DEPARTMENT OF ECONOMICS Uppsala University Bachelor Thesis (a Minor Field Study) Authors: Anders Larsson and Martin Nybom Supervisor: Anders Forslund Spring 2006

Government response to poverty and unemployment in South Africa A micro-level evaluation of the Expanded Public Works Programme

1

Abstract Using data from the Labour Force Survey conducted by Statistics South Africa twice yearly this thesis intend to evaluate the Expanded Public Works Programme regarding its effectiveness in creating employment and raising income in households with participating individuals. The South African labour market is well known for its high rates of unemployment and also its segregation, primarily between black and white people, but also young people are having a hard time finding jobs. In order to fight these problems the South African government has launched the Expanded Public Works Programme (EPWP) which provides low- semi-skilled labour with short term employment, the primary target groups being black and coloured people, women, disabled people and young people. Our findings indicate that the EPWP does not significantly enhance the individual’s probability of being employed, nor does it raise the per capita income of households with participating individuals.

Keywords: South Africa, Labour market, EPWP, Propensity score matching, unemployment.

Acknowledgements The field work for this thesis was carried out in Pretoria during November-January 2005-2006. This would not have been possible without the economic support from SIDA. We would like to thank Louise Kennerberg at IFAU for helping us with contacts at Statistics South Africa, Wanda Steyn for helping us in Pretoria and accommodating us at Statistics South Africa and also Astrid and Hans Näslund for their hospitality in Pretoria. We would also like to thank Statistics South Africa for providing us with the Labour Force Survey datasets.

2

1. Introduction ............................................................................................ 4 2. Unemployment and poverty in post-apartheid South Africa ................ 6 2.1 Public Works Programmes and the EPWP .......................................................... 8 2.2 Earlier research on the EPWP ........................................................................... 10 2.3 Earlier research on PWPs and job training in general ..................................... 11

3. Data ....................................................................................................... 13 3.1 Sample reduction .................................................................................................15 3.2 The variables .......................................................................................................15

4. Evaluation and the Propensity Score Matching Method ..................... 18 4.1 How to tackle the evaluation problem ............................................................... 18 4.2 Propensity Score Matching................................................................................ 20 4.3 The various methods to select the controls and our choice ...............................21

5. Results: the effect of participation on employment and income ......... 22 5.1 The appropriateness of the EPWP with respect to its objectives ...................... 26

6. Conclusions ...........................................................................................28 References................................................................................................. 29 Appendix ................................................................................................... 31

3

1. Introduction Unemployment has significant economic and social costs for individuals and households, as well as for the larger society. Unemployment and the inability to earn a regular income is closely related to why people end up in poverty and also why they are unable to move out of poverty, especially in developing countries which fail to provide a social safety net (State of the Nation, 2005, p. 488). This relationship between poverty and the inability to earn a regular income through employment leads up to the conclusion that if you want to alleviate poverty, then you should promote employment and job creation. In South Africa today, more than a decade after the fall of apartheid and the re-opening of the economy, the unemployment rate as well as the number of people living in poverty1 remain at very high levels. With this in mind it is understandable why unemployment and policies concerning employment has been frequently debated in post-apartheid South Africa. In this paper we make an attempt to evaluate the effects of one of the most important and symbolic measures taken by the South African government, namely the Expanded Public Works Programme (EPWP). The employment policy measures taken by the government, i.e. the ruling party ANC, could roughly be divided into indirect and direct ones. The indirect part concerns the macro-level, and there mainly the promotion of economic growth and overall macroeconomic stability. The direct part concerns the micro-level and deals mainly with short-term job creation and skills training of unemployed individuals within the EPWP. The EPWP is implemented by local government in the provinces and contains four different sectors (infrastructure, social sector, environmental sector and economic sector). This leads to a diversity among both the job opportunities provided and the characteristics of the participants, which in turn complicates the analysis and the generalisation of the programme effects. Furthermore, the EPWP aims at satisfying several different goals simultaneously, namely improving infrastructure and economic growth on the macro-level, and creating jobs and alleviate poverty on the microlevel. Earlier research on the EPWP has criticised it arguing that this fourfold objective leads to difficulties, especially when it comes to the targeting of participants (McCord, 2004a, p 72). Anna McCord argues that if the main objective is to alleviate poverty the target group should mainly be female heads of households in rural areas. When looking at sustainable job creation and investment in the training of coming labour market participants, on the other hand, it is 1

Poverty level as measured by the UN-standard of less than one US-dollar per day

4

preferable to target the youth, she argues. The widespread objectives are therefore a key concern one has to bear in mind when evaluating the EPWP. This paper will have a closer look at the achievements of the EPWP when focusing strictly on its micro-level objectives from the viewpoint of the programme participants. Does EPWPparticipation increase the probability that people manage to find a sustainable job? Does EPWP-participation help households to raise their living standard? Our overarching objective is then to attempt to reach a conclusion on the question whether the EPWP is an appropriate intervention when looking at its capacity to accomplish the announced micro-level objecitves. We will thus not address the macro-level objectives of the EPWP. The research question will be answered by using data from the Labour Force Survey (LFS), a nationwide survey of the South African labour market released by Statistics South Africa twice a year. To be able to separate the ‘true effect’ of participation a matched group of controls will be created by the means of propensity score matching. By comparing the participants with the matched controls we then aim at isolating the participation effect on the chance of employment and household income. This evaluation distinguishes itself from earlier ones performed, first and foremost since it is based on a large, nationwide dataset, but also since it is the first one evaluating the programme capacity within the renewed EPWP-framework. The research performed by e.g. Anna McCord will be helpful when it comes to interpretations etc., but our findings will at the same time serve a new purpose in that they are based on a larger sample size within a reshaped programme context. To begin with we will give an account of the economic-political situation in the country, with concentration on the labour market and the unemployment problem, as well as the situation regarding poverty and income structure. We also present the direct policy measures recently taken to confront the unemployment problem and to alleviate poverty, i.e. first and foremost the EPWP, and summarise the existing research on the EPWP, mainly conducted by Anna McCord at the University of Cape Town, as well as some research of other public works programmes. In the following section some considerations on our data and the LFS in general will be given before we present the evaluation problem and the method we apply to deal with it. Further, in section 5 we present our results and discuss them. Finally in section 6 we conclude and summarize.

5

2. Unemployment and poverty in post-apartheid South Africa The rate of unemployment in South Africa is by any measure very high. According to the official definition, which excludes the prevalent group of discouraged work seekers (people who wants a job but has stopped actively looking for one), the rate of unemployment in September 2005 was 26,7 percent. Since the labour force participation rate was 56,5 percent at the same time, the estimated absorption rate of the South African labour market was only 41,4 percent. An estimated 11,2 percent of the working age population was considered as discouraged work seekers, which gives an unemployment rate close to 40 percent when using the expanded definition (LFS Statistical Release, 2005). Although the unemployment rate has stagnated and even marginally declined the last years the trend after the 1994 liberation is that of an increasing rate of unemployment. In 1995 the official rate was 16,9 percent and the expanded rate 26,5 percent to compare with the figures of today. The explanation to this development focuses on insufficient economic growth, a rapidly growing working age population and thus a larger labour force, and, analyses arguing that this development is an unavoidable consequence in a period of difficult transition (State of the Nation, 2005, p 425). Another popular explanation claims that the South African labour market suffers from a structural mismatch where high-skilled labour is being demanded at the same time as lowskilled labour is being supplied. The labour force grows rapidly, mainly from low-skilled immigration and youth entrants, and simultaneously less low-skilled labour is being demanded due to the fierce competition from in the first place the Chinese counterparts (McCord, 2004b, p 7). After a decade of democracy South Africa is also still a country with high levels of poverty and income inequality. Despite advances in measurement, the development and dynamics of poverty has become the subject of much debate, to a large extent due to insufficient recognition that different ways of measuring yield different results. Another barrier to reliable inter-temporal estimates of poverty has been the questionable quality and comparability of data. This gives rise to the concern that observed changes may be technique driven, rather than reflecting the actual dynamics in the society. The somewhat contrasting results given by the Income and Expenditure Survey (IES) in 2002, concluding that “South Africans, on average, became poorer between 1995 and 2000”, and the All-Media Products Survey (AMPS), revealing that the population share falling in the lowest category of the living standard measure had decreased from 20 to 5 per cent, exemplifies the ambiguity. Despite the diverging views on the recent development one has to come to the conclusion that South 6

Africa still faces a significant poverty problem (State of the Nation, 2005, pp 483). There seem to be little signs of a declining monetary poverty and the food insecurity among children remains high. Data from the General Household Survey reveals that in 2002 30,8 per cent of children under 18 went hungry because of a lack of food. For a poor province like Eastern Cape the same number was close to 50 per cent (State of the Nation, 2005, p 491). Besides the poverty, contemporary South Africa is a country with a massive income inequality contested only by Brazil after international comparisons of GINI-coefficients are being made. Data from Statistics SA over the period indicates that inequality has widened within all four population groups2 (where blacks record the highest GINI), as well as overall. This suggests a development, not of deracialisation of equality, but rather a “deracialisation of inequality” (State of the Nation, 2005, p 494). To a large extent the economy of today’s South Africa could be seen as dual. A “first economy” consisting of people working in the formal sector with written contracts, pension funds, unemployment insurances and access to financial institutions, and a “second economy” consisting of people with informal employment earning a low and insecure income and without access to the benefits of the “first economy”. The post-apartheid development has, as it seems so far, rather stalemated the duality of the South African economy, instead of integrating it. Since the ANC gained power in 1994 their policy strategy, as well as the wider ideological framework affecting their economic decisions, have been far from on a steady course. The Reconstruction and Development Programme (RDP) initiated by the Mandela-government in 1994 were based on the traditional ANC-agenda with an interventionist state idle to see radical progress when it comes to poverty alleviation and employment. The same government changed course quite dramatically in 1996 when they dropped the Keynesian outlook and announced the Growth, Employment and Redistribution Strategy (GEAR) based on a strictly neo-classical analysis of the economy. The developmental ethos of the RDP was now neglected and the overarching objective was fiscal discipline as a means to deficit reduction and stimulating economic growth (State of the Nation, 2005, p 481). In 1999 Thabo Mbeki replaced Mandela in office and by this time the “new ANC” had started to show signs of serious preoccupation with the non-improving situation in the country when looking at unemployment and poverty alleviation. 2 The traditional population group division, as constructed by the apartheid-system, are still of most common usage and consists of blacks/Africans, coloureds, Indians/Asians and whites.

7

Mbeki hosted the Growth and Development Summit (GDS) in 2003 where the government emphasised that the main route to reducing poverty and unemployment was by the means of an Expanded Public Works Programme. Within the GEAR there had been no place for neither a universal unemployment insurance nor a Basic Income Grant (BIG) provided by the state, which in combination with the massive number of unemployed partly could explain the nondeclining poverty level. The new EPWP was now going to serve as a substitute to a BIGpolicy, more in line with the ideological stance of the “new ANC”, as well as a programme for job creation and skills development. The micro-level objectives of the EPWP is thus to lift individuals and households out of poverty and unemployment by offering short- to mid-term jobs, skills training and wage income which in turn will enable people to move from the second economy to the first. In the subsequent section we will briefly present the cornerstones of EPWP and some findings of recent research on the subject.

2.1 Public Works Programmes and the EPWP Public Works Programmes (PWPs) are labour market interventions traditionally used countercyclical in order to provide unskilled and semi-skilled labour with short-term employment during periods when unemployment and also poverty are critical problems. PWPs are seen to be most effective when unemployment has the characteristics of being transitional rather than structural. The intention of these programmes is both to alleviate poverty and to create sustainable jobs. The thought is that instead of just transferring money as a subsidy the people that are employed within the programmes are actually performing something and the programmes leave something behind, for example an improved infrastructure. The workers within the programmes are employed with wages lower than the market wage in order to attract its target group, in programmes that are labour intensive, so that as many people as possible can be employed. PWPs have traditionally been concentrated on infrastructure, such as road construction and maintenance but also in social sectors such as childcare (Subbarao, 2003). PWPs have, since the first nationwide program was launched in 1994, played an important role in the South African government’s aim to create job opportunities in order to fight the massive unemployment problem that the country is facing. The EPWP is a part in this strategy and is a nationwide PWP which was launched in 2004. It sprung out of the Community Based Public Works Programme (CBPWP) which was launched in 1994. However, the CBPWP 8

focused mainly on enhancing the country’s infrastructure and not as much on job creation. The government wanted to change this and at the GDS in 2003 it was decided that the country should create a PWP that focused more on job creation and skills development. The following year this idea was concreted and the EPWP was born. The EPWP covers all spheres of government and state-owned enterprises and the goal is to create one million job opportunities until the end of 2009 (www.epwp.gov.za, 2006). The micro-level aim of the EPWP is to draw poor, unemployed people into work opportunities and to give them some sort of training that will put them in an enabling position to move from the second to the first economy (Interview with Stanley Henderson3, 2005). The objective is to achieve this by offering short-term employment, and to provide the participants with a certificate that they have undergone training which can be used when applying for formal jobs in the future. The length of the employment differs between projects and sectors but the average time is six months (Interview with Lucky Mochalibane4, 2005), usually with longer employment in the social sector (up to two years) and shorter in the infrastructure sector (normally four months). The projects within the programme should use labour intensive methods and focus on skills development. There is no additional funding for the EPWP, instead the government has told the municipalities that they should use a certain amount of their spending using labour intensive methods in production rather than machine intensive. The projects should preferably be placed in areas where the output, for example a new road or a child nursery, also enhances the living-standards of the local population. There is a set of benchmarks for how the combination of participants shall be within the projects. The guidelines say that 40 percent should be women, 30 percent youth, and 2 percent people with disabilities. Among races blacks are prioritised because of the vast joblessness within the group. The unemployment among the youth5 is a big problem in South Africa and that group is a prioritised target within the EPWP. Furthermore, the EPWP consists of four sectors: the infrastructural sector, the social sector, the economic sector and the environmental sector. In the infrastructure sector the emphasis is on creation of job opportunities in a labour intensive way, for example road construction and maintenance. In these projects the workers are paid according to their performance, for example per metre of road built in a certain road 3

Stanley Henderson is the chief director of monitoring and evaluation for the EPWP at the Department of Public Works. 4 Lucky Mochalibane is the chief director of communications and marketing at the Department of Public Works.

9

project. The goal for this sector is that by the end of 2009 having constructed approximately 37,000 km of roads, 31,000 km of pipelines, 1,500 km of storm water drains and 150 km of urban sidewalks. (www.capegateway.gov.za, 2006) In the social sector the focus is on the Home Community Based Care programme (HCBC) and the Early Childhood Development programme (ECD). The HCBC basically concerns relieving hospitals in care giving for people suffering from HIV or AIDS. The ECD focuses on childcare for vulnerable children, often AIDS orphans. In the environmental sector there are a whole range of programmes. One of the major programmes is the Working for Water programme. This programme focuses on the eradication of invasive alien vegetation which consumes enormous amounts of water, leading to damages on the indigenous ecosystems. Here are job opportunities created involving for example cutting down trees in a labour intensive way. The last sector, the economic, focuses on the creation of small businesses through encouragement of entrepreneurship. One of these projects is the so called New Venture Creation Learnership Programme, where poor unemployed people are provided with training and funds to be able to start their own businesses. The overall budget of the EPWP from 2004-2009 is around 50 billion South African rands, which responds to approximately 63 billion SEK (www.epwp.gov.za, 2006). 2.2 Earlier research on the EPWP Due to the freshness of the EPWP, the research performed on it is not of large scale. However, there is some recent research, most of it carried out by Anna McCord at the University of Cape Town and their research unit Southern Africa Labour and Development Unit (SALDRU). In a paper from 2004 she argues that the unemployment in South Africa is of structural type rather than transitional because of the sustained high unemployment rates that the country is facing (McCord, 2004b). According to her this makes the EPWP, which increases the supply of semi-skilled labour, a non efficient tool in the ambition to reduce unemployment. According to McCord focus should be on trying to create a demand for labour instead of increasing the supply. McCord continues by stressing that the EPWP is of too small scale in order to give significant effect on unemployment. In another paper, which contains a thorough examination of two South African EPWPs, McCord argues that there is a mismatch between the type of labour demanded in the country and the supply of labour which the

5

Youth are persons 16-34 years of age, according to the national definition (Stanley Henderson)

10

EPWP provide (McCord, 2004a). She argues that the type of labour demanded in South Africa is people with higher education and not semi-skilled labour, which there is already an abundance of. Regarding the poverty alleviation aspect McCord’s findings indicate that participating in the EPWP had a potential of being significantly advantageous concerning the non-income poverty for these households (e.g. when looking at the number of skipped meals among adults, school attendance by children in the household and opportunities to purchase better clothes). Looking at this McCord believes that the EPWP can be a part of building human and social capital, though she argues that the results could be even more significant if the EPWP targeted even poorer people. When it comes to poverty measured in income McCord claims that the employment period within the EPWP is too short to make a sustainable difference in the participating household’s economy, and that the additional income was quickly consumed instead of invested. McCord is also critical to the way that the EPWP is described in the popular discourse as the blessing comprehensive solution to South Africa’s unemployment predicament. This has lead to policy makers being influenced and the EPWP becoming too broad. She argues that the policy makers should concentrate the EPWP on core areas that give the greatest results and let other areas, like encouraging entrepreneurship and promoting HIV and AIDS awareness, be run by other bodies (www.cssr.uct.ac.za, 2006). McCord also investigates how individuals having participated in the two examined PWPs perform on the labour market. Her conclusion is that partaking in the programme does not enhance the individual’s chances of finding a sustainable job. This she says is because of the extreme levels of rural (where most projects are placed) unemployment and the small demand for low- and semi-skilled labour. In her conclusions she claims that “the gap between policy expectation and programme reality is significant” and that “there is an urgent need to open up the policy space and to seek alternative responses to this critical problem”.

2.3 Earlier research on PWPs and job training in general PWPs as a labour market intervention tool is a policy that has been used in both developed and developing countries. Examples of other developing countries where these types of programmes have been used are Bangladesh, India, Ethiopia, Kenya, Zimbabwe, Tanzania, 11

and Ghana. A paper by Kalanidhi Subbarao at the World Bank summarises how PWPs in different countries has affected e.g. poverty and welfare, as well as how they have succeeded in targeting the poorest (systemic shocks and social protection: role and effectiveness of public works programs). Subbarao argues that there are a few important features in PWPs that must be carefully designed in order to maximise the benefits to the poor. Firstly, he points out the importance of the chosen wage rate. In this matter, the chosen wage rate between programmes used in different countries varies a lot in relation to the minimum wage and the market wage. For example, in the Kenyan ‘cash for work’ programme 1992-1993, the wage rate was set similar to the minimum wage, which was higher than the market wage. This becomes non-efficient in targeting the poorest, because it may attract low-paid employed who crowd out the objective group. Subbarao claims that in order to attract the poorest, the wage rate should be set below the market wage (Subbarao, 2003, p 6). Another important issue is to what extent the programme is labour intensive. The labour intensity is measured by wage cost in relation to total cost of the programme and is often referred to as the cost effectiveness of a project. The objective to achieve as high labour intensity as possible is, according to Subbarao, of vital nature because it maximises the gain to the participants, which is often poor people. Also on this matter, the degree of labour intensity or cost effectiveness differs between PWPs and countries. Subbarao’s study shows that in the examined programmes, labour cost in relation to total cost was lower in road construction projects (around 40-50 percent) than in for example reforestation projects where the relation was around 70-80 percent (Subbarao, 2003, p 13). Subbarao also looks at how well different programmes succeed in targeting the poorest, and mentions contrasting examples like a PWP in Argentina, where 80 percent of the participants came from the poorest 20 percent and 60 percent came from the poorest 10 percent, and a programme in the Philippines where a somewhat high wage rate attracted mostly marginally poor and non-poor people. Earlier studies with intention to evaluate how participation in a labour market programme improves the chance of finding employment seldom show evidence of positive results. Worldwide, the usefulness of active labour market policy and PWPs is a subject of scepticism among researchers and politicians as well as employers and job-seekers themselves, although the bulk is concentrated on already developed countries (e.g. Heckman et al, 1999; Grubb & Martin, 2001). A study conducted in Russia, a transitional economy which in some respects resembles South Africa, show that vocational training programmes had no overall employment effect on participants relative to non-participants (Nivorozhkin & Nivorozhkin, 12

2005). However, when looking at blue-collar workers solely they find a discernible positive effect in the short run, although the effect seems to decrease in the longer run. The explanation according to the authors could be the high labour demand from the industries resulting in an excess of vacancies at the time of the study. The conclusions made by Grubb and Martin are that public works programmes can have a positive effect, if well designed, but the impacts are not large.

3. Data The Labour Force Survey (LFS) is a twice-yearly rotating panel household survey conducted by Statistics South Africa, which examines the extent of employment in both the formal and informal sectors, and the extent of unemployment, according to the standard definitions of the ILO. The survey gathers detailed information on some 70000 adults of working age living in some 30 000 dwelling units across the country. The rotating panel sample involves visiting the same dwelling units on a number of occasions before replacing them with new ones. A proportion of 20 percent is being replaced each survey round, which in turn allows for both longitudinal and cross-sectional analysis (LFS Statistical Release, 2005). The selection of households is based on the Stats SA master sample, which is processed by dividing the country into 53 District Councils (DCs) from where they through the Power Allocation Method generate some 3000 PSUs. The population density of each DC decides the probability6 of a PSU being included in the survey (Interview with Dr Jacques De Klerk7, 2005). There are various potential, as well as inevitable, weaknesses with using the LFS data that are worth being highlighted. Firstly, there is a problem with the freshness of the inclusion probabilities in the master sample since it is based on the numbers of the 2001 census. Although this is the most recent census performed one can expect some biases in the sample’s representativity of the population. This problem becomes more obvious when one considers the diverging tendencies in population growth in between different income and population 6

The probability of a PSU being included is calculated by weighting of the PSU´s.

PPSU =

nPSU .nS N PSU

Where nPSU is the number of households in the PSU selected. NPSU is the total number of households in the stratum, in this case the District Council. The ns is the number of PSUs in the stratum.

13

groups in combination with the large scale geographical segregation among these different groups’ dwelling units (De Klerk, 2005). It is therefore likely that households in areas with a faster than average population growth since 2001 are over-represented in the survey. Since these areas also tend to consist of poorer than average households this could bias our results. Worth noting are also a couple of potential problems with the household interviewing method that is used to carry out the LFS. An apparent concern is the fact that in many of the visited dwelling units one or several of the household members were absent by the time of the interview. In these cases the present members had to answer on behalf of the absentees. This raises a question of the preciseness of the data collected since the other members’ awareness of the absentee’s working life characteristics in many cases could be insufficient (Interview with Peter Buwembo8, 2005). Further, there is a quite complex language situation in South Africa which raises a potential problem with the survey. Although English reaches out to most people it is very seldom people’s native tongue. Especially blacks in rural areas are unfamiliar with English, and some might not even have heard it being spoken. Since the questionnaire is in English an important responsibility regarding translation and interpretation is being put on the interviewers. Worth being noticed is also that the interviewers are not full-time hired employees of Stats SA, but unemployed people from the different provinces hired for a shorter time (around two weeks) and with high school diploma as minimum education (Buwembo, 2005). Despite the above mentioned caveats we consider the LFS to be the most appropriate source of quantitative data available for answering our research question. An option, for example, is the quarterly Survey of Employment and Earnings (SEE), which collects information on formal employment in South Africa. The SEE obtains data direct from formal sector businesses registered for VAT, in contrast with the LFS’ household-based survey methodology which covers the whole work force. With respect to the subject of this paper the LFS, considering its rich content of variables and observation units as well as its wider coverage of the working age population, seems to be the more appropriate data source to use. The then remaining alternative to conduct a survey by our own is both too costly and time consuming considering the large set of observations required.

7 8

Dr Jaques De Klerk is the Stats SA Manager of Methodology and Audit Peter Buwembo is the Statistics South Africa Census Manager

14

3.1 Sample reduction

The datasets used in the study are from the LFS 2004:2, the LFS 2005:1 and the LFS 2005:2. From the datasets we have excluded those individuals that we believe are not of relevance for our evaluation, i.e. do not share the characteristics of the participants in the EPWP. We started off by excluding people of age younger than 16 and older than 65. This because there is no participant of the age of 15, and we chose to draw the upper age limit at the South African retirement age which is 65. We have also excluded people who are apparently not looking for a job over the time period that we study. These contain e.g. students, retired, handicapped and disabled people. After excluding these we remain with people who want to have a job, either unemployed or discouraged work seekers. Since these are the two important target groups for EPWP-participation this sample reduction is suitable in order to create a control group. Within the treatment group we also excluded people who answered that they are still within the project. This since we want a dichotomous dependent variable, i.e. either employed or not employed, and ongoing participation is hard to categorise within this dichotomy. After the sample reduction there is 39 288 individuals remaining in total where 374 are treated individuals. 3.2 The variables

In table 1 below there is descriptive statistics of all variables where all individuals in the sample are included. In our study the variables employed and household income per capita are the dependent variables. It differentiates whether people have a job or not after participation, according to the official unemployment definition and it also indicates the income per capita in households with EPWP participants. The explaining variables are, as mentioned above, divided into groups. In the group of personal characteristics age, which ranges from 16 to 65, gender, black and coloured are included. Black and coloured are included because they are the main target groups concerning population groups and the two remaining groups of whites and indians thus together functions as the omitted dummy. Gender and age are also fundamental when estimating our model and are therefore also included. The second group is schooling and living standard. Included here are the variables education, tertiary and poor. The variable education ranges from no schooling, which also includes informal schooling such as pre-school and reception school in line with the Stats SA recommendation 9, to finished high school. The variable tertiary is a dummy considering whether the person has

9

www.statssa.gov.za/census01/html/Persons.pdf, 2006

15

any type of finished higher education, e.g. college or university diploma, certificate of degree etc. Those categorised as poor are people living in households with an income per capita below or equivalent to 250 South African Rands per month10, which is approximately corresponding to the UN-definition of poverty. To create this variable we had to use the mean incomes from each income group alternative in the LFS-questionnaire. We then added these together for each household and divided the total sum by the number of people living in the household. Further, there is one important group consisting of variables which describe the individuals’ working life characteristics at the time before participation. The variable unemployed explains if the person was unemployed or not prior to participation and the variables informal and formal regards whether the person was employed in the informal or the formal sector. Time ago contains information on how long ago it was since the individual worked, ranging from 0 for those who had a job up to 36 (months) for those who either had been unemployed for more than 3 years or answered that they never worked. We chose to add these together and give them the value 36, and to be able to separate the effects for those who had never worked we created a dummy for this. We also created a dummy for those who had been unemployed for more than three years. This we had to do in order to make these variables compatible with the way that the questions in the LFS were asked.

9 1 ZAR≈1,25 SEK.

16

Table 1: Describing statistics of all participants and non-participants. Variable Employed Household inc/pc EPWP Age Gender Black Coloured Education Tertiary Poor Unemployed Formal Informal Time ago More than three Never Worked Unemployment in DC Limpopo Gauteng North West Kwazulu-Natal Free State Northern Cape Eastern Cape Western Cape

Description

Mean

Std Dev

Min

Max

.4975368

0

1

1275.946

0

20833.33

.0971034

0

1

12.628 .4979285 .4134169 .3549373

16 0 0 0

65 1 1 1

3.809071

0

12

.2057376

0

1

.4996897

0

1

.3708744 .4355739 .353186 17.31538

0 0 0 0

1 1 1 36

.318561

0

1

.4847406

0

1

26.34023

7.164873

11.2

48.3

.0949654 .10871 .0891875 .2574832 .0711413 .0661525 .1257636 .1120953

.2931709 .311279 .2850179 .4372533 .257064 .2485517 .331587 .3154877

0 0 0 0 0 0 0 0

1 1 1 1 1 1 1 1

=1 if person employed .4503665 Monthly Income per capita in households with 673.3377 participants =1 if person participated in EPWP .0095194 Personal characteristics age in years 36.06956 =1 if is a man .4544645 =1 if person is black .7812309 =1 if person is coloured .1478314 Schooling and living standard completed years of primary and secondary school 8.033114 =1 if person has some sort of post-secondary .0442883 education =1 if household income per capita ≤R250/month .5177917 Working life characteristics (before participation) =1 if person was unemployed .1646559 =1 if person was employed in the formal sector .2544797 =1 if person was employed in the informal sector .1460751 time ago it since the person last worked, in months 19.24684 =1 if person has work-experience, but from more .1146151 than three years ago =1 if person has never worked before .3773926 Economic environment unemployment rate in district council where person resides =1 if person lives in Limpopo province =1 if person lives in Gauteng province =1 if person lives in Northwest province =1 if person lives in Kwazulu-Natal province =1 if person lives in Free State province =1 if person lives in Northern Cape province =1 if person lives in Eastern Cape province =1 if person lives in Western Cape province

In the group of economic environment we created a variable including the unemployment rate in each District Council (DC), which we find important for the probability for an individual of finding a job after participation. In this group there is also dummy variables for South Africa’s provinces, this since their labour market characteristics and the number of participants in each province varies a lot. The intention is that the variables in the group of economic environment will help to control for the important effect on employment probability caused by economic surroundings and regional differences in labour demand. In Table 1 above the mean values and standard deviations of the variables are being outlined.

17

4. Evaluation and the Propensity Score Matching Method Our study aims at evaluating the effect of EPWP-participation on short-run labour market performance and the income per capita in households with participants. Therefore we face the inevitable evaluation problem constantly pursuing social science. In an ideal situation one would want to compare the participating (or treated) individuals’ post-treatment performance with the performance the very same individuals would have recorded without participation (or treatment11) over the same time period. Because of the self-evident impossibility to create such a situation, we have to find a non-experimental comparison as similar as possible to the one above. More explicitly, we want to create a control group of the non-treated consisting of individuals as identical as possible to the ones undergoing treatment when considering relevant and observable characteristics affecting the selection to the program as well as the outcome variables. If we have a randomised selection process and access to comparable data on both treated and non-treated individuals, there would be an experimental-like situation, sometimes referred to as a natural experiment, which could be regarded as ‘second-best’ to the ideal situation. In the EPWP-case the selection process is random, but unfortunately there is currently no available data on the programme applicants not selected to treatment. 4.1 How to tackle the evaluation problem

Since there is no information on EPWP-application among the non-treated we have to create an appropriate control group based on observable data on the non-treated and thus we have a non-randomised situation. A direct comparison between the two groups in a non-randomised situation is likely to be misleading since the individuals exposed to the treatment generally differ systematically from the ones not treated. With our large reservoir of potential controls the matching sampling method can be used to produce a control group in which the distribution of covariates is similar to the distribution in the treated group. Although there exist alternatives to matched sampling there are a few reasons why matching is appealing, listed for example by P.R. Rosenbaum and D.B. Rubin (Rosenbaum & Rubin, 1983). These are, for example, that matched treated and control pairs allow a relatively broad public to appreciate the equivalence of treatment and control groups, and to perform simple matched pair analyses. The matched sampling also yields an estimate of the average treatment effect with a lower variance since the distribution of the covariates in treated and control groups are

11 ’Treatment’ is the methodological term generally used in the literature to indicate participation in a programme.

18

more similar in matched samples than otherwise. A more detailed examination of the method is presented in the following subsection. This study of the benefits of the EPWP thus generally follows the mainstream theory on how to evaluate the causal effect of some treatment on some outcome Y experienced by units in a population of interest (Bryson et al, 2002; Sianesi, 2001). This theory, which has been briefly discussed above, can use some further explanatory outline. Y1i → the outcome of unit i if i were exposed to the treatment Y0i → the outcome of unit i if i were not exposed to the treatment Di ∈ {0,1} → dummy variable of the treatment actually received by unit i Yi = Y0i + Di (Y1i − Y0i ) → the actually observed outcome of unit i X → a set of pre-treatment characteristics

Thus, the causal effect of the treatment on unit i will equal Y1i − Y0i , that is the difference between the outcome unit i would experience if treated and if not treated. As discussed earlier, this difference is impossible to measure since we can only observe either Y1i or Y0i , never both. Therefore, not testable ‘identifying assumptions’ about Y0i have to be made to make causal inference of the treatment effect. The average treatment effect on the treated (ATT) can be written in the following way E (Y1i − Y0i Di = 1) = E (Y1i Di = 1) − E (Y0i Di = 1)

where the unobserved counterfactual E (Y0i Di = 1) - the outcome the treated on average would have experienced, had they not been treated – need to be estimated based on these ‘identifying assumptions’. The creation of this estimate can be characterised as an artificial way of creating a randomised selection process to the treatment. By controlling for the differences in X that has an effect on the selection to treatment between the treated and non-treated, the above mentioned, randomised ‘second-best’ situation can be approached. The set of assumptions in question are different depending on the method used to approach this situation. The method of matching by the means of propensity score estimation will be further developed below.

19

4.2 Propensity Score Matching The matching method is, in principle, straightforward and simple. For every treated individual a match, dictated by observable characteristics, is selected from among the non-treated. The effects of treatment can then be calculated as the difference between the differences in outcome between the treated and their non-treated matches. Although the method has a motivated appeal it is also worth noting the critical assumptions it rests on. Firstly, it is assumed that if one can control for observable differences in characteristics between the two groups, then their outcome would be identical in the absence of treatment. This assumption is known as the Conditional Independence Assumption (CIA). This requires a rich dataset so that all variables affecting selection and outcome can be observed, and thereby any differences between treated and non-treated can be applied to the treatment effect. If our vector X do not contain all the variables affecting both selection and outcome, then the CIA is violated since the treatment effect will be accounted for partly by unobserved information. Other assumptions matching makes are that an individual’s participation decision does not depend on the decisions of others, and that the impact of the treatment on one person does not depend on whom else, or how many others, are being treated (the so-called stable unit value treatment assumption, SUTVA) (Bryson et al, 2002). An important constraint with matching is that as the number of characteristics that are being matched increases, the possibility of finding a match decreases. Rosenbaum and Rubin have overcome this problem by introducing the method of ‘propensity score matching’ (Rosenbaum & Rubin, 1983). They show that matching can be performed on a propensity score, i.e. a single index reflecting the probability of being selected to treatment, with consistent results of the outcome in the same way as a large set of covariates are being used. However, the problem of decreasing possibility to find a match while increasing the number of covariates does not entirely disappear with the application of the propensity score. The socalled ‘support problem’ appears when there is nobody among the non-treated with a propensity score similar to that of a particular treated individual. There are several methods of various complexities dealing with this issue, but normally treated individuals without support among the non-treated are omitted from the analysis. If a sizeable proportion of the treated are being omitted then the relevance of the results will suffer since the estimated treatment effect is based on a sub-sample and thus only valid for a ‘supported’ sub-population. Whether this

20

will be a problem in the analysis below will depend upon the proportion of the treatment group lost (Bryson et al, 2002).

4.3 The various methods to select the controls and our choice After the selection of variables and the calculation of the propensity score we face one last decision before the matching can be performed, namely what technique(s) to use when selecting the controls. The first, and probably most straightforward, is the traditional nearest neighbour (or one-to-one) method, where the controls are selected based on their distance to the treated observations with respect to the propensity scores. Every observation is thus being compared with the nearest neighbour among all potential controls, i.e. the control with the least deviating propensity score. This technique is very intuitive, but could in some cases yield fairly poor estimates since the propensity score distance could be much greater for some nearest neighbours than for others and they nevertheless contribute to the estimation of the treatment effect independently of this difference (Heckman et al, 1997; Heckman et al, 1998). One alternative to the nearest neighbour method is kernel-based matching, where all treated are matched with a weighted average of all controls with weights that are inversely proportional to the distance between the propensity scores of treated and controls. With this technique a function is being used that weights the contribution of each control group member, so that more importance is being attached to those controls providing a better match. The application of kernel-based matching is therefore a way to deal with the potential problem in the nearest neighbour method discussed above. Other frequently used techniques include e.g. radius matching (where the controls within a certain propensity score radius are selected) and two-to-one matching (where the two nearest neighbours are used as controls). Since we have access to a large set of potential controls the propensity score distances between the treated and their controls were consistently small. In the first place we will thereby apply the traditional nearest neighbour method due to its appealing intuition. The kernel-based method will also be applied in order to further test the consistency of our results (Heckman et al, 1997; Heckman et al, 1998).

21

5. Results: the effect of participation on employment and income The aim of the following analysis is to look at the efficiency of the EPWP when asking if participation improves the employment probability of the participants and the per capita income level in their households. We apply propensity score matching and because of the large quantity of potential controls available all treated individuals are supported despite our fairly rich set of covariates. Thus there is no ‘support problem’ weakening the results in this case. The controls are identified as every single observation’s nearest neighbour when comparing propensity scores. The same control can be used more than once in cases where two or more observations share the same nearest neighbour. There is also a possibility that an observation could have two or more neighbours with equally close propensity scores. The latter occurred for 8 observations and in these cases we let the software randomise the selection of controls from these equally close neighbours. The following straightforward task is to compare the mean values of the EPWP-participants with the mean values of their matched controls. If the critical assumptions of the matching method hold, then the matched means function as proxies for the unobservable means the participating individuals would have recorded, had they not participated. In Table 2 below we present the results received when comparing the differences of the means between the participants and their respective controls. The results reveal that there is an insignificant positive effect on employment probability of just over 0,5 percentage points. The participation effect on per capita income in the household is, according to our results negative (the participants had on average a per capita household income about 56 Rands lower than their matched controls), but also this effect is statistically insignificant. A more thorough presentation of the test results is presented in Table 3 in the appendix. From the test results we notice that the matching method was successful in balancing the characteristics of the controls with the ones of the participants. The significant differences between the means of the participants and the unmatched controls were in all cases turned into insignificant differences or no difference at all (in one case) through the matching process. This indicates that the comparability of the control group was enhanced by the matching process. Concerning the results when using kernel matching method instead of nearest neighbour the results indicates that participation has a small negative effect of -0,6 percentage points on employment probability and household income per capita for participants being about 96 Rands lower than for their matched controls.

22

Table 2: Mean values of characteristics for non-participants, participants, and the matched control group (by nearest neighbour and also kernel matching method for the dependent variables, for all independent variables matched means are from nearest neighbour matching). Variable

Mean Non-EPWP

EPWP Employed Household inc/capita

0 .4504292 674.264759

Mean EPWP

Matched mean nearest neighbour 1 0 .4438503 .438502674 576.875307 632.867806 Personal characteristics

Difference between the treated and their matched controls nearest neighbour (p-value*) 0 1 .449946203 0,0053 (0,941) 672.462918 -55,9925 (0,445) Matched mean kernel

Age Gender Black Coloured

36.0689 .4543352 .7804903 .1482243

36.13904 36.026738 .4679144 .475935829 .8582888 .855614973 .1069519 .112299465 Schooling and living standard

Education Tertiary Poor

8.027291 8.639037 8.64973262 .0441486 .0588235 .07486631 .5168834 .6122995 .582887701 Working life characteristics (before participation)

-0,0107 (0,968) -0,0160 (0,380) 0,0294 (0,413)

Unemployed Formal Informal Time ago More than three Never worked

.1641312 .2549211 .1460914 19.2359 .114329 .3772935

0,0134 (0,655) 0 (1,000) -0,0134 (0,610) 0,9146 (0,469) 0,0374 (0,123) -0,0107 (0,765)

Unemployment in DC Limpopo Gauteng Northwest Kwazulu-Natal Free State Northern Cape Eastern Cape Western Cape

26.3239

28.03984

27.9294118

0,1104 (0,830)

.0950044 .1089325 .0894537 .256463 .0708485 .065709 .1259444 .1127101

.0909091 .0855615 .0614973 .3636364 .1016043 .1122995 .1069519 .0481283

.109625668 .109625668 .053475936 .35026738 .098930481 .122994652 .080213904 .056149733

-0,0187 (0,395) -0,0241 (0,268) 0,0080 (0,638) 0,0134 (0,703) 0,0027 (0,903) -0,0107 (0,650) 0,0267 (0,210) -0,0080 (0,622)

.2192513 .205882353 .2085561 .20855615 .144385 .157754011 20.38496 19.4703877 .144385 .106951872 .3877005 .398395722 Economic environment

0,1123 (0,897) -0,0080 (0,826) 0,0027 (0,917) -0,0053 (0,815)

* A p-value greater than 0,10 indicates that there is no statistically significant difference between the mean values of the EPWP-participants and the respective means of their matched controls

The performed evaluation of the effects of EPWP-participation thus suggests that there is no statistically significant effect, neither when it comes to post-participation labour market performance nor when looking at household income per capita. But are the critical assumptions satisfied? The most crucial uncertainty normally concerns the assumption that all characteristics affecting the selection to a programme and its outcome are observable (the CIA). This is certainly a pressing issue in our case as well. A concern over the LFS’ ability to cover all the theoretically observable factors of interest, as well as our application of the

23

received data, also needs to be raised before drawing conclusions from the results. Factors which normally are of great relevance when evaluating the effects of labour market programmes are the ones concerning pre-participation labour market performance, i.e. time in unemployment, time out of employment, search intensity etc. This sphere tends to catch up a lot of relevant unobservable effects, e.g. motivation and effort in the post-participation jobsearch process, and it is therefore of great importance that the accessible data covering the effects of these factors are as exhaustive as possible and treated accurately by the researcher. The pre-participation labour market characteristics are included in our matching, but the coding from the LFS data is not as straightforward as one would desire. The design of the variables is explained in a previous section and there the complexity regarding the labour market characteristics variables is being discussed. In the analysis we included both registered unemployed and discouraged work seekers since they both are important target groups within the EPWP, but only the unemployed answered to the questions regarding unemployment in the LFS. Because of this we used the more general question asking all non-employed for how long time they had been without a job by the time of the survey. Although we differentiate between if individuals were registered unemployed or outside of the labour force at the time of the survey, where the latter could mean a lower degree of motivation or search intensity, some effect otherwise caught by the duration in the unemployment register could bias our matching. There is also a potential problem with how the answering alternatives in the LFS are designed for some questions. The alternatives are sometimes pre-defined intervals as e.g. in the question discussed above. The alternatives were of the form “6-12 months”, “2-3 years”, “>3 years” etc. As explained in section 3.2 we coded these answers by using the mid-point of the intervals (with the logical exception of the last example above). But what about the ones answering “>3 years” then? And how do you compare these individuals with the ones answering that they had never worked? One individual might have answered that she had never worked, but since she is only 19 years old and entered the labour force a short time ago, she could hypothetically be “better of” than an older person answering “>3 years”. Our solution to differentiate the question into one discrete variable indicating how many months the person had been out of work ranging up to 3 years (i.e. 36 months) and two dummies for “never worked” and “>3 years” might not be optimal, but we also feel that this type of information is indispensable when evaluating the effects of a labour market programme. There is also important to keep in mind the possibility of heterogeneous effects within our sample. By this we mean that the effect of participation could differ in between groups of the 24

population when looking at ethnic group, gender, age etc. It could have been plausible for us to control for these effects by performing our evaluation on subgroups based on these characteristics instead of using the larger sample. Unfortunately we were unable to do this since it then would have given to small samples of participants in each subgroup, but one should however keep the potential heterogeneous effects in mind when concerning our results. A further important aspect when analysing evaluations of programme participation is the time in treatment, and then foremost the participants’ comparability vis-à-vis their controls regarding starting and finishing date. The LFS does not reveal the exact starting and finishing dates of the participation, which is a weakness. What we instead could do was to select as participants those that answered that they participated during the year between the LFS 2004:2 and LFS 2005:2, but had neither started by the time of the LFS 2004:2 nor were still participating by the time of the LFS 2005:2. The data from LFS 2005:1 also enabled us to make sure that the participants at least participated over this point of time. This leaves us with an uncertainty of the exact length of the participation, as well as the exact date of graduation from the programme, although the former (participation length) will be restrained to maximum one year, and the latter (time since graduation) will go from very recently to a maximum of 6 months. The matching of the controls was then based on the characteristics in 2004:2. Each pair of participant and control is thus not matched based on their characteristics at the point of time when the participation started, but on the point of time for the LFS 2004:2. However, since the programme’s starting and finishing dates varies in between the participants, we would have needed information on the potential controls for each and every of these potential dates in order to control for the issue above. Since there was no access to such information, we had to exclusively rely on the using of the biannual information from the LFS. Our main concerns, to sum up, are the satisfaction of the CIA and the design of a few critical variables. Concerning the sample size we feel that our 374 observations are enough to draw some basic, statistically secured conclusions. We believe that the broader concept of using the large and comprehensive LFS data set and applying a matching method is appropriate in our case. We have a broad base of potential controls, as well as a rich set of information on the observations. The LFS provides data that enables us to approach e.g. the CIA better than other currently available sources. The matching method gives results who allow them to be easily examined and interpreted, although one has to consider the possibility of heterogeneous 25

effects when interpreting these results. In our case the matching process also helped us to balance the characteristics of the controls with the ones of the participants. Since our covariates were chosen with respect to what affect selection to the programme and its outcome, the matching process logically must have improved the estimation of the “true” participation effects. Although it is important to stress the uncertainty in our results, we will round off with some normative considerations regarding the EPWP and its capacity to comply with the micro-level objectives.

5.1 The appropriateness of the EPWP with respect to its objectives On the micro-level the EPWP aim at alleviating poverty and joblessness by offering shorttime job opportunities and skills development that in turn will improve the possibility to find a regular job and earn a regular wage income. In this paper we have focused on these two micro-level objectives and our evidence indicates that the participation has no significant effect on these objectives. The conclusions resemble the ones drawn by Anna McCord in her interview-based study of two projects in Limpopo and KwaZulu-Natal (McCord, 2004a). She argues that there already is an excessive supply of low-skilled labour in South Africa and that it would be more efficient to take measures on the demand side. The EPWP also targets individuals mainly in poor, rural areas with a low amount of vacancies of importance for the EPWP-participants. The participants are hence often graduating from the EPWP only to find that there are no suitable jobs to apply for in their local area. The asymmetric relationship between the demand and the supply when it comes to low-skilled labour could therefore be an important explanation to the results received here, as well as the results received by Anna McCord. When discussing the probable shortcomings of the EPWP to satisfy its micro-level objectives and the asymmetry between labour supply and demand, one, though, has to remember that the demand side is being addressed on the macro-level, mainly through the promotion of economic growth. The difference is that the government in this sphere has a much less interventionist policy strategy and that the demand side is expected to allocate according to strict market mechanisms. Simultaneously the “policy package” intends to mobilise and activate the already abundant supply of low-skilled labour. We argue that a wiser strategy instead might be to make the policies focusing on the micro- and macro-level, as well as the labour market’s demand and supply side, more compatible with each other. This could imply either a more active state on the macro-level creating an institutional setting stimulating the 26

demand for low-skilled labour on a long-run basis, or less direct interventions in the labour supply aiming to approach a more market based allocation process of low-skilled labour, e.g. combined with a broader strategy to raise the education level in the long run. The idea to raise the living standard of poor households through the EPWP instead of applying e.g. some sort of basic income grant can also be criticised. Our evidence indicates that EPWP-participation on average fail to help poor households increase their living standard. As already suggested, a plausible explanation could be the evidence of a status quo on the labour market for the group in question. If the participation fails to increase the probability to find a sustainable job and earn wage income, then one would expect the monetary living standard to remain at low levels. Instead of helping poor households to increase their postparticipation living standard, our evidence suggests that it on average returns to the preparticipation level. McCord argued that the EPWP for many households functions as a temporary “wage shock” without lasting effects, however it is yet uncertain what the utility of this contribution looks like. Further research looking at whether it leads to e.g. valuable long run investments in capital goods and education, or if it goes to less sensible purposes, is needed. Our evidence also focused strictly on monetary living standard, and McCord has argued that there are non-monetary benefits coming from EPWP-participation like an increased awareness of various diseases (in the first place HIV/AIDS), the importance of schooling, and the privileges within the welfare system (a large part of poor households have little awareness of their rights to e.g. child benefits and old age grants). Although the effects on these “soft” factors need further investigation, they are important to bear in mind when judging the EPWPs ability to improve the living standard of poor households (McCord, 2004a). As mentioned earlier, the EPWP does neither consist of homogenous projects, nor homogenous participants. The projects differ in what experiences they provide as well as the length of the job opportunities. Our evidence suggests that the EPWP on the whole fails to comply with its micro-level objectives, but individual projects could be more successful. The normative discussion is of course dependent on the validity of our conclusions. The EPWP is complex in its character and therefore hard to unambiguously praise or reject. The wide variety of objectives on different levels and their place within a broader economic context makes it futile, if not impossible, to place a general verdict on the EPWP based on our study. Nevertheless, we believe that it is relevant to study certain aspects of more complex

27

phenomenon and that is what our approach to evaluate two of EPWPs key objectives has tried to do.

6. Conclusions The result from our study indicates that participating in the EPWP has limited influence on the two micro-level objectives, namely to increase participants probability of being employed and raising the income of the participants’ households. However, one should be careful when drawing any definite conclusions. As mentioned before there are a number of issues that could influence why we get the results that we get, especially the design of the LFS questionnaire and the potential inability to observe all characteristics influencing selection to and outcomes from participation. A plausible reason is that the participants are mostly low-skilled and that the challenge in creating lasting jobs for this group is huge, especially when considering that there already is an abundance of these types of workers and that the number of jobs in this sector is growing very slowly. Concerning the poverty alleviation objective it seems as if the EPWP does not raise the income in households with participants. According to our results, the income per capita in households with participants is lower than the households of their matched controls, although this difference is statistically insignificant . When drawing any conclusions from these results one can speculate in that the employment time that the EPWP offers in most cases is too short in order to be able to create any lasting change in income level, rather has the income from the EPWP been described as a short-term “wage shock”. Depending on how this contribution is used, the EPWP could function as a short-run poverty relief which makes them able to buy food or clean clothes or otherwise invest in capital goods or education. The diversity of the projects within the EPWP, as well as the target groups, aggravates the analysis. Our study aimed at evaluating the EPWP on the whole, but by separating individual projects or dividing the population into subgroups one could come to different conclusions.

28

References Bryson, Alex, Dorsett, Richard and Susan, Purdon, (2002). ”The use of propensity score matching in the evaluation of active labour market policies”. Working paper, Department for Work and Pensions, UK. Buwembo, Peter. Statistics South Africa Census Manager. Interview in December 2005. Daniel, John, Southall, Roger and Lutchman, Jessica (eds.), (2005). State of the Nation: South Africa 2004-2005. Cape Town: HSRC Press; East Lansing: Michigan State University Press. De Klerk, Jacques. Statistics South Africa Manager of Methodology and Audit. Interview in December 2005. Grubb, D and Martin, J. P, (2001). “What works and for whom? A review of OECD countries´ experiences with active labour market policies.” Swedish Economic Policy Review, 8, 9-56. Heckman, J.J., Ichimura, H. and Todd, P.E. (1997). “Matching As An Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme”, Review of Economic Studies, 64, 605654. Heckman, J.J., Ichimura, H. and Todd, P.E. (1998). “Matching as an Econometric Evaluation Estimator”, Review of Economic Studies, 65, 261-294. Heckman, J., LaLonde, R., and Smith, J. (1999). “The Economics and Econometrics of Active Labor Market Programs.” In Handbook of Labor Economics, vol. 3A, North-Holland, pp.1865-2097.

29

Henderson, Stanley. Chief Director of monitoring and evaluation of the EPWP, Department of Public Works. Interview in December 2005. McCord, Anna, (2004a). “Policy expectations and programme reality: the poverty reduction and labour market impact of two public works programmes in South Africa”, ESAU public works research project, SALDRU, School of Economics, University of Cape Town. McCord, Anna, (2004b). “Public works and overcoming under-development in South Africa”, SALDRU, University of Cape Town. Mochalibane, Lucky. Chief Director of Communications and Marketing, Department of Public Works. Interview in December 2005. Nivorozhkin, Anton and Nivorozhkin, Eugene, (2005). “Do government sponsored vocational training programs help the unemployed find jobs?” Upjohn Institute, working paper no 05115. Rosenbaum P.R., Rubin D.B. (1983). “The Central Role ofThe Propensity Score in Observational Studies for Causal Effects.” Biometrika, 1983. Sianesi, Barbara, (2001). ”Implementing propensity score matching estimators with stata”, University College London, Institute for fiscal studies, UK Stata users group, London. Statistics South Africa, (2006). Statistical Release of the Labour Force Survey 2005:2. Available at www.statssa.gov.za. Subbarao, Kalanidhi, (2003). “Systemic shocks and social protection: Role and effectiveness of public works programs”, Social protection unit, Human development network, The World Bank. www.anc.org.za/images/maps/samapo.gif, online May 10 2006. www.epwp.gov.za, online January 10, 2006.

30

www.cssr.uct.ac.za/media/saldru_anna1.pdf, online January 10 2006, published November 3, 2005. www.capegateway.gov.za/eng/directories/projects/7319/86919, online January 10 2006. www.statssa.gov.za/census01/html/Persons.pdf, online January 12 2006

Appendix Table 3. Descriptive statistics of test results with propensity score test.

31

Image 1. Map of South Africa. Variable

Sample

Employed

Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched Unmatched Matched

Household inc/pc Epwp Age Gender Black Coloured Education Tertiary Poor Unemployed Formal Informal Time ago More than three Never worked Unemployment/DC Westerncape Easterncape Northerncape Freestate Kwazulu-Natal Northwest Gauteng Limpopo

Mean Treated Control .44385 .45043 .44385 .44118 576.88 674.26 576.88 642.46 1 0 1 0 36.139 36.069 36.139 36.027 .46791 .45434 .46791 .47594 .85829 .78049 .85829 .85561 .10695 .14822 .10695 .1123 8.639 8.0273 8.639 8.6497 .05882 .04415 .05882 .07487 .6123 .51688 .6123 .58289 .21925 .16413 .21925 .20588 .20856 .25492 .20856 .20856 .14439 .14609 .14439 .15775 20.385 19.236 20.385 19.47 .14439 .11433 .14439 .10695 .3877 .37729 .3877 .3984 28.04 26.324 28.04 27.929 .04813 .11271 .04813 .05615 .10695 .12594 .10695 .08021 .1123 .06571 .1123 .12299 .1016 .07085

.1016 .09893 .36364 .25646 .36364 .35027 .0615 .08945 .0615 .05348 .08556 .10893 .08556 .10963 .09091 .095 .09091 .10963

32

Bias reduction % -1,3 0,5 -8,2 -5,5 Ua Ua 0,6 0,9 2,7 -1,6 20,3 0,7 -12.4 -1,6 16,8 -0,3 6,6 -7,3 19,3 6,0 14,0 3,4 -11,0 0 -0,5 -3,8 6,6 5,3 9,0 11,2 2,1 -2,2 24,4 1,6 -23,9 -3,0 -5,9 8,3 16,4 -3,8 11,0 1,0 23,3 2,9 -10,6 3,0 -7,9 -8,1 -1,4 -6,4

t-test bias 59,4 32,7 Ua Ua -60,1 40,9 96,6 87,0 98,3 -9,3 69,2 75,7 100,0 -683,5 20,4 -24,5 -2,8 93,6 87,6 -40,8 77,0 91,3 87,5 71,3 -3,0 -357,0

t

P>t

-0,25 0,07 -1,47 -0,76 Ua Ua 0,11 0,13 0,52 -0,22 3,62 0,10 -2,24 -0,23 3,09 -0,04 1,37 -0,88 3,68 0,82 2,86 0,45 -2,05 0 -0,09 -0,51 1,28 0,72 1,82 1,54 0,41 -0,30 4,61 0,21 -3,94 -0,49 -1,10 1,26 3,61 -0,45 2,30 0,12 4,72 0,38 -1,89 0,47 -1,45 -1,11 -0,27 -0,85

0,799 0,941 0,142 0,445 Ua Ua 0,915 0,897 0,600 0,826 0,000 0,917 0,025 0,815 0,002 0,968 0,170 0,380 0,000 0,413 0,004 0,655 0,040 1,000 0,926 0,610 0,202 0,469 0,069 0,123 0,679 0,765 0,000 0,830 0,000 0,622 0,270 0,210 0,000 0,650 0,021 0,903 0,000 0,703 0,059 0,638 0,148 0,268 0,788 0,395

33