On the basis of the protocol established by Valenzuela et al. (2006) in their nation-wide study of day labourers in the United States, the researchers conducted a census of day labourers in the various cities and towns in South Africa in 2005 and 2006. The researchers recorded the names of the streets where the day labourers congregated and counted the number of day labourers at each site.²

An important inference to be made is that in a census of this nature, the numbers might not reflect the total size of the day labour workforce. For example, exclusions would be workers who were hired prior to the count and workers who regularly used the hiring site but for some reason did not search for work on the day in question (Valenzuela et al. 2006:4). There were close to 1000 locations in South Africa where people were picked up, while a minimum of 45 000, mostly African men, stood at these sites looking for work each day.³

Table 1 provides a summary of the estimated number of day labourers in the nine provinces in South Africa during the 2005–2006 research and counting process, compared with the provincial population distribution at the time.

Table 1 shows that the provinces with the highest number of day labourers were the wealthier ones (Gauteng, Western Cape and KwaZulu-Natal). The interviews conducted with the day labourers confirmed these provinces as the most popular destinations for foreign and South African day labourers alike.

Regarding the sampling procedure, the capitals of all provinces and important hubs in the rural areas had to be covered proportionally according to the number of day labourers present. A cluster sampling approach was used (Rubin & Babbie 1997), with clustering taking place in terms of both the number of day labourers interviewed in each centre and the size of the various hiring sites. Had this not been done, there would have been an over-representation of rural versus urban areas and big hiring sites versus small hiring sites.

Aiming for a representative sample size of around 10% (to control for possible sampling error) meant that between 2500 and 4000 interviews had to be conducted across South Africa (De Vos et al. 2004:200). The fieldwork commenced in February 2007 and was concluded at the beginning of 2008.⁴ A total of 3830 questionnaires were accepted for the analysis, representing 8.5% of the estimated research population.

From an ethical standpoint, it was crucial not to deprive any day labourer of work because he was participating in the interview process. In some cases, therefore, interviews were not conducted. This meant that the data had to be cleaned so that a cross-sectional regression analysis could be conducted. Thereafter, the size of the sample was 2645.

The following descriptive statistics were compiled on the participants in the study.

Table 2 provides a summary of the basic demographic characteristics of the day labourers who participated in the survey.

The survey results for gender and age of the day labourers reflect the racial composition of the broader informal sector in South Africa (Saunders 2005). Notably, the age distribution highlights the persistent unemployment and under-employment among the country’s youth (Statistics South Africa 2015). In addition, each day labourer supported an average of four people, excluding themselves. At least 170 000 people survived on the income generated through this type of informal employment.

A key element in the study of subjective well-being among day labourers is the income that they earn (Blaauw et al. 2013). Table 3 shows the average income earned by survey participants in a ‘good week’ and ‘bad week’ respectively, across the different provinces.

However, merely looking at the averages does not tell the full story. Harmse et al. (2009) observed differences in attitude about income-earning prospects among day labourers in the various cities and towns across South Africa. An analysis of income earned at a more disaggregated level is also necessary. Table 4 and Table 5 disaggregate the distribution of income earned by day labourers in South Africa, in a good and bad week respectively, at the provincial level.

Table 4 reveals that in (what survey participants perceived as) good weeks, the majority of day labourers earned in excess of ZAR 400 per week in 2007. This applied to South Africa as a whole as well as to six of the nine provinces. Income levels in good weeks were clearly lower in the Free State, Northern Cape and Limpopo provinces.

Table 5 reveals only one income category (less than ZAR 100 per week) applicable to the majority of day labourers in all provinces during a bad week. Again, it seems that the situation in the Free State, Northern Cape and Limpopo provinces was worse than at the national level. This in turn appears to mirror general inequality patterns in South Africa and highlights relative provincial economic performance at the time. This would have important implications for the power relations within the competitive day labour market. Given the persistent depressed economic conditions in these provinces, day labourers had virtually no power to negotiate wages with prospective employers (Theodore et al. 2015). Theory would suggest that these factors have an undeniably negative influence on the subjective well-being of day labourers in South Africa (Sen 1999, 2003). This deduction constitutes one of the key building blocks for the model selection and empirical analysis described in the next section.

To investigate the impact of geographical location on the well-being of day labourers in South Africa, we employed a cross-sectional analysis using dummy variables for the nine provinces – which the exploratory study by Blaauw et al. (2013) did not make provision for. The purpose of these dummy variables was to account for differences in the well-being of day labourers between provinces. The model can be specified as follows:

y_i is the dependent variable for a certain cross-section (i); D is the subgroup in the cross-section – in this case, a province – and β is a k × 1 vector of parameters to be estimated on the k × 1 observations of the explanatory variables (X); i is 1,…N (Brooks 2008). As a robustness check, we also estimated a cross-sectional panel model. The results corroborated our findings. However, given that t = 1 in the panel analysis, we only report on the cross-sectional analysis in the ensuing paragraphs.

Two functions were specified, namely a subjective and an objective function. The difference between these two functions was the dependent variable, which was a proxy for the subjective and objective measure of well-being. We based our choice of the dependent variable for the subjective function on the work of Knight, Song and Gunatilaka (2009), which concluded that satisfaction with household income is one possible measure of subjective well-being (SWB). The dependent variable for the subjective function was the satisfaction of the day labourer with their level of income (Is your income worse, or good or better than expected?). This was a binary type variable with two categories (D = 0 if it is worse; D = 1 if it is good or better than expected). The variable may, at first glance, appear to be merely an income satisfaction variable. However, there is an argument that even if the actual income is as good as expected, it does not necessarily mean that the individual is satisfied or happy with it. The socio-economic construct of day labourers potentially counters this conclusion.

Since day labouring is a survival strategy, traditional economic considerations such as optimality and wealth creation do not feature directly. Therefore, within this unique context, the dependent variable does not merely constitute a comparison between expectations and reality; indeed, it is a measure of subjective well-being among labourers.

Day labourers are well aware that the supply of their type of work is increasing constantly while the demand is not keeping pace. The proverbial cake must therefore be shared among an ever-growing day labour force, while a constantly diminishing level of real earnings is expected (Blaauw et al. 2013). This view is supported by the fact that reservation wages are steadily falling in many informal labour markets (Theodore et al. 2009). If a day labourer says his or her income is as good as or better than expected, it shows a level of subjective satisfaction, given the typical uncertainty and income variances in the day labour market. The fact that this was a recurring theme in discussions with the survey participants convincingly motivates the use of this variable in the subjective function. We therefore argue that the dependent variable does not merely mean a comparison between expectations and reality; it is a measure of subjective well-being among labourers.

According to the literature, income is a very important indicator of well-being in poor communities (Diener et al. 1993, 2002). The dependent variable for the objective function was the log of the best wage earned per day. When day labourers achieve their best income levels, it brings about a feeling of happiness and improved well-being (Blaauw et al. 2013).

Table 6 and Table 7 provide descriptive statistics of the dependent variables for the objective and subjective functions of the two models.

The two dependent variables were the ‘Best wage’ and ‘Good or better expectations’, respectively. The highest mean best wage was in the Western Cape, followed by KwaZulu-Natal and Gauteng. The maximum best wages recorded were in KwaZulu-Natal, Gauteng and Limpopo. The lowest mean best wage was in the Free State. There was high dispersion in the data, as revealed in the large differences between the minimum and maximum best wages per province as well as the high kurtosis values.

For the subjective measure (looking at the percentage of total day labourers per province), the Eastern Cape recorded the most labourers with good or better income expectations, followed by Gauteng and the Western Cape. Together these accounted for about 50% of the sample per province. Limpopo recorded the lowest number of labourers satisfied with their income (only 3%), followed by the Free State and KwaZulu-Natal.

Table 7 indicates the best wage divided into categories. In addition, it shows the number of day labourers whose income expectations were worse, or good or better, according to these wage categories. It is evident that the first two best wage categories were dominant with lower numbers recorded as the income levels increased. Furthermore, it is in these two best wage categories that the income expectations were also the worst. Looking at the full sample, 58% of day labourers were not satisfied with their income as opposed to 42% who were satisfied.

The basic approach to specifying a well-being function is to include basic variables, conventional economic variables, comparison variables, community variables and attitudinal variables as explanatory variables (Cummins 2000; Kingdon & Knight 2004; Knight et al. 2009). The following variables from the day labourer survey were used and classified according to these broad categories:

The basic demographic variables are standard throughout the literature. They include gender, education, marital status and age. In this survey, age was categorised and was not used as a continuous variable. See Tables 8–11 below.

The comparison between current income and past circumstances and aspirations formed the subjective perception of the participants of their financial situation. These comparison variables were primarily a comparison of an individual at different times, or with other day labourers that the participant knew.

The attitudinal variables were included to control for behaviour, which affected subjective well-being. The three attitudinal variables reflected the psychological state of mind or attitude shaping their perception of their job or situation. The community variables captured the unobserved provincial differences, which could have been related to government funding, infrastructure in rural areas, governance, environment and culture (Knight et al. 2009). The variables were grouped according to the nine provinces in South Africa.

The objective and subjective functions were estimated on a cross-section with 2645 (n) observations within the dummy variable framework. The two-stage least-squares method (2SLS) was used in the objective function, but due to the binary nature of the dependent variable for the subjective function a probit model was selected as the most appropriate model for this function.

The income variable, log of the wage in a good week (GOODWEEK), was instrumented because of endogeneity problems that occur with income when estimating SWB functions (Kingdon & Knight 2004; Senik 2005). The endogeneity test indicated that the income variable (GOODWEEK and the change in income) was endogenous in the objective function, but not in the subjective function. The J-statistic probabilities were 0.0 and 0.58 for the objective and subjective functions, respectively. In the case of the objective function, the null hypothesis (exogeneity) was rejected, indicating the endogeneity of the variable.

The log of the lowest wage (LOWWAGE), change in the log of the lowest wage and the number of months working as a day labourer (MONTHS) were used as possible instruments. The orthogonality C-test confirmed the exogeneity of these variables where the probability of the J-statistic was >0.05 and showed acceptance of the null hypothesis (exogeneity). The weak instrument test showed that the number of months working as a day labourer was weaker than the other two instruments. The estimation results for the objective function and the subjective function are presented in Table 12.

Most demographic variables were significant in the 2SLS model, which estimated objective well-being across South Africa. The twenties age group was significant. The education variables were significant with an unexpected sign for completed secondary school. The economic variables were all significant and the signs were according to expectations, except for previously having a full-time job which recorded a positive sign. Possibly, former full-time workers were more confident of their future prospects at the time of being interviewed as opposed to being dejected or angry about losing their employment status. Only the job opportunity and food variables were significant for the comparison variables – the signs were unexpected. For example, having enough food might reduce the opportunity cost of not standing on the street corner for days on end, leading to less employment and income. This might well be a cyclical and therefore non-linear relationship which needs to be tested in future research. The attitudinal variables were all significant, except for staying with family.

When accounting for the differences in provinces in the sample (2SLS with dummies), the significance of some of the variables changed. Most of the variables that were significant in the 2SLS model were still significant with the same sign, except for the economic variable turned down a job which became insignificant, and the comparison variables having food and better job opportunities at this site became insignificant. The attitudinal variable, staying with family, became significant. All the economic and attitudinal variables became significant (except turned down a job) in the dummy variable model. The dummy variables for the nine provinces indicating the geographical links and community variables were all significant. This shows that there were significant differences between the provinces and it confirms the importance of community variables as explanatory variables for well-being functions. The bigger size effects for provinces such as the Western Cape and Gauteng possibly reflected their status as the powerhouse economic regions in South Africa.

The subjective function was first modelled using the probit model to estimate subjective well-being in South Africa, while the second probit model used dummy variables to account for the differences between the provinces in South Africa. The results of the two models were very similar. Only gender, primary schooling and the twenties age group (only in the first probit) were significant. The economic variables, income and whether the participant previously had a full-time job were significant with the expected sign. All the comparison variables were significant with the expected sign. No attitudinal variables were significant, except for the support variable in the dummy variable model with an unexpected sign. The support variable was only significant at the 90% confidence level and might reflect the fact that day labourers saw the need for support as detracting from their perceptions of their own capabilities and levels of agency.

The dummy variables for the nine provinces indicating the geographical links and community variables were all significant. This shows that there were significant differences between the provinces and also confirms the importance of community variables as explanatory variables for well-being functions. The negative signs in the case of the subjective function might reflect the disappointment linked to anticipated better opportunities in the migrant destination provinces as well as the prevailing negative perceptions in the poorer provinces from which the migrants originated.

Comparing the objective and subjective functions, there were clear differences in the results. The attitudinal variables were significant in the objective function, whereas comparison variables were significant in the subjective function. The economic variables were significant in both functions; however, only two of the four variables were significant in the subjective function. This confirms what the literature says about income being an important factor in determining well-being within poor communities.

The objective function was mainly driven by the tangible aspects of day labouring, such as staying with family, injuries occurring and being part of a support group, while the dependent variable was the actual income. The subjective function was mainly driven by comparison variables, including whether there were better job opportunities at the site, whether day labourers had enough food and changes in income. These variables were based more on perceptions and the function’s dependent variable was based on income expectations. The community variables and the dummy variables for the provinces were significant in both functions. The objective function’s coefficients were positive and indicated that the provinces differed in terms of income (Knight et al. 2009). The subjective function’s coefficients were negative, showing that perceptions of well-being had a negative impact on subjective well-being – which was in contrast to the positive impact on well-being in the objective function. This possibly also explained the different signs for the full-time job variable in the two functions. This is because the perception of a full-time job was negatively perceived in terms of subjective well-being, in contrast to the positive sign in the objective function, which was linked to a certain income rather than the perception.

Mulcahy and Kollamparambil (2016) found that rural-to-urban migration has a negative impact on subjective well-being in South Africa due to unmet expectations and aspirations, and emotional fallout. For day labourers in South Africa, this may also be the case. The richer provinces have more job opportunities but also more competition, and therefore the income may not be as good as day labourers expect. This is supported by the evidence presented in a study by Krugell and Blaauw (2011): in metropolitan areas, greater occupation density was negatively associated with the earnings of day labourers, which reflected the competition for jobs among day labourers in the big cities. A general increase in the number of day labourers in South Africa’s major cities since the global economic downturn and slower growth in the construction industry have given rise to even greater competition.

On a methodological level, the results open up new avenues for refining the study of subjective well-being as it relates to the informal economy. The question arising from the application of Sen’s CA is the extent to which it is not the income per se that influences the level of subjective well-being but rather what people can afford with the income and the welfare difference they feel afterwards (such as freedom from hunger and other forms of marginalisation) that determine their levels of subjective well-being. Although the literature supports income as a measure of objective well-being, it cannot speak to the capabilities, functionings and freedoms it brings. Measuring the dimensions of subjective well-being is the methodological challenge for future research flowing from this study.

Well-being in the informal economy is a multifaceted phenomenon. It is imperative to use both the objective and subjective measures of well-being when dealing with well-being among day labourers and, by implication, other marginalised groups. For many vulnerable groups such as day labourers, waste pickers and car guards, the informal economy offers both opportunities and certain freedoms to the unemployed. Even with limited capabilities, these people are able to access work and income through some form of agency. This offers them at least some self-respect and provides an alternative to a life of begging or, worse, crime. A day labourer in Port Elizabeth summed it up very well: ‘We cannot sit at home and watch the hungry family’.

Finding ways to enhance and deepen the agency and capabilities offered by the informal economy will provide an important new field of research for academics and other research professionals. This in turn will lay the foundation for more robust urban and social development policies and implementation plans in South Africa, and strengthen hopes for a much more equal society.