Synthetic Data for an Imaginary Country, Sample, 2023
A synthetic hierarchical dataset for simulation and training purposes
World, 2023
Get MicrodataIdentification
WLD_2023_SYNTH-SVY-EN_v01_M
Synthetic Data for an Imaginary Country, Sample, 2023
A synthetic hierarchical dataset for simulation and training purposes
| Name | Country code |
|---|---|
| World | WLD |
Synthetic data
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".
| Type | Identifier |
|---|---|
| DOI | https://doi.org/10.48529/MC1F-QH23 |
The full-population dataset (with about 10 million individuals) is also distributed as open data.
ssd
Household, Individual
Version
V. 2023-05-01 8K HH EN
2023-05-01T04:00:00.000Z
World Bank, Development Data Group
Dataset generated using RealTabFormer (sample of 8,000 households), with post-processing. English version.
Scope
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.
province (admin1), district (admin2)
The dataset is a fully-synthetic dataset representative of the resident population of ordinary households for an imaginary middle-income country.
Producers and sponsors
| Name | Affiliation |
|---|---|
| Development Data Group, Data Analytics Unit | World Bank |
| Name | Abbreviation | Grant number | Role |
|---|---|---|---|
| UNHCR-World Bank Joint Data Center on Forced Displacement | JDC | KP-P174174-TF0B5124 | Sponsored research work for the development of synthetic data for the purpose of assessing statistical disclosure risk measures. |
Sampling
Sample frame
Synthetic Data for an Imaginary Country, Full Population, 2023 (WLD_2023_SYNTH-CENS-EN_v01_M)
World Bank
Household
true
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.
This is a synthetic dataset; the "response rate" is 100%.
Sample weights were calculated that take the stratification into account. See the R script provided as an external resource.
Survey instrument
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.
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.
Data collection
| Start | End |
|---|---|
| 2023 | 2023 |
| Start date | End date |
|---|---|
| 2023 | 2023 |
- other
Data processing
| Type | Description |
|---|---|
| synthetic data generation | The synthetic data generation process is described in detail in a technical document "Generating a relational synthetic dataset for an imaginary country - Technical documentation" provided as external resource. |
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.
Quality standards
The synthetic dataset is intended to provide a realistic representation of a middle-income countries. A set of summary indicators/tables was produced to ensure the realistic aspect of the data.
Access policy
World Bank Microdata Library
2 (one at the household level, one at the individual level). The two data files can be merged using variable "hid" as merging key.
Data available as open data (CC BY 4.0 license)
Data Access
| Name |
|---|
| World Bank, Microdata Library |
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.
Disclaimer and copyrights
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
| Name | Affiliation |
|---|---|
| OD | World Bank |
2023-05-01T04:00:00.000Z
Metadata version
1.0 EN
2023-05-01T04:00:00.000Z