The Quarterly Labour Force Survey (QLFS) is a household-based sample survey conducted by Statistics South Africa (Stats SA). It collects data on the labour market activities of individuals aged 15 years or older who live in South Africa.
Kind of Data
Sample survey data [ssd]
Unit of Analysis
v3.0: Edited, anonymised dataset for public distribution
The current version of the QLFS data was downloaded from the Statistics South Africa (Stats SA) website in April 2014. Stats SA updated the QLFS results (2008-2013) to reflect the new population benchmarks from Census 2011. Although the weighting changes are not clearly documented by Stats SA, users are advised to remain aware of these slight calibration differences when employing weights. These updates are in addition to the following changes to the previous versions:
The previous version of the QLFS 2008 Q3 was downloaded from the Statistics South Africa (Stats SA) website by DataFirst in January 2012. The version differs in a number of ways from the version that was obtained by DataFirst (from Stats SA) at some undeteremined time prior. The first of these differences is the way in which observations that fit into "unspecified", “not applicable” or "missing" type categories are coded for certain variables. For example, in the older version of the QLFS 2008 Q3 the "Q26ETIME" variable is coded 88, with the associated label "Not applicable", for 93,694 observations. In the newest version this category of responses is assigned the code 0 and is not labelled (as it was in the previous version) for the same 93,694 observations. This recoding process has been applied to a large number of categorical variables in the datafile. A few other categorical variables have instead been recoded in a similar vein but as different (non-zero) values. For example, values of 88 for Q4213MONHRSWRK have been redefined as having the value 8.
Second, there is an apparent difference between the definitions of underemployment ("underempl") between versions. Note that this variable was also subjected to the abovementioned recoding procedure. 2691 observations shift status from "underemployed" to "not underemployed" between versions. Encouragingly, the "not applicable" coded observations are consistent.
The metadata accompanying the release of the QLFS 2008 Q3 is somewhat ambiguous with its description of the variable derivation, which is supposedly constructed based on the following criteria (taken directly from the Stats SA metadata):
Underemployment (underempl) (@233 1.)
If hours usually work is less than 35 or total hours usually work is less than 35 and if additional hours that could have been worked is between 0 and 3 and if available to start work in the next four weeks should extra work become available then that is underemployment.
with the relevant variables named as follows:
Q418HRSWRK (Hours usually worked, Question 4.18)
Q420TOTALHRS (Total hours usually work, Question 4.19)
Q422MOREHRS (Like to be able to work more hours, Question 4.22)
Q425STARTXWRK (Able to start extra work, Question 4.25)
The above definition provided could have a number of possible interpretations. This complicated the process of checking for the source of the between version discrepancy. One plausible interpretation could be that workers were defined as unemployed if they worked fewer than 35 hours per week in one job (Q418HRSWRK < 35) or less than a total of 35 hours a week on more than one job (Q420TOTALHRS < 35). Simultaneously. they must also have expressed some interest in working more (1 <= Q422MOREHRS <= 3) and confirmed that they were available to work more (Q425STARTXWRK == 1) to fulfil all criterion and be defined as underemployed. Note that this definition (in terms of the numerical ranges specified) would only apply to the original version, as the recoding of missing, non-applicable or unspecified variable values as 0 would alter the listed mathematical inequalities that comprise the logical tests assigning status to observations.
This definition produces results that agree with Stats SA's derived version of "underempl" in the later version of the datafile only. In the older version, entries instead appear to be erroneously assigned into the underemployed category largely on the basis of their answer for Q422MOREHRS and Q425STARTXWRK. More specifically, having values in the ranges defined for both of these two variables is a necessary and sufficient condition for being assigned into the underemployed category. However, the ostensible cutoffs for the hours worked variables (Q418HRSWRK and Q420TOTALHRS) are irrelevant in the underemployment calculation.
This issue was fixed in the newer version, which has a definition congruent with the one detailed above. Users looking to check this themselves are advised that the redefinition of the "not applicable" category to zero valued entries for the Q418HRSWRK and Q420TOTALHRS variables must be taken into account when generating their versions of the underemployment variable.
Third, a number of extra variables were introduced in the later version. It is unclear why these are not present in the older version of the datafile as they are detailed in metadata that was released at the same time as the original data:
1) "Geo_type" - Geography type (e.g. urban formal, rural informal, etc.)
2) "Hrswrk" - Hourse worked. A derived variable that was probably aimed at getting around problems created by the recoding of the hours worked variables used in the derivation of the underemployment variable
3) "Metro_code" - Metropolitan area code (e.g. Cape Town, eThekwini, Johannesburg, etc.)
4) "Status_Exp" - Expanded unemployment status.
5) "Stratum" - 6 digit number representing stratum formed during master sample 2006 where digit 1 represents province, based on 2005 provincial boundaries, digits 2-3 represent the metro/non-metro area and digit 4 confers geography type.
Finally, the two versions have different weights. To DataFirst's knowledge, the weighting changes are not clearly documented by Stats SA. The most likely explanation for the difference between the two sets of weights is that the newer version is calibrated to an updated set of mid-year population estimates. Users are advised to remain aware of these slight calibration differences when employing weights.
INDIVIDUALS: labour market activity, labour preferences, labour market history, demographic characteristics, marital status, employment status, education, grants, tax.
The QLFS sample covers the non-institutional population except for those in workers' hostels. However, persons living in private dwelling units within institutions are 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 QLFS frame has been developed as a general purpose household survey frame that can be used by all other household surveys irrespective of the sample size requirement of the survey. The sample size for the QLFS is roughly 30 000 dwellings per quarter.
The sample is based on information collected during the 2001 Population Census conducted by Stats SA. In preparation for the 2001 Census, the country was divided into 80 787 enumeration areas (EAs). Stats SA's household-based surveys use a Master Sample of Primary Sampling Units (PSUs) which comprises of EAs that are drawn from across the country.
The sample is designed to be representative at the provincial level and within provinces at the metro/non-metro level. Within the metros, the sample is further distributed by geography type. The four geography types are: urban formal, urban informal, farms and tribal. This implies, for example, that within a metropolitan area the sample is representative at the different geography types that may exist within that metro.
The current sample size is 3 080 PSUs. It is divided equally into four sub-groups or panels called rotation groups. The rotation groups are designed in such a way that each of these groups has the same distribution pattern as that which is observed in the whole sample. They are numbered from one to four and these numbers also correspond to the quarters of the year in which the sample will be rotated for the particular group.
The sample for the QLFS is based on a stratified two-stage design with probability proportional to size (PPS) sampling of primary sampling units (PSUs) in the first stage, and sampling of dwelling units (DUs) with systematic sampling in the second stage.
Stats SA updated the QLFS results (2008-2013) to reflect the new population benchmarks from Census 2011. Users are advised to remain aware of these slight calibration differences between the previous version and the current (revised) data version when employing weights.
The sampling weights for the data collected from the sampled households are constructed so that the responses could be properly expanded to represent the entire civilian population of South Africa. The weights are the result of calculations involving several factors, including original selection probabilities, adjustment for non-response, and benchmarking to known population estimates from the Demographic division of Stats SA. The base weight is defined as the product of the provincial Inverse Sampling Rate (ISR) and the three adjustment factors, namely adjustment factor for informal PSUs, adjustment factor for subsampling of growth PSUs, and an adjustment factor to account for small EAs excluded from the sampling frame (i.e. EAs with fewer than 25 households).
In general, imputation is used for item non-response (i.e. blanks within the questionnaire), and edit failure (i.e. invalid or inconsistent responses). The eligible households in the sampled dwellings can be divided into two response categories: respondents and non-respondents, and weight adjustment is applied to account for the non-respondent households (e.g. refusal, no contact, etc.). The sampled dwellings with no eligible households, e.g. foreigners only, or no households, (i.e. vacant dwellings), do not contribute to the survey. The non-response adjusted weight is the product of the base weight with the non-response adjustment factor given above. If the PSU level non-response rate is too high, the non-response adjustment is applied at the VARUNIT level, where two VARUNITs have been created by grouping PSUs within strata. PSU level non-response adjustment is applied only if the corresponding adjustment factor is less than 1,5.
Final survey weights
The final survey weights are constructed using regression estimation to calibrate to the known population counts at the national level population estimates (which are supplied by the Demography division) crossclassified by 5-year age groups, gender and race, and provincial population estimates by broad age groups are used for calibration weighting. The 5-year age groups are: 0–4, 5–9, 10–14, 55–59, 60–64, and 65 and over. The provincial level age groups are: 0–14, 15–34, 35-64, and 65 years and over. The final weights are constructed in such a manner that all persons within a household would have the same weight.
The user of the data acknowledges that the original collector of the data, the authorized distributor of the data, and the relevant funding agency bear no responsibility for use of the data or for interpretations or inferences based upon such uses.
Copyright, Statistics South Africa
DDI Document ID
University of Cape Town
Date of Metadata Production
DDI Document version
Version 02 (April 2014) - Adapted version of the DDI "DDI-ZAF-DATAFIRST-QLFS-2008-Q3-V2" received from Data First.