The LFS is a twice-yearly rotating panel household survey, specifically designed to measure the dynamics of employment and unemployment in South Africa. It measures a variety of issues related to the labour market,including unemployment rates (official and expanded), according to standard definitions of the International Labour Organisation (ILO).
All editions of the LFS have been updated (some more than once) since their release. These version changes are detailed in a document available from DataFirst (in the "external documents" section titled "LFS 2000-2008 Collated Version Notes on the South African LFS").
Kind of Data
Sample survey data [ssd]
Unit of Analysis
Households (dwellings) and individuals
v2.1: Edited, anonymised dataset for licensed distribution
The South African September 2001 LFS dataset was originally released in March 2002 as 4 data files (household, worker, person and stratum psu). A second version was downloaded from the Statistics South Africa website subsequent to that in January 2006, which differed slightly from the originally obtained release in that it had weights that were aimed at reflecting the February 2005 population estimates. This version was benchmarked to the 2001 South African census (whereas previously, it had been benchmarked to the 1996 South African census). As a result, the weight variables in each data file differ between versions 1.0 and 2.0.
A third version (version 2.1) was downloaded by DataFirst on 11 August 2011, which differed slightly from version 2.0 in the following ways:
1. No suffix ” Sep2001” attached to the variable names
2. Year and Month variables included
3. Variable labels were altered. Previously, all variable labels were literal questions. Now the variable labels describe the variables.
4. A number of variables were renamed. This could prove confusing when comparing between current and previous versions of the data file.
Version 2.1 and 2.0 also have some substantive differences:
Worker data file
Sector: The variable representing employment sector has had a number of observations recoded into various other categories. There are 40 differences between versions.
Industry: The variables representing ”Main industry” (one is specific, the other more general) have been recoded for a number of observations. There are 151 differences between versions for the general (14 unique values) industry variable and 40 differences between versions for the specific (191 unique values) industry label. This would suggest that the differences between the more general industry variables are not driven exclusively by changes to observations in the specific industry variable entry.
Main Occupation: The variables representing ”Main occupation” (one is specific, the other more general) have been recoded for a number of observations. There are 129 differences between versions for the general (13 unique values) occupation variable and 40 differences between versions for the specific (371 unique values) occupation variable. This would suggest that the differences between general occupation variable are not driven exclusively by changes to observations in the specific occupation variable entry.
Employment Status: The variables representing employment status have several substantive differences. The two variables reflect two definitions, narrow and expanded, of employment status. Some observations that were previously defined as having one particular (within both variables!) are now defined as another. Official employment status (STATUS1) has 253 differences, whereas expanded employment status has 524 differences.
Household/general data file
The household/general file in version 2.0 has 17 extra observations. Each of these observations are coded missing or unspecified for each variable except Province and Household Weight.
Household characteristics, household listing, demographics, education, economic activity, work for pay, business ownership, unemployment, employers, main work activity in the past week, wages, salary, employment, migration
The LFS sample covers the non-institutional population except for workers' hostels. However, persons living in private dwelling units within institutions are also enumerated. For example, within a school compound, one would enumerate the schoolmaster's house and teachers' accommodation because these are private dwellings. Students living in a dormitory on the school compound would, however, be excluded.
Producers and sponsors
Statistics South Africa
The LFS is a twice-yearly rotating panel household survey. A rotating panel sample involves visiting the same dwelling units on a number of occasions (in this instance, five at most), and replacing a proportion of these dwelling units each round. New
dwelling units are added to the sample to replace those that are taken out. The pilot round of LFS fieldwork took place in February 2000, based on a probability sample of 10 000 dwelling units. This survey took place six months later, using a larger probability sample of 30,000 dwelling units. Among the 10,000 households visited in February, approximately 40% were re-visited in September 2000. The fieldworkers had some difficulty in identifying certain dwelling units in the sample, particularly in those areas where there are no addresses.
The Master Sample is based on the 1996 Population Census of enumeration areas (EA) and the estimated number of dwelling units from the 1996 Population Census. All 3000 PSUs included in the Master Sample were used in the Labour Force Survey. A PSU is either one EA or several EAs when the number of dwelling units in the base or originally selected EA was found to have less than 100 dwelling units. Each EA had to have approximately 150 dwelling units but it was discovered that many contained less. Thus, in some cases, it has been found necessary to add EAs to the original (census) EA to ensure that the minimum requirement of 100 dwellings, in the first stage of forming the PSUs, was met. The size of the PSUs in the Master Sample varied from 100 to 2445 dwelling units. Special dwellings such as prisons, hospitals, boarding houses, hotels, guest houses (whether catering or self-catering), schools and churches were excluded from the sample.
Explicit stratification of the PSUs was done by province and area type (urban/rural). Within each explicit stratum, the PSUs were implicitly stratified by District Council, Magisterial District and, within the magisterial district, by average household income (for formal urban areas and hostels) or EA. The allocated number of EAs was systematically selected with "probability proportional to size" in each stratum. Once the PSUs included in the sample were known, their boundaries had to be identified on the ground. After boundary identification, the next stage was to list accurately all the dwelling units in the PSUs.
The second stage of the sample selection was to draw from the dwelling units listing whereby a systematic sample of 10 dwelling units was drawn from each PSU. As a result, approximately 30,000 households (units) were interviewed. However, if there was growth of more than 20% in a PSU, then the sample size was increased systematically according to the proportion of growth in the PSU.
The first pilot round of LFS fieldwork took place in February 2000, based on a probability sample of 10 000 dwelling units. The sample was increased to 30 000 dwelling units in September 2000. Both of these surveys were published as discussion documents. The third round took place in February 2001, using the same 30 000 dwelling units. The fourth round of the LFS, which took place in September 2001 drew a new sample of 30 000 dwelling units were visited. Rotation of 20% of this commenced with the fifth round being conducted (February/March 2002)
The initial weights (household weights), based on the sample design, were equal to the inverse of the probability of selection. The initial weight for each member of the household was the same as the weight for the household itself. Further adjustment factors were then calculated within PSUs to account for non-response. To adjust for under-enumeration and to align survey estimates with independent population estimates, the weights were calibrated against Person benchmarks. A software package called CALMAR was used to perform this calibration. Using an iterative procedure, CALMAR adjusted the weights so that Person estimates conformed as closely as possible to external Person benchmarks. Gender, race and age group parameters were used for the Person cross-classification of the population.
Stats SA revised their population model to produce mid-year population estimates in the light of mortality data released in 2005 (see Stats SA Statistical Release P0309.3, 2005). The benchmarks for the LFS discussed in this statistical release have been adjusted accordingly. Weights were then adjusted according to those 2005 population estimates.
Statistics South Africa. Labour Force Survey: September 2001. [dataset]. Version 2.1. Pretoria: Statistics South Africa [producer], 2002. Cape Town: DataFirst [distributor], 2011. DOI: https://doi.org/10.25828/w038-1k46