August 2010

A Youth Wage Subsidy Experiment for South Africa

Concept Note

  1. Description of the Intervention

The project seeks to investigate the impact of wage subsidy on the probability of employment and post-unemployment earnings of South Africa’s youths. The project will select a random sample of 4,000 unemployed young workers in the age group 20 to 24. This sample will be split into a treatment and control group, using pair-wise matching technique as the randomization strategy. The treatment group will receive a wage subsidy, and the control group will not. These wage subsidies will be paid directly to the firms that employ the young people. The total subsidy available for each individual will be R5,000 (approximately $670), constituting up to half the individual’s wage per month and covering at least six months of employment. The monthly amount available per month is thus capped at approximately R800 (just over $100). The national median monthly wage for youth aged 20-24 was R1,500 (approximately $200) in 2007. Gauteng median wages are higher than this whilst those in the other two provinces of Limpopo and KwaZulu-Natal are lower. Thus the wage subsidy is of the correct magnitude to be attractive to firms based on the wages paid to the currently employed youth.

It is expected that the wage subsidy will help improving employment chances of young workers by offsetting some of the training costs as well as costs associated by hiring first-time workers whose productivity has not been revealed. The project will provide evidence to inform policy makers on the potential for a national youth wage subsidy to lower youth unemployment in South Africa – the country where persistently high unemployment is one of the most pressing socio-economic challenges facing the Government, and where three quarters of the unemployed are young people (in June 2009, the total unemployment rate was 23.6 percent, and the rate for young workers was much higher – for example, 46.4 percent for the 20-24 year olds).

  1. Main Research Questions and Hypotheses

The main research questions of this study are as follows:

  1. How effective is the wage subsidy program in improving the probability of obtaining and keeping employment?
  2. How effective is the program in improving the quality of post-unemployment jobs?

Note that the answers to above two questions, together with program costs, will also allow addressing the question whether the program is cost-effective, that is, whether its benefits outweigh the costs.

To primary outcome indicators to be used by this study are:

  1. Duration of unemployment of program participants, as measured by time spent in unemployment
  2. Quality of post-unemployment jobs, as measured by (i) wages, (ii) duration of post-unemployment job, and (iii) type of appointment (fixed-term vs. permanent).
  1. Impact evaluation
  1. Methodology

The study will consist of the following steps:

  1. Selecting a random sample of unemployed and economically non-active youths (4,000 20 to 24 year olds, residing in Gauteng, KwaZulu-Natal, and Limpopo provinces).
  2. The sample is drawn in two ways. Random sampling from clusters in each of the provinces will account for approximately 2,500 of the sample. The remaining 1,500 will be drawn from Department of LabourLabourCentres. This will provide inferences into the role and effects of labour market institutions.
  3. Approximately 2,000 will come from Gauteng (Johannesburg), and 1,000 from KwaZulu Natal and Limpopo. The Limpopo sample includes more rural areas to gather inferences on how such an intervention will affect rural as well as urban areas.
  4. Carrying out a baseline survey of persons included in the study. Data is collected through an electronic (hand-held computer) questionnaire and includes:
  5. Contact details
  6. Demographic information
  7. Education
  8. Family relationships and characteristics
  9. Aspirations, life satisfaction and crime
  10. Activities (current, previous and secondary) with specific questions depending on whether the activity was:
  11. working for someone else including details about remuneration
  12. self employment including details about remuneration
  13. further or higher education
  14. unemployed but searching
  15. unemployed but not searching
  16. Randomly allocating the sample into the treatment and control groups using a pairwise matching randomization strategy. This will be split by:
  17. Province and sampling strategy (random sampling vs. labourcentres)
  18. Gender
  19. Matching will be run on age, cluster, education (most likely a binary variable for those with and without matric), intensity of search, networks, and potentially raven’s matrix scores.
  20. Giving individuals in the treatment group a subsidy/voucher letter indicating eligibility for the subsidy and explaining how the subsidy works.
  21. Paying wage subsidies to employers who provide jobs to the participants included in the treatment group. The total amount allocated to each individual in the form of the wage subsidy will be R5,000. The subsidy will be valid for at least six months subsequent to employment. The subsidy can be claimed for up to half the individual’s wage. This will be paid to the firm that employs a person with a wage subsidy on a monthly basis, until the subsidy is exhausted. For example, a firm which employs an individual at a wage of R2,000 per month may claim R833 per month for six months provided that they continue to employ this person for the full six month duration. Only firms that are registered for tax will be eligible for the wage subsidy.
  22. Carrying out a follow-up survey of persons included in study. All reasonable attempts will be made to re-interview all 4,000 individuals in the sample. This will be facilitated through the capture of contact details, including contact details of friends and family, and through the capture of location data.
  23. Carrying out a survey of firms that employ persons included in study that find a job.
  24. Analyzing baseline, follow up and firm surveys and identifying the impact of the program using the difference-in-differences approach.
  25. Formulating recommendations for the potential for scaling up a targeted youth wage subsidy to the national level.
  1. Sample selection

The project will select a random sample of 4,000 unemployed young workers in the age group 20 to 24. These individuals will be drawn from the following clusters.

  • Gauteng:

Alexandra (two clusters); Diepkloof Zone 6; Dlamini; Eastbank; Hillbrow; Ivory Park (two clusters); Jabulani; Johannesburg CBD; Meadowlands Zone 8; Mofolo Central; Naledi Ext 1; Orange Farm Proper; Orlando East; Protea Glen; Thulani.

  • KwaZulu-Natal:

Durban CBD; Kwadabeka R; Kwambiza; Mozambique B; Qhodela; South Beach; Sydenham; Umbumbulu; Umlazi.

  • Limpopo:

Dikgale (four clusters); Luthuli; Madiba Park; Seshego (three clusters).

These clusters were sampled based on the 2001 Census. The probability of clusters being drawn was proportional to the number of young people living in these clusters at the time of the 2001 Census.

Part of the sample will also be drawn from the Department of Labour’sLabourCentres to provide some inferences into the role of institutions in job creation. Some of the functions of these LabourCentres include help with job search, assistance with c.v. preparation and matching of firms and workers. LabourCentres that are the closest to these clusters were selected. In total 4,000 individuals will be interviewed in the first round. Approximately 2,500 of these will be randomly drawn from the clusters and approximately 1,500 will be randomly selected from the LabourCentres.

  1. Power calculations

The experiment design was conditioned on a sampling scenario for randomisation based on achieving a power of 80% – i.e. we will correctly identify a difference in groups with a probability of 0.80. Furthermore, we set a significant level of 0.05 and the 95% confidence interval for the transition rate (φ) extends 5 percentage points above and 5 percentage points below the expected transition rate for the control group (i.e. φ lies in the centre of a 10 percentage point band). Power calculations were undertaken in programme called ‘Optimal Design’ (Spybrook, Raudenbush, Liu, Congdon and Martinez, 2008).With the sample size of 4000 and the panel data evidence on transition rates, into employment, the implied transition rate of the treatment group needs to be 4-5 percentage points higher than observed in the control group in order to achieve a power of 80%. Pairwise-matching or stratifying (blocking) prior to randomization will increase the power since it reduces the heterogeneity within the ‘blocks’, ‘bins’ or pairs and increases the precision of the treatment effect estimate. This approach also allows for closer analysis of how the treatment would affect sub-groups. Unfortunately ‘Optimal Design’ does not allow for power calculations on binary outcomes that take into account this ‘blocking’. We are using a greedy Mahalanobis matching algorithm adapted from a programme supplied by David McKenzie at the World Bank to match pairs, after stratifying by gender, sample and geographical location.

  1. Validity of identification strategy for identifying causal impacts

Attrition: Significant effort has been taken to collect information from respondents , this includes the provision of multiple contact numbers and contact will take place by means of mobile phones. Id numbers of participation have been collected for verification of identify of participants. Trials to verify whether these individuals can be contacted through these telephone numbers, carried out approximately 10 months after the interview, indicate that 95% could be reached. The remaining 5% will have to be tracked by visiting their homes.

Re-interview process: Re-interview w ill take place as follows; those in the treatment group who received the sample will be interviewed at community centers, while the control group interviews will be conducted through “house-to-house” interviews. No financial incentives have been included to encourage re-interview participation, but this may be considered depending on the funding available.

Labor Centre sample: The inclusion of the Labour Centre sample allows for a comparison to be made between the unemployed in general and those that use the Labour Centre, and between the subsidy by itself and the subsidy coupled with the support for job seekers given by the LabourCentres, such as the matching of potential employers with potential employees. Thus the aim is to investigate the role of job matching and job searching assistance in increasing the probability of finding employment. Given that the Labour Centre sample is not drawn from the same population as the general sample – individuals that attend the LabourCentres clearly self-select themselves into this method of job search – these results are not directly comparable. However, the information on the Labour Centre sample is of interest to the Department of Labour for them to understand how these help the unemployed to find jobs. Furthermore, those in the general sample are asked questions about whether they know of and have used a Labour Centre in order to collate the results.

Extending the number of treatment groups: Extending the treatment groups to include a group comprising of wage subsidy recipients and Labour Centre will be problematic given the small sample of 18-24 year olds identified in LabourCentres which in turn has implications for the power of the randomised experiment and thus the robustness of the results obtained.

  1. DATA SOURCES
  1. Baseline survey

The first round survey included sample individuals from three provinces: Gauteng; KwaZulu-Natal and Limpopo (see above). It was concluded in May 2010.

  1. Follow up surveys

Surveys have already been conducted with firms who are currently employing young personsselected in the study to determine factors influencing their hiring decisions. During the process of allocation of vouchers, interviews will be conducted with all individuals to determine their current labour market status, search behaviour and other key variables.

The follow-up to the baseline survey will re-visit individuals interviewed in the baseline survey. It is planned to be implemented May to July 2011. The impact evaluation report will be produced and disseminated by December 2011.

  1. TIMELINE

Activity / Period
Baseline survey / April -- July 2010
Follow-up survey / May -- July 2011
Analysis and dissemination / August -- December 2011
  1. EXPENDITURES AND RESOURCES

Expenditures / Amount (in $1000)*
First Round Survey:
Survey design, first round of individual and firm surveys / 640.4
Subsidies:
Budget for 800 subsidies / 556.4
Subsidy management:
Staff and other costs associated with managing the subsidy payments / 88.1
Follow-up Survey:
Tracking of individuals, follow-up individual and firm surveys / 544.0
Administration:
Management fees / 253.4
TOTAL / 2,082.4
Resources / Amount (in $1000)*
Department of Labour / 206.6
National Treasury / 1,015.8
3ie / 688.6
PSPPD / 96.4
SIEF** / 75.0
TOTAL / 2,082.4

*Based on exchange rate $1 = 7.26091 ZAR.

**SIEF funding will cover part of the costs of the first round survey.