authority
| Name |
|---|
| World Bank, Microdata Library |
Location of Data Collection
World Bank Microdata Library
Disclaimer and copyrights
Disclaimer
The data are to be used for training or simulation purposes only. It is not intended to be representative of any particular country, and should not be used for inference purpose.
Metadata production
Producers
| Name | Affiliation |
|---|---|
| OD | World Bank |
Date of Metadata Production
2023-05-01T04:00:00.000Z
DDI Document version
1.0 EN