https://orcid.org/0000-0003-4511-1305
![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
Wage subsidy-based job retention policy has served as a dominant tool used to mitigate job losses in the context of COVID-19. In South Africa, such a policy served as a core component of the government’s policy response: the Temporary Employer-Employee Relief Scheme (TERS). We make use of longitudinal survey data to analyse aggregate and between-group TERS receipt during the pandemic as well as the relationship between receipt and job retention. We find that the policy reached millions of workers but coverage was highest during the beginning of the pandemic. Although several groups disproportionately benefited, vulnerable groups were over-represented amongst recipients over time. Benefits were higher in relative terms for lower-wage workers. Although not causally identified, we find evidence of a significant, positive relationship between TERS receipt and job retention, consistent with the policy being successful in its aim of minimising job losses, however only during the most stringent lockdown period.
Acknowledgments
The authors would like to express their gratitude to Vimal Ranchhod based in the School of Economics at the University of Cape Town for his helpful feedback on an earlier version of this paper. They are also thankful to an anonymous reviewer for very thorough comments and suggestions. This paper draws from the following two working papers by the authors as part of the NIDS-CRAM Wave 5 study: Köhler, T. and Hill, R., 2021, ‘Wage subsidies and COVID-19: The distribution and dynamics of South Africa's TERS policy’, Development Policy Research Unit Working Paper 202109; Köhler, T. and Hill, R., 2021, ‘The distribution and dynamics of South Africa's TERS policy: Evidence from NIDS-CRAM Waves 1-5’, NIDS-CRAM Wave 5 Policy Paper 7.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Correction Statement
This article was originally published with errors, which have now been corrected in the online version. Please see Correction (http://dx.doi.org/10.1080/0376835X.2023.2224694).
Notes
1 An exception to this is when an employer employs fewer than 10 workers. In this case, the UIF paid these workers directly upon receipt of their bank account details from the employer. However, this changed from May 2020 when the UIF introduced direct payments to employee bank accounts although applications still needed to be submitted by the employer.
2 Calculated using QLFS 2021Q1 (StatsSA) microdata.
3 Namely, the Casual Workers Advice Office, the Women on Farms Project, and Izwi Domestic Workers Alliance.
4 The relevant clauses for the calculation of TERS benefits were clauses 3.4, 3.5, 3.6, and 5.3. Importantly, the TERS could only cover the cost of salaries and no other firm expense.
5 Because of the NIDS-CRAM sampling design, the sample is regarded as ‘broadly’ representative of the adult South African population in 2020. Specifically, the weighted NIDS-CRAM estimates are only representative of the outcomes in 2020 of those aged 15 years and older who were surveyed as part of the NIDS in 2017 and were followed up on 3 years later.
6 Wave 1 included only the relevant TERS question for regular and casual workers. One possible implication of this may be that the Wave 1 estimate of the number of recipients is underestimated. However, the magnitude of such an underestimate is not expected to be large given that in Waves 2 to 5, the weighted estimate of TERS receipt among those who run a business and the self-employed is between just 2.3% and 5.2% of TERS, respectively recipients.
7 The questions for both samples are as follows, respectively: ‘Did you receive any money from the UIF’s TERS in [reference month]?’ and ‘Did you receive any money from the UIF’s TERS for yourself or any staff in this business in [reference month]?’.
8 These are as follows: ‘No because my company is not eligible’, ‘No because I do not know where or how to apply’, or ‘No because I get support from other private institutions’.
9 This is because the sample sizes for each relevant ‘No’ response are too small for any reliable statistical inference (1–30 observations in a given wave). The coding of reports of ‘Don’t know’ and refusals to answer the relevant question also do not significantly affect our estimates given the very small subsample affected (1 to 36 observations in a given wave).
10 For instance, due to payment delays, a worker may report receiving a TERS payment in June 2020, but this payment may have been intended to cover the worker for May 2020. We are unable to identify the extent of this discrepancy given data limitations.
11 For instance, in an analysis of South African household survey data, Wittenberg (Citation2017) shows that individuals who respond in brackets rather than with point estimates tend to have higher incomes.
12 Using this method, the following number of values per wave were identified as outliers: 3 for February 2020 wages in Wave 1, 2 for April 2020 wages in Wave 1, 2 for June 2020 wages in Wave 2, 5 for October 2020 wages in Wave 3, 1 for January 2021 wages in Wave 4, and 4 for March 2021 wages in Wave 5.
13 For each wave, the relevant sample sizes are as follows: n = 2 650 for February 2020 (78% of the employed sample); n = 2 110 for April 2020 (77%); n = 1 756 for June 2020 (79%); n = 2 357 for October 2020 (83%); n = 2 217 for January 2021 (90%); n = 2 445 for March 2021 (89%).
14 Note, however, that due to a lack of data on employer contributions to worker wages, our benefit estimates assume zero employer wage contributions. This approach necessarily overestimates total benefit figures, however, given the incentive-compatibility structure of the programme, we believe firms would be incentivised to contribute as little as possible towards worker wages, thus making this overestimation minor. We do emphasise that the reader should bear this in mind when considering these results, however.
15 These include creating a smaller, matched sample of treatment and control observations with similar propensity scores, stratifying observations on their propensity scores and estimating effects within strata, or inverse probability weighting.
16 Researchers often use ad hoc methods for sample trimming to address a lack of overlap in the distribution of propensity scores by treatment group. Crump et al. (Citation2009) develop a systematic approach and propose trimming all units with propensity scores outside the range [0.1, 0.9]. Given the already small sample size of the NIDS-CRAM, we however choose not to trim our observations given that doing so results in a significant sample size reduction.
17 Our probit results are not shown here but are available upon request.
18 The occupation codes used follow the International Standard Classification of Occupations (ISCO-08). The industry codes used are those found in the Statistics South Africa’s General Household Survey (2005) industry code list. The 1-digit level is the lowest level of disaggregation in the data.
19 However, it should be noted that the NIDS-CRAM data does not permit us to estimate TERS receipt across the whole period (that is, in every month) but only in given reference months. As such, this aggregate estimate is likely underestimated.
20 We choose to use pre-pandemic (February 2020) wage data in this calculation given that self-reported wages in every period thereafter in the NIDS-CRAM likely includes any benefit received among TERS recipients.
21 For the individual proportions of TERS recipients and employment shares used in the calculation of these ratios, refer to and in the appendix.
22 Considering the estimates in panel (b) rely on the imputed TERS benefits estimates, they are subject to the same caveats discussed earlier in the paper.
23 We compare TERS benefit amounts to pre-pandemic (February 2020) wages here because, as outlined above, we make use of this data to estimate TERS benefit amounts.
24 Recall that Wave 2 outcomes for the NIDS-CRAM are reported for June 2020, which was after the South African economy had begun a staggered reopening by moving to lockdown level 3 on 1 June 2020. As a result, a number of individuals that had lost their jobs during the ‘hard lockdown’ could have found themselves rehired into new positions by the time they were surveyed in Wave 2.
25 In an online survey of 2 688 individuals during April/May 2020, Statistics South Africa (Citation2020) reports that 21.3% of respondents who were employed during the lockdown reported decreased wages as a result of the pandemic. Although this study was not nationally representative, it serves to illustrate how businesses and institutions may have reacted to the pandemic.