Identifying the Wage Differential in the Temporary Employment Services Sector: Evidence for South Africa using Administrative Tax Records

Abstract

Although the temporary employment services (TES) or labour broker sector has been growing in recent decades, and there has been much heated public debate on whether the jobs offered constitute ‘decent work’, there has been little empirical research on this sector in developing countries. In this paper, we use a unique administrative panel dataset based on income tax records for the period 2011–2015 in South Africa, to explore the wage and benefits differentials between TES and non-TES workers. We find a substantial gross wage differential of around 88 per cent, which remains high at 34 per cent even after accounting for worker fixed effects and controlling for the individual and job characteristics available in the data. We also show that TES workers are much less likely to report benefit contributions than non-TES workers, and when they do, their contributions as a percentage of the gross wage are on average much lower than among non-TES workers. These results add substance to the arguments that TES workers are in a more precarious position than non-TES workers, and that this form of employment contributes to high levels of labour market inequality in South Africa.

JEL CLASSIFICATION CODES:

• J31
• J41

Keywords:

• Temporary employment services
• wage differentials
• administrative data
• South Africa

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Other reasons suggested are that employers might prefer to use TES workers because they would have already gained some experience working under organisational rules or because they could use the opportunity to screen TES candidates before offering permanent positions.

2 Menon (2019) reports a wage penalty of 21 per cent to short-term workers when compared to long-term workers, while Das (2019) finds that temporary workers or casual workers earn a wage premium at the lower end of the earnings distribution but suffer a penalty of around 30 per cent at the upper end of the distribution.

3 For example, for South Africa, Wittenberg (2017) finds that the Quarterly Labour Force Survey under-reports income by around 40 per cent on average compared to data from the tax records.

4 To try and identify TES workers from South Africa’s Quarterly Labour Force Surveys (QLFS), Benjamin, Bhorat, and van der Westhuizen (2010) and Bhorat et al. (2016) used the standard industry classification code 889, Business Activities Not Elsewhere Classified, which falls under the broader category Finance and Business Services, and which includes ‘labour recruitment and provision of staff; activities of employment agencies and recruiting organisations; hiring out of workers (labour broking activities)’. However, this code also lists another 10 activities which are not distinguishable from the labour broker sector, including employment in security services and debt collecting/credit rating agencies which have grown rapidly in South Africa over recent decades. In fact Budlender (2013) shows that a large percentage of the workers in this category had permanent contracts.

5 Another benefit of the tax data is that because of the very large number of observations (and because individuals can be matched over the tax years), we are able to estimate worker fixed effects with much more accuracy than with household survey data.

6 Based on data from the Quarterly Labour Force Survey (a household-level survey), Quarter 4 of 2019.

7 The Labour Relations Act 66 of 1995 is one of South Africa’s overarching pieces of labour legislation. It has been amended many times to reflect the changing dynamics of the labour market and provide protection to different types of workers. The amendments governing temporary workers were contained in Labour Relations Amendment Act No.6 of 2014.

8 While the IRP5 data is now available up until 2018, the indicator for TES firms is incomplete after the 2015 tax year.

9 For a more detailed discussion of the structure of the data, see Pieterse et al. (2016).

10 This is the equivalent of US$156.50 using an exchange rate of R12.78/$for 2015. If workers earn below this amount and no tax was deducted, then the employee is not issued with either an IRP5 or an IT3(a) form.

11 The levy is paid as a portion of an employer’s salary bill to the revenue service. The levy is then distributed to encourage skills training and development.

12 This affects 5 per cent of the sample amounting to around 2.9 million non-TES job contracts and 190,000 TES job contracts (out of a total of ∼58 million job contract observations).

13 Many of these overlapping contracts at the same firm have the same start and end dates and earnings information and are therefore likely to be duplicates. Where time period or earnings information differs, it is likely that they are IRP5 revisions, but because we are unable to identify which is the most recent version, we adopt the averaging approach. Around 5 per cent of the full sample of 58 million observations were assumed to be duplicates.

14 More detailed information on the precise cleaning and construction of the main job sample can be found in the published working paper version of this article (Cassim & Casale, 2018). In this process, we closely followed the conventions used by others who have worked with the tax data.

15 Removing observations where individuals earn in excess of R10 million per annum excludes only around 10 contracts. Removing observations where individuals earn below R2000 results in a loss of around 2.7 million observations, of which just under 10 per cent involve TES jobs.

16 Relative to other developing countries, the proportion of people employed by TES firms in South Africa may seem fairly low. In China, 20 per cent of all urban workers were ‘dispatched workers’ (2011), 33.9 per cent of India’s organised manufacturing sector was ‘contract labour’ (2010–2011), and 16 per cent of all non-agricultural employment in the Philippines was contract or project-based labour (2012) (ILO, 2015). However, it should be noted that data on TES work specifically is hard to come by, and these studies analysed different kinds of temporary employment relationships to what is captured in the South African tax data, namely workers employed through Temporary Employment Services agencies.

17 We use similar methods to Segal and Sullivan (1998) whose administrative employee data for the US are most similar to ours.

18 For simplicity, we assume that education is largely time-invariant here as survey data show that only 4 per cent of the employed are enrolled in education at the same time as being in employment (Kerr, David, & Wittenberg, 2016).

19 When age and its quadratic are included as continuous variables instead, the expected signs are obtained and the other regression coefficients remain largely the same in the analysis.

20 As in the descriptive table, job contract length is included as a series of dummies in the regressions, with the top category truncated at one year or more, as this is how it is collected in the IRP5 forms. Using the panel nature of the data we can disaggregate this top category of ‘a year or more’ further into the number of years spent at the same firm. However, doing so has very little effect on the main variable of interest. For example, the coefficient on the TES variable in our final specification falls from −0.292 to −0.284. We prefer not to include this estimated ‘tenure’ variable in the main regressions though, as there is some ambiguity around how to code those who have multiple contracts of less than a year and who reappear in the same firm in subsequent years.

21 These dummies also take on a value of zero for the TES spell.

22 The transformation of the TES coefficient is based on the estimation method used in Halvorsen and Palmquist (1980) to interpret dummy variables in semilogarithmic equations.

23 In addition, we ran the regressions with a sample including only individuals with one job contract per year (37,630 974 observations compared with 41,389,932 in Table 4), to see if those who switched frequently within years were driving the results. However, the coefficient using specification 3 was −0.295, only marginally different from the coefficient of −0.292 observed in Table 4.

24 Further, because there is no information on hours worked per day in the tax data, it is not possible to measure how much of the wage differential is due to TES workers potentially working fewer hours per day in their contracts compared to non-TES workers.

25 Earnings quintiles are estimated using the full earnings distribution, i.e. including both TES and non-TES workers.

26 Also, some of these regressions by industry are estimated with a much smaller number of observations and therefore should be viewed with caution as there will be fewer switchers.

27 Around 3.7 million observations were dropped, or 9 per cent of the sample from Table 4. Half of all TES workers had one TES spell between 2011 and 2015 as opposed to having many consecutive TES contracts over the period of assessment.

28 The coefficient on the TES dummy in Column 2 (−0.372) is larger than that in Column 1 (−0.308) because the jobs just before and just after the TES spell, during which wages tend to be lower, are removed from the non-TES comparison group and accounted for by the dummies. In other words, the estimate in Column 1 reflects the wage differential between the TES job and all non-TES jobs, while in Column 2 it reflects the differential between the TES job and only the non-TES jobs outside of the ‘prior/post’ period.

29 The 2009 Sanlam Survey suggests that on average, employee contributions were 5.9 per cent of gross earnings while the employer contributions were 9.9 per cent of gross earnings. Further, in a note on retirement fund contributions, Momentum (2016) consistently uses the example of employers contributing double that of employees to retirement schemes with an upward limit of 15 per cent of gross earnings.

30 Note that the gross differential between TES and non-TES workers for this group of positive-contributors is lower than for the full sample because they would include a less precarious group of TES workers if they are reporting some benefits.