Organisation

Statistics South Africa (Stats SA)

SAE motivation

The production of small area poverty statistics in South Africa has been reported since 2000 (Stats SA, 2000; Alderman et al., 2002). In 2018, Statistics South Africa published poverty and inequallity estimates by applying small area estimation methods using the 2011 Census and survey data from the Income & Expenditure Survey (IES) 2010/11. According to the office, the main motivation for developing model-based poverty estimates is to support the implementation of local level government programmes and the absence of reliable survey estimates at the required administrative level.

Stats SA collects data from households by means of sample surveys consisting of approximately 30,000 dwelling units. This is a nationally representative sample that may be reliably disaggregated down to a provincial level. The Census is one of the few enumeration exercises that is able drill down to lower levels of geography. However, it does not cover some social and economic phenomena in detail. Furthermore, it has only been conducted in ten-year intervals in South Africa, as in many other countries.

There are three spheres of Government in South Africa, namely National Government, Provincial Government and Local Government. Whilst – in the main – planning and decision-making take place at national and provincial government level, the implementation of programmes takes place at local government level. The poverty surveys that Stats SA conduct are designed to produce reliable estimates at national, provincial and metropolitan area levels. This leaves a gap in relevant poverty data to direct the implementation of programmes at local levels. It is against this background the decision to embark on a project to indirectly estimate the poverty levels at local level was made. Direct estimation at this level would have increased the survey costs substantially.

Indicators in the scope of the study

The main indicators produced included the following:

  • Poverty headcount
  • Poverty gap
  • Severity of poverty
  • Inequality using the Gini-coefficient

Poverty indicators were derived using the national poverty lines and all indicators were decomposed to produce estimates at local government level.

SAE work within the organisation

For the Poverty Mapping 2011 project – Stats SA collaborated with the World Bank. A technical expert was procured by the World Bank to assist the Stats SA team in data organization, computation, analysis and compilation of the report. According to the report (Statistics South Africa, 2018), the project was developed within Stats SA by the Poverty and Inequality Statistics team.   

Input data

The Income and Expenditure Survey (IES) 2010/11 (Statistics South Africa, 2012a)  was the source of direct estimates whereas the 2011 Census (Statistics South Africa, 2012b) provided auxiliary data.

The main objective of the IES is to produce relevant statistical information on household consumption expenditure patterns for updating the consumer price index (CPI) basket of goods and services and the production of money-metric poverty and inequality estimates. The information was collected in 25,328 households across the country over a period of 12 months from 23 August 2010 to 4 September 2011 (Statistics South Africa, 2012a). The survey used a combination of diary and recall methodology to collect household economic information.

Building the SAE model/Model Building

The Poverty Map 2011 was constructed based on the standard method of small area estimation (SAE) proposed by Elbers, Lanjouw, and Lanjouw (2000 and 2003) - also known as the ELL method.  Information on household per capita consumption from IES 2010/11 was used to produce welfare indicators’ estimates at provincial, district and municipal levels, corresponding to 9, 52, and 234 areas, respectively.

The ELL method is composed of three stages. First, the comparability between census and survey variables is verified to select eligible predictors. Then, a regression model to relate the welfare indicator and predictors is developed using survey data. Finally, welfare indicators (such as the poverty headcount ratio and Gini coefficient) are computed applying the model prediction equation (estimated coefficients derived from the survey) to Census 2011 records.

The model dependent variable was the logarithm of the household per capita consumption and model predictors were: head of household socioeconomic characteristics (sex, population group, education, and income), dwelling characteristics (type, construction materials, water and electricity supply, toilet facilities, and tenure), assets, urban-rural classification of geographical area, and province indicator.

Benchmarking/data validation

Whilst there was no register/administrative data used for benchmarking, the patterns observed using non-income poverty measures based on Census 2011 were compared with the results of the poverty map.

The South African Multidimensional Poverty Index (SAMPI) was computed for years 2001, 2011 and 2016, using 2011 Census and the Community Survey 2016 data to observe changes in poverty and to validate the poverty map estimates. The SAMPI is an index comprised of four dimensions: health, education, living standards and economic activity. The results of this comparison indicated a strong correlation. According to Statistics South Africa (2018, p. 39), “both consumption-based and multidimensional measures reveal similar geographical dispersion of household poverty rates”.

Update of SAE methods and of small area estimate

Stats SA will be conducting a population Census in February 2022 and further developments in SAE projects will follow. Stats SA plans to update small area estimates using the Census 2022 data and the upcoming Income and Expenditure Survey planned to start by the end of 2022 (Statistics South Africa, 2020).

Use of SAE estimates

Stats SA has established a demand for reliable small area statistics. To provide relevant, timely statistics and expand the statistical knowledge base, there is a need for statistics that offer in-depth insight into the social and economic dynamics of the population, taking into account the spatial aspects that prevail. The estimates are used to inform planning and allocations at local government level.

Capacity building for SAE production

The capacity building component was included in the collaboration with the World Bank through the appointment of technical experts. This included training on data management, development of models, use of the PovMap software, etc. However, there is a need for further capacity building in this area at Stats SA.   

Future work on SAE

Stakeholders have recommended that Stats SA provides estimates at lower levels for informed decision-making. Such consideration may require an increase in sample size or other methods to estimate different social and economic phenomena at lower levels of geography. However, an increase in the sample size would put a strain on the organisation’s existing budget and in the absence of additional funding; hence, an increase in the sample size is unlikely. Therefore, the organisation will update the report on Measuring Poverty in South Africa with the Census 2022 data.

In addition, Stats SA plans to use SAE methods to estimate the response in subpopulations, particularly in lower geographic locations, using regular household surveys, the decennial Census, and administrative records.  

Challenges

Stats SA indicates that a main challenge is SAE capability and technical competencies. The organisation seeks to build capacity and competence around SAE because it is a valuable tool for producing SDGs and other development indicators at sub-provincial levels

Sources/References

Alderman, H., Babita, M., Demombynes, G., Makhatha, N., and Ozler, B. (2002). How Small Can You Go? Combining Census and Survey Data for Mapping Poverty in South Africa. Journal of African Economies 11: 169–200. https://doi.org/10.1093/jae/11.2.169

Elbers, C., Lanjouw, J.O. and Lanjouw, P.F. (2000). Welfare in Villages and Towns: Micro-estimation of Poverty and Inequality. Technical Report. Discussion Paper TI 2000-029/2. Tinbergen Institute, Amsterdam. Accessed on 19 Jan 2022.  <https://papers.tinbergen.nl/00029.pdf>

Elbers, C., Lanjouw, J.O. and Lanjouw, P.F. (2003). Micro–level Estimation of Poverty and Inequality. Econometrica 71(1):355–64.

Statistics South Africa (2000). Income and expenditure of households, 2010/2011.  Pretoria: Statistics South Africa. Accessed on 19 Jan 2022. <http://www.statssa.gov.za/publications/P0100/P01002011.pdf>

Statistics South Africa (2012a). Measuring poverty in South Africa.  Pretoria: Statistics South Africa. Accessed on 19 Jan 2022. <https://www.gov.za/sites/default/files/gcis_document/201409/povertyrep.pdf >

Statistics South Africa (2012b). Census 2011 Statistical release – P0301.4.  Pretoria: Statistics South Africa. Accessed on 19 Jan 2022. <https://www.statssa.gov.za/publications/P03014/P030142011.pdf>

Statistics South Africa (2016). Community Survey 2016, Statistical release P0301.  Pretoria: Statistics South Africa. Accessed on 19 Jan 2022. <http://cs2016.statssa.gov.za/wp-content/uploads/2016/07/NT-30-06-2016-RELEASE-for-CS-2016-_Statistical-releas_1-July-2016.pdf>

Statistics South Africa (2018). Poverty Mapping in South Africa: Applying small area estimation techniques using IES 2010/11 and Census 2011.  Pretoria: Statistics South Africa. Accessed on 19 Jan 2022. <http://www.statssa.gov.za/publications/Report%2003-10-00/Report%2003-10-002011.pdf>

Statistics South Africa (2020). Work Programme 2020/21 Statistics South Africa.  Pretoria: Statistics South Africa. Accessed on 19 Jan 2022.

<http://www.statssa.gov.za/wp-content/uploads/2013/05/Work_Programme_2020-2021.pdf>

Information provided by Ms Nozipho Shabalala, Ms Patricia Koka, Mr Davyson Chauke, Ms Kerotse Mmatli and Mr Werner Ruch, Statistics South Africa, Poverty and Inequality Chief Directorate 

Meeting with Ms Nozipho Shabalala, Ms Malerato Mosiane and Mr Solly Molayi, December 2021

  • No labels