The Impact of the Agricultural Minimum Wage on Farmworker Employment in South Africa: A Fixed Effects Approach - Draft Chris Garbers - 16294335 March 21, 2015

Abstract Through the use of the fixed effects estimator, this paper estimates the impact of the agricultural minimum wage on farmworker employment in South Africa. My identification strategy makes use of a "natural experiment"-type setting where I have identified periods of exogenous variation in the minimum wage to identify the causal effect of interest. The findings indicate that formal unskilled farm employment decreased by approximately 16% as a result of the regulation of which 7.5% is directly attributable to higher unskilled labor costs resulting from the wage floor. There is also evidence of skill and capital intensification resulting from the minimum wage.

1

Contents 1 Introduction

4

2 Literature Review 2.1 Minimum Wage Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Empirical Findings Regarding Minimum Wages . . . . . . . . . . . . . . . . . . .

5 5 6

3 Data and Methodology 3.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Econometric Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10 10 13 16

4 Empirical Analysis 4.1 The effect of the minimum wage on formal farmworker employment . . . . . . . .

19 19

5 Conclusion

25

A Appendix

30

2

1

Introduction

Since the early 1990s research regarding the employment effects of minimum wages has garnered increased attention in both developed and developing economies. This interest was driven mainly by disparate findings as to the likely employment outcome when minimum wages are introduced into labour markets. Whilst most research indicated towards unemployment resulting from a binding wage floor, others have found evidence indicating toward the possibility of employment gains (Katz and Krueger (1992); Card and Krueger (1995); Neumark and Wascher (1991)). Subsequent research has shown that positive employment outcomes are most likely due to studies focusing on a narrow sub-group or specific industry as well as being unable to incorporate the importance of region specific demand shocks to labour market outcomes. Therefore, on aggregate, unemployment is the expected result of minimum wage legislation (Neumark and Wascher (2007)). Most minimum wage studies have concerned a diverse range of developed economies with developing country evidence concentrated amongst Latin American economies (Neumark, Salas, and Wascher (2013)). In South Africa, minimum wages are a relatively recent addition with the first legislation implemented in 1999 and covering only the Contract Cleaning Sector. Since then several sectoral minima have been introduced with coverage spreading to all the major sectors of South Africa (Bhorat, Kanbur, and Mayet (2013)). From 1 March 2003 minimum wage legislation was extended to the agricultural sector, a major source of formal employment for unskilled labour (StatsSA (2008)). Since then, a few studies concerning the impact of this legislation on agricultural employment have emerged with varying results. Whilst Conradie (2005) as well as Bhorat, Kanbur, and Stanwix (2012) estimated disemployment effects, Murray and Van Walbeek (2007) found that no significant employment changes resulted from the agricultural minimum wage. The difference between these results could be due to narrow focus of the Conradie (2005) and Murray and Van Walbeek (2007) analyses where survey data on specific sub-groups of agricultural workers were used. The large structural unemployment rate of the South African economy and its implications for political and economic stability emphasizes the importance of the agricultural sector as a major source of formal employment for the unskilled (Bhorat et al. (2012); StatsSA (2008)). Reaching a consensus on the employment effects of the agricultural minimum wage in South Africa is therefore of vital importance, especially when one considers the economic costs of protests over farmworker wages as evidenced by the events in De Doorns at the end of 2012 (SABC (2013)). This paper aims to contribute to the discussion by estimating the employment outcomes of the agricultural minima using a district level panel dataset. This dataset allows the analysis to incorporate the importance of region specific conditions to labour market outcomes whilst simultaneously avoiding estimation bias that may result from narrowly focused studies. Additionally, an interview was conducted with Prof. Nick Vink1 who was present during discussions resulting in the path set for the minimum wage. He provided testament to the exogeneity of the minimum wage, thus 1

Professor of Agricultural Economics at the Univeristy of Stellenbosch.

3

providing a sound foundation for econometric analysis regarding the employment impact of the minimum wage since feedback is limited to one direction (Vink (2013)). The second section of this paper reviews the minimum wage theory after which empirical findings regarding minimum wages are considered. Section 3 regards the data and methodology of the analysis as well as a descriptive overview of the dataset. Section 4 contains a discussion of the econometric results of the analysis after which Section 5 concludes.

2

Literature Review

2.1

Minimum Wage Theory

The employment consequences associated with minimum wage legislation has been hotly debated by academics since they were first implemented in the early 1900s (Obenauer and von der Nienburg (1915); Stigler (1946); Lester (1947)). Neoclassicals argued that wages were determined by worker productivity and so a binding minimum wage would realise unemployment effects. On the other hand, progressive arguments made use of efficiency wage theory and the effect minimum wages could have on aggregate demand through increased consumer buying power. They also voiced concern over the exploitation of workers in imperfectly competitive labour markets and saw minimum wages as a means to address this issue (Neumark et al. (2013)). From a theoretical point of view, the employment consequences of minimum wage legislation depend on the structure of the affected labour market. Employment losses can be expected to result from a binding minimum wage in a perfectly competitive labour market whereas employment gains may be realized in monopsonistoc labour markets (Borjas (2010)). Under the assumptions of perfect competition, a minimum wage set above the market clearing wage (a binding minimum wage) will result in employment losses as a result of inadequate demand at the higher wage rate (Borjas (2010); Brown, Kilroy, and Kohen (1982); Rocheteau and Tasci (2007)). In a monopsonistic labour market where firms have to pay all individuals the same wage, the labour supply curve Ls no longer depicts the marginal cost of labour M Cl . This results from the fact that the marginal cost of an additional worker needs to include a pay rise to all existing workers in order to maintain a homogenous wage rate across employees. Therefore in a monoposonistic labour market the M Cl curve lies above the Ls . Combined with the profit maximizing behavior of firms, the fact that M Cl > Ls implies that individuals’ wages are less than their value of marginal product (Rocheteau and Tasci (2007))2 . Within this framework, the imposition of a minimum wage mitigates firm influence over wage rates by keeping marginal costs constant until all individuals willing to work at the minima are employed. This feature allows for employment gains to take place under monopsonistic conditions in the face of minimum wages. Furthermore, it is possible to employ the same number of individuals as would have occurred in a competitive market if the government knew and was able to set the minimum wage such that it coincides 2

Firms maximize profits when they employ workers such that their value of marginal product equates to their marginal cost.

4

with the competitive market clearing wage ((Borjas, 2010, p. 195); Rocheteau and Tasci (2007); Stigler (1946); (Manning, 2006, p. 87)). As shown above, it is possible to generate employment opportunities with a minimum wage policy, provided the underlying labour market structure is monopsonist. This boost to employment occurs through mitigating the negative effects of monopsony power by reducing firm influence over the wage level. On the other hand, employment losses are also a possibility in a perfectly competitive labour market. Given the interlinked and complex structures of eoconomies, it is more likely that a mixture of the two market structures will prevail, making ex ante hypotheses regarding the employment effects of minimum wages a highly contentious issue (Manning (2006)). When assessing the potential impact of minimum wages in a developing country, ex ante hypotheses become even harder to formulate as one has to account for the legislation’s impact on both the formal and informal sector (Ham (2013)). Labour market institutions - e.g. the minimum wage - only cover formal workers and so, minimum wages could have an impact on conditions prevalent in the informal labour market. These dynamics are most often explained in terms of the Harris-Todaro two sector model. Assuming a perfectly competitive labour market, one would expect a wage gain and employment loss to occur in the formal sector whilst in the informal sector one would expect wage rates to fall and employment to rise. This narrative results from the migration of workers toward the informal sector as surplus formal labour seeks employment (Harris and Todaro (1970); Ham (2013)). From the theory discussed above it is evident that case specific knowledge is required in order to make predictions as to the possible employment effect of a minimum wage. In light of this the next section reviews the empirical evidence on the relationship between minimum wages and employment, drawing on results from both developed and developing countries.

2.2

Empirical Findings Regarding Minimum Wages

Research regarding the employment effects of minimum wages in developed economies is comprehensive with empirical work dating back to 1915 (Obenauer and von der Nienburg (1915); Neumark et al. (2013); Bhorat et al. (2013)). In 1982 Brown, Gilroy, and Kohen published a summary of the existing literature on the employment effects of minimum wages, concluding that minimum wages and employment typically exhibit a negative relationship with an associated elasticity of -0.3 (Brown et al. (1982); Neumark and Wascher (2007)). These results developed the consensus that negative employment effects will be realised where minimum wages are implemented. In the early 1990s this consensus wavered amidst critique over the econometric techniques used in the earlier minimum wage research as well as the publication of new findings that were in conflict with the aforementioned consensus. (Card and Krueger (1995); Kim and Taylor (1995)). Critique concerning the accuracy of the estimations was based on a lack of variation in minimum wages during the sample period, misspecification of the model, possible endogeneity of the minimum wage variable, as well as the lack of a well-defined counterfactual. In addition to this, new minimum wage research using a more plausible identification strategy found mixed results regarding 5

employment effects (Neumark and Wascher (2007)). Time series analysis of US state level panel data by Neumark and Wascher (2007) found a negative relationship between minimum wages and employment and estimated an employment elasticity of around -0.2 regarding the minimum wage. On the other hand, both Katz & Krueger (1992) and Card & Krueger (1993) found positive employment effects in the US fast food industry resulting from an increase in the minimum wage, estimating an employment elasticity of about 0.35 regarding changes to the minimum wage. The blame for this disparity in empirical outcomes has been largely attributed to differing measurement techniques used by the researchers. Katz and Krueger (1992) as well as Card and Krueger (1993) both measured the employment effects of minimum wage changes using differencein-difference analysis compared to Neumark & Wascher’s (1992) time series approach. The difference-in-differences approach allows for the minimum wage change to be viewed as a type of natural experiment by comparing fast-food restaurant employment changes in a state where the minimum wage was raised to that of a state where no such change was implemented. Brown, Freeman, Hamermesh, Osterman, and Welch (1995) question the accuracy of these case study findings on grounds of unreliable survey data and inappropriate control groups. Card and Krueger (1993) compared employment outcomes in the New Jersey fast-food industry following an increase in the minimum wage to that of the fast-food industry in Pennsylvania where no such change occurred. However, employment trends between the two states diverged significantly prior to the minimum wage change bringing the appropriateness of their control group (and thus their results) into question (Neumark and Wascher (2007)). With regards to the reliability of their data: Neumark and Wascher (2000) replicated the Card and Krueger (1993) study using payroll data (as opposed to survey data) and found that minimum wage increases were associated with decreased employment in the fast food sector with an estimated elasticity between -0.1 to -0.25 (Neumark and Wascher (2007)). Brown et al. (1995) also question the accuracy of the estimates when using difference-in-differences to gauge the employment effects of minimum wage changes. The authors argue that state specific demand shocks dominate as a source of variation in employment, and so, relative demand shocks will swamp the effect of a higher minimum wage. For example, region specific economic growth could buoy demand for products produced by an industry with a minimum wage, dampening and possibly even eliminating the employment consequences associated with the legislation. Similarly, in regions where such economic growth did not take place the minimum wage employment effects may be more pronounced. This indicates that narrow studies which focus on a specific industry or region are more likely to find minimum wage effects not in line with competitive theory. In the absence of a consensus as to the aggregate employment effects that arise from minimum wage changes, Neumark and Wascher (2007) published a comprehensive summary of the new minimum wage research in an attempt to consolidate the various findings and techniques. Their summary showed that on aggregate, minimum wage legislation is associated with negative employment outcomes in developed economies with an associated elasticity of around -0.1 to -0.3 (Neumark and Wascher (2007)). An important finding of their summary regards the inclusion of a business cycle and relative wage variable. The business cycle variable aids in improving 6

the accuracy of the minimum wage coefficients by capturing dynamics present in the broader economy; thus enabling the researcher to discern between unemployment resulting from minimum wages and that resulting from an economic downswing. The inclusion of a relative wage variable has a similar role to the business cycle variable in that it removes an important source of omitted variable bias from the minimum wage coefficients by controlling for the opportunity cost of labour. The relative wage variable is a ratio expressing wages in the sector covered by the minimum wage law to those wages not covered by the law. Thus one should find a negative relationship between employment in the covered sector and the relative wage. Neumark and Wascher (2007) also find that longer sample periods allow for easier identification of a negative associated relationship between minimum wages and employment. Since minimum wages usually apply to the unskilled labour force, the implementation of a binding wage floor incentivises production to becoming more skill and/or capital intensive. Such adjustments may take time to be realised, and so identifying the impact of minimum wages on employment should be better suited to long sample period studies (Brown et al. (1995); Literature on the employment effects of minimum wages in developing countries is sparse relative to its developed counterpart; however substantial research has been done regarding the Latin American experience. Empirical evidence has indicated that in Latin American economies the informal sector accounts for around 30-70% of the labour market (Almeida and Terrell (2008); Gasparini and Tornarolli (2009)). As discussed earlier, minimum wage legislation can have an impact on labour conditions in the informal sector through the migration of displaced formal workers as per the two-sector model (Harris and Todaro (1970)). Latin American research by Fajnzylber (2001)and Maloney and Mendez (2004) question the validity of the above hypothesis with their results indicating that minimum wages tend to increase both formal and informal sector wages. However, research on Honduras by Ham (2013) indicates that although informal wages may initially rise following changes to the wage floor, the long run adjustment sees a reduction in informal wages. This concurs with the developed country evidence in that it takes time for the full employment effect of minimum wages to be realised (Neumark and Wascher (2007); Ham (2013)). Latin American evidence also suggests that minimum wages tend to be more binding in developing economies with wage floors accounting for around 20-60% of the average wage (Maloney and Mendez (2004); Boeri, Helppie, and Macis (2008)). This evidence implies that modifications to the wage floor should realise changes to the prevailing labour market conditions. Fajnzylber (2001) tests this hypothesis by estimating the employment consequences of the minimum wage in Brazil, and finds an associated elasticity of -0.05 to -0.08 for the formal sector and -0.05 to -0.15 for the informal sector. Bell (1997), Maloney and Mendez (2004), and Arango and Pachón (2007) find similar unemployment effects in Columbia with an estimated employment-minimum wage elasticity of around -0.02 to -0.12 and in Costa Rica, Gindling and Terrell (2007)estimate an elasticity of around -0.11. From a South African perspective, minimum wage legislation has been a relatively recent occurrence. The first minimum wage in post-Apartheid South Africa (SA) was implemented in 1999 and covered the contract cleaning sector. Since then several sectoral sub minima have been legislated with coverage extending to all the major sectors of South Africa (Bhorat et al. (2013)). The South 7

African literature on the employment effect of minimum wages is somewhat limited; however the impact of domestic worker and farmworker minimum wages are relatively well-documented cases. Bhorat (2000) and Hertz (2005) both found that the number of employed domestic workers in SA decreased following the implementation of sectoral minimum wages with Hertz (2005) estimating an associated elasticity of -0.46. On the other hand, Dinkelman and Ranchhod (2011) find no significant proof of unemployment effects in the domestic worker industry following the minimum wage legislation. The disparity in the findings could to some extent be attributed to differing samples with Dinkelman and Ranchod choosing to remove rural domestic workers from their sample (Dinkelman and Ranchhod (2011)). With regards to the agricultural minimum wage, Conradie (2005) uses a survey of 190 Western Cape grape farmers to estimate and elasticity of about -0.3 to -0.6 regarding farmworker employment responses to minimum wage changes. In another case study based on 109 sugar industry farmworkers, Murray and Van Walbeek (2007) found that in response to the wage floor, employers substituted away from lower-skilled labour and reduced the number of hours worked. Bhorat et al. (2012) used difference-in-differences analysis to compare the employment outcomes in the agricultural sector to that of a control group during 2000-2007. Their control group consists of unskilled, non-unionised individuals of working age who are not covered by another sectoral minimum wage. The results from Bhorat et al. (2012) indicate that agricultural employment dropped by approximately 15% as a result of the minimum wage. Bhorat et al. (2013) investigate the impact of the retail, domestic worker, forestry, taxi, and private security sectoral minima on employment within their sectors during the years 2000-2007. Their analysis consisted of a difference-in-differences estimation as per Bhorat et al. (2012) and Card and Krueger (1995) with the control group being the narrow labour force. Their results indicate employment gains took place post minimum wage implementation in the retail, domestic worker, and private security sectors whereas forestry employment did not significantly change. Only taxi worker employment was significantly adversely affected by the minimum wage (Bhorat et al. (2013)). The literature reviewed above indicates that on average, both developed and developing economies experience employment losses as a result of minimum wage legislation, proving the competitive labour market model to be most applicable on aggregate. Most of the South African literature has analysed the impact of minimum wages using region specific case studies or difference-indifferences analysis with varying results. However, these results may be inaccurate due to the dominance of region specific demand shocks in explaining employment outcomes. Also, only the Bhorat et al. (2012) and Bhorat et al. (2013) studies have considered the minimum wage impact on SA employment over a longer time span and so, SA studies over a shorter time period may be biased as they cannot account for the long-run dynamics of minimum wages. The dominance of the difference-in-differences approach to measuring South African minimum wage effects and the lack of longer time period studies create scope for contribution through the use of an alternative estimation technique with a long sample period. This paper aims to participate in the minimum wage debate by estimating the impact of the agricultural minimum wage on employment through the use of a district level panel dataset running from 1997-2007. 8

This dataset enables the researcher to control for the dominance of region specific demand shocks through the use of the fixed effects estimator. The long time span over which the dataset runs also allows for the incorporation of long-run minimum wage dynamics in the estimates. The next section discusses the creation of this dataset as well as the methodological approach followed in measuring the employment impact of the agricultural minimum wage.

3

Data and Methodology

3.1

Dataset

The dataset used for this analysis contains household, person, and labour force data from the October Household Survey (OHS) and Labour Force Survey (LFS) for the period 1997-2007. The OHS was conducted in October every year from 1995-1999 whereas the LFS was conducted during March and September each year during 2000-2007. The 1995 and 1996 waves of the OHS were excluded as only 7 farmworkers were surveyed during this period. This translates into a pooled dataset consisting of 19 surveys administered over 11 years with 1 220 445 observations of which 46 629 work in the commercial farming sector and 31 790 are classified as farmworkers. Farmworkers are identified as those with elementary occupations in the agricultural sector, but only working in commercial agriculture. This definition allows focus to be placed on those individuals to whom the minimum wage is most likely to apply as they are the least skilled agricultural workers. In order to estimate the vailidity of the two-sector model, focus will be placed on the unskilled informal workers as well as the unskilled subsistence farmworkers of which account for 24% and 5% of the unskilled workforce respectively. All observations in this dataset are of working age i.e. older than 15 years and younger than 65 years. In order to estimate the impact of the agricultural minimum wage on farmworker employment the relevant hourly minimum wage is assigned to all observations in the dataset3 . From 1 March 2003 the agricultural minimum wage was implemented with the path for farmworker minimum wages being determined by Sectoral Determination 8 (SD8) for the period 1 March 2003 - 28 February 2006 and Sectoral Determination 13 (SD13) for the period 1 March 2006 - 28 February 2009. According to this legislation, the minimum wage applicable to farmworkers depended on the local municipality in which the worker resided. A higher minimum wage was applicable in local municipalities classified as area A municipalities whilst a lower minimum wage applied to municipalities not in area A (declared area B municipalities). The value of the minimum wages were converted into year 2000 Rands using historical CPIX data from Stats SA. The path for minimum wages in the agricultural industry as per SD8 and SD13 in 2000 Rands is shown in table 3.1 below. Unfortunately the OHS and LFS datasets do not contain any information regarding the local municipality of participants; however they do indicate which magisterial district or district council 3

An expected minimum wage of R 0.1 per hour is assigned to all observations prior to 1 March 2003 indicating that all individuals expect to earn some positive real hourly wage if they were to be employed.

9

Table 3.1: Minimum wage for the agricultural sector as per SD8 and SD13 in year 2000 Rands Area A Area B CPIX (nominal wage) (nominal wage) (2000=100) LFS 2003a (Period 14) 4.1 3.33 123.5 LFS 2003b (Period 15) 4.1 3.33 125.6 LFS 2004a (Period 16) 4.47 3.66 128.9 LFS 2004b (Period 17) 4.47 3.66 130.3 LFS 2005a (Period 18) 4.87 4.03 133.5 LFS 2005b (Period19) 4.87 4.03 136.4 LFS 2006a (Period 20) 5.1 4.54 138.6 LFS 2006b (Period 21) 5.1 4.54 143.3 LFS 2007a (Period 22) 5.34 5.07 146.2 LFS 2007b (Period 23) 5.34 5.07 152.9 Source: Government Gazette No. 24114; No. 28518 and Stats SA. Survey

Area A (2000 Rands) 3.32 3.26 3.47 3.43 3.65 3.57 3.68 3.56 3.65 3.49

Area B (2000 Rands) 2.7 2.65 2.84 2.81 3.02 2.95 3.28 3.17 3.47 3.32

the surveyed inhabited. Each district council contains a number of magisterial districts and local municipalities. For the period October 1997 - March 2004 magisterial district data was captured whereas from September 2004 - September 2007 district council data was captured. For simplification, magisterial districts were recoded to show which district council they fell under and thus each observation in the dataset has a district council value attached. Since each district council (DC) contains a number of local municipalities, a DC may contain only area A municipalities; only area B municipalities; or a mixture of both. For those DCs that contained both area A and B municipalities, population estimates from the Community Survey 2007 were used to determine the proportion of the DC population that reside in area A municipalities. A new variable, area weight, was then assigned to each individual in the dataset. This variable was dependent on the DC in which the individual resided and its value indicated the proportion of the DC population that lived in area A municipalities. Table 3.2 below shows the different area weight outcomes based on the calculation explained above, for the full sample as well as for farmworkers only. Table 3.2 indicates that 83.65% of the sample resides either in DCs containing only area A municipalities or in DCs containing only area B municipalities with the remaining 16.35% of observations residing in DCs with an area weight somewhere in-between. For farmworkers it is very much the same; however a smaller majority of farmworkers reside in DCs containing only area A or area B municipalities. Assigning the applicable minimum wage to observations living in DCs consisting of only area A or B municipalities is a relatively simple task. On the other hand, for those individuals residing in mixed DCs, assigning the appropriate minimum wage becomes problematic due to municipal data not having been recorded. In response to this constraint an expected minimum hourly wage is assigned to all observations whose value depends on the DC in which they reside. This expected minimum hourly wage was calculated as per equation 1 below, where Area A M in W age and Area B M in W age represent the hourly minimum wage applicable

10

Table 3.2: Area weight outcomes for the full sample and farmworker sub-sample Area weight 0 0.09 0.12 0.16 0.33 0.39 0.61 0.69 0.71 0.81 0.89 0.9 1 Total

Full Sample Frequency Percent 689 381 56.49 19 993 1.64 11 024 0.9 9 893 0.81 16 623 1.36 15 725 1.29 15 136 1.24 20 324 1.67 30 278 2.48 22 062 1.81 16 543 1.36 21 956 1.8 331 507 27.16 1220445 100

Farmworkers Frequency Percent 13 421 42.22 1 882 5.92 571 1.8 110 0.35 360 1.13 1 252 3.94 1 160 3.65 519 1.63 1 087 3.42 632 1.99 2 295 7.22 175 0.55 8 326 26.19 31790 100

Source: Own calculations using OHS and LFS datasets.

in each period in year 2000 Rands: E[M in W age] = Area W eight × Area A M in W age + (1 − Area W eight) × Area B M in W age

(1)

Using equation 1 above it is possible to assign an agricultural minimum wage to each observation in the dataset for the entire sample period 1997-2007. This minimum wage is constant prior to March 2003 whereas from March 2003 individuals experience variations in their expected minimum wage as per SD8 and SD13. Since the area A and area B minimum wages changed at different rates and were DC dependent it will be more informative to estimate the impact of the minimum wage on employment at a DC level as opposed to an individual level. For example, if you were residing in a DC containing only area A municipalities during 2006-2007 your expected minimum wage would have increased by less than that of an individual living in a DC containing only area B municipalities. One could then hypothesise that the unemployment effects will be stronger in area B DCs than in area A DCs during this period (assuming a perfectly competitive labour market and elastic labour demand). One can also expect vast differences in the employment capacity of area A (relatively rich) DCs compared to area B (relatively poor) DCs. This shows that there exists a large degree of heterogeneity between DCs and so the employment effects of the minimum wage will be DC dependent. In lieu of this observation and the importance of region specific demand shocks to labour market outcomes, the dataset was aggregated to a DC level to form a panel dataset. The panel dataset consists of 53 District Councils sampled over 19 periods from 1997-2007 and contains a total of 994 observations. The panel dataset is unbalanced meaning that all DCs were not observed in every time period. There were no observations for the Dr Ruth Segomotsi District Municipality (DC39) prior to September 2004 and the Uthungulu District Municipality (DC28) was not sampled during OHS 1997. All the remaining 51 DCs 11

were surveyed in every period from 1997-2007. This translates into 96.23% of the data having been observed in every time period; 1.89% missing an observation in 1997; and 1.89% only being observed after September 2004. The omissions should therefore not have any significant impact on the inference drawn from the analysis.

3.2

Econometric Methodology

The impact of the agricultural minimum wage on farmworker employment will be estimated through the application of a fixed effects model to the panel dataset. The effect of the minimum wage will be estimated on four variables of interest, namely: 1. the log number of unskilled commercial farmworkers 2. the log real wage of unskilled commercial farmworkers 3. the log number of commercial farmworkers that were not unskilled 4. the log number of tractor sales4 The model specification that will be used borrows from the panel data experience of Neumark and Wascher (2007; 2013) and is as per equation 2 below: Yi,t = β0 + β1 M in W age + ΨXi,t + εi,t

(2)

Where Yi,t denotes the variable of interest in DC i in period t. β0 denotes the intercept term; β1 denotes the contemporaneous effect of the logged real agricultural minimum wage. Ψ is a vector of coefficients for the vector Xi,t which includes controls for time fixed effects, DC fixed effects, education, experience, race, gender, DC population, the opportunity cost of unskilled farm labour, as well as the level of unionisation in each DC in each period. εi,t denotes the error term. The use of fixed effects rather than OLS stems from the fact that OLS makes use of between-DC variation to derive the estimates. This raises issues of endogeneity due to the possibility of omitted variable bias being present in the coefficients. Oft cited examples of this relates to important explanatory factors that are inherently difficult to measure such as natural ability. The omission of these covariates could introduce significant bias into the coefficients due to their influence on both the dependent and independent variables. For example, DCs in more developed regions could possibly appeal to a labour force that is relatively more productive since the economic opportunities in these areas will be quite plentiful, thus making these regions more attractive to naturally gifted individuals. In turn, the natural ability of individuals is likely to affect their occupation, education level, and experience. Through this channel the natural ability of individuals residing within a DC could influence both the number of farmworkers present within DCs as well as the control variables as per equation 2 above. 4

The data for tractor sales in each DC was provided by NAAMSA (The National Association of Automobile Manufacturers of South Africa).

12

With the fixed effects estimator it is assumed that all omitted variables have either time invariant values or time invariant effects, and so it is possible to partial out these unobserved effects (Allison (2009)). This is done by including DC and time dummies into the model specification allowing the researcher to control for the fact that economic opportunities and circumstances differ between DCs, thus removing an important source of bias from the coefficients. The time dummies are included to capture general trends present in the South African economy. For instance, as the modernisation of the SA economy continues and labour shifts toward the tertiary sector, the steady state level of farm labour may decrease. Such trends are likely to have the same impact across DCs and could describe some of the variation observed in employment outcomes. Additionally, including time fixed effects affords controlling for the business cycle, an important covariate identified by Neumark and Wascher (2007). Similarly, the fixed effects regression command in Stata automatically includes DC dummies; the purpose of which is to control for omitted variables whose values are time invariant but whose effects are time variant. The DC dummies therefore allow the estimator to utilise only within-DC variation in deriving the coefficients. As explained above, between-group variation is very likely to suffer from omitted variable bias; however since the fixed effects estimator focuses only on within-group variation this bias is removed and so the estimated coefficients are more likely to be accurate (Allison (2009)). With regards to the other control variables: Controls for the opportunity cost of unskilled farm labour is included as per the evidence from the minimum wage literature. The opportunity cost of farm labour is incorporated via a relative farm wage variable. Ideally, the relative farm wage should indicate unskilled farm wages relative to skilled farm wages in an effort to incorporate the employer’s option to change the composition of their workforce in response to changing opportunity costs of labour; however too few skilled farmworkers were surveyed from too few DCs for this to be a viable option. In both the OHS and LFS surveys there are approximately 200 occurrences where no skilled farmworkers were surveyed in certain DCs. Thus, these DCs would not have a relative farm wage if only skilled farm workers were used as an alternative to unskilled farm labour. Due to this lack of skilled farmworkers in the dataset, the opportunity cost of unskilled farm labour is calculated as a fraction showing the average unskilled farm wage as a proportion of the average wage for all commercial farm labour which is not unskilled. With this formulation only 52 observations do not have a value for the relative farm wage compared to 200 when only skilled farm workers are used. Controls for education, experience, gender, race, and unionisation are included to control for demographic changes which took place within DCs during the sample period. Finally, in the regressions concerning the number of employed, a control for DC population is included to aid in improving the accuracy of the minimum wage coefficients due to its explanatory power in terms of the workforce numbers within DCs. Using equation 1 to assign the relevant real hourly minimum wage to individuals residing in mixed DCs introduces some measurement error into the estimable model. The degree of measurement error will be largest in DCs with a relatively even mix of area A and area B municipalities. From table 3.2 it is evident this measurement error will only be significant for around 15% of the farm13

worker observations. For approximately 68% of the observations there is no measurement error and the remaining 17% of observations should exhibit a relatively small amount of measurement error. The degree to which this affects the accuracy of the coefficients as per equation 2 can be clarified via a discussion on the properties of the expected minimum wage as defined by equation 1. There is no measurement error in the minimum wage for individuals residing in DCs containing only area A or area B municipalities. For these observations the expected minimum wage equals the legislated minima and the estimable model is as per equation 2. However, for those individuals residing in DCs composed of a mixture of area A and B municipalities, the minimum wage is observed imprecisely. In this case equation 2 becomes: Yi,t = β0 + β1 [Ei,t (M in W age) + ui,t ] + ΨXi,t + εi,t = β0 + β1 Ei,t (M in W age) + ΨXi,t + β1 ui,t + εi,t

(3)

Where ui,t denotes the degree of measurement error in the minimum wage for DC i in period t. The specification of equation 1 dictates that this measurement error be positive for individuals residing in area A municipalities contained within mixed DCs whilst it will be negative for those living in area B municipalities located in mixed DCs. Efficient and unbiased estimates of the coefficients require the following two assumptions to hold: 1. Ei,t [εi,t |Ei,t (M in W age); Xi,t ] = 0 2. Ei,t [ui,t |Ei,t (M in W age); Xi,t ] = 0 Given the list of controls as indicated under equation 2, it is unlikely that there will be any remaining conditional correlation between εi,t and Xi,t . With regards to the second assumption, a key factor to consider relates to the relationship between Ei,t (M in W age) and ui,t . Substituting in for Ei,t (M in W age) from equation 1, condition 2 above becomes: Ei,t [ui,t |Area W eighti,t ; M in W agei,t ; Xi,t ] = 0

(4)

Where Area W eighti,t and M in W agei,t refer to the applicable area weights and minimum wages in each DC per period as discussed in section 3.1. If Ei,t [ui,t |Area W eighti,t ] 6= 0 then the estimates of β1 will suffer from attenuation bias as per the classical errors in variables problem5 . As stated above, the majority of the sample does not suffer from measurement error in the minimum wage variable limiting the scope for bias in the coefficients6 . Additionally, the nature of ui,t is such that there is no correlation between the measurement error and the remaining components of the control vector Xi,t . Of more pressing concern is whether the variation in the agricultural minimum wage can be viewed as exogenous in nature. In order to establish the exogeneity of the wage legislation, a review of the documentation utilized by the Department of Labor (DoL) in setting the path for minimum wages was performed 5 6

By construction Ei,t [ui,t |M inW agei,t ] = 0, refer to equation 1. A robustness check in section 4 shows measurement error not to be of concern in the estimates.

14

in addition to an interview with Prof. Nick Vink who was present during official discussions relating to the agricultural minimum wage. Prior to implementing SD8, the DoL commissioned an investigation into the state of affairs in the agricultural labor market of South Africa, the findings of which is contained in DoL (2001). The report concluded in favor of a minimum wage for the agricultural sector, citing its usefulness as a tool for poverty alleviation among farmworkers. As a result of these findings, the DoL implemented SD8. This implementation can be seen as being exogenous in nature since the impact of the legislation on employment carried little weight in the policy decision. In evidence of this, the report stated that "...minimum wage cannot be opposed purely on grounds of its adverse effect on employment." (DoL (2001)). This underlying theme of social equity concerns being the principle motivation behind the implementation of the legislation provides the basis for my assumption of exogeneity regarding the variation of minimum wages as per SD8. The convergence period described by SD13 was also the result of research commissioned by the DoL, this time completed by the Employment Conditions Commission (ECC). In their 2004 report on the implementation of SD8 and the industry perceptions thereof, the ECC found that the administrative incapacity of market participants prevented an efficient implementation of the two-tier wage system. For example, it was administratively difficult for the DoL to effectively monitor farms which stretched over both area A and B municipalities (ECC (2004)). Following these findings, SD13 was implemented in 2006 which saw the two-tier wage system converge to a single national minimum wage for the agricultural sector by 2008. This convergence can also be seen as exogenous in nature due to the fact that it was the result of administrative incapacities as opposed to a convergence between economic conditions prevailing in area A and B municipalities (Vink (2013)). Thus we have established the exogeneity of SD8 and SD13, and as a result, any remaining omitted variables from equation 2 should be characterized by both time invariant values and time invariant effects. Therefore the regressions should comply with the fixed effects assumptions and the resulting minimum wage coefficients should be unbiased. With this in mind, the next section aims toward formulating a hypothesis as to the likely effect the agricultural minimum wage had on farmworker employment by considering evidence from the descriptive statistics.

3.3

Descriptive Statistics

Table A.1 in the appendix includes a few key descriptive statistics regarding the dataset. The discrete jump in the number of observations when moving from the OHS to the LFS surveys illustrates the difference in design and scope between these surveys; however one can still infer some of the basic conditions that prevailed in the agricultural labour market leading up to, and after, the implementation of SD8 and SD13. It is evident that throughout the sample period, most farmworkers were non-unionised black males with low levels of education. Table A.1 and the mode of figure A.2 also show that farmworker employment is concentrated among the bottom end of the age distribution; however the a sizable right tail in figure A.2 reveals a significant 15

presence amongst the older age profiles. Additionally, females account for roughly a third of the agricultural labour force and that ratio has remained very stable throughout 1997-2007. When considering the impact of the minimum wage on wages present in the agricultural sector it is evident from table A.1 that the minimum wage was much more binding in area B DCs compared to area A DCs. From 2003 to 2007 the average farmworker wage in area A DCs comfortably exceeded the legislated minimum with the difference between actual wages and the minimum wage increasing over the period. On the other hand, area B average wages were consistently below the legislated minimum. Figure A.3 in the appendix provides a clearer picture by plotting the kernel densities of the logged hourly nominal farmworker wage for both area A and area B in 2003 and 2007 with reference lines to indicate the applicable minimum wage in each area. From figure A.3 it is apparent that over time the minimum wage has become less binding in both areas as the wage distribution shifts further away from the legislated minima in both areas from 2003 to 2007. Figure 3.1 below plots both the area A and area B minimum wage for farmworkers during the period March 2003 to September 2007 in both nominal and real terms. Figure 3.1 reveals that although farm labour costs increased significantly in nominal terms over the period in which SD8 and SD13 were implemented, the real cost of farm labour remained relatively stable. This is due to persistent moderate inflation being present in the SA economy with the South African Reserve Bank (SARB) mandated to keep inflation within a target band of 3-6% (SARB, 2013). It is also evident that the real cost of farm labour increased more in area B DCs than in area A DCs. This is due to the fact that area B minimum wages increased more rapidly than area A minima following the implementation of SD13. Figure 3.1: Minimum wages for farmworkers over the period 2003-2007

16

When one looks at farmworker employment as per table A.1, one can see a steady increase in the number of farmworkers leading up to the implementation of SD8. From 2002 to 2005 farmworker employment decreased by approximately 250 000; however this trend seems to have reversed after 2005 so that there was a net loss in farmworker employment of around 125 000 from 2002 to 2007. As stated above, there is evidence to support the notion that the agricultural minimum wage realised more job losses in area B DCs than in area A DCs. Following this intuition, figure 3.2 below plots the number of farmworkers in different area weighted DCs from 2000-2007. Figure 3.2: Number of farmworkers in different area weighted DCs from 2002-2007

From figure 3.2 it is apparent that relatively richer farming areas (high area weight value) started experiencing declines in their labour force approximately 1-2 years earlier than the poorer farming areas. Additionally, the richer DCs experienced more moderate changes to the number of farmworkers present in their area compared to the poorer DCs. This disparity could partly be attributed to the modernisation of the South African economy where primary sector workers shift toward employment in the tertiary sector (Bureau for Food and Agricultural Policy (2012)). Such a shift could take place more easily in urban areas where the demand for tertiary labour is higher, explaining the relatively earlier decline of farmworker employment in high area weight DCs. Regardless of structural labour demand changes taking place in the SA economy, it is interesting to note that all the series depicted in figure 3.2 exhibit a steeper downward slope in the period 2003 - 2005 compared to previous periods. The fact that this occurred simultaneously with the implementation of SD8 as well as one of the longest business cycle upswings in SA history provides a tentative indication that the institutional constraints (SD8 and SD13) imposed on the agricultural labour market may have led farmworker employment to diverge from employment trends present in the broader economy so that the minimum wage realised a net employment loss in agriculture (Venter (2009)). This effect is much more pronounced in the series depicting the poorer DCs, and so provides further evidence toward the employment effect of the minimum wage being stronger in the poorer DCs. 17

Reducing the weekly working hours of farmworkers is another technique employers can make use of in order to ease the financial burden of the minimum wage; however from table A.1 it seems that farmworker hours have remained relatively stable over the sample period at around 50 hours per week. Figure A.3 in the appendix plots the kernel density function of the average weekly working hours in the agricultural sector and illustrates the stability of farm working hours during the sample period since the distributions do not shift significantly over time. Although adjustments to the hours worked does not seem to have taken place as a result of the minimum wage, it is plausible that employers could have adjusted the composition of their workforce due to the changing opportunity cost of labour resulting from the minimum wage. Additionally, employers could look toward capital investment not only to increase the productivity of their existing labour force but also to complement the shift toward skill intensive production. Figure A.1 in the appendix provides some indication of this, showing a strong upward trend in tractor sales for the period 2000 - 2007. A shift toward skill intensive production is less evident in skilled commercial farmworker numbers as evidenced by the clear downward trend depicted by the second series in figure A.1; however there is a tentative upward reversal around 2006 which could be indicative of a time lag in adjustment. Therefore, from the evidence described above it seems that agricultural employers adjusted the size of their unskilled labour force in response to the minimum wage legislation. Furthermore, the descriptive statistics point toward this unemployment effect being especially pronounced in poorer DCs. This potential disparity in outcomes between DCs could be the result of moderate inflation in the SA economy which saw area A real minimum wages remain relatively constant whereas area B real minimum wages displayed a clear upward trend. Another potential factor could be the relatively non-binding nature of the minimum wage in area A DCs compared to area B DCs. There is also tentative evidence that the minimum wage led to more capital intensive production means; however the impact on the skilled commercial farm labour force is less clear.

4

Empirical Analysis

All of the regressions discussed in this section were estimated with OLS and are weighted according to the person weights assigned in the OHS for the period 1997-1999 and the LFS for the period 2000-2007.

4.1

The effect of the minimum wage on formal farmworker employment

Table 4.1 below summarises the results from various regressions on the log number of farmworkers per DC during the period 1997 to 2007. The full regression output can be found in table A.2 in the appendix.

18

Table 4.1: Regressions on the Log Number of Unskilled Commercial Farmworkers

Log Minimum Wage

OLS1

OLS2

FE1

FE2

FE3

FE4

0.051**

0.043*

-0.12

-1.067**

-0.576

-0.599

1.99

1.65

-0.64

-2.03

-0.91

-0.95

Log Minimum Wage (-1)

-0.574

-0.595

-0.74

-0.77

Log Minimum Wage (-2)

-0.187**

-0.173**

-2.36

-2.37

Log Minimum Wage (-3)

0.022 0.50

Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

No No 0.051** 942 0.004 3207

No No 0.043* 890 0.226 2712

Yes No -0.12 890 0.358 1390

Yes Yes -1.067** 890 0.402 1348

Yes Yes -1.315** 888 0.403 1348

Yes Yes -1.367** 890 0.404 1349

Source: Own calculations using OHS and LFS datasets. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

OLS1 is just a simple estimate of the relationship between the log real minimum wage and the log number of unskilled commercial farmworkers in the different DCs. The minimum wage coefficient therefore reveals the partial correlation between the number of farmworkers and the minimum wage. Given the information contained in figure 3.2 above, there seems to be a large degree of omitted variable bias in the minimum wage coefficient as it is positive and significant. In OLS2 we attempt to remove some of this bias through the inclusion of demographic controls for gender, education, experience, unionisation, the relative farm wage, as well as the population present in each DC. One can see that these controls do little to remove the bias from the minimum wage coefficient; although it does improve the model’s fit considerably as shown by the R-squared statistic. FE1 attempts to improve the estimates by controlling for unobservable but time invariant heterogeneity through the inclusion of DC dummies. As indicated by the change in the minimum wage coefficient, this removes an important source of omitted variable bias. The inclusion of these dummies allows the researcher to control for DC specific characteristics and conditions that influence labour market outcomes within DCs. Thus, the inclusion of DC dummies controls for district specific variation in labour demand, agricultural productivity, skill composition, and institutional and cultural differences. The estimates from FE1 now depict the negative associated relationship between the minimum wage and farmworker employment we expected to find. Additionally, there has been a significant reduction in the Akaike information criterion (AIC) from OLS2 to FE1 as well as a much improved model explanatory power. This reveals that the model has become more parsimonious when DC fixed effects are controlled for. Upon the inclusion of time fixed effects in FE2 the AIC drops further and the minimum wage effect becomes much more pronounced and significant. The minimum wage coefficient is now significant at a 5% level and the minimum wage - employment 19

elasticity for farmworkers is estimated as -1.067. The model specification as per FE2 includes all of the necessary control variables indicated by the minimum wage literature; however this literature also provides evidence of minimum wages impacting labour market conditions in the medium to long run as opposed to having a contemporaneous effect. Following this intuition FE3 includes lags of up to 3 years of the real minimum wage7 . From FE3 it is evident that the minimum wage effect is strongest during the first year post-implementation; however only the second year lag is significant. The inclusion of a more dynamic minimum wage effect has also lead to a stronger minimum wage elasticity: -1.315 in FE3 compared to -1.067 in FE2. Upon the removal of the positive and insignificant third minimum wage lag in FE4 there is a loss in parsimony as indicated by the marginally higher AIC. When looking at the total minimum wage effect, we can see that this omission marginally increases the estimated minimum wage elasticity to -1.367 and does not affect the significance of the second period lag. Regardless, the preferred specification remains FE3 as it does not restrict the three year lag to exhibit a zero coefficient in addition to FE3 being more parsimonious than FE4. The estimated elasticities in table 4.1 of around -1.3 seems high by international standards, so it is worth calculating what kind of employment losses is implied by our estimates and comparing that to what have been found in other studies. During 2003-2007 approximately 26% of unskilled farmworkers resided in area A DCs and 42% resided in area B DCs. This period also saw area A and B real minimum wages increase by about 5.2% and 23% respectively. With a minimum wage - farmworker employment elasticity of -1.3, this translates into a decline in the number of unskilled farmworkers of approximately 16.6%. These results are in line with the findings of Bhorat et al. (2012) where they estimated a decline of approximately 15% in the probability of being employed as a farmworker due to the minimum wage. In order to assess whether the estimates suffer from bias due to measurement error in the minimum wage, the regressions from table 4.1 will be re-estimated on a sample of DCs containing only area A or area B municipalities. Recall that the minimum wage is measured imprecisely for DCs that contain a mixture of both area A and area B DCs. However, there is no measurement error present in the minimum wage variable for DCs that contain either only area A or only area B municipalities. These “pure” DCs provide the means to assess the degree of measurement error present in DCs which contain a mixture of area A and B municipalities. If similar estimates are obtained to those found in table 4.1, the degree of measurement error will be small and thus prove the estimates to be robust. Table A.3 in the appendix provides these results. The degree of sample retention in table A.3 is positive indicating that 77% of the observations included in table 4.1 were area A or area B only DCs, thus reducing the risk of bias arising from measurement errors in the minimum wage. Furthermore, table A.3 shows the control variables to be consistent to those of table A.2. With regards to the minimum wage effect, the significance of the coefficients has improved significantly, as shown by FE3 where both the contemporaneous 7

Lags of more than 3 years were excluded in order to preserve the sample size. The quadratic experience term was dropped from the specification to preserve parsimony given the minute size of its coefficient as well as the statistical insignificance thereof.

20

and two year lag are significant at a 5% level. However, the estimated minimum wage elasticity has dropped from -1.315 to -1.143 from table 4.1 to table A.3. This provides evidence against the presence of attenuation bias since the estimated elasticities from table 4.1 should be lower than those of table A.3 if attenuation bias was present. Therefore the results from table A.3 provide testament to the robustness of the fixed effects estimator for this analysis. Seeing as the minimum wage legislation was implemented during one of South Africa’s longest business cycle upswings, there could be some endogeneity concerns with regards to the period during which SD8 was implemented. In order to address this possible confounding factor, table 4.2 below replicates the regressions of table 4.1 but on a reduced sample size consisting of the 2003-2007 period only. This removes the possibility of confounding factors relating to the date of SD8’s implementation as we now focus only on the convergence period. As the results from table 4.2 show, the negative and significant result remains. Table 4.2: Regressions on the Log Number of Unskilled Commercial Farmworkers (2003-2007)

Log Minimum Wage

FE1

FE2

FE3

FE4

FE5

-0.783

-0.413

-0.298

-0.250

-0.106

-1.46

-0.49

Log Minimum Wage (-1) Log Minimum Wage (-2) Log Minimum Wage (-3)

-0.32

-0.28

-0.12

-1.131

-1.153

-1.378**

-1.60

-1.65

-2.11

-0.094

-0.119

-1.16

-1.58

-0.040 -0.89

Time Fixed Effects Long Run Minimum Wage Effect N R2 AIC

No -0.783 487 0.245 619.745

Yes -0.413 487 0.261 617.237

Yes -1.563*** 487 0.269 617.765

Yes -1.522*** 487 0.269 615.876

Yes -1.484** 487 0.268 614.506

Source: Own calculations using LFS datasets. t-statistics reported below coefficients. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

A final point of concern regarding the unskilled farmworker results relates to the large size of the estimated long run minimum wage-employment elasticity compared to the existing literature. It is unlikely that the minimum wage had a direct impact on farmworker employment, the channel of influence would rather be via its effect on wages paid to farmworkers in addition to its impact on the opportunity cost of farm labor. To embody this line of thought, I estimate a 2SLS on the log number of unskilled commercial farmworkers over the entire sample period, instrument log farm wages with the minimum wage. The results of this estimation are contained in table 4.3 below.

21

Table 4.3: 2SLS Number of Unskilled Commercial Farmworkers Instrumented: Instruments:

Log Unskilled Real Farm Wage Xi,t st 1 Stage Regression Regressors Coef. Log Minimum Wage 0.0127 Log Minimum Wage (-1) 0.0441*** Log Minimum Wage (-2) 0.0234* Log Minimum Wage (-3) 0.0323** Controls as per Xi,t Yes Instrument Relevance? F(4, 52) = 15.34 N 888 2 R 0.2494 2nd Stage Regression Regressor Coef. Log Unskilled Real Farm Wage -0.5848*** Controls as per Xi,t Yes N 888 R2 0.2631 Source: Own calculations using OHS and LFS datasets. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively.

The results of table 4.3 show firstly that the minimum wage had a significant and positive impact on the wages paid to farmworkers; and secondly that the minimum had a significant and negative influence on unskilled farm employment as a result thereof. Furthermore, the estimated elasticity of -0.5848 falls in line with previous estimates obtained in the minimum wage literature. Combined with the elasticity of -1.3 obtained in table 4.1, this indicates that approximately half of the impact of the minimum wage can directly be attributed to its direct influence on the cost of unskilled farm labor. In turn, this points toward changes in the opportunity cost of farm labor as the other channel through minimum wages had an influence of unskilled farm employment. In lieu of this, the next set of regressions concerns the impact of the minimum wage on the employment of commercial farmworkers that are not unskilled as shown by table 4.4 below. The results reveal a positive employment effect from the agricultural minimum wage for farmworkers that are not unskilled (henceforth referred to as skilled farmworkers). The estimates also indicate this effect to be statistically significant at a 1% level with the total minimum wage - employment elasticity estimated at approximately 0.5. This reveals that skilled farmworker numbers rose by approximately 6.5% as a result of the implementation of the agricultural minimum wage8 . It is evident from the results thus far that the minimum wage had a significant impact on skill intesification within the agricultural sector. In order to investigate whether this adjustment may have been complemented via capital accumulation by farm owners, table 4.5 below displays the 8

The author did apply the 2SLS estimator to skilled farmworkers as in table 4.3; however, no significant result was found. This was attributed to the minimum wage effect being very diluted since it first influenced unskilled wages which then had an impact on relative farm wages, and subsequently, on skilled farm employment.

22

results from fixed effects regressions run on the log number of tractor sales within each DC910 . Table 4.4: Regressions on the Log number of all other Commercial Farmworkers

Log Minimum Wage

FE1

FE2

FE3

FE4

-0.042*

0.334

0.379

0.37

-1.85

0.61

0.56

0.55

Log Minimum Wage (-1)

-0.131

-0.144

-0.13

-0.14

Log Minimum Wage (-2)

0.223**

0.243**

2.48

2.85

Log Minimum Wage (-3)

0.031 0.62

Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

Yes No -0.042* 890 0.286 1516

Yes Yes 0.334 890 0.301 1517

Yes Yes 0.502*** 888 0.302 1519

Yes Yes 0.469*** 890 0.302 1519

Source: Own calculations using OHS and LFS datasets. t-statistics reported below coefficients. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

Table 4.5: Regressions on the Log Number of Tractor Sales

Log Minimum Wage

FE1

FE2

FE3

FE4

-0.019

1.203

1.177

1.183

0.023

0.992

1.247

1.225

-0.078

-0.08

1.055

1.047

0.335**

0.263*

0.149

0.149

Log Minimum Wage (-1) Log Minimum Wage (-2) Log Minimum Wage (-3)

-0.103* 0.0443

Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

Yes No -0.019 813 0.011 2000

Yes Yes 1.203 813 0.242 1805

Yes Yes 1.331** 811 0.245 1802

Yes Yes 1.376 813 0.243 1807

Source: Own calculations using OHS and LFS datasets. Standard errors clustered at a DC level and reported below coefficients. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. 9

The total number of commercial farmworkers was included among the explanatory variables for these regressions to capture the fact that workers are needed to operate the machinery. 10 The full regression output can be found in the Appendix under table A.5.

23

Table 4.5 reveals a positive and significant minimum wage effect on tractor sales with the elasticity estimated at 1.54. Note that in table 4.5 standard errors are reported below the coefficients as opposed to the t-statistics. As a result of the large magnitude of these standard errors we are more concerned with the significant positive result as opposed to the point estimate, taking it as evidence of capital intensification resulting from the minimum wage. Thus, not only did skilled employees become better represented in the agricultural sector, but this was accompanied by increased capital investment by farmers. The impact of the minimum wage on tractor sales is especially interesting since more tractors should increase the productivity of workers across the skill spectrum, mitigating the higher cost of the remaining unskilled portion of the farm workforce.

5

Conclusion

In summary, the empirical analysis has shown that the agricultural minimum wage was accompanied by job losses amongst the formal unskilled of around 16% of which 7.5% can directly be attributable to the impact of the legislation on increasing unskilled farm wages. The change in opportunity costs resulting from the implementation of the wage floor also saw formal employers shift toward skill intensive methods of production, realising a 6% rise in skilled farmworker employment. The minimum wage was also found to have led to capital intensification as shown by the positive and significant relationship between the legislated minima and tractor sales. Formal sector employers therefore made use of workforce compositional changes as well as capital investments in order to increase productivity and in doing so, mitigate the costs of higher wages for their remaining unskilled labour. Given that the South African economy is characterised by a largely idle labour force, labour intensive production methods should be incentivized where plausible. Improving the degree of skill improvement and attainment amongst the unskilled labour force therefore appears to be of principle concern and provides scope for further government intervention. In this sense, the implementation of policies that aim to bridge the divide between agricultural labour demand and the existing stock of unskilled farm labour could prove to be a fertile policy landscape.

24

References Allison, P. D. (2009). Fixed Effects Regression Models - Quantitative Applications in the Social Sciences (1 ed.). SAGE Publications Inc. Almeida, R. K. and K. Terrell (2008). Minimum wages in developing countries: Helping or hurting workers? World Bank Working Papers. [Online]. Available at: http://documents.worldbank.org/curated/en/2008/12/10158898/ minimum-wages-developing-countries-helping-or-hurting-workers. Arango, C. A. and A. Pachón (2007). The minimum wage in colombia1984-2001: Favoring the middle class with a bite on the poor. Ensayos sobre política económica. Baker, M., D. Benjamin, and S. Stanger (1999). The highs and lows of the minimum wage effect: A time-series cross-section study of the Canadian law. Journal of Labor Economics 17 (2), 318–350. Bell, L. A. (1997). The Impact of Minimum Wages in Mexico and Colombia. Journal of Labor Economics 15 (3), 102–35. Bhorat, H. (2000). Are minimum wage adjustments an effective mechanism for poverty alleviation? some simulations for domestic and farm workers. In Trade and International Policy Secretariat (TIPS) 2000 Annual Forum, Development Policy Research Unit (DPRU). Bhorat, H., R. Kanbur, and N. Mayet (2013). The impact of sectoral minimum wage laws on employment, wages, and hours of work in South Africa. IZA Journal of Labor & Development 2 (1), 1–27.

Bhorat, H., R. Kanbur, and B. Stanwix (2012). Estimating the impact of minimum wages on employment, wages and non-wage benefits: the case of agriculture in South Africa. DPRU Working Papers No 12149. [Online]. Available at: http://www.dpru.uct.ac.za/ wp-12149-estimating-impact-minimum-wages-employment-wages-and-non-wage-benefits-case-a Boeri, T., B. Helppie, and M. Macis (2008). Labor regulations in developing countries: a review of the evidence and directions for future research. World Bank Social Protection Discussion Paper No 833. [Online]. Available at: https://ideas.repec.org/p/wbk/hdnspu/46306.html. Borjas, G. (2010). Labour economics Fifth Edition. New York: McGraw-Hill/Irvin. Brown, C., R. B. Freeman, D. S. Hamermesh, P. Osterman, and F. R. Welch (1995). Comments by reviewers. Industrial and Labor Relations Review 48 (4), pp. 828–849. Brown, C., C. Kilroy, and A. Kohen (1982). The effect of the minimum wage on employment and unemployment. Journal of Economic Literature 20 (2), 487–528. Bureau for Food and Agricultural Policy (2012). Farm sectoral determination: an analysis of agricultural wages in South Africa. [Online]. Available at: http://www.bfap.co.za/ documents/research%20reports/BFAP%20farm%20sector%20determination%20report% 20draft%2017%20Dec.PDF. 25

Burkhauser, R. V., K. A. Couch, and D. C. Wittenburg (2000). A reassessment of the new economics of the minimum wage literature with monthly data from the current population survey. Journal of Labor Economics 18 (4), 653–680. Card, D. (1992). Using regional variation in wages to measure the effects of the federal minimum wage. NBER Working Papers No 4058. [Online]. Available at: http://www.nber.org/papers/ w4058. Card, D. and A. B. Krueger (1993). Minimum wages and employment: A case study of the fast food industry in New Jersey and Pennsylvania. NBER Working Papers No 4059. [Online]. Available at: http://www.nber.org/papers/w4059. Card, D. E. and A. B. Krueger (1995). Myth and Measurement: The New Economics of the Minimum Wage. Princeton University Press. Conradie, B. (2005). Wages and wage elasticities for wine and table grapes in South Africa. Agrekon 44 (1), 138–156. Dinkelman, T. and V. Ranchhod (2011). Evidence on the impact of minimum wage laws in an informal sector: Domestic workers in South Africa. CEPR Discussion Papers No 8682. [Online]. Available at: http://ideas.repec.org/p/cpr/ceprdp/8682.html. DoL (2001). Determination of employment conditions in south african agriculture. Government Gazette No. 22648. ECC (2004). Investigation into the farm worker sector with a view to amend the sectoral determination 8. Government Gazette No. R.821. [Online]. Available at: http://www.labour.gov. za/DOL/downloads/documents. Fajnzylber, P. (2001). Minimum wage effects throughout the wage distribution: evidence from Brazil’s formal and informal sectors. Anais do XXIX Encontro Nacional de Economia (098). Gasparini, L. and L. Tornarolli (2009). Labor Informality in Latin America and the Caribbean: Patterns and Trends from Household Survey Microdata. Desarrollo y Sociedad 14 (63), 13–80. Gindling, T. H. and K. Terrell (2007). The effects of multiple minimum wages throughout the labor market: the case of Costa Rica. Labour Economics 14 (3), 485–511. Ham, A. (2013). Revisiting the effects of minimum wages in developing countries: Evidence from a particular policy change in Honduras. SSRN Working Papers No 2217835. [Online] Available at: http://dx.doi.org/10.2139/ssrn.2217835. Harris, J. R. and M. P. Todaro (1970). Migration, unemployment and development: a two-sector analysis. The American Economic Review 60, 126–142. Hertz, T. (2005). The effect of minimum wages on the employment and earnings of South Africa’s domestic service workers. DPRU Working Paper No. 05099. [Online]. Available at: https: //ideas.repec.org/p/ctw/wpaper/05099.html#biblio.

26

Katz, L. F. and A. B. Krueger (1992). The effect of the minimum wage on the fast food industry. NBER Working Papers No 3997. [Online]. Available at: http://www.nber.org/papers/w3997. Keil, M., D. Robertson, and J. Symons (2001). Minimum Wages and Employment. Centre for Economic Performance, London School of Economics and Political Science. Kim, T. and L. J. Taylor (1995). The Employment Effect in Retail Trade of California’s 1988 Minimum Wage Increase. Journal of Business Economic Statistics 13 (2), pp. 175–182. Lester, R. A. (1947). Marginalism, minimum wages, and labor markets. The American Economic Review 37, 135–148. Maloney, W. and J. Mendez (2004). Measuring the impact of minimum wages. evidence from Latin America. In Law and Employment: Lessons from Latin America and the Caribbean, pp. 109–130. University of Chicago Press. Manning, A. (2006). A Generalised Model of Monopsony. The Economic Journal 116 (508), 84–100. Murray, J. and C. Van Walbeek (2007). Impact of the Sectoral Determination for farm workers on the South African sugar industry: Case study of the KwaZulu-Natal North and South Coasts. Agrekon 46 (1), 94–112. Neumark, D., J. Salas, and W. Wascher (2013). Revisiting the minimum wage-employment debate: Throwing out the baby with the bathwater? NBER Working Papers No 18681. [Online]. Available at: http://www.nber.org/papers/w18681. Neumark, D. and W. Wascher (1991). Evidence on employment effects of minimum wages and subminimum wage provisions from panel data on state minimum wage laws. NBER Working Papers No 3859. [Online]. Available at: http://www.nber.org/papers/w3859. Neumark, D. and W. Wascher (2000). Minimum wages and employment: A case study of the fast-food industry in new jersey and pennsylvania: Comment. American Economic Review 90, 1362–1396. Neumark, D. and W. Wascher (2007). Minimum wages and employment. Foundations and Trends in Microeconomics 3 (1-2), 1–182. Obenauer, M. L. and B. M. von der Nienburg (1915). Effect of Minimum-wage Determinations in Oregon: July, 1915. Number 6. US Government Printing Office. Rocheteau, G. and M. Tasci (2007). The minimum wage and the labor market. Federal Reserve of Cleveland Economic Commentary. [Online]. Available at: http://ideas.repec.org/a/fip/ fedcec/y2007imay1.html. SABC (2013). Farmworkers’ Strike Taking Its Toll on Economy. [Online]. Available at: http://www.sabc.co.za/news/a/b92ae9004e3aad73914fb7f251b4e4e2/ Farmworkers-strike-taking-its-toll-on-economy-20131801.

27

SARB (2013). Inflation targeting framework. [Online]. Available at: http://www.resbank.co. za/MonetaryPolicy/DecisionMaking/Pages/default.aspx. StatsSA (2008). Gross domestic product: Annual estimates per sector 2000-2007. [Online]. Available at: http://www.statssa.co.za. Stigler, G. J. (1946). The economics of minimum wage legislation. The American Economic Review 36, 358–365. Venter, J. (2009). Business cycles in South Africa during the period 1999 to 2007. South African Reserve Bank Quarterly Bulletin, 61–69. Vink, N. (2013, Oct. 1). Personal interview.

28

A

Appendix

Figure A.1: The Number of Tractor Sales and Commercial Farmworkers that are not Unskilled

Figure A.2: Kernel Density of the Age Distribution of Farmworkers

29

Figure A.3: Kernel Densities of Logged Hourly Wages by Area

30

Figure A.4: Kernel Density of Hours Worked by Year

31

32

0.02

Farmworkers as proportion of total sample

0.26 0.39 0.4 -

Proportion of total farmworkers in area A

Proportion of population in area B

Proportion of total farmworkers in area B

Area A nom min wage

51.4

-

3.76

-

0.46

0.45

0.23

0.42

0.08

0.39

0.64

5.11

34.22

0.26

952,454

0.02

2,164

89,325

2000

Ave. hours worked per week by farmworkers 51 52.53 52.59 Source: Author’s calculations using data from the OHS and LFS surveys.

3.82

-

4.57

-

0.48

0.41

0.19

0.46

0.14

0.4

0.68

4.97

33.94

0.23

470,973

0.03

1,698

65,995

1999

3.76

3.9

Ave. farmworker wage in area B

-

3.3

-

0.53

0.42

0.16

0.46

0.1

0.36

0.7

4.64

34.33

0.25

423,600

0.02

959

49,560

1998

3.61

-

Area B nom min wage

2.61

0.48

Proportion of population in area A

Ave. farmworker wage in area A

0.11

0.55

Proportion of farmworkers that are Black

Proportion of farmworkers that are unionised

4.98

Ave. farmworker education

0.35

34.23

Ave. farmworker age

Proportion of farmworkers that are female

0.27

Narrow unemployment rate (weighted)

275,935

1,257

Number of farmworkers

Number of farmworkers (weighted)

82,613

Number of observations

1997

51.96

2.57

-

4.14

-

0.49

0.44

0.22

0.43

0.07

0.34

0.65

5.13

35.75

0.28

1,037,227

0.03

3,546

134,420

2001

51.46

2.31

-

4.3

-

0.46

0.44

0.24

0.43

0.07

0.35

0.65

5.15

34.67

0.3

1,147,954

0.03

3,755

133,522

2002

49.96

2.95

3.33

4.73

4.1

0.48

0.43

0.22

0.45

0.07

0.35

0.66

5.45

34.66

0.3

1,122,443

0.03

3,551

126,694

2003

Table A.1: Dataset characteristics for the period 1997-2007

50

3.25

3.66

4.82

4.47

0.52

0.41

0.22

0.47

0.06

0.36

0.69

5.49

35

0.27

1,036,135

0.03

3,626

131,129

2004

50.46

3.83

4.03

5.85

4.87

0.52

0.4

0.24

0.49

0.08

0.35

0.71

5.98

34.8

0.27

791,085

0.03

3,686

137,370

2005

49.09

4.32

4.54

6.62

5.1

0.5

0.4

0.27

0.5

0.08

0.38

0.7

6.12

34.74

0.26

800,008

0.03

3,768

135,253

2006

49.28

4.9

5.07

6.85

5.34

0.53

0.39

0.25

0.51

0.09

0.36

0.73

6.22

35.15

0.24

827,336

0.03

3,780

134,564

2007

Table A.2: Regressions on the Log Number of Unskilled Commercial Farmworkers

Log Minimum Wage

OLS1

OLS2

FE1

FE2

FE3

FE4

0.051**

0.043*

-0.12

-1.067**

-0.576

-0.599

1.99

1.65

-0.64

-2.03

Log Minimum Wage (-1) Log Minimum Wage (-2) Log Minimum Wage (-3)

-0.91

-0.95

-0.574

-0.595

-0.74

-0.77

-0.187**

-0.173**

-2.36

-2.37

0.022 0.50

Log DC Population

0.588***

Experience 2

Experience Union

Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

0.956***

0.941***

8.15

11.93

11.86

10.76

12.05

-2.447

-1.161

-1.212

-1.19

-0.41

-1.63

-0.83

-0.86

-0.84

-0.268***

-0.015

-0.092

-0.092

-0.101

-3.86

-0.2

-1.11

-1.1

-1.21

0.5***

-0.084

0.026

0.035

0.031

4.65

-0.81

0.24

0.32

0.28

-0.004

0.003

0 -0.981**

-0.971**

-1.43

1.31

0.16

-1.657**

-1.425**

-0.956**

-3.2

-3.33

-2.17

-2.19

-2.2

-0.059

-0.032

-0.027

-0.028

-0.028

-1.35

-1.19

-1.05

-1.08

-1.08

2.727***

-5.731***

-1.895

-5.874**

-6.63**

-6.423**

61.77

-3.79

-1.09

-2.74

-2.75

-2.7

No No 0.051 942 0.004 3207

No No 0.043 890 0.226 2712

Yes No -0.12 890 0.358 1390

Yes Yes -1.067 890 0.402 1348

Yes Yes -1.315 888 0.403 1348

Yes Yes -1.367 890 0.404 1349

Log Relative Farm Wage Constant

0.949***

-0.63

Female Education

0.899***

Source: Own calculations using OHS and LFS datasets. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

33

Table A.3: Regressions on the Log Number of Unskilled Farmworkers in Area A and B only DCs

Log Minimum Wage

OLS1

OLS2

FE1

FE2

FE3

FE4

0.07**

0.054*

0.005

-1.122*

-1.729**

-1.656**

2.39

1.78

0.23

-1.84

Log Minimum Wage (-1) Log Minimum Wage (-2) Log Minimum Wage (-3)

-2.13

-2.15

1.077

0.938

1.13

0.94

-0.538**

-0.54**

-3.48

-3.01

0.047 1.04

Log DC Population

0.53***

Female Education Experience 2

Experience

0.89***

0.94***

0.957***

0.936***

6.69

9.81

9.71

9.18

10.46

-2.627

-1.995

-0.468

-0.606

-0.584

-1.51

-1.22

-0.32

-0.41

-0.46

-0.299***

-0.024

-0.153

-0.131

-0.07

-4.1

-0.27

-1.67

-1.47

-0.63

0.26***

-0.058

0.101

0.13

0.131

2.05

-0.43

0.76

0.99

0.91

0.001

0.003

-0.001 -1.306**

-1.331**

0.38

0.9

-0.38

Union

-1.363**

-1.753**

-1.163**

-2.36

-3.34

-2.17

-2.41

-2.44

Log Relative Farm Wage

-0.113**

-0.048

-0.04

-0.038

-0.04

-2.38

-1.68

-1.4

-1.35

-1.46

2.727***

-2.688

-2.43

-6.661**

-7.212**

-7.424**

61.77

-1.55

-1.14

-2.73

-2.85

-2.43

No No 0.07 737 0.008 3207

No No 0.054 685 0.235 2099

Yes No 0.005 685 0.36 1152

Yes Yes -1.122 685 0.406 1121

Yes Yes -1.143 685 0.424 1106

Yes Yes -1.258 685 0.433 1099

Constant Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

Source: Own calculations using OHS and LFS datasets. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

34

Table A.4: Regressions on the Log number of All Commercial Farmworkers that are Not Unskilled FE1

FE2

FE3

FE4

-0.042*

0.334

0.379

0.37

-1.85

0.61

0.56

0.55

-0.131

-0.144

-0.13

-0.14

Log Minimum Wage (-2)

0.223**

0.243**

2.48

2.85

Log Minimum Wage (-3)

0.031

Log Minimum Wage Log Minimum Wage (-1)

0.62

Log DC Population

0.871***

0.902***

0.923***

0.903***

9.92

8.84

7.92

9.31

-1.54

-1.875

-1.853

-1.839

-1.08

-1.34

-1.33

-1.32

-0.138**

-0.026

-0.018

-0.023

-2.17

-0.28

-0.2

-0.26

Experience

0.245**

0.197

0.203*

0.196

2.03

1.6

1.69

1.59

Experience2

-0.004*

-0.003

-1.78

1.01

-1.345**

-1.673**

-1.683**

-1.652**

-2.71

-3

-3.01

-2.96

-0.061**

-0.064**

-0.062**

-0.063**

-2.19

-2.28

-2.25

-2.24

-3.866**

-3.822

-3.88

-3.583

-2.91

-1.65

-1.41

-1.36

Yes No -0.042 890 0.286 1516

Yes Yes 0.334 890 0.301 1517

Yes Yes 0.502 888 0.302 1519

Yes Yes 0.469 890 0.302 1519

Female Education

Union Log Relative Farm Wage Constant Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

Source: Own calculations using OHS and LFS datasets. t-statistics reported below coefficients. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

35

Table A.5: Regressions on the Log Number of Tractor Sales FE1

FE2

FE3

FE4

-0.019

1.203

1.177

1.183

-0.84

1.21

0.94

0.97

-0.078

-0.08

-0.07

-0.08

Log Minimum Wage (-2)

0.335**

0.263*

2.24

1.77

Log Minimum Wage (-3)

-0.103**

Log Minimum Wage Log Minimum Wage (-1)

-2.33

Log DC Population Log Relative Farm Wage Constant Controls for DC fixed effects Controls for time fixed effects Long-run Minimum Wage Effect N R2 AIC

0.170**

-0.122

-0.177**

-0.12

2.13

-1.55

-2.05

-1.56

0.054

0.053

0.055

0.054

1.36

1.46

1.53

1.49

2.351***

8.07**

9.123***

8.433**

4.07

3.32

3.62

3.35

Yes No -0.019 813 0.011 2000

Yes Yes 1.203 813 0.242 1805

Yes Yes 1.331** 811 0.245 1802

Yes Yes 1.376 813 0.243 1807

Source: Own calculations using OHS and LFS datasets. t-statistics reported below coefficients. *, **, and *** denote statistical significance at 10%, 5%, and 1% respectively. Standard errors clustered at a DC level.

36