The impact of basic and social infrastructure investment on South African economic growth and development

ABSTRACT

Basic and social infrastructure investment can assist in addressing widespread inequality and divided societies by promoting economic growth and social development. The aim of this study is to determine whether basic and social infrastructure investment differently affect economic growth and social development indicators of urban and rural municipalities. We used a balanced panel dataset containing infrastructure, economic, demographic and social indicators for rural and urban municipalities for the period from 1996 to 2012. Principal component analysis was used to construct synthetic indices of basic and social infrastructure. Restricted within least squares dummy variable estimation techniques are used to evaluate the differences between urban and rural municipalities. The elasticities of basic and social infrastructure investment generally are more pronounced for economic growth and social development indicators in rural municipalities. These findings could potentially influence policy decisions in terms of infrastructure investment in favour of rural municipalities to increase economic growth and social development.

KEYWORDS:

• Basic and social infrastructure
• economic growth
• social development
• principal component analysis

JEL CLASSIFICATION:

• I130
• I250
• I380
• O110

1. Introduction

The main aim of this research article is to analyse the effect of basic and social infrastructure investment¹ on economic growth and social development and to compare the returns of these investments in urban and rural municipalities. Infrastructure delivery is of real concern in the everyday lives of the people in rural areas. As a result, this analysis can potentially assist in focusing limited resources on areas where the returns can be the most significant.

South Africa’s post-democratisation period was marked by a significant decentralisation of economic decision-making and service delivery, resulting in a system of local government that is constitutionally responsible for the economic and social development of their areas (Krugell & Naudé, 2005). During the past two decades, however, limited progress has been made, with widespread inequality and divided societies inherited from the previous government dispensation and spatial policies still being prevalent in the country (Booysen, 2003; Tregenna & Tsela, 2012; Adams et al., 2015).

To address the socio-economic challenges and inequalities in the country, the government of South Africa has implemented various programmes, the most recent being the National Development Plan. The National Development Plan recognises poor education outcomes, a divided community, uneven public service performance, divided spatial patterns and a crumbling infrastructure as some of the challenges that have to be addressed in order to overcome persistent poverty and inequality. Central to the aforementioned challenges are infrastructure delivery constraints that inhibit economic growth, social development and the reduction of poverty and inequality across the country (NPC, 2011:19). Given the different levels and concentration of inequality and poverty in rural and urban areas, it is likely that basic and social infrastructure investment could affect economic growth, the disposable income of households and social development in these regions differently, and this warrants in-depth analyses.

Research has found that insufficient infrastructure in informal settlements is a key obstacle to economic development (McRae, 2015:36). Furthermore, it has been shown that infrastructure investment and economic growth have a strong positive relationship (De la Fuente & Estache, 2004:5; Foster & Briceño-Garmendia, 2009:10), while the exact impact of infrastructure investment on social development remains inconclusive. In saying this, we deduce that sustained economic growth and social development are a necessary if not sufficient condition to reduce poverty and inequality. Consensus has therefore been reached that, under the right conditions, basic and social infrastructure investment does contribute to increased economic growth, social development and the reduction of inequality and poverty (Calderón & Servén, 2008:1). The collective impact of basic and social infrastructure investment on economic growth and social development in rural and urban municipalities, respectively, has remained largely understudied, mainly due to a lack of data availability and quality (Jerome & Ariyo, 2004:39; Bogetic & Fedderke, 2005:12). Furthermore, there is little empirical evidence of the direct impact that infrastructure investment has on income. Related studies have investigated the relationship between the demand for infrastructure investment and income (Komives et al., 2001), the relationship between infrastructure investment and savings (Estache et al., 2002), and the relationship between infrastructure investment and poverty alleviation (Brenneman & Kerf, 2002).

This article contributes to the existing literature by addressing the previously mentioned gaps in the literature in four ways: by measuring the effect of basic and social infrastructure investment on economic growth and social outcomes in urban and rural municipalities, respectively, and furthermore comparing these effects to determine whether the returns to urban and rural municipalities are similar – this, according to the authors’ knowledge, is the first article of its kind; the analysis is done at a sub-national (municipality) level, which is often a challenge due to data constraints; the study investigates the direct relationship between basic and social infrastructure investment and the disposable income of households in rural and urban municipalities, respectively, not analysed before; and the study uses panel data analysis not often used in these types of studies, which has an advantage over cross-sectional data in that it can address endogeneity issues.

The method followed is to compare the derived basic and social infrastructure investment elasticities of urban and rural municipalities with regard to various economic growth and social development indicators. To derive the elasticities, we make use of panel estimation techniques and a balanced panel dataset.

Being able to quantify the impact of basic and social infrastructure investment on economic growth and social development in urban and rural areas, respectively, can contribute to the development of policy to reduce overall and spatial inequality, and furthermore direct investment spending towards those spatial regions with prioritised needs (Calderón & Servén, 2004:26). The rationale for this argument is the indirect positive relationship between increased levels of economic growth and social development and the reduction of spatial inequality.

The rest of the article is set out as follows. In Section 2, literature on the effects of basic and social infrastructure investment on various socio-economic indicators is reviewed. In Section 3, the methodology and data used in the research article are discussed. Section 4 then reports the results, and in Section 5 the results are discussed and conclusions drawn.

2. Literature review

An increasing body of literature studies the social and economic impact of advances in physical infrastructure in developing countries (McRae, 2015). Chong et al. (2007:344) confirm that when a community has access to a comprehensive set of basic infrastructure services, the welfare effect is greater when compared with communities where certain components of infrastructure services are missing. Metwally et al. (2007:61) add that the basic infrastructure also lays the foundation for effective social infrastructure delivery, such as schools, hospitals and police stations. Social infrastructure in itself also has the ability to increase the economic growth and social development of a nation’s citizens and ensures that the basic infrastructure is better utilised (ESCAP, 2006:5). Economic growth and social development, in turn, can play an important role in addressing long-term growth challenges in South Africa, including double-digit unemployment and the poor quality of human capital (Simo-Kengne, 2016).

Understanding the channels through which basic and social infrastructure affect economic growth and social development is essential in order to optimise infrastructure investment efforts. The literature review presents the research conducted on the effect that basic and social infrastructure investment has on economic growth and social development, utilising various empirical studies, and it will be discussed according to the conceptual framework shown in Figure 1.

Figure 1. Conceptual framework for literature review. Source: Authors’ own construct.

Notes: GDP = gross domestic product; HH = household.

2.1. Interaction between basic and social infrastructure investment

The addition of basic and social infrastructure services not only has a direct economic growth and social development effect on a household, but also allows for the better utilisation of other infrastructure services (Chong et al., 2007:344). Electrification reduces indoor air pollution, and allows for safer food storage and cooking practices, which, in turn, increases health (Barnes et al., 2004:16). Electricity, water and sanitation connections also increase learners’ ability to attain an education by reducing incapacity due to illness. The benefits of this for human capital accumulation, economic growth and social development are obvious.

2.2. Interaction between basic and social infrastructure investment, economic growth and social development

Consensus has been reached that, under the right conditions, basic infrastructure investment contributes to reducing inequality and poverty via the channel of economic growth and social development (Calderón & Servén, 2008:1). There are various ways in which basic and social infrastructure have been found to affect economic growth and social development. For example, increasing electricity infrastructure has a strong impact on the productivity of a business by reducing the loss of output resulting from power outages and surges. Water and sanitation infrastructure has a lesser but still significant impact on the productivity of a business by protecting and even improving the health of the employees, thereby increasing their productivity. A number of studies have also found basic infrastructure to have a strong impact on the efficiency of education and health facilities (Brenneman & Kerf, 2002:5). This is important given the fact that urban–rural disparities regarding access to healthcare services have a persistent and more pronounced adverse effect on the poor (Booysen, 2003). Calderón & Servén (2008:16) add that basic and social infrastructure investment is also associated with reduced income inequality.

2.3. Basic and social infrastructure investment and its impact on disposable income

Estache (2004:5) confirms that little evidence even exists on the direct impact of infrastructure on household income, and cites only two other empirical studies in his research. The first is the work of Komives et al. (2001:20), who comment on how the demand for infrastructure changes as the income increases, as opposed to the mere impact of infrastructure on income The second is a study by Estache et al. (2002:90), which focuses on savings, rather than increases in income, resulting in higher disposable income levels. Brenneman & Kerf (2002:5) also comment on how basic and social infrastructure increases the disposable income of households as opposed to increasing household income itself. This study will therefore also investigate the impact of basic and social infrastructure investment on disposable income.

2.4. Basic and social infrastructure investment and its impact on poverty through increased economic growth and social development

More research attention has been directed towards the impact of basic infrastructure investment on poverty and inequality in recent years (Estache et al., 2002:15). De la Fuente & Estache (2004:2) note that basic and social infrastructure could reduce poverty, even though empirical literature has been noted to be far from conclusive on the exact impact that basic infrastructure investment has on poverty and inequality. Nevertheless, consensus has been reached that, under the right conditions, basic infrastructure investment does contribute towards alleviating inequality and poverty through higher levels of economic growth and social development (Calderón & Servén, 2008:1).

2.5. Basic and social infrastructure investment and its impact on education

Increasing the availability and quality of basic infrastructure services for the poor in developing countries has a significant and positive impact on the education of the poor and, therefore, potentially their income and welfare (Leipziger et al., 2003:7).

Basic infrastructure investment affects literacy through a number of channels. Brenneman & Kerf (2002:5) indicate that increased water and sanitation infrastructure improves education performance due to the reduction of water-related diseases, thereby also decreasing absenteeism in schools. Electricity infrastructure also increases literacy, due to lighting that enables students to study into the night in addition to making use of technology (Bond, 1999:47). Increasing water, sanitation and electricity infrastructure also reduces the time needed to collect wood for lighting, heating and cooking, which increases the available time to study in addition to increasing the likelihood of children attending school (Bond, 1999; Brenneman & Kerf, 2002). Attending school is of course a prerequisite for improved levels of human capital and consequently higher economic growth and poverty reduction.

2.6. Assessment of literature review

The results of the impact of basic and social infrastructure on economic growth and social development vary across studies due to the respective infrastructure indicators used, due to the methodologies employed and according to the country or group of countries on which the analyses focused.² However, the literature rarely comments on whether the basic and social infrastructure investment would impact differently on economic growth and social development in urban and rural areas, respectively. In some of the reviewed studies, the authors did comment that basic and social infrastructure could theoretically have a proportionately different effect on the rural poor as opposed to those from the urban areas (ADB, 2012:68). This forms the rationale for the research question for this research: what is the impact of basic and social infrastructure investment on economic growth and social development in urban and rural areas, respectively?

Furthermore, the impact of basic and social infrastructure investment on outcome variables such as social capital (the value added by investing in schools, hospitals and policing), the benefits of proximity of social infrastructure delivery and improved governance, implying the establishment of effective and efficient policy to address socio-economic challenges and spatial inequalities, and implementation and the monitoring of these policies, is rarely discussed in the literature. We will argue these matters in the concluding Section 5.

3. Research design and methodology

3.1. Data

The selected basic and social infrastructure, demographic, economic growth and social development indicators were sourced from the Information Handling Services Information and Insight Regional explorer databank, which contains infrastructure, economic, demographic and socio-economic data for each of the municipalities in South Africa from 1996 to 2012 (IHS, 2013).³ The respective municipality boundary sets are in accordance with the Demarcation Board revision used for the 2012 municipal elections. The urban/rural municipality classifications will be made according to information obtained from the National Department of Cooperative Governance and Traditional Affairs. For the purpose of this study, infrastructure and control variables will be assumed to be homogeneous due the legislation that governs infrastructure delivery and the respective surveys that calculate the variables being homogeneous.

The basic infrastructure index will be based on the number of households that have access to water, electricity and sanitation, while the social infrastructure index will use proxy variables for health, education and safety – due to the lack of direct measures on a municipal level – for each of the municipalities from 1996 to 2012. The number of households is used to normalise the synthetic index (Calderón & Servén, 2004; Romp & De Haan, 2007; Straub, 2010). Taking direction from a number of mentionable studies, such as Calderón (2009:9) and D’emurger (2001:97), real output per capita is used to determine economic output.

Household disposable income as opposed to household income will be used for the purposes of this empirical analysis. This will allow not only for capturing the direct cost saving stemming from the lower unit costs of receiving service, but also for the increased potential to earn higher incomes resulting from higher education, productivity and the increased availability of hours per day to actually work. Household disposable income (HHINC) is derived from total income for all households in a municipality, excluding taxes.

Research on the impact of infrastructure on poverty by Estache et al. (2000) and Jerome & Ariyo (2004:1) relied on standard $2/day and $1/day income poverty lines for their empirical analysis, respectively. This study, however, will employ an income poverty estimate as calculated by the Information Handling Services Regional eXplorer for the sake of consistency and the lack of availability of the dollar estimates at a municipal level. The percentage of people in poverty (PPOV) is defined as the number of people living in households who have a combined household income that is less than the respective household poverty income divided by the total population.

Jerome & Ariyo (2004:38) use variations of literacy (adult, male and female) when analysing the impact of infrastructure investment on education. We use a similar approximation of education in the form of functional literacy, which is similar to adult literacy. Functional literacy (PLIT) is defined as the literacy level of people older than 20 years who have completed their primary education (Grade 7). People who are functionally literate are assumed to have a level of reading and writing skills that enable them to manage daily life and employment (Korotayev et al., 2011).

3.2. Calculating the basic and social infrastructure indices

Principal component analysis is used to construct synthetic basic and social infrastructure indices. The respective infrastructure stock indices will provide an indication of the extent to which basic and social infrastructure is delivered in each of the municipalities in the country. While the method has been used in cross-country analysis, and in a few sub-national studies, it has not been deployed to analyse urban and rural differences on a sub-national level. The estimations of the synthetic basic infrastructure index are as follows:where BINF is the synthetic index of basic infrastructure, SAN is the number of households with hygienic toilets, WATER is the number of households with water connections above Reconstruction and Development Programme (RDP) level, ELEK is the number of households with electricity connections and HH is the number of households.

Each of the three basic infrastructure indicators carries approximately the same weight in the newly generated synthetic basic infrastructure index. The first principal component accounts for 85% of the total scaled variance in the synthetic index and is highly correlated with the underlying infrastructure measures. The correlations with dependent variables conform to the expectations detailed in the literature review.

The estimated synthetic social infrastructure index, as the first principal component, was calculated as follows:where SINF is the social infrastructure synthetic index, Functional Lit is the number of people (Pop) over the age of 20 years with Grade 7 completed, Nr Crimes is the actual number of crimes reported, Med Spending is the medical expenditure per household in nominal rand values and HH is the number of households.

The education and safety components of the social infrastructure carry approximately the same weights in the social infrastructure index, while health carries a smaller weight. The first principal component accounts for 41% of the overall variance and is highly correlated with the underlying infrastructure measures. The synthetic social infrastructure index also correlates strongly with all dependent variables and conforms to the expectations detailed in the literature review.

3.3. Model estimations and validation

Choosing the correct model estimation technique would involve testing whether restrictions (dummy variables) and fixed effects (FEs) are statistically significant. Dummy variables in the unrestricted (between) least squares dummy variable (LSDV) estimation in favour of unrestricted ordinary least squares (OLS) models will be used. The use of the restricted (between) LSDV estimation would then be compared with the FE within LSDV estimation to determine whether period and/or cross-section effects are significant (Hausman, 1978; Baltagi, 2005:66). The constructed panel dataset used in the empirical analysis assumes that T is sufficiently large and will produce consistent and sufficiently biased estimates within LSDV estimation techniques. The respective models and validation tests are detailed in the following.

Unrestricted OLS regression estimation: where represents the respective development indicators, indicates the specific municipality (1 to 234) and indicates the period (1996–2012). represents the synthetic index for basic infrastructure, while SINF denotes the synthetic index of social infrastructure. The error term, which varies over and t, is denoted by . The dependent variables will comprise the log of gross domestic product per capita (LGDPPC), household income (LHHINC), percentage of people in poverty (LPPOV) and functional literacy (LPLIT).

The restricted (between) LSDV regression estimations are as follows: where represents the dummy variable for rural (one) and urban (zero) municipalities, with representing the basic infrastructure interaction dummy variable calculated as and SRU being the social infrastructure interaction dummy variable calculated as .

The restricted/unrestricted t test performed on the efficiency and validity of use of the slope and dummy variables is defined as follows (Greene & Hensher, 2010:363):where indicates the number of restrictions, while represents the number of pooled cross-sections and the number of years. RSSR would be the restricted sum of square residuals and USSR the unrestricted sum of square residuals, while indicates the degrees of freedom. The hypotheses being tested are defined as follows (δ being the coefficient of the dummy variables): The FE within LSDV two-way error component estimation is detailed as follows (Baltagi, 2005:33):where andwhere

represents unobserved individual effects, represents unobserved time effects and represents the stochastic disturbance term, with being the sum of the three components. The average of the error term is zero, and its variance is fixed and distributed normally, independent and identically, or . In order to determine whether the restricted (between) LSDV or FE within LSDV models provide better estimates, it is required that the joint Chow FE test (F test) should be conducted. The null hypotheses for a two way-error component model are defined as follows (Baltagi, 2005:33):

The Chow test statistic for a two-way error correction model, assuming Gaussian errors, is defined as follows (Thomas 2004:32):Should the null hypothesis be rejected, it can be assumed that cross-section and/or time effects exist between the municipalities and that the within LSDV estimation will produce more efficient and precise estimates. However, the test is only valid if individual cross-section and time effects are judged to be individually significant. The individual cross-section specification is defined as follows (Thomas, 2004:32):

The Chow test statistic for the one-way FE model with cross-section effects is defined as in Thomas (2004:32):The individual period specification is defined as follows (Thomas, 2004:32): The Chow test statistic for the one-way FE model with period effects is defined as:Rejecting the null hypothesis that joint and individual period and cross-sectional effects are significant will signal the use of the FE within the LSDV model.

The validated estimation will then undergo specification tests for serial correlation, heteroskedasticity and endogeneity in order to assess whether measurement concerns, collinearity among infrastructure assets, identification and heterogeneity concerns have been addressed (Romp & De Haan, 2007; Calderón & Servén, 2008; Straub, 2010).

Regressing the economic growth and social development variables against basic (BINF) and social infrastructure (SINF) will provide the coefficients needed to compile the respective urban and rural basic and social infrastructure equations for each of them on the economic growth and social development variables.

4. Results

The respective restrictive (OLS), unrestricted (between) LSDV and FE within LSDV two-way error correction estimation results, the respective model validation tests and the specification tests are presented in Tables 1 and 2. The validated model and its corresponding values are used to construct the respective urban and rural economic growth and social development equations for the ensuing basic and social infrastructure (Table 3). The results will be used to indicate whether, and to what extent, basic and social infrastructure affects urban and rural gross domestic product per capita (LGDPPC), household disposable income (LHHINC), the percentage of people in poverty (LPPOV) and functional literacy (LPLIT).

Table 1. Summary of basic infrastructure regression results.

Row	Dependent variable	LGDPPC			LHHINC			LPPOV			LPLIT
	Modelling technique	OLS	Between LSDV	Within LSDV	OLS	Between LSDV	Within LSDV	OLS	Between LSDV	Within LSDV	OLS	Between LSDV	Within LSDV
A	C	9.6874 11313.18 (0.000)	9.6079 1005.11 (0.000)	9.6913 3825.02 (0.000)	10.9352 1835.10 (0.000)	10.8936 1615.05 (0.000)	10.9410 4286.40 (0.000)	−0.7341 −209.18 (0.000)	−0.6901 −182.24 (0.000)	−0.7587 −66.07 (0.000)	−0.5636 −243.788 (0.000)	−0.6028 −255.53 (0.000)	−0.5446 −128.57 (0.000)
B	Basic infrastructure (BINF)	0.3905 72.8107 (0.000)	0.3612 62.7905 (0.000)	0.0874 18.3913 (0.000)	0.2332 62.5024 (0.000)	0.2112 52.0178 (0.000)	0.04340 9.06318 (0.000)	−0.1521 −69.1967 (0.000)	−0.1293 −56.727 (0.000)	−0.1217 −26.186 (0.000)	0.11518 79.4868 (0.000)	0.0994 72.9024 (0.000)	0.0295 19.3944 (0.000)
C	Urban rural dummy (RUDUM)		0.3868 13.7612 (0.000)			0.1045 5.2684 (0.000)			−0.1179 −10.5890 (0.000)	0.1087 2.1571 (0.031)		0.1734 33.3210 (0.000)	−0.0646 −3.71501 (0.000)
D	Interaction variable (BRU)		−0.0348 −1.7068 (0.090)	−0.0172 −1.8665 (0.062)		0.0781 5.4313 (0.000)	−0.0254 −2.7239 (0.000)		−0.0749 −9.2780 (0.000)				−0.0187 −5.9971 (0.000)
E	R^2 adjusted	0.5713	0.6000	0.9889	0.4955	0.5159	0.9745	0.5462	0.60423	0.9379	0.61368	0.6979	0.9860
F	F statistic	5301.39 (0.000)	1989.51 (0.000)	1423.32 (0.000)	3906.55 (0.000)	1413.52 (0.000)	568.07 (0.000)	4788.19 (0.000)	2024.95 (0.000)	224.580 (0.000)	6318.15 (0.000)	4595.59 (0.000)	1118.04 (0.000)
	Model validation and specification tests
G	Restricted/unrestricted t test CV		143.5291 4.6105			84.7342 4.6105			292.4047 4.6105			555.0050 4.6105
H	Chow two-way test CV			592.0962 1.2294			271.2847 1.2294			82.5489 1.2294			331.1312 1.2294
I	Chow cross-section test CV			615.8460 1.2371			107.5373 1.2371			74.4449 1.2371			335.25989 1.2371
J	Chow period test CV			49.7675 2.0048			1231.2304 2.0048			1906436 2.0048			341.1893 2.0048
K	Serial correlation given fixed effects CV:			48.8123 2.326			48.7697 2.326			46.9653 2.326			46.5778 2.326
L	Heteroskedasticity CV:			3974.8326 286.1389			2712.0122 286.1389			3035.0114 286.1389			2876.2384 286.1389
M	Hausman test for endogeneity CV:			18.11598 9.2103			264.1117 9.2103			3.2077 9.2103			4.8682 9.2103

Notes: Rows A–F: coefficient, t-statistic (probability). CV: critical value.

Table 2. Summary of social infrastructure regression results.

Row	Dependant variable	LGDPPC			LHHINC			LPPOV			LPLIT
	Modelling technique	OLS	Between LSDV	Within LSDV	OLS	Between LSDV	Within LSDV	OLS	Between LSDV	Within LSDV	OLS	Between LSDV	Within LSDV
A	C	9.6973 855.651 (0.000)	9.5995 769.884 (0.000)	9.7060 4609.398 (0.000)	11.0004 1432.62 (0.000)	10.9382 1272.56 (0.000)	10.9320 1992.13 (0.000)	−0.7246 −173.83 (0.000)	−0.6797 −148.50 (0.000)	−0.7334 −378.859 (0.000)	−0.5521 −178.21 (0.000)	−0.5960 −188.1260 (0.000)	−0.5970 −202.4447 (0.000)
B	Social infrastructure (SINF)	0.4312 2.2821 (0.000)	0.4181 35.7245 (0.000)	0.0692 13.5245 (0.000)	0.2172 31.4375 (0.000)	0.1966 24.36512 (0.000)	0.1677 32.3911 (0.000)	−0.1790 −47.716 (0.000)	−0.1563 −36.3808 (0.000)	−0.0787 −16.7290 (0.000)	0.1270 45.57664(0.000)	0.1146 38.5383 (0.000)	0.1099 39.4923 (0.000)
C	Urban rural dummy (RUDUM)		0.5995 19.7923(0.000)			0.3330 15.9468 (0.000)	0.3201 243.0000 (0.0260)		−0.2120 −19.0619 (0.000)			0.2431 31.5900 (0.000)	0.2404 33.5356 (0.000)
D	Interaction variable (SRU)		−0.2289 −9.1939 (0.000)	−0.0534 −5.2922 (0.000)		−0.0809 −4.7154 (0.000)	−0.0245 −2.2273 (0.000)		0.0185 2.20249 (0.000)	0.0536 5.7769 (0.000)		−0.0681 −10.7641 (0.000)	−0.0570 −9.8188 (0.000)
E	R^2 adjusted	0.3374	0.4038	0.9911	0.2196	0.2729	0.7038	0.393582	0.4563	0.94913	0.3717	0.5116	0.5769
F	F statistic	1787.77 (0.000)	793.116 (0,000)	1571.286 (0.000)	988.3181 (0.000)	440.0593 (0.000)	491.4256 (0.000)	2276.793 (0.000)	982.7116 (0.000)	263.9667 (0.000)	2077.230 (0.000)	1226.0360 (0.000)	282.490 (0.000)
	Model validation and specification tests
G	Restricted/unrestricted t test CV		222.4682 4.6105			146.9876 4.6105			231.1745 4.6105			570.2751 4.6105
H	Chow two-way test CV			1044.5964 1.2294						154.2475 1.2294
I	Chow cross-section test CV			1089.3636 1.2371						114.6273 1.2371
J	Chow period test CV			318.5478 2.0048			365.2682 2.0048			301.1434 2.0048			39.7010 2.0048
K	Serial correlation given fixed effects CV:			0.06283 2.326			59.9917 2,326			42.0981 2.326			56.2817 2.326
L	Heteroskedasticity CV:			4522.4004 286.1389			1513.7893 286.1389			3043.2874 286.1389			1599.2874 286.1389
M	Hausman test for exogeneity CV:			7.9528 9.2103			17.5266 9.2103			24.5197 9.2103			3.8097 9.2103

Notes: Rows A–F: coefficient, t-statistic (probability). CV: critical value.

Table 3. Urban–rural municipality results.

Variable	Area	Result
Basic infrastructure
LGDPPC	Urban
	Rural
LHHINC	Urban
	Rural
LPPOV	Urban
	Rural
LPLIT	Urban
	Rural
Social infrastructure
LGDPPC	Urban	LGDPPC = 9.7060 + 0.0158SINF
	Rural	LGDPPC = 9.7060 + 0.0692SINF
LHHINC	Urban	LHHINC = 11.2521 + 0.1432SINF
	Rural	LHHINC = 10.9320 + 0.1677SINF
LPPOV	Urban	LPPOV = −0.7334−0.0251SINF
	Rural	LPPOV = −0.7334−0.0787SINF
LPLIT	Urban	LPLIT = −0.3566 + 0.0529SINF
	Rural	LPLIT = −0.5970 + 0.1099SINF

The Levin et al. (2002) t* test for unit roots was conducted on levels with individual intercept and trend included in the test equation. Given BINF (−8.1207, p = 0.000), BRU (−4.99627, p = 0.000), LGGDP (−15.8509, p = 0.000), LHHINC (−16.5707, p = 0.000), LPPOV (−15.4276, p = 0.000) and LPLIT (−16.2228, p = 0.000), the results indicate that the calculated test statistic is smaller than the critical value ∼ N(0,1) (one-tail) of −1.645, all p < 0.05. However, the SINF (−1.6402, p = 0.0505) and SRU (−1.6588, p = 0.0486) test statistics resulted in the null hypothesis being rejected in favour of the alternative null hypothesis, at a 10% and a 5% level of significance, respectively. The individual series in levels are therefore stationary.

Each of the calculated restricted/unrestricted t-test values (rows G) was greater than the critical value, resulting in the null hypothesis being rejected in favour of the use of the restricted (between) LSDV regression results. The test therefore confirms that basic infrastructure investment has a statistically significant and different effect on economic growth and social development in urban and rural development, respectively.

The Chow specification F-test for two-way error correction models was used to determine whether fixed (period and/or cross-section) effects are significant. The calculated F-stat (rows H) is greater than the critical value in each of the respective regressions. Period and cross-section effects are therefore present that should be controlled for. Testing for individual cross-section effects individually also rejects the null hypothesis of no individual cross-section. Cross-sectional heterogeneity should therefore be controlled for. Lastly, testing for individual period effects individually resulted in the null hypothesis also being rejected (rows J). It is therefore necessary to control for period effects with dynamic adjustments over time. The Chow specification tests comply with all three requirements of rejecting the joint and individual null hypothesis of no period and/or cross-sectional effects in favour of using the FE within LSDV model.

The FE (within) LSDV is then subjected to specification tests for serial correlation (rows K), heteroskedasticity (rows L) and endogeneity (rows M), in order to determine whether the model is correctly specified and produces unbiased and consistent estimates. Each of these specification tests is now discussed.

The joint LM test for serial correlation confirmed that the FE within LSDV basic infrastructure (Table 1) regression is not stationary. All calculated F-stats (rows K) were smaller than the cited critical values. The joint LM test for serial correlation confirmed that the social infrastructure within LSDV LGDPPC regression presented in Table 2 is stationary. All other calculated F-stats (rows K) were greater than the respective critical values. Therefore, the null hypothesis of no serial correlation is rejected. Serial correlation is not expected to affect the unbiasedness or consistency of the estimates, only their efficiency.

Heteroskedasticity was tested as suggested by Greene (2013:714) with the joint LM test being distributed as chi-square with N − 1 degree of freedom. The calculated LM statistic (rows L) was greater than the critical value in each of the estimations. This resulted in the null hypotheses of homoskedasticity being rejected, indicating the presence of heteroskedasticity in the residuals. The presence of heteroskedasticity in the residuals can be corrected with the white period coefficient covariance method to correct for regular residual heteroskedasticity in light of N > T (Arellano, 1987:431).

Exogeneity of the explanatory variables was tested using the Hausman specification test, which is distributed chi-square with degrees of freedom. Within Table 1 (basic infrastructure), the calculated chi-square statistic (row M) was greater than the critical value in the LGDPPC and LHHINC estimations, resulting in the null hypothesis of exogeneity being rejected. Therefore, the two models are either miss-specified or correlations exist between individual effects and exogenous variables in the respective economic growth and social development estimations. However, the calculated chi-square statistic was smaller than the critical value in the case of the LPPOV and LPLIT estimations, resulting in the null hypothesis of exogeneity being accepted. In Table 2 (social infrastructure), the calculated chi-square statistic (row M) was greater than the critical value in the case of LHHINC and LPERPOV, resulting in the null hypothesis of exogeneity being rejected. The calculated chi-square statistics (row M) for LGDPPC and LPLIT are smaller than the critical value, resulting in the null hypothesis of exogeneity being accepted.

Baltagi (2005) indicated that if the T in the estimations is sufficiently large, the coefficients are considered consistent and sufficiently unbiased, and this would validate the use of the FE within LSDV estimates, even though it might not be optimal. Therefore, the coefficients produced in the within LSDV estimation will be used to estimate the respective urban and rural economic growth and social development equations. Additionally, the FE within LSDV estimation sweeps out individual and/or time-specific effects. The estimates should therefore not be biased in the presence of endogeneity of the explanatory regressors. Consequently, the coefficients will be used to calculate the urban and rural economic growth and social development equations presented in Table 3.

The results conform to expectations detailed in the literature review that basic and social infrastructure delivery has a positive impact on economic growth and social development. They also conform to the view that the impact of basic and social infrastructure economic investment on economic growth and social development would be greater in rural municipalities. However, the results provide empirical evidence which supports the sentiment that was previously only postulated normatively.

5. Discussion and conclusion

It is noteworthy that the empirical results of basic and social infrastructure investment in South Africa generally indicate lower economic growth and social return elasticities when compared with other countries as cited in the literature review. The results obtained from the respective BINF and SINF (Table 3) urban and rural equations range between 0.02 and 0.09, compared with a range between 0.05 and 0.39 as indicated in the literature review. A study by Sahoo & Dash (2008:19) suggests income elasticity of infrastructure (BINF and SINF) investment to be between 0.20 and 0.25. The calculated elasticities range between 0.02 and 0.17. Suescún (2007) indicates the infrastructure elasticity of poverty to be −0.32 using a $2/day poverty estimate. The LPPOV results obtained in BINF and SINF (Table 3) suggest elasticities ranging from −0.02 to −0.12 for urban and rural municipalities. Literacy-induced elasticities of infrastructure of 0.12 calculated by Suescún (2007) are also higher than the derived urban and rural BINF and SINF elasticities, which range between 0.02 and 0.11.

The pre-mentioned results imply that investment in social and basic infrastructure in South Africa has a relatively smaller effect compared with other countries on outcome variables such as social capital formation, exploiting the benefits of proximity and improved governance. The generally lower economic growth and social development returns affecting social capital formation could be accounted for by including quality of investment measures for basic and social infrastructure, respectively (Calderón & Servén, 2008). The lower elasticities could also underline governance concerns (Hemson, 2004:17), ill-considered spatial implementation such as the benefits of proximity (Luo & Wang, 2003:876), and the inability of planners to understand the cultural aspects required to optimise social capital returns (Putnam, 1995), resulting in a generally lower economic and social return of infrastructure investment when compared with other countries. Many of these factors have been identified by the National Planning Commission as binding constraints for South Africa becoming a growing and inclusive society. The fact that the public sector has a central role in providing collective goods, also coining on the benefits of the proximity of planned social infrastructure, places them in an ideal position to influence infrastructure policy and planning programmes aimed at inclusive economic growth and social development. Using detailed economic growth and social development elasticities of basic and social infrastructure investment for urban and rural municipalities, respectively, would assist such planning initiatives and optimise investment returns. However, the need for an all-encompassing development plan, which considers not only the economic growth and social development at a micro, sub-national level, but also at a provincial and national level, is a prerequisite of inclusive growth and the development of spatial equality. Nevertheless, policy developed at a sub-national level has the benefit over ‘one policy fits all implemented at national and provincial level’ that it addresses the specific needs of a community.

The empirical research confirms that basic and social infrastructure affects urban and rural economic growth and social development differently. The economic growth and social development return would be greater in rural municipalities than in similar infrastructure investments in urban municipalities. The government should therefore consider this finding in its basic and social infrastructure delivery plan as a means to reduce the economic growth and social development inequality experienced between urban and rural municipalities.

The presented results should be interpreted in light of a number of limitations experienced in conducting the research. Basic and social infrastructure quality indices should preferably be included in the analysis, as suggested by Calderón & Servén (2008). The qualitative information, however, is not available and will most probably not be compiled in the foreseeable future. Straub (2010:692) also suggests that inside (lagged and differenced) instrumental variables should be used in a generalized method of moments (GMM) framework to correct for endogeneity. The complicity of finding valid instruments in addition to the restricted modelling methodology – restricted (within) LSDV – unnecessarily complicates the estimation of results for the purpose of the research.

This study lays the foundation for further research on the topic. In the future, with the availability of longitudinal data that stretch over a longer time span than the medium term, opportunities open up for the analysis of inter-generational transfers of benefits derived from social and basic infrastructure investment. This type of analysis would probably capture the inter-generational spill-over effects of basic and social infrastructure investment. In addition, a modelling framework that estimates the combined impact of basic and social infrastructure on economic growth and social development in urban and rural municipalities, respectively, needs to be constructed. This could empirically validate the hypothesis of Metwally et al. (2007:61) and ESCAP (2006:5) that basic infrastructure lays the foundation for effective social service infrastructure implementation and that social infrastructure is necessary for the optimal utilisation of basic infrastructure. The model should also be integrated into a municipal planning framework that calculates the economic growth and social development returns of planned basic and social infrastructure investment. Such a return on investment estimation could, firstly, ensure the optimal utilisation of available resources and, secondly, serve as an indicator of where basic and social infrastructure should be increased to create a more inclusive and equal society on a spatial level in order to provide the practical realisation of the vision of South Africa’s Constitution.

Acknowledgements

The authors appreciate the valuable comments provided by the anonymous referees on the earlier draft submitted. The authors wish to acknowledge Economic Research Southern Africa for financial assistance with this research article. All mistakes and omissions remain our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

Economic Research Southern Africa provided financial assistance.

Notes

1 The choice of basic and social infrastructure indicators to measure infrastructure investment has been somewhat contentious. Certain studies use physical or expenditure approximations of infrastructure (Calderón & Servén, 2004; Romp & De Haan, 2007), whereas other use physical measures of basic and social infrastructure (Straub, 2010). However, concerns about the validity of physical or expenditure approximations have swayed researchers in recent studies to use physical measures. Therefore, in this study, in line with recent literature, we used physical measures – namely electricity, water and sanitation provision – as indicators of basic infrastructure investment and the provision of schools, hospitals and police stations as indicators of social infrastructure investment.

2 The categorising between rural and urban regions differs between countries and the categorising can be based on different variables, for example: administratively, morphologically, demographically and functionally. In South Africa, the Organisation for Economic Cooperation and Development (OECD) bases the categorisation on demographic approaches and distinguishes between three types of regions, grouped according to the share of the regional population living in rural communities (Ballas et al., 2003:7). They classify a region as: ‘Predominantly Rural’ if more than 50% of the population lives in rural communities; ‘Significantly Rural’ if between 15 and 50% of the community stays in rural communities; and ‘Predominantly Urbanised’ if less than 15% of the community stays in urban areas (Schmidt & du Plessis, 2013). However, rural areas (this also applies to urban areas), notwithstanding the exact base of categorising, share similar characteristics across countries, such as not having the same quality of amenities as offered in urban areas, lower per-capita income, fewer employment opportunities, higher levels of poverty and lower levels of social and basic infrastructure. In this research article, we focus on addressing these shared characteristics of rural and urban areas across countries, irrespective of the method used to categorise the areas.

3 One of the reviewers of the journal made the valid point that people should be provided with the basic services allowing them to gain access to increasing opportunities, thereby improving their well-being, and an important dimension of this is mobility and the transport infrastructure and services that are necessary to promote and support this aspect. This aspect constitutes a limitation to the dataset because data on transport infrastructure were not available in the dataset.