Reza C. Daniels. Principal Investigator: National Income Dynamics Study – Coronavirus Rapid Mobile Survey (NIDS-CRAM), SALDRU Research Associate & Associate Professor, School of Economics, UCT
February 17th 2021 marked the official release of the National Income Dynamics Study – Coronavirus Rapid Mobile Survey (NIDS-CRAM) Wave 3 data. Several researchers highlighted important employment trends that suggested a partial but incomplete recovery to pre-pandemic (February, 2020) employment levels.
One of the most important findings was that in October 2020 (Wave 3 of NIDS-CRAM), the employment rate for 18-64 year olds saw a partial recovery to pre-lockdown levels in February 2020.
This led to a robust debate about whether this partial recovery was plausible, given other things that we know happened in the South African economy during the reference period of 2020. Inevitable comparisons were made to Statistics South Africa’s (StatsSA) Quarterly Labour Force Survey (QLFS) in 2020, which is a survey designed to estimate nationally and provincially representative employment and unemployment levels and rates for four quarters of the year.
But this is not a strictly valid comparison because the NIDS-CRAM and the QLFS are designed to do different things.
In particular, the QLFS is designed to estimate nationally representative employment and unemployment rates, unlike the NIDS-CRAM survey. In this article, we explore what the NIDS-CRAM data can tell us about measuring employment in South Africa.
Both SALDRU, in the case of the National Income Dynamics Study (NIDS) & NIDS-CRAM, and StatsSA, in the case of the QLFS, provide documentation with the release of the survey data that is important for users to read. In it, technical information concerning the survey design, sample, fieldwork and weights are discussed. SALDRU has always taken the view that it is better to be transparent about the overall survey process, so that researchers are better equipped to understand the limitations and strengths of the data. The Panel User Manual is the key document here for both NIDS and NIDS-CRAM, and each release of data is accompanied by an update to this manual that captures pertinent events in the data production process for each wave of data collected.
For data production organisations, a key phrase used that helps us to understand whether the quality of the data is sufficient in each release of data is “fit for purpose”. This phrase captures the idea that all data production efforts have limitations associated with them. It is the job of each data production organisation to ensure that, from a statistical perspective, the data production process minimizes the mean square error (MSE, which consists of bias and variance) of the estimates of parameters from that data (e.g. means, medians, modes or totals). The ability of any data production organisation to do this is proportional to the budget and inversely proportional to the time-frame – meaning that MSE goes up if the time-frame is short. That isn’t necessarily a problem because not all components of MSE threaten the data’s characterisation as “fit for purpose”.
The purpose of NIDS, and thus by extension NIDS-CRAM, is to follow a sample of individuals that was nationally representative of South Africa in 2008 over the course of their lives.
It is a voluntary longitudinal (panel) survey, which means that respondents can choose to participate in each round (wave) of the survey. Because respondents drop out of the survey with time, which we call attrition, the sample is periodically topped up (refreshed) with new individuals to ensure that it remains representative of the country as time elapses. The NIDS sample was refreshed in 2017 (Wave 5), and it was that sample that constituted the sampling frame of NIDS-CRAM.
It was an explicit decision of the sampling team of NIDS-CRAM to weight each wave of NIDS-CRAM data to the South African population in 2017, as captured by NIDS Wave 5. Therefore, estimates of parameters (e.g. employment rates or population totals) from any wave of the NIDS-CRAM data reflect the 2020/2021 experiences of a representative sample of South Africans drawn in 2017, not in 2020 or 2021. From a statistical perspective, this limitation is necessary because of the sampling frame. While the sampling team could have weighted the NIDS-CRAM population to 2020/21 population totals, this would have been incongruent with drawing a sample from a 2017 sample frame, resulting in biased parameter estimates; thereby reducing the quality of the data by increasing the MSE.
With this caveat in mind, let us review the Wave 3 NIDS-CRAM findings for employment dynamics.
Figure 1: Employment outcomes for 18-64 year olds in NIDS-CRAM Waves 1 – 3
Figure 1 shows that between February and April 2020, we had previously found a substantial increase in those who were not employed (defined as broadly unemployed plus not economically active) – from 43% to 52%, as well as an increase in furloughed workers (i.e. employed with partial or no pay). The reason the population of unemployment and not economically active are grouped together is because of questionnaire design constraints for the month of February in Wave 1 of NIDS-CRAM.
Figure 1 also shows that by Wave 3 it was evident that the percentage of people employed in October 2020 was much closer to its February pre-pandemic level. This employment rate is best thought of as the employment to population ratio because it is not restricted to the sub-population who are in the labour force (defined as the total number of people who are currently employed plus the number who are unemployed and seeking employment). The fraction of people employed (including furloughed workers) changed from 57% in February, to 48% in April and June, to 55% in October. This is what has been described by different researchers as a “bounce back” and a “partial recovery”, but it is important to remember that the rates are being compared to a February pre-pandemic baseline.
These results are robust to a number of data checks in NIDS-CRAM. The employment recovery described here remains broadly the same after testing a number of alternative approaches to measuring employment, including using different definitions of employment that take account of different age categories or what counts as “employment” (i.e. using full-time equivalent work based on days worked in the week rather than the count of the employed). Similarly, the results are unchanged after using a number of different weighting schemes. All NIDS-CRAM authors analyzing the employment data find the same broad trends.
Notwithstanding these trends, the South African labour market looks very different in October 2020 compared to before lockdown (February).
Of those who lost their jobs in April, only around half were employed again by October, while about a third of those without employment in February were employed in October. There were also occupational and industrial sector level changes accompanying the employment changes, though it is too early to tell if these changes are temporary or a more permanent feature of the South African economy.
The employment recovery was positively correlated with educational attainment among prime-age adults and youth. On the other hand, an April job loss was more likely to be persistent for youth relative to older groups. Employment history correlated strongly with 2020 employment outcomes: individuals with a more extensive history of employment were more likely to remain stably employed, or, among the non-employed, to find work.
Further disaggregating the data by gender provides insight into the different experiences of men and women over this period. In Waves 1 and 2 of NIDS-CRAM it was found that women were particularly hard-hit by the initial lockdown phases and school closures, both in terms of labour market outcomes and childcare responsibilities.
With the move from Level 3 lockdown in June to the much less restrictive Level 1 lockdown in October, the recovery in employment was shared almost equally between men and women. However, given how much larger the fall in women’s employment was as a result of the initial shock to the labour market, women still remained behind men in terms of reaching their pre-pandemic employment levels in October.
Lastly we remind readers that caution should be exercised in generalizing NIDS-CRAM results to the overall 2020/21 population. The NIDS-CRAM 2020/21 data is drawn from a sample based on NIDS 2017, which had a higher labour force absorption rate to start with (55% in 2017) compared to StatsSA’s QLFS 2017 (48%) (see Ranchhod & Daniels, 2021).
 Bassier, I., Budlender, J. & Zizzamia, R. (2021) The labour market impacts of COVID-19 in South Africa: An update with NIDS-CRAM Wave 3.
Casale, D. & Sheperd, D. (2021) The gendered effects of the Covid-19 crisis and ongoing lockdown in South Africa: Evidence from NIDS-CRAM Waves 1-3.
Espi, G., Ranchhod, V. & Leibbrandt, M. (2021) Age, employment history and the heterogeneity of Covid era employment outcomes.