![Sample our Area Studies journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5922/TOC-Subject-Sample-2021-Area-Studies.jpg"})
ABSTRACT
In this paper we use data from waves 1–5 of NIDS-CRAM to investigate labour market outcomes in 2020/1 for four age groups: youth (aged 18–24), prime-age adults (aged 25–39), middle-age adults (aged 40–54) and older adults (aged 55–64). We contrast outcomes just before and just after the advent of the COVID-19 pandemic and lockdown (February and April 2020) with outcomes one year later (March 2021), and study transitions between the periods. We find that although the NIDS-CRAM employment-to-population ratio was near identical in February 2020 and March 2021 (56.4% versus 56.6%), there had been extensive churning between the two periods. By March 2021, 23% of the February 2020 employed had lost work and 30% of the non-employed had found work. Amidst these changes, youth experienced the largest employment-to-population ratio increase, while older adults suffered the largest decrease in employment and a decline in participation rates (changes not statistically significant).
Acknowledgments
This paper draws from a working paper by the authors funded by the CRAM study (wave 5 policy papers; available on their website). The authors would like to thank Debra Shepherd, the NIDS-CRAM sampling team and other authors for their feedback and advice on that paper, and Ihsaan Bassier for providing data pertaining to the QLFS panels. In addition, we would like to thank our anonymous Development Southern Africa reviewers for their valuable feedback and help in refining the paper. Any errors or omissions remain our own.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 In addition to outright employment loss, there was a substantial further decrease in active employment in April 2020, with many workers maintaining an employment relationship but being furloughed or put on paid leave. However, these phenomena had ceased to be unusually prevalent by June and October 2020 (Jain, Bassier et al. Citation2020; Bassier et al. Citation2021).
2 While there has been some disaggregation of employment rates by age and a limited focus on middle-age adults (Espi et al. 2021), the outcomes for older adults approaching retirement age remain largely unexplored.
3 This was done to counterbalance attrition between waves of NIDS-CRAM (Ingle et al. Citation2021).
4 Occupation data was only collected for April 2020 onwards and industry data only for June 2020 onwards.
5 The QLFS panels were made available on a confidential basis to some members of SALDRU to collaborate with Stats SA on a data quality exercise. Matching data included demographic information, as well as respondents’ first and last names. Permission was obtained from Stats SA for project participants to also use the data for research purposes.
6 There are some differences between NIDS-CRAM and the QLFS that will affect both the historical benchmarking and any comparison made between the surveys the 2020/1 period. In NIDS-CRAM any work performed in the space of an entire month (e.g. April or June) was sufficient for a respondent to be classified as employed, whereas in the QLFS a reference period of one week (immediately preceding the interview) is used to classify employment. This means that it is possible that more casual and temporary work is captured in NIDS-CRAM relative to the QLFS, a fact that could explain part of the difference in the extent of churning observed in comparison to the historical QLFS panel. Secondly, employment information in each wave of NIDS-CRAM uniformly covers a single month, whereas employment information contained in the QLFS can be based on any week within a three-month period.
7 The fact that these estimates are based on cross-sectional weights from the first period in question, and not on specially created balanced panel weights, means that they do not account for panel attrition over the period, or for the sample rotation that is part of the QLFS design. This may cause some bias in the estimates.
8 None of these changes within age groups over time are statistically significant. All changes are not statistically significant unless explicitly stated.
9 The changes over time for these alternate age groupings were not statistically significant.
10 The unemployed (searching or discouraged) cannot be identified in February because of a lack of questions concerning job search for February in the wave 1 questionnaire.
11 For both youth and older adults (as well as other age groups) differences between February 2020 and later employment-to-population ratios are not statistically significant.
12 In comparing these proportions one must keep in mind that the periods do not line up perfectly, with interviews in QLFSs coming from any point in 3-month quarter periods, and in NIDS-CRAM aligned with particular months. However, the April 2020 to March 2021 period covered by NIDS-CRAM waves 1–5 matches the QLFS 2020 Q2 to 2021 Q1 period quite well. In addition, the majority of these demographic characteristics are unlikely to change rapidly over time (unless driven by changes in the sample in between waves).
13 The fact that these estimates are weighted using calibrated weights means that differences in the samples may be attenuated by things like adjustment for non-response and calibration to external population estimates. This is especially true for demographic characteristics like race and gender studied here.
14 The use of the 2010–14 period was determined by data availability. It is possible that rates of churning were different (higher) in the 2015–2019 period but the extent of the differences in transition rates observed in this section mean that we are confident in the exceptionality of the employment dynamics depicted by NIDS-CRAM relative to historical estimates.
15 As discussed in the ‘Data and methods’ section, these benchmark estimates are based on the balanced panel of individuals remaining in the QLFS for four consecutive periods of the rotating panel in the reference periods. If the smaller sample of people who remain in the sample for the full four periods is not a random subset of the full sample, this will affect the reliability of these estimates as a benchmark for the NIDS-CRAM results. In particular, if attrition differentially selects on lower churn patterns, and if this differential selection is greater than any existing in NIDS-CRAM, this will be an issue. Earlier work by the authors (Espi et al. Citation2021b) compared transitions in NIDS-CRAM and the same QLFS data but for just two consecutive quarters or periods. Large discrepancies in churning rates, including in job finding rates, remained when looking at this shorter period. This shows that it is unlikely that the effects of attrition (which are likely to be small when studying a panel over only two consecutive periods) are responsible for the observed discrepancies between contemporary and historical rates of churning.
16 These ranges cover the job finding rates for different non-employed groups (the searching unemployed, discouraged work-seekers and the not economically active).
17 Disaggregating by age group and distinguishing between four employment statuses means that the sample sizes for estimates of transitions within groups are quite small in some cases (see ), and the uncertainty of these estimates should be borne in mind.
18 It would be possible to identify April 2020 lockdown job losers, but it is not possible to identify those who are discouraged or not economically active as a result of the lockdown.
19 Substantial additional job loss after April 2020 has been observed elsewhere (Espi et al. Citation2020).
20 A range of occupations fall under the elementary occupation banner, including cleaners and helpers, food preparation assistants, and labourers in agriculture, forestry, mining, construction, manufacturing and transport (ILO Citation2012).
21 The community, social and personal services sector includes public administration and government activities, education services, health and social work, professional organisations and trade unions, along with recreational, sport and entertainment activities (Stats SA Citation1993).
22 There were no clear downward trends in industry or occupation employment totals. This is because the first period in both cases was around the peak of the COVID-19 employment shock and so we only capture recoveries (upward trends) thereafter, and cannot look at declines relative to pre-crisis levels.
23 The QLFS can be treated as a panel for several waves from the beginning of 2020 if information matching individuals across waves is provided. We did not perform this analysis in this paper as this information has not been made available by Stats SA.