Multidimensional Indicator of Energy Poverty in South Africa Using the Fuzzy Set Approach

Abstract

The electricity supply in South Africa is characterized by load-shedding. This study analyzed the determinants of the multidimensional energy poverty index (MEPI) in South Africa. The data, which were taken from the 2019–2021 General Household Survey (GHS), were analyzed using Tobit regression. The results showed that between 2019 and 2021, the use of clean energy for cooking declined from 85.97% to 85.68%, respectively, whereas the use of clean energy for water heating declined from 87.24% in 2020 to 86.55% in 2021. Space heating with clean energy declined from 53.57% in 2019 to 50.35% in 2021. The average fuzzy MEPI was 0.143 and Western Cape and KwaZulu-Natal provinces had the highest average values with 0.180 and 0.176, respectively. In the combined dataset, the Tobit regression results showed that, compared to Western Cape, the fuzzy MEPI significantly decreased (p < 0.01) by −0.038, 0.028, 0.045, 0.023, 0.029, 0.038, 0.037, and 0.042 for residents in Eastern Cape, Northern Cape, Free State, Kwazulu-Natal, North West, Gauteng, Mpumalanga, and Limpopo provinces, respectively. In addition, the fuzzy MEPI for the Black, Coloured, Asian, and White respondents decreased by 0.042, 0.062, and 0.084, respectively. The fuzzy MEPI for male-headed households and the number of social grants increased, whereas the fuzzy MEPI significantly decreased (p < 0.01) for the monthly salary and age of household heads. It was concluded that energy poverty in South Africa manifests through unclean energy utilization for space heating. The promotion of clean energy utilization should focus on deprived provinces, farms, and tribal areas.

1. Introduction

Energy remains a very important component of household expenditure. It is also an essential indicator to assess the effectiveness of economic policies aimed at reducing poverty in developing countries [1]. The type of energy used by households is an important measure of poverty given the intricate linkages between energy and consumers’ utilities [2]. A consensus now exists among policymakers on the interactive relationships between the energy supply and economic development. This has propelled the concurrent inclusion of access to clean energy as one of the foremost Sustainable Development Goals (SDGs) [3]. Energy poverty is defined as inadequate access to affordable and reliable clean energy. More importantly, an objective evaluation of the SDGs has revealed the interrelatedness and multifunctionality of clean energy. It had been concluded that access to clean energy is fundamentally essential to achieving some SDGs such as the alleviation of poverty, zero hunger, climate change mitigation, and the promotion of children’s health [4,5]. Regrettably, the achievement of SGDs in many sub-Saharan African (SSA) countries may suffer some setbacks due to a decline in the supply of electricity over the past few decades [6].

Available statistics have shown that in 2019, 90% of the world’s population had access to electricity [3]. Moreover, between 2010 and 2019, this number declined from 1.2 billion to 759 million, although 2.6 billion still had no access to clean cooking energy [3,6]. A recent estimate revealed that about 860 million people lacked access to electricity in 2021 [7], of whom 600 million were in Africa [8]. In SSA, it has been emphasized that policy revisions are desperately required to address clean energy poverty given its role in achieving the goals of the Paris Agreement [9]. More importantly, African economic resilience, sustainability, improvements in general health, poverty reduction, human security, and migration disincentives largely hinge on economic policies that are meant to deliver improvements in access to clean energy [10,11,12]. However, improving access to clean energy remains a herculean task given that an estimated 81% of the SSA population derives their cooking energy from solid biomass and wood-based sources [13].

In South Africa, the gaps between energy demand and supply, along with intermittent load-shedding, are development challenges that urgently require the government’s attention [14]. The constitutional role of Eskom as the ultimate power generator and electricity supplier is now being called into question as load-shedding progresses from one stage to the next and there are reports of unscheduled blackouts in certain provinces. Achieving the goal of low carbon emissions in power generation, as specified in the Paris Agreement, will be difficult for South Africa given the dominance of coal in the current electricity generation system. Some statistics have shown that more than 80% of the total electricity generation in South Africa is currently sourced from the thermal combustion of coal [15]. Moreover, compliance with the Paris Agreement will result in an approximately 50% reduction in the amount of electricity that is generated from coal over the next few decades [15]. The need to reduce the impact of emissions from coal burning on human health, as well as other associated environmental consequences, mandates an energy policy transition that deprioritizes the emission of greenhouse gases during electricity generation [16]. It has been noted that, based on current arrangements, electricity supply cannot match demand due to certain prevailing circumstances that influence the availability of coal and the dwindling combustion capacity of some existing power plants [17,18].

There is conventional wisdom in understanding the persistent apprehension among stakeholders in the power sector towards the need to provide sustainable solutions to South Africa’s looming energy crises [15]. The intermittent power transmission lapses over the past few years are ticking time bombs that may ephemerally explode beyond any statistical predictions or forecasts [14]. Conventional wisdom also holds that, in addition to the need to address tactical mismanagement in the energy sector, the international debate surrounding green economic growth reaffirms the need to reduce dependence on the burning of coal as the traditional method of generating electricity [18]. As a result, discussions about energy stability have pinpointed the need to diversify the country’s energy sources to include underutilized and cleaner renewable sources [19].

A proper understanding of the factors associated with energy poverty is fundamentally essential to encourage both stakeholders’ engagement and economic reforms to enable short-term access to clean energy. The literature is replete with statistical approaches and assumptions for computing household energy poverty indices based on energy expenditure, housing characteristics, primary sources of energy, and utilization of electrical assets [20]. A more direct method classifies households as being energy poor if they are unable to maintain an indoor temperature of between 18–21 degree Celsius, as required by the World Health Organization (WHO) [21]. This approach suffers significant setbacks from data availability given that most household surveys do not collect information about indoor temperatures [20]. On the other hand, the energy poverty line is utilized for the income/expenditure approach [20,22]. This approach considers the burden of energy expenditure within a household by evaluating the proportion of total income devoted to energy. However, a major setback is the inadequacy of many surveys to comprehensively cover energy expenditure and the fundamental limitations of income data in many surveys [20].

To address some of these limitations, the multidimensional energy poverty index (MEPI) was proposed [23,24] as an offshoot of the conventional multidimensional poverty analysis [25,26,27,28]. Although multidimensional energy poverty indices have been computed by several authors using the Alkire–Foster approach [23,24,29] with a focus on the incidence and severity parameters, the major drawbacks of this methodology are its arbitrariness in defining the weights of the indicator, arbitrary definition of the poverty line, and mandatory coding of attributes as either zero (0) for the least deprived or one (1) for the most deprived. Given the nature of data for the computation of the MEPI, several other approaches, such as Principal Component Analysis (PCA), Multiple Correspondence Analysis (MCA), and fuzzy set, can be used. The fuzzy set approach has previously been used by some authors to compute the MEPI [30,31,32]. This approach was used in this study because of its very rare applicability by researchers despite its notable advantages of not assuming arbitrary weights for the welfare attributes, not assigning the poverty line arbitrarily, and allowing coded indicators to assume any value between zero (0) for the least deprived and one (1) for the most deprived.

Furthermore, several studies conducted in South Africa have shown that significant disparities exist in energy poverty across households’ heads gender, provinces, and economic sectors [33,34]. In addition, affordability is one of the major factors in the use of clean energy [35]. Therefore, poor households may not be able to use clean energy for all their domestic needs due to its high cost, even though they are connected to the main electricity grid [35,36]. Millions of South Africans living in rural and urban areas have access to electricity but are unable to afford it [37]. Therefore, being connected to the national electricity grid does not always guarantee its utilization for improving households’ welfare [38]. In other studies conducted in South Africa, energy poverty was positively related to being African, household size, transportation, and education expenditure, but negatively related to food expenditure, being female, Coloured, and obtaining a tertiary degree [39]. Other studies have found that being a male-headed household [40], household income [39,41], living in rural areas [42,43], marital status [43], household size [44], employment status [41], and agricultural income [45] were found to significantly influence energy poverty. Several studies conducted outside South Africa have found that energy poverty is influenced by marital status [46], household size [47], and employment status [48].

This study hypothesizes that households’ demographic variables do not significantly influence the MEPI. Furthermore, this paper seeks to add to the existing body of knowledge by using nationally representative South African datasets from three different years to analyze multidimensional energy poverty using the fuzzy set approach. The datasets also provide a veritable platform to understand the magnitude of energy poverty during the COVID-19 pandemic. The study will provide an empirically reliable platform to track South Africa’s progress towards SDG 7.

2. Materials and Methods

2.1. Data and Sampling Procedures

This study used the 2019–2021 General Household Survey (GHS) data for South Africa. The data collection procedures followed the Master Sample (MS) framework that was developed in 2013 and has been in use since 2015. The framework utilizes the 103,576 enumeration areas (EAs) that were derived from 3324 primary sampling units (PSUs). A total of 33,000 dwelling units (DUs) were selected using a representative sampling technique that progressed from the provinces to the metropolitan and geographic areas. A stratified two-stage sampling method was used, and samples were allocated based on probability proportional to size. The sample weights were also computed for each of the respondents [49,50,51] and were utilized for the analyses. The collection of the 2019 data commenced in January and ended in December. The interviews were conducted using Computer-Assisted Personal Interviews (CAPI) and 19,649 households were successfully interviewed. In 2020 and 2021, due to the COVID-19 pandemic, Computer-Assisted Telephone Interviews (CATI) were used for data collection. The number of successfully interviewed households was 8896 in 2020 and 9626 in 2021.

2.2. Fuzzy Set Multidimensional Energy Poverty Index (MEPI)

Following Cerioli et al. [52], this study adopts the fuzzy set approach to compute the MEPI. The fuzzy set theory was proposed by Zadeh [53] and is based on the notion that a class of objects can be defined by some degree of their set membership. The theory notes that every element belongs to the poverty set to some degree. The efficiency of fuzzy sets in the multidimensional analysis of poverty was demonstrated and emphasized by Dagum and Costa [54]. Based on the welfare attributes $j=1,\dots .,m$, the degree of poverty displayed by a household can be presented as ${\mu }_B\left[X_j\left(a_i\right)\right]=X_{ij},0\le X_{ij}\le 1.$ Unlike other multidimensional poverty measures, the fuzzy approach assigns weights to attributes that are proportional to the magnitude of deprivations exhibited by all respondents. The cutoff points for each of the indicators were first determined by denoting deprived households as 1 and non-deprived as 0. The dimensions of energy poverty were explored from twelve attributes, which are the ownership of functioning electrical appliances (television, DSTV/M-Net subscription, computer/laptop, refrigerator/freezer, home security services, geysers, and cell phone), access to electricity, and the type of energy used for cooking, lighting, water heating, and space heating.

The multidimensional poverty ratio ${\mu }_B\left(a_i\right)$, which highlights the levels of clean energy deprivation, is defined as the weighted average of $X_{ij}$, which is presented as:

$${\mu }_B\left(a_i\right)={\displaystyle \sum }_{j=1}^m{X}_{ij}{w}_j/{\displaystyle \sum }_{j=1}^m{w}_j$$

(1)

In Equation (1), $w_{j }$ is the weight attached to the $jth$ welfare attribute, which is an inverse function of the degree of deprivation [55]. The computation of the attributes’ weights can be expressed as:

$$w_j=\left[log{\displaystyle \sum }_{i=1}^{n}g \left(a_i\right)/{\displaystyle \sum }_{i=1}^n{X}_{ij }g\left(a_i\right)\right]\ge 0$$

(2)

Therefore, $w_j$ is small when many of the households are deprived of a welfare attribute and high when only a few are deprived. Another important advantage of the fuzzy poverty index (${\mu }_B\left(a_i\right)$) is that it can be decomposed across respondents’ demographic variables. This is expressed as:

$${\mu }_B\left(a_i^k\right)=\frac{{\sum }_{j=1 X_{ij}^k}^m{w}_j}{{\sum }_{j=1}^m{w}_j}$$

(3)

2.3. Tobit Regression Model of the Determinants of MEPI

The Tobit regression model was used to analyze the demographic factors explaining the MEPI. This model can be specified as:

$${\mu }_B\left(a_i\right)={\alpha }_k+{\beta }_{jk}{\displaystyle \sum }_{j=1}^{22}{H}_{ik}+e_{ik}$$

(4)

where i denotes the households, k denotes the time periods of the analyses, ${\alpha }_k$ are the constant terms for the kth period, ${\beta }_{jk}$ are the estimated jth parameters, $H_{ik}$ are the independent variables, and $e_{ik}$ denotes the error terms. The independent variables are the province (Eastern Cape (yes = 1, 0 otherwise), Northern Cape (yes = 1, 0 otherwise), Free State (yes = 1, 0 otherwise), KwaZulu-Natal (yes = 1, 0 otherwise), North West (yes = 1, 0 otherwise), Gauteng (yes = 1, 0 otherwise), Mpumalanga (yes = 1, 0 otherwise), and Limpopo (yes = 1, 0 otherwise)), population group (Coloured (yes = 1, 0 otherwise), Asian/Indian (yes = 1, 0 otherwise), White (yes = 1, 0 otherwise)), gender (male headed households (yes = 1, 0 otherwise)), age of household head, number of members under 5 years, number of members aged 5 to 17 years, number of adult members over 60 years, monthly salary (ZAR), total monthly grants (ZAR), geography (tribal area (yes = 1, 0 otherwise), farm (yes = 1, 0 otherwise)), and year of data collection (2020 (yes = 1, 0 otherwise) and 2021 (yes = 1, 0 otherwise)).

3. Results

3.1. Energy Poverty Attributes and Deprivation Levels

Figure 1 shows the distribution of the households that were deprived of the selected attributes. It shows that the average number of households without access to electricity was 5.19% during the period 2019–2021. However, there was a slight increase in the proportion of deprived households between 2020 and 2021 from 4.27% to 4.57%, respectively. The results also showed a drastic increase in the percentage of households that were deprived of space heating from 46.43% in 2019 to 49.67% in 2021. Moreover, there was a decline in the proportion of households that were deprived of the use of clean energy for cooking, lighting, and water heating between 2019 and 2020, although there were some slight increases between 2020 and 2021.

3.2. Distribution of Fuzzy MEPI

Table 1. Average MEPI across provinces, races, genders, and economic sectors in South Africa (2019–2021).

Variables	2019		2020		2021		All
	Mean	Std. Dev.	Mean	Std. dev.	Mean	Std. Dev.	Mean	Std. Dev.
Province
Western Cape	0.200	0.209	0.150	0.164	0.169	0.180	0.180	0.192
Eastern Cape	0.141	0.193	0.112	0.162	0.117	0.177	0.129	0.183
Northern Cape	0.138	0.218	0.116	0.203	0.112	0.190	0.128	0.210
Free State	0.154	0.193	0.116	0.152	0.125	0.163	0.137	0.177
KwaZulu-Natal	0.190	0.180	0.167	0.155	0.161	0.158	0.176	0.168
North West	0.175	0.207	0.149	0.176	0.156	0.191	0.164	0.196
Gauteng	0.173	0.231	0.123	0.177	0.130	0.191	0.151	0.211
Mpumalanga	0.145	0.198	0.074	0.097	0.115	0.169	0.121	0.175
Limpopo	0.091	0.124	0.071	0.093	0.066	0.089	0.081	0.111
Race
Black	0.175	0.209	0.138	0.174	0.144	0.181	0.158	0.195
Coloured	0.106	0.161	0.064	0.087	0.071	0.097	0.090	0.137
Asian/Indian	0.049	0.112	0.028	0.051	0.048	0.112	0.045	0.105
White	0.031	0.056	0.029	0.068	0.026	0.048	0.030	0.057
Gender
Male	0.157	0.213	0.123	0.176	0.129	0.183	0.143	0.199
Female	0.155	0.185	0.129	0.158	0.134	0.163	0.143	0.174
Sector
Urban	0.128	0.194	0.097	0.164	0.100	0.165	0.114	0.181
Tribal	0.197	0.189	0.162	0.153	0.169	0.166	0.180	0.175
Farm	0.302	0.305	0.234	0.265	0.238	0.269	0.271	0.290
All	0.156	0.202	0.126	0.168	0.131	0.174	0.143	0.188

Source: GHS datasets (2019–2021).

shows the distribution of the MEPI across the selected demographic variables. The average fuzzy MEPI was 0.14 between 2019 and 2021. However, between 2019 and 2020, the fuzzy MEPI slightly declined, whereas a slight increase was recorded between 2020 and 2021. Across the provinces, Limpopo had the lowest average MEPI, whereas Western Cape, KwaZulu-Natal, and North West provinces had the highest average MEPI. The table further shows that White households had the lowest average MEPI. However, Black respondents had the highest average MEPI over the study period, followed by Coloured respondents and those of Asian origin. In the combined data, the average fuzzy MEPI was the same for male and female respondents. Additionally, respondents from urban areas had the lowest average MEPI over the study period, whereas those on farms had the highest.

Figure 2 shows the distributions of the MEPI for households in 2019, 2020, and 2021 using the fuzzy set approach. The results showed that across the period studied, the majority of respondents had MEPI scores less than 0.50. Specifically, 79.63%, 84.41%, and 83.60% of respondents had MEPI scores < 0.25 in 2019, 2020, and 2021, respectively.

Table 2. Distribution of MEPI across the selected demographic variables in 2019–2021.

	2019				2020				2021
	<0.25	0.25 < 0.50	0.50 < 0.75	≥0.75	<0.25	0.25 <0.50	0.50 <0.75	≥0.75	<0.25	0.25 < 0.50	0.50 < 0.75	≥0.75
Province
Western Cape	73.28	17.81	4.45	4.45	83.61	11.09	2.86	2.44	78.59	14.74	3.54	3.14
Eastern Cape	86.53	7.49	1.27	4.70	90.17	4.82	2.50	2.50	89.63	5.56	0.74	4.07
Northern Cape	84.94	6.43	2.27	6.36	88.40	4.20	2.18	5.22	89.05	4.45	2.61	3.89
Free State	80.02	13.33	3.01	3.64	87.01	9.46	1.70	1.83	84.93	10.87	2.02	2.18
KwaZulu-Natal	64.28	29.65	4.20	1.87	66.69	30.08	2.36	0.87	68.78	27.86	2.19	1.17
North West	74.43	17.82	2.58	5.16	77.97	16.84	1.65	3.54	79.74	14.06	1.31	4.89
Gauteng	78.94	11.46	2.83	6.77	86.90	7.99	1.19	3.91	86.42	7.19	2.24	4.15
Mpumalanga	82.12	11.65	2.77	3.46	95.80	2.62	1.31	0.26	87.65	6.17	4.20	1.98
Limpopo	91.52	6.84	0.90	0.74	96.32	2.91	0.61	0.15	96.92	2.22	0.62	0.25
Race
Black	76.69	15.02	3.22	5.08	82.46	12.33	2.18	3.02	81.46	12.80	2.51	3.22
Coloured	88.21	8.76	1.16	1.87	96.63	2.62	0.37	0.37	95.31	3.44	1.10	0.16
Asian/Indian	96.93	1.53	1.02	0.51	100.00	0.00	0.00	0.00	97.35	1.32	0.00	1.32
White	98.82	0.97	0.21	0.00	98.95	0.42	0.63	0.00	99.24	0.76	0.00	0.00
Gender
Male	79.62	12.34	2.97	5.07	85.16	9.67	2.11	3.07	83.73	10.50	2.56	3.22
Female	79.63	14.41	2.54	3.41	83.51	12.49	1.80	2.19	83.46	12.30	1.89	2.35
Area
Urban	86.16	7.69	1.89	4.26	91.64	3.82	1.54	3.00	91.03	4.63	1.49	2.85
Tribal	68.68	24.08	4.13	3.10	74.73	21.56	2.16	1.55	73.99	20.97	3.04	2.00
Farm	57.72	17.60	7.22	17.46	67.16	14.55	7.84	10.45	67.49	14.86	5.88	11.76
All	79.63	13.22	2.79	4.36	84.40	10.97	1.97	2.66	83.60	11.34	2.24	2.81

Source: GHS datasets (2019–2021).

shows the distribution of the computed MEPI across the provinces, races, genders, and economic sectors in 2019, 2020, and 2021, respectively. Among the provinces, KwaZulu-Natal, Western Cape, North West, and Gauteng had the lowest proportions of respondents with MEPI scores < 0.25 at 64.28%, 73.28%, 74.43%, and 78.94%, respectively, in 2019. Similar results were obtained in 2020 and 2021. In 2020, KwaZulu-Natal and North West provinces had the lowest proportion of respondents with MEPI scores < 0.25 at 66.69% and 77.63%, respectively. Figure 3 also shows similar results for the combined data, with Limpopo province having the highest proportion (93.84%) of respondents with MEPI scores < 0.25.

Table 2 shows that Black respondents had the lowest proportion of MEPI scores < 0.25. Specifically, 76.69%, 82.46%, and 81.46% of Black respondents had MEPI scores < 0.25 in 2019, 2020, and 2021, respectively, as opposed to 98.82%, 98.95%, and 99.24% of White respondents. Figure 4 also shows that in the combined dataset, Black respondents had the lowest proportion of MEPI scores < 0.25. With regard to gender, the proportions of respondents with MEPI scores < 0.25 were almost the same in 2019, with 79.62% for male and 79.63% for female respondents. Table 2 shows that in 2020 and 2021, the proportion of male respondents with MEPI scores <0.25 was slightly higher than that of females. Similar results are presented in Figure 4. Moreover, Table 2 shows that in 2019, 2020, and 2021, respondents from farm and tribal areas had the lowest proportion of MEPI scores < 0.25. In addition, the table shows that 17.46%, 10.45%, and 11.76% of farm residents had MEPI scores ≥0.75 in 2019, 2020, and 2021, respectively. In Figure 3, similar results are presented, with respondents from farm areas having the highest proportion of MEPI scores ≥0.75.

3.3. Determinants of Fuzzy Set Multidimensional Energy Poverty Index

Table 3. Tobit results of the determinants of fuzzy MEPI (2019–2021).

	2019		2020		2021		All Respondents
	Coefficient	t Stat.	Coefficient	t Stat.	Coefficient	t Stat.	Coefficient	t Stat.
Province
Eastern Cape	−0.048 ***	−7.14	−0.020**	−2.27	−0.035 ***	−4.15	−0.038 ***	−8.39
Northern Cape	−0.039 ***	−7.88	−0.005	−0.71	−0.026 ***	−4.17	−0.028 ***	−8.42
Free State	−0.050 ***	−9.66	−0.036 ***	−5.78	−0.044 ***	−7.52	−0.045 ***	−13.55
KwaZulu-Natal	−0.032 ***	−5.62	−0.001	−0.22	−0.026 ***	−4.02	−0.023 ***	−6.35
North West	−0.038 ***	−6.19	−0.012 *	−1.66	−0.025 ***	−3.54	−0.029 ***	−7.18
Gauteng	−0.040 ***	−6.14	−0.029 ***	−3.66	−0.043 ***	−5.50	−0.038 ***	−8.88
Mpumalanga	−0.038 ***	−4.87	−0.049 ***	−4.98	−0.027 ***	−2.80	−0.037 ***	−7.14
Limpopo	−0.056 ***	−8.51	−0.023 ***	−2.63	−0.037 ***	−4.38	−0.042 ***	−9.35
Population Group
Coloured	−0.043 ***	−7.09	−0.031 ***	−3.59	−0.042 ***	−5.01	−0.042 ***	−9.83
Asian/Indian	−0.087 ***	−8.63	−0.044 ***	−2.76	−0.040 ***	−2.85	−0.069 ***	−9.49
White	−0.106 ***	−18.14	−0.040 ***	−4.53	−0.056 ***	−6.40	−0.084 ***	−19.86
Female	−0.013 ***	−4.65	−0.009 ***	−2.58	−0.009 ***	−2.59	−0.011 ***	−5.82
Age of Head	−0.001 ***	−8.21	−0.001 ***	−4.34	−0.001 ***	−3.58	−0.001 ***	−10.23
Child under 5	0.005 *	1.93	0.009 ***	2.96	0.007 **	2.10	0.006 ***	3.61
Child 5 to 17	−0.017 ***	−12.02	−0.008 ***	−4.72	−0.010 ***	−5.54	−0.013 ***	−13.98
Adult over 60	−0.023 ***	−5.91	−0.020 ***	−4.16	−0.024 ***	−5.00	−0.022 ***	−8.54
Salary	−0.001 ***	−14.73	−0.002 ***	−14.50	−0.001 ***	−11.70	−0.001 ***	−22.16
Total grants	0.016 ***	8.94	0.010 ***	4.57	0.011 ***	5.30	0.013 ***	11.62
Geography
Tribal	0.046 ***	12.13	0.049 ***	10.51	0.049 ***	10.87	0.048 ***	19.02
Farm	0.180 ***	23.80	0.140 ***	13.81	0.136 ***	14.14	0.161 ***	31.08
Year
2020	−	−	−	−	−	−	−0.034 ***	−15.03
2021	−	−	−	−	−	−	−0.030 ***	−13.42
Constant	0.263 ***	37.39	0.183 ***	19.35	0.192 ***	21.20	0.243 ***	49.83
var(e)	0.036		0.025		0.027		0.031
Number of observations	19,649		8896		9626		38,174
LR chi2(22)	2337.5 ***		1004.3 ***		1011.3 ***		4402.8 ***
Log-likelihood	4748.7		3743.3		3686.7		11,839.93

Note: ***—statistically significant at 1%, **—statistically significant at 5%, *—statistically significant at 10%; Sources: GHS datasets (2019–2021).

presents the results of the Tobit regression of the determinants of the fuzzy MEPI. The results showed that the estimated models produced a good fit for the data since the likelihood ratio Chi-Square parameters were all statistically significant (p < 0.01). Therefore, the estimated parameters in the models were not jointly equal to zero and the research hypothesis should be rejected. The results indicate the statistical significance of many of the included variables and that the signs of the parameters remained consistent over the study period and in the pooled results. Aside from KwaZulu-Natal and North West provinces in 2020, all estimated models showed that the parameters for the provinces were statistically significant (p < 0.01) with negative signs. The results in Table 3 show that holding other variables constant, the fuzzy MEPI among respondents from Eastern Cape significantly decreased (p < 0.01) by 0.048, 0.020, 0.035, and 0.038 in 2019, 2020, 2021, and the combined analysis, respectively, compared to those from Western Cape. The fuzzy MEPI among respondents from Northern Cape decreased by 0.039 in 2019, 0.005 in 2020, 0.026 in 2021, and 0.028 in the combined datasets compared to those from Western Cape.

Table 3 also shows that in comparison to those from Western Cape, the fuzzy MEPI of respondents from Free State significantly decreased (p < 0.01) by 0.050 in 2019, 0.036 in 2020, 0.044 in 2021, and 0.045 in the combined dataset. The fuzzy MEPI of respondents from KwaZulu-Natal significantly decreased (p < 0.01) by 0.032, 0.026, and 0.023 in 2019, 2021, and the combined dataset compared to those from Western Cape. The fuzzy MEPI of respondents from North West province significantly decreased (p < 0.01) by 0.038 in 2019, 0.025 in 2021, and 0.029 in the combined dataset compared to those from Western Cape. Compared to those from Western Cape, the fuzzy MEPI of respondents from Gauteng significantly decreased (p < 0.01) by 0.040, 0.029, 0.043, and 0.038 in 2019, 2020, 2021, and the combined dataset, respectively. Table 3 also shows that the fuzzy MEPI of respondents from Mpumalanga significantly decreased (p < 0.01) by 0.038, 0.049, 0.027, and 0.037 in 2019, 2020, 2021, and the combined dataset, respectively, compared to those from Western Cape. Finally, compared to respondents from Western Cape, the fuzzy MEPI of those from Limpopo significantly decreased (p < 0.01) by 0.056 in 2019, 0.023 in 2020, and 0.037 in 2021. In the combined dataset, the fuzzy MEPI of respondents from Limpopo significantly decreased (p < 0.01) by 0.042 compared to those from Western Cape.

The results in Table 3 further show that across all the models, the parameters of the racial groups had negative signs and were statistically significant (p < 0.01). Specifically, Table 3 shows that, on average, the fuzzy MEPI of Coloured respondents was significantly lower than that of Black respondents by 0.043, 0.031, 0.042, and 0.042 in 2019, 2020, 2021, and the combined dataset, respectively. The fuzzy MEPI of respondents of Asian origin decreased by 0.087, 0.044, 0.040, and 0.069 in 2019, 2020, 2021, and the combined dataset, respectively, compared to Black respondents. Table 3 also shows that in comparison to Black respondents, the fuzzy MEPI of White respondents significantly decreased (p < 0.01) by 0.106 in 2019, 0.040 in 2020, 0.056 in 2021, and 0.084 in the combined dataset.

Furthermore, in line with expectations, the fuzzy MEPI of male respondents significantly decreased (p < 0.05) by 0.013 in 2019, 0.009 in 2020, 0.009 in 2021, and 0.011 in the combined dataset compared to that of female respondents. The number of children aged under 5 increased the fuzzy MEPI in all the estimated models. Specifically, as the number of children aged under 5 increased by one unit, the fuzzy MEPI significantly increased (p < 0.05) by 0.009 in 2020, 0.007 in 2021, and 0.006 in the combined dataset. Moreover, an increase of one unit in the number of household members aged 5 to 17 years significantly reduced (p < 0.01) the fuzzy MEPI by 0.017 in 2019, 0.008 in 2020, 0.010 in 2021, and 0.013 in the combined dataset. Additionally, a unit increase in the number of respondents aged above 60 years significantly reduced (p < 0.01) the fuzzy MEPI by 0.023 in 2019, 0.020 in 2020, 0.024 in 2021, and 0.022 in the combined dataset.

Table 3 shows that in line with expectations, the parameters for monthly salary in all the estimated models were statistically significant (p < 0.01) with negative signs. These results showed that an increase in monthly salary by one unit will significantly reduce (p < 0.01) the MEPI over time. Similarly, as expected, the parameters for social grants were statistically significant (p < 0.01) with positive signs in all the estimated models. These results imply that an increase in social grants by ZAR 1000 will result in an increase in the MEPI. In addition, in line with expectations, the parameters for tribal and farm areas in the estimated models were statistically significant (p < 0.01) with positive signs. Specifically, the fuzzy MEPI of respondents who lived in tribal areas significantly increased (p < 0.01) by 0.046 in 2019, 0.049 in 2020, 0.049 in 2021, and 0.048 in the combined dataset compared to those in urban areas. The fuzzy MEPI of respondents who lived in farm areas significantly increased (p < 0.01) by 0.180 in 2019, 0.140 in 2020, 0.136 in 2021, and 0.161 in the combined dataset compared to those living in urban areas.

4. Discussion

It was found that the MEPI decreased between 2019 and 2020 before it slightly increased in 2021. In addition, in the combined results, the parameters for the 2020 and 2021 dummy variables implied a significant decrease in the MEPI in 2020 and 2021 compared to 2019. It should be noted that since 2008, the electricity supply in South Africa has been characterized by load-shedding. Load-shedding impacts different economic sectors negatively with undertones of aggravated welfare deprivation and inequality [56]. This finding, however, underscores the positive impact of South African energy policies and the expansion of the energy infrastructure on access to clean energy before and during the COVID-19 pandemic [57]. More importantly, the government is now seeking a sustainable energy development pathway post-pandemic through the promotion of Eskom efficiency to ensure a drastic reduction in load-shedding [58]. South Africa has made significant progress in access to electricity from 36% in 1994 to 87% in 2012 [56]. A target of 97% electrification was set by the Department of Energy for 2025 with the goal of connecting an additional three million households to the grid and another 300,000 households to solar energy [56].

KwaZulu-Natal, Western Cape, and North West provinces had the highest average MEPI. Additionally, the MEPI was lower in other provinces than in Western Cape. Although access to electricity is just one of the twelve indicators that were utilized to compute the MEPI, the findings can be juxtaposed with a report that indicated that access to electricity was lowest in KwaZulu-Natal, North West, and Gauteng provinces [59]. These findings also show that Limpopo had the lowest average MEPI, which can be linked to a report that indicated it has the highest electricity coverage in South Africa [59]. The findings can also be compared to those of Ismail and Khembo [39], which showed that residents in Gauteng and Limpopo reduced the MEPI among South African provinces. Additionally, Mbewe [60] found that Northern Cape and Western Cape provinces made the lowest contributions to the number of energy households, whereas Eastern Cape and Gauteng provinces made the highest contributions.

The results also showed that Black households were the most energy poor among the races in the surveys. The results are consistent with the findings of Ismail and Khembo [39]. These findings can be explained by the fact that poverty is still concentrated among Black South Africans despite several years of economic interventions by the government [61]. More importantly, inequality of access to economic resources, lapses in human capital development, and locational factors can explain the differences in poverty among the different races in South Africa [61]. In addition, the MEPI of female-headed households was significantly lower compared to male-headed households. This finding underscores certain gender dimensions in energy poverty in South Africa [35] and is inconsistent with the findings of Ismail and Khembo [39] and Annecke [62]. The finding is, however, consistent with the findings of Tchereni, et al. [44]. The dependence on dirty and hazardous domestic energy sources among male-headed households, as well as a generally high MEPI, may also be caused by other economic deprivations. In many instances, household heads would need to prioritize the purchase of food and other essentials, whereas allocations for energy can be cut back [63,64,65].

The results also showed that as the age of household heads increased, the MEPI significantly decreased. This gives some indication that younger household heads were poorer. This is contrary to previous studies that found that age either showed statistical insignificance [66] or was positively associated with energy poverty [67]. The literature emphasizes the importance of the household head’s age in explaining energy poverty with the notion that younger household heads are expected to be more energy secure [68,69]. However, the contextual aggregation of certain welfare attributes into the MEPI, as carried out in this study, is different from a unidimensional poverty estimation that focuses on energy expenditure and its insecurity, as adopted in some other studies.

In addition, the composition of household members also significantly influenced energy poverty. The results indicate that although the number of household members aged under 5 years increased energy poverty, the number of household members aged 5–17 years and >60 reduced it. The results are in tandem with the findings of Drescher and Janzen [70]. Specifically, the positive association between the number of children aged under 5 and the MEPI may be a reflection of the expected high financial commitments needed to take care of these children and the fact that parents with many household members would prioritize the food needs of the children. In addition, the number of household members aged above 60 years reduced the MEPI. This reflects the higher likelihood of households in this category having many of the electrical assets that were used in this study.

The results further showed the role of income in the reduction of the MEPI. This was expected because high-income earners will prioritize the utilization of clean energy and some electrical assets. Some authors have highlighted the negative correlation between energy poverty and household income [41,48]. However, income from social grants increased energy poverty because of the very high likelihood that the recipients of such grants do not spend them on the provision of clean energy. It should also be noted that the recipients of social grants in South Africa are largely selected based on certain welfare deprivation parameters such as unemployment, disabilities, and low income. In a related manner, households from tribal and farm settings had a higher MEPI compared to their urban counterparts. This was expected due to the prevailing poverty in tribal and farm areas that would often compel the utilization of unclean energy sources. Moreover, dirty energy sources such as wood and animal dung are always in abundant supply in tribal and farm areas.

5. Conclusions

Access to clean energy is a development indicator that is perpetually linked to several SDGs. The aftermath of several climatic hazards and environmental pollution currently deemphasizes electricity generated from coal. Given the high reliance of South Africans on this crude technology, significant investment will be required to shed electricity production from coal to other efficient renewable sources. The current deficits in power generation that have culminated in national load-shedding provide some insights into the looming crises that may befall the South African energy sector. This study analyzed the multidimensional energy poverty index using the fuzzy set approach. The results showed some fluctuations in the level of deprivation with respect to certain energy poverty attributes, with space heating using clean energy being one of the most deprived. The findings from this study reemphasize the spatial interventions needed to identify the energy poor with a greater focus on tribal and farm settings. Similarly, Black South Africans require more interventions to catch up with other races in terms of access to clean energy. Interventions that create employment and reduce the reliance of the majority on social grants promise significant steps towards reducing multidimensional energy poverty. There is also the need to promote equity of access to clean energy across genders, with interventional aid provided to male-headed households.

Finally, this study is limited by the COVID-19 pandemic, which led to a drastic reduction in the number of respondents for the 2020 and 2021 datasets. Moreover, the quality of the data in the 2020 and 2021 periods may have been affected due to the use of telephone interviews, unlike the 2019 data that were collected through face-to-face interviews. Besides the data limitations, further studies may consider a comparison of the results from some notable methodologies in multidimensional poverty analyses to ascertain their uniformity in energy policy implications and robustness.