Can Energy Efficiency Promote Human Development in a Developing Economy?

Abstract

It has recently been underscored that access to energy has adverse impacts upon human development in South Asia. In this paper, we apply different variants of the autoregressive distributed lag (ARDL) model to explain how improved access to energy might adversely impact human development in India over 1980–2018. From the basic ARDL model, a 1% increase (decrease) in energy efficiency will increase (lower) human development by 6.1% in the long run. We note that the causality runs from energy efficiency to human development. The application of the novel dynamic ARDL simulations offers two insights; first, it confirms the importance of energy efficiency for driving human development. Secondly, it shows asymmetric effects: we find that a 10% increase in energy efficiency boosts human development from 7% to 12% in the long run, while a 10% decrease in energy efficiency lowers human development from 7% to 3%. Using the frequency domain causality analysis, we establish that energy efficiency drives human development in India. We also explore the symmetric and asymmetric impacts of several control variables on human development in India. Our findings establish that energy efficiency will not only help India reduce its environmental footprint but also propel human development.

1. Introduction

Over the last decade, energy has posed a serious challenge to sustainable development. On the one hand, [1] highlight endemic energy poverty, especially in the developing world, as several billion people across the globe live without electricity. [2] argue that energy poverty has become a “global phenomenon with unprecedented economic, political, and social concerns”. Hence, under Sustainable Development Goal 7 (SDG7), access to energy is labeled as a key ingredient for promoting economic development [3] and pro-poor growth [4]. On the other hand, it has been long argued that clean energy and energy efficiency are the major weapons to fight climate change by reducing carbon footprints [5,6,7,8,9,10]. As [11] note, although international agencies such as the World Bank stress the role of energy efficiency for mitigating energy poverty, little progress has been made to “demonstrate the value of energy efficiency”. Our goal is to demonstrate the value of energy efficiency in the energy sector for promoting human development to fight energy poverty.

The principal target areas of SDG7 are threefold: energy access, energy efficiency, and clean (renewable) energy. Our main contribution is to establish the primacy of energy efficiency over energy access for promoting human development. In other words, with deteriorating energy efficiency, improvements in energy access might fail to mitigate energy poverty. We seek to establish, for the first time—to the best of our understanding, a missing link between energy efficiency vis-à-vis human development using the experience of India.

For exploring this missing link, we have applied the standard ARDL (autoregressive distributed lag) model and extend our analysis by using the novel dynamic ARDL simulations (dynardl) model and the frequency domain causality test. The dynardl is a machine learning-based algorithm for testing cointegration after controlling for nonlinearity, additivity and heterogeneity [12]. We thereby highlight the inner dynamics of energy efficiency as a major determinant of the effects of access to energy upon human development. The dynardl model can simultaneously explore short-run and long-run relationships, in both levels and differences in variables of interest, and extract the individual effects of explanatory variables on the dependent variable. It also offers visual confirmation of the consequence of a counterfactual change in an explanatory variable on the dependent variable holding all other explanatory variables unchanged (ceteris paribus).

Energy’s special role in human well-being has, hence, piqued serious interests among researchers in recent years [5,13,14,15,16,17,18,19,20,21,22]. Although access to energy has been labeled as a key driver for promoting human development, in their recent work, [13] highlight a counter-intuitive finding that improved access to sustainable energy and electricity adversely impacted human development in South Asia. However, human development in countries from other developing regions—such as sub-Saharan Africa and Caribbean-Latin America—have immensely benefitted from increased access to energy and -clean energy.

The goal of this paper is to revisit the aforementioned counter-intuitive finding by examining the implications of energy efficiency for human development in India that homes 75% of South Asia’s population and generates 77% of the regional GDP. We will apply the time series analysis instead of a panel model of the previous literature to avoid the panel bias raised by [23] (It is useful to highlight that economic development and per capita GDP change slowly over time—if not at a glacial pace. The average per capita GDP of our globe takes about four (4) decades to double, while developing countries experience much slower increases in their per capita GDP. As an example, countries in Sub-Saharan Africa take 93 years to double their per capita GDP, while India and South Asia take about two decades to achieve the same feat. The per capita income takes about four (4) decades to double for nations in the Latin America and Caribbean group. Hence, instead of a short panel dataset, with two decades of data, it is advisable to use long time series data for capturing the dynamics of economic development. Our dataset of about four (4) decades can offer useful insight for India and South Asia. To uncover potential effects of energy efficiency on development through a variety of potential transmission channels, large panel data models capture both within-country and cross-country variation. Yet, short panel data models reduce the majority of variations in the data and, thereby, worsen measurement error biases. Hence, short panel models lose efficacy even though long panel data models are much more effective than time series models [24]. Our choice of the time series modeling is dictated by unavailability of a long panel dataset). In doing this, we will highlight the implications of efficiency in the generation of electricity—hitherto neglected in the current studies—to explain why measures of access to energy might not create a silver bullet for promoting human development in India. On the one hand, improving energy access lowers energy poverty and promotes human development. On the other hand, deterioration in energy efficiency lowers human development and adversely impinges on energy poverty. Thus, the net effect depends on the relative strengths of these two opposite effects.

In other words, if improvements in access to energy are accompanied with decreases in energy efficiency—we argue—improved access to energy will have a deleterious effect on human development. Against the above backdrop, our paper assesses the nature of the relationship between per capita GDP in India, as a proxy for human development, vis-a-vis efficiency in the electricity sector, growth in labor and capital inputs in the electricity production along with a control variable (globalization) in India over a time period from 1980 to 2018.

The remainder of the paper is laid out as follows: We provide a literature review in Section 2. The data and methodology part of our study is discussed in Section 3. In Section 4, we summarize the empirical findings before offering policy recommendations and conclusions in Section 5.

2. Literature Review: Energy Vis-à-Vis Human Development and Per Capita GDP

Human development, in the context of energy economics, is a ‘multidimensional concept’ since human development is critically predicated on health and educational outcomes as well as the GDP of a country [13]. The per capita GDP of a country remains a key driver of human development and, hence, economic growth acts as a spur to human development. [25] argues—building upon the earlier work of [26,27]—that energy, being a key factor of production like labor and capital, promotes economic growth.

The importance of energy in human development has long been recognized since access to quality energy services improves productivity, impacts health and educational outcomes and also helps the diffusion of communication technologies—as noted in the early work of [26,27]. According to [13], for fully modeling human development, it is imperative to understand the impacts of access to energy on human development. Increases in economic growth—above the population growth of a country—will result in a rising per capita GDP. Increases in per capita GDP will trigger improvements in living standards and, thereby, advance the quality of life for millions of people in the developing world. Such advances will lead to improved health and educational outcomes and human development see [28]: the rising purchasing power of people—due to increased rising per capita GDP—will enable people to substitute away from inefficient, polluting and hazardous energy sources. It has been noted by the [29] that indoor and outdoor pollutants are a major cause of respiratory illness killing 600,000 children in the developing world. Thus, the energy-GDP growth nexus and its impacts on the per capita GDP are important for understanding the role of energy in driving human development. Yet, a relatively unexplored issue is the role of energy efficiency—for improving access to energy –to promote economic growth and, thereby, human development.

We will consider the linkage between energy efficiency and per capita GDP since per capita GDP impacts human development. The nexus between energy use and economic growth plays an important role in the determination of the per capita GDP: if the increased energy use fails to raise the economic growth of a country above its population growth, increased energy consumption will fail to improve its human development.

2.1. Energy Efficiency and Access to Energy: Energy Poverty and Human Development

The role of energy efficiency in promoting global sustainability is discussed in [30,31,32,33]. Ref. [33] highlight various methodologies for measuring energy efficiency in high energy-consuming industries to lower carbon footprints. [34] utilized state-level data for the US to explore how fuel types in the generation of electricity and electricity consumption of the residential sectors impact carbon emissions and what role energy efficiency has to capture the adverse effects of energy use on carbon footprints. [35,36] re-evaluated the load forecasting models to highlight how better forecasts can improve US energy efficiency and, thereby, lower carbon footprints and achieve sustainability. [37] examine the effect of energy efficiency on the competitiveness of small and medium-sized enterprises in northern Italy. They advocate for the targeted measures of popularizing energy efficiency for non-energy benefits as a means of improving mitigation strategies.

Access to energy is critically enmeshed with energy poverty and, hence, increases in per capita GDP can have a lasting influence on economic and social development. Due to the nexus between energy poverty and access to energy, we posit that the efficacy in the production of energy can influence per capita GDP and, thereby, determines whether increased access to energy will bring out increased human development. In a handful of recent work, this link between energy and per capita GDP has been seriously examined. In their work, [38], using a panel of seven (7) South Asian countries over 1995–2017, note that energy poverty bears a long-term relationship with an index of economic development, measured by the per capita GDP, in the region. The main corollary of the recent work is that energy poverty, by lowering per capita GDP, can worsen human development if access to electricity and clean energy fails to be ‘affordable’. In this context, we argue that energy affordability is dependent partly on the per capita GDP. Thus, it is imperative to understand the nexus between energy and the per capita GDP to explore the effects of energy access and energy production on human development in the region. In their work, [39] posit that industrial value-added is an important factor for promoting economic development. For exploring the nexus between economic development and energy use, they model the relationship between per capita GDP, as a measure of development, and electricity consumption. In related work, energy poverty has been found responsible for triggering and perpetuating deeper economic and social problems, compromising social well-being, health outcomes and productivity growth, among others [17,18,22,40,41,42,43]. In an earlier study, [2] link energy poverty to energy affordability by defining energy poverty as the incapability of households to meet their domestic energy needs. They further argue that the affordability of energy can be a major problem for both developing and developed countries and note that 40% of the households in some European countries fail to purchase adequate energy to meet their energy needs see [2].

2.1.1. Economic Growth–Energy Nexus

In the existing literature on the relationship between economic growth and electricity/energy consumption, four hypotheses have emerged [44]: (i) neutrality hypothesis, (ii) conservation hypothesis, (iii) growth hypothesis, (iv) feedback hypothesis. Therefore, if there are unidirectional causality flows from energy consumption to economic growth, energy policies are important to promote economic growth and economic development. Yet, the findings from India are mixed: by using the Granger causality approach, [45] shows that there is a unidirectional causality that flows from per capita GDP to electricity consumption per capita in India during the period 1950–1996. Using the same methodology, [46]; Ref. [47] have found that the direction of causality is from electricity consumption to economic growth. In other words, energy policies can drive economic development in India. This finding from India is in consonance with the observations made on other countries: [46,48,49,50,51,52] argue that electricity consumption leads to economic growth, i.e., the existence of the growth hypothesis was accepted by them. Yet, for other group of countries, [53,54,55,56] have shown the tenability of the neutrality hypothesis, i.e., no causal relationship between economic growth and electricity consumption. Some studies also reported the bidirectional causality see [34,40,57,58] between energy use and economic growth. In his study, [59] finds support for the neutrality hypothesis and argues that as the energy cost is negligible in Tunisia. Tunisia’s economic growth is not driven by energy costs and energy policies. Therefore, the direction of causality between the energy use and economic growth is not conclusive; rather, it is conflicting due to the heterogeneity in data as well as in the methodology. In early studies, the importance of the nexus for policy makers has been noted. [60,61] have demonstrated that an assessment of the true direction of the causal relationship between energy use and economic growth is essential for the policymakers to enact the proper energy policy for achieving sustainable economic growth.

We posit that the causal relationships between human development and energy efficiency in the production of energy is an important element for designing suitable energy policies to improve human development. Thus, it is imperative that policy makers have a prior knowledge of the pattern and trend of energy efficiency in in the production of electricity for devising appropriate investment strategies to boost energy supply. Our paper aims to fill this gap in the literature by considering human development, efficiency in electricity generation, increased uses of labor and capital inputs in the Indian electricity sector, and globalization in a single econometric framework.

2.1.2. The Role of Electricity in the Indian Context

Energy, more specifically electricity, as an input of production has been considered as a major driver of economic growth and development starting from the early work of [62]. Energy uses can also improve the productivity of other inputs of production and technology in general, thereby spurring economic growth and development. India has been at the crossroads in transitioning its energy sector to a lower emissions future with the purported goal of ensuring the robust delivery of reliable and affordable energy to customers for improvements in human development. It is highly unlikely, amidst the continuing financial crises for electricity generators since 2015, that gains in energy efficiency in the generation of electricity in India can be maintained due to the new electricity distribution system (Toward this end of improving human development, the privatization of the electricity distribution sector was initiated by launching electricity distribution companies (DISCOMs) in 2015. Within a few years, the financial health of DISCOMs plunged, and the distributors of electricity are unable to pay the electricity generators. Consequently, DISCOMs are not able to manage their massive losses—as of May 2020, DISCOMs’ overdue payments to generators rose to US$16b. In response to the mounting losses, the Government of India injected US$12.03b in 2021 to keep the DISCOMs afloat.) In this paper, we examine the adverse effects of potential decreases in energy efficiency, in the electricity sector, on human development in India. India is a major producer and user of electricity in the globe. Indian policy makers are also aware of the role of renewable electricity for managing the ecological footprints of the electricity sector: roughly about 37% of the installed capacity of electricity generation in India is from renewable sources such as hydroelectric, nuclear and thermal plants. During the fiscal year (FY) 2019–2020, India produced 1383.5 TWh by utilities, and the total electricity production at 1598 TWh is the third largest in the world. For the FY 2019–2020, the per-capita electricity consumption at 1208 kWh is low compared to the international standard. With increased uses of modern technologies—both for production and consumption—Indian policy makers anticipate a massive increase in the demand for electricity.

Given the installed capacity of electricity generation—against the backdrop of a massive spurt in the demand for electricity—the poles and wires are not a valid measure of access to electricity in India. Due to massive spikes in excess demand for electricity during business hours, providers are forced to adopt ad hoc rationing regimes for allocating electricity. As a result, electric poles and wires in most areas of India are often without electricity for hours during the peak hours. Moreover, when electricity is available, many households are priced out, as they cannot purchase adequate energies for their household needs.

To resolve the energy access problem, India will have to increase its capacity of electricity generation with a clear perspective on the purchasing power of households. Rapid increases in the capacity of production calls forth innovation, invention and diffusion in technologies, and they can be easily understood by the intensity and growth of the inputs of production and effect of globalization on that particular sector. Yet, the recent issues with DISCOMs highlight that access to electricity—without an adequate purchasing power for consumers and producers—creates energy poverty, as users will be unable to pay their bills. Consequently, distributors will fail to pay their dues to the generators. The energy system will be unsustainable.

3. Data and Methodology: The Nexus between Electricity Sector and Per Capita GDP in India

The following estimation model, modifying the model developed [63] and [13], for the Indian case is expressed as the following:

X_2t = f (EFF_t, GEL_t, GEK_t, LX_1t)

(1)

In Equation (1), time is indexed by t, and the per-capita GDP (X₂) for India is postulated to be a function efficiency in electricity generation (EFF), GEL is growth in labor input, GEK is growth of capital input for generating electricity in India. LX₁ is the KOF index of globalization in India by a year see [64] (The extant models specify in Equation (1) that human development—proxied by either human development index or human capital or health and education outcomes—is a function of economic growth, energy variables and other control variables (see [13], Equation (3), on p. 3). The control variables are meant to represent important structural changes within an economy due to financial development, urbanization, foreign direct investment, trade, and remittances. For the purpose of a parsimonious model, we choose the KOF globalization index as an explanatory variable that captures the inner dynamics of a country and its evolving relationships with the rest of the world). This index of globalization is a summary measure of economic, political and social dimensions of transformation due to globalization in a country see [64]. Annual data on the Indian economy over the period 1980 to 2017 are collected from the Reserve Bank of India (RBI)’s Handbook of Statistics, RBI-KLEMS database. Equation (1) is linearized and written as—after converting X₂, GEK, and GEL into their (natural) logarithmic values:

X_2t = γ₀ + γ₁ EFF_t + γ₂ GEL_t + γ₃GEK_t + γ₄ LX1_t + u_t

(2)

In Equation (2), the time component is presented as ‘t’, the disturbance term is presented as u_t, and the parameters that need to be estimated are γs.

Note that the HDI, which has been used by [13], is available from the early 1990s. On the contrary, the per capita GDP for India is available for about four (4) decades. Hence, we choose the time series analysis to understand the long-time dynamics of developmental changes as explained in introduction.

3.1. Estimation Strategy

The short-run and long-run relationships between the posited variables in Equation (2) will be investigated using the autoregressive distributed lag (ARDL) model. We will then use the ARDL limits testing strategy of [65] after establishing that the data series has the essential statistical qualities, because this (ARDL) methodology can handle both stationary and non-stationary variables (integrated of up to order 1 or even fractionally integrated). Because they are resilient to the potential of autocorrelation, this method can produce unbiased and efficient estimators. Two key steps comprise the ARDL approach: As suggested by the existing theory, we will first analyze whether there is a long-run relationship involving our variables of interest, after which we will evaluate the key parameters and the framework for error correction in step 2 if a long-run association exists.

3.1.1. Basic Statistical Properties of the Data Series:

Unit Root Tests: To check if our variables of concern had any unit roots, the ADF test and PP test were run as per the modified Akaike Information Criterion (AIC) and as reported by [21]. Now, assuming the variables are stationary at the level (I(0)) or first difference (I(1)), the ARDL limit testing technique can be utilized to find cointegration.

Table 1. Descriptive statistics of the chosen variables.

Tests	X2	EFF	GEL	GEK	LX1
Mean	6.711728	0.814003	0.356346	0.423923	3.712708
Median	6.663949	0.814159	0.417855	0.434673	3.621348
Maximum	7.594553	0.895847	0.704378	0.699802	4.131051
Minimum	6.047145	0.771561	0	0	3.430381
Std. Dev	0.463878	0.030415	0.256177	0.165195	0.265169
Skewness	0.311098	0.561938	–0.28758	–0.20378	0.398794
Kurtosis	1.87854	2.834384	1.525773	2.492414	1.515318

shows that all series are I(0) or I(1), indicating that the ARDL approach is the best fit.

3.1.2. ARDL Modeling for Estimation

According to [65], the ARDL model’s error correction representation is as follows:

$$\mathsf{\Delta }{\mathrm{X}}_{2\mathrm{t}}={\alpha }_0+{\sum }_{i=1}^p {\alpha }_1 \mathsf{\Delta }{\mathrm{X}}_{2\mathrm{t}-\mathrm{i}}+{\sum }_{i=0}^q{\alpha }_2 \mathsf{\Delta } {\mathrm{EFF}}_{\mathrm{t}-\mathrm{i}} +{\sum }_{i=0}^q{\alpha }_3 \mathsf{\Delta } {\mathrm{GEL}}_{\mathrm{t}-\mathrm{i}} +{\sum }_{i=0}^q{\alpha }_4 \mathsf{\Delta } {\mathrm{GEK}}_{\mathrm{t}-\mathrm{i}} +{\sum }_{i=0}^q{\alpha }_5 \mathsf{\Delta } {\mathrm{LX}1}_{1\mathrm{t}-\mathrm{i}}+{\beta }_1 {\mathrm{EFF}}_{\mathrm{t}}+{\beta }_2{\mathrm{GEL}}_{\mathrm{t}}+ +{\beta }_3 {\mathrm{GEK}}_t+{\beta }_4 {\mathrm{LX}}_{1\mathrm{t}}+{\epsilon }_{\mathrm{t}}$$

(3)

where the intercept term is represented by α₀. In this regard, the short-run coefficient for the mentioned variables in our study are presented by α_i (i = 1, 2, …, 5), and the parameter βi (for i = 1, 2, …, 5) is the long-run coefficient; lastly, the disturbance term is presented by ε_t.

Hence, the ARDL bound testing approach permits us to construct a model with the presence of both I(1) and I(0) variables together.

Now, the null hypotheses (H0) can be illustrated as: α₁ = α₂ = … = α₅ = 0. Therefore, the above-mentioned null hypothesis suggests the presence of no cointegration; on the other hand, the alternative one asserts for a cointegration. In other words, alternative hypothesis (H1) postulates that a minimum one parameter of αi is not zero. The Wald test will be used to compare the F-statistics to the critical values of [65]. Now, if the computed F-statistics are greater than the critical value’s upper bound, the ARDL mechanism detects cointegration.

The ARDL model in the preceding equation implies a linear combination of dependent and independent variables, implying a symmetric adjustment of y to any shock in our interest variables in both the long and short run. [65] developed a linear cointegration autoregressive distributed lag model (ARDL) to simultaneously examine long and short-run impacts. In their model, both increases and decreases in the independent variable symmetrically affect the dependent variable. To do this, we employ the ARDL bounds testing method of [65]. Because the coefficient estimates in the presence of cointegration exhibit the super consistency quality, we do not need to address the problem of endogeneity between variables when utilizing the time series approach [66]. To put it another way, endogeneity has no bearing on the ARDL results. Furthermore, even if there are omitted stationary variables, the estimates’ super consistency characteristic persists see [67,68].

3.1.3. Novel Dynamic ARDL Simulations: An Extension

The intricacies of the auto regressive distributed lag model have been well-articulated in current work see [39]. To overcome these complexities, [69] introduced the novel dynamic ARDL model to use simulations to capture the impact of regressor variations on the independent variable see [12]. The simulations will also offer diagrams to enable us to visually confirm the altered time profile of the independent variable following changes in the regressors. The novel dynamic ARDL simulations will be based on the model, using Equations (2) and (3), given by Equation (4) as postulated by [69]:

∆X_2t = α₀ + α₁ ∆EFF_t + α₂ ∆GEK_t + α₃ ∆GEL_t + α₄ ∆LX_1t + β₁ EFF_t-1 + β₂ GEK_t-1 + β₃ GEL_t-1 + β₄ LX_1t-1 + ε_t

(4)

Note that here, the intercept is α₀. Other α_is are the coefficients for the short run and β_is are the coefficients for the long run of the ARDL model, as stated in the context of Equation (3).

3.1.4. Frequency Domain Causality Test

To examine the direction of causal nexus, we assess the causal linkages between efficiency in electricity generation (EFF) and human development (X₂) through the frequency domain causality test approach. The frequency domain causality test method of [22], as [39] emphasize, will enable us to confirm the stability condition of the model. Regarding the relevant variables of our model, namely—efficiency in electricity generation (EFF) and economic development (X₂)—the equation of the frequency domain causality test is reduced to:

X_2t = θ₀ + θ₁ X_2t-1 + θ₂ X_2t-2 + …. + θ_p X_2t-p + λ₁ EFF_t-1 + λ₂ EFF_t-2 + …. + λ_p EFF_t-p + error_t

(5)

In Equation (5), we have to estimate and examine θ_is and λ_is to assess the postulated causal flows running from EFF to X₂ and error_t is the error term.

4. Results and Discussion

We have presented the descriptive statistics in Table 1, which includes the estimated values of mean, median, maximum and minimum values, and standard deviation on the selected variables.

Unit root tests were performed to determine the order of integration among the variables and prevent any false findings because, as mentioned in the methodology, the bounds test framework is suited for the variables which are either integrated of order zero, I(0), or integrated of order one, I(1). Now, by applying the Augmented Dickey–Fuller (ADF) and Phillips–Peron (PP) tests, the alternative stationarity hypotheses were compared to the presence of a unit root as the null hypotheses. The outcomes of the unit root tests are displayed in

Table 2. Unit root tests.

ADF Test Results
Variables	Intercept				Trend and Intercept
	I (0)		I (1)		I (0)		I (1)
	t-Value	Prob.	t-Value	Prob.	t-Value	Prob.	t-Value	Prob.
X2	3.549883	1.0000	–4.637527	0.0007	0.620532	0.9717	–5.271436	0.0012
EFF	–1.780784	0.3835	–4.066172	0.0032	–1.122112	0.9114	–4.592234	0.0041
GEL	–1.912871	0.3230	–7.406791	0.0000	–1.880689	0.6443	–7.524634	0.0000
GEK	–3.595382	0.0107	–10.83325	0.0000	–3.876973	0.0233	–10.42670	0.0000
LX1	1.120773	0.9970	–3.703450	0.0082	–2.530046	0.3128	–3.996790	0.0178
PP Test Results
X2	15.93985	1.0000	–4.637527	0.0007	0.727703	0.9995	–15.13879	0.0000
EFF	–2.874343	0.0581	–4.051137	0.0033	–1.420426	0.8381	–4.590645	0.0041
GEL	–2.285087	0.1819	–7.195004	0.0000	–2.266149	0.4410	–7.575756	0.0000
GEK	–3.909380	0.0047	–10.83352	0.0000	–4.212910	0.0104	–10.42670	0.0000
LX1	0.625066	0.9886	–3.750291	0.0073	–1.833926	0.6678	–4.068937	0.0150

below.

We discovered that all of the variables chosen in our analysis are stationary after the first difference, with the exception of GEK, which is stationary at I(0). As a result, we can use the ARDL framework to estimate the dependent variable (X₂). The AIC was utilized in this work to identify an appropriate lag for the ARDL model based on the restricted observations. The calculated ARDL model, as shown in

Table 3. Estimated ARDL models and bounds F-test for cointegration.

ARDL Model		F-Statistics	CV 1%		CV 5%
			I(0)	I(1)	I(0)	I(1)
X2, EFF, GEL, GEK, LX1	(4,4,3,4,0)	9.1915024	3.29	4.37	2.56	3.49

, has passed various diagnostic tests (see

Table 4. Diagnostic Analysis.

Diagnostic Test	Chi2 (p-Value)	Result
Breusch–Godfrey LM	0.8934	Serial correlation problem is not found
Breusch–Pagan–Godfrey	0.8728	Heteroscedasticity problem is not found
Ramsey RESET Test	0.7605	Correct model specification
Jarque–Bera Normality Test	0.9326	Normal distribution of the residual

).

The estimated result on the F-test is presented in Table 3 and Figure 1, and it confirms the cointegrating relationship or a long-run relationship between human development vis-à-vis efficiency in electricity generation (EFF), growth rates in labor (GEL) and capital inputs (GEK) in the electricity sector, and globalization (LX₁) of the aggregate Indian economy, which is significant at a 1% level of significance. Now, the ARDL estimates clears all diagnostic tests, according to Table 4.

The ongoing policy changes in the Indian economy might have caused several structural breaks in the data, giving unreliable results. In order to confirm the stability of the short- and long-run coefficients, we apply the CUSUM (cumulative sum) and CUSUMSQ (cumulative sum of squares) tests from [70]. The figures of CUSUM and CUSUMSQ confirm that the test statistics are totally contained within the CV at 5% significance and presented in Figure 2. This indicates that the extracted parameters have remained steady over time and that the statistical results are reliable.

The short-run ECM model and long-run cointegration model are presented in

Table 5. The short-run ECM model.

ECM Regression
Restricted Constant and No Trend
Variable	Coefficient	Std. Error	t-Statistic	Prob.
∆X2t-1	–0.237332	0.143138	–1.658061	0.1195
∆X2t-2	–0.391068	0.129973	–3.008834	0.0094
∆X2t-3	–0.230195	0.105675	–2.178330	0.0470
∆EFFt	–0.017044	0.351853	–0.048440	0.9621
∆EFFt-1	–1.504839	0.380388	–3.956058	0.0014
∆EFFt-2	–0.721881	0.341624	–2.113087	0.0530
∆EFFt-3	–0.529368	0.339985	–1.557035	0.1418
∆GELt	–0.145987	0.023160	–6.303292	0.0000
∆GELt-1	–0.049690	0.027333	–1.817951	0.0905
∆GELt-2	–0.078590	0.026600	–2.954483	0.0105
∆GEKt	–0.017350	0.028731	–0.603891	0.5556
∆GEKt-1	0.070881	0.026647	2.660030	0.0187
∆GEKt-2	0.018036	0.027107	0.665367	0.5166
∆GEKt-3	–0.053284	0.018434	–2.890580	0.0119
ECT/CointEqt-1 *	–0.159171	0.018398	–8.651306	0.0000
R-squared	0.837222	Mean dependent var		0.042773
Adjusted R-squared	0.717280	S.D. dependent var		0.020224
S.E. of regression	0.010753	Akaike info criterion		–5.926756
Sum squared resid	0.002197	Schwarz criterion		–5.253361
Log likelihood	115.7548	Hannan–Quinn criter.		–5.697109
Durbin–Watson stat	2.008306

*: ECT gives the error correction term. It measures how quickly the dependent variable returns to equilibrium after a shock to other variable.

and

Table 6. Long-run cointegration model.

Levels Equation Dependent Variable (X2)
Restricted Constant and No Trend
Variable	Coefficient	Std. Error	t-Statistic	Prob.
EFF	6.314552	2.434584	2.593689	0.0212
GEL	–0.428576	0.161478	–2.654089	0.0189
GEK	–0.569571	0.329425	–1.728984	0.1058
LX1	2.960027	0.708177	4.179784	0.0009
C	–8.578739	4.005812	–2.141573	0.0503
EC = X2 − (6.3146 × EFF − 0.4286 × GEL − 0.5696 × GEK + 2.9600 × LX1 − 8.5787)

Note: EC labels the error correction model.

, respectively.

As the ECT records the annual percentage of correction of disequilibrium, Table 5 finds that approximately 16% error correction takes place every year. In this regard, we have found highly significant and negative ECT; thus, we may conclude that the adjustment process towards equilibrium converges to the long-run equilibrium, which also confirms causality running from EFF (efficiency in electricity production) to X₂ (human development). For the ECM, we undertook the Wald Test to examine the joint significance of the coefficients of this estimated model, which notes that all the variables are jointly significant in short run (see Appendix A.2). In long-run analysis, as presented in Table 6, we find that the independent variables—except GEK—are significant at least at the 5% level. From Table 6, we have the critical finding: ceteris paribus, the efficiency in the generation of electricity sector drives the per capita GDP in India. We note that the per capita GDP (X₂) bears an inverse and statistically significant relationship with GEK and GEL.

4.1. Evidence from Novel Dynamic ARDL Simulations: Discussion

The estimated results from novel dynamic simulated ARDL error correction models are reported in

Table 7. Results of Dynamic ARDL Simulations Analysis.

∆X2	Coef.	Std. Err.	T	p > t
ECT^	–0.18	0.08	–2.25 **	0.04
L1_X2	–0.04542	0.050588	–0.9	0.377
∆_EFF	–0.0547	0.437537	–0.13	0.901
L1_EFF	0.391297	0.166946	2.34 **	0.027
∆_GEL	–0.12877	0.035345	–3.64 ***	0.001
∆_GEK	0.018681	0.032272	0.58	0.567
∆_LX1	0.435469	0.157616	2.76 **	0.01
L1_GEL	–0.01679	0.023562	–0.71	0.482
L1_GEK	–0.04934	0.029101	–1.7	0.102
L1_LX1	0.147595	0.096101	1.54	0.136
_cons	–0.49854	0.178583	–2.79 **	0.01
R-Squared		0.5593	F	3.81
Adj. R-Squared		0.4124	Prob>F	0.0033
Number of Obs.		37	Simulations	1000

Note: L1 posits the first lag of the concerned variable, ∆ represents the first difference mechanism, ***, **, implies 1%, 5% level of significance, ECT ^ implies the error correction term.

: one of the main findings is that EFF increases human development in the long run and not in the short run: a 1% increase in EFF will result in a 0.39% increase in human development, measured by the per capita GDP (X₂), in the long run. The short-run effect is not statistically significant. The effects of EFF from the standard ARDL model, given in Table 5 and Table 6, are different from the findings from the novel dynamic ARDL simulations mainly for the short run. The result is in consonance with the findings of [71]. This result has a special significance for the finding of [13], who note that energy access worsens human development in South Asia. This is feasible, as our work highlights, if energy efficiency declines with rising energy access. We, hence, provide an economic rationale for the findings of [13].

Growth in labor input (GEL) in the electricity sector—from the novel dynamic ARDL simulations—has an adverse short-term impact on human development that is statistically significant. There is no statistically significant impact of GEL on human development in the long run. The impact of employment in the energy sector on overall human development is expected. In the extant literature, there is no prior work to shed light on this effect. Intuitively, the results from the ARDL model for GEL can be justified in terms of low technological frontier see [72,73]: this result implies that, ceteris paribus, increases in production of energy by increasing employment lower energy efficiency and increase the cost of energy production, which results in lower human development. We find that growth in capital input (GEK) in the energy sector, ceteris paribus, has neither long-term nor short-term impacts on human development from the ARDL model. The findings of the novel dynamic ARDL simulations are different from the findings from the standard ARDL models for the long run as well as for the short run. From the novel dynamic ARDL model, we see that GEK has an adverse impact on human development. This result can be justified in terms of the crowding out effect of increases in investment in the energy sector—as shown in [74,75]. Hence, we find that the key driver of human development is energy efficiency in the electricity sector of India.

Globalization (LX₁) has a strong and positive short-run effect on human development, although there is no evidence of any long-run impact. The effect of globalization from the standard ARDL model is different from the predictions of the dynamic ARDL simulations for the long run. The dynamic ARDL model concurs with the findings of [13] that the external sector promotes human development.

The error correction term (ECT) is negative and statistically significant, although the error correction rate is low (about 20%). The R-squared value shows that 55% of the variability in the dependent variable (X₂) is explained by the chosen regressors. The projected F-statistics and the associated P value show that the proposed model is a good fit.

The dynamic ARDL simulations plot the effects of a change in a regressor, ceteris paribus, on the dependent variable. The impulse response plot, as presented in Figure 3, shows human development in India following a 10% increase and 10% decrease in efficiency in electricity generation, EFF, which confirms that the decrease (increase) lowers (increases) human development in the long run. There is a minor decrease in human development in the short run, which is not statistically significant. The dots specify predicted values, whereas the dark blue to light blue line specifies 75%, 90%, and 95% confidence intervals, respectively. The sub-panels of Figure 3 show how an increase (decrease) in energy efficiency (EFF) by 10% increases (decreases) human development (X₂). Figure 4 and Figure 5, respectively, trace the impulse response plot of human development following 10% decreases and 10% increases in growth in labor input and capital input in the electricity sector. Figure 6 shows the impulse response plot of human development following a 10% decrease/increase in the globalization index (LX₁) for India.

Figure 4 shows how a 10% increase (decrease) in growth in labor input (GEL) in the electricity production impacts human development (X₂). Once again, the dots specify average prediction value, whereas the dark blue to light blue line specifies 75, 90, and 95% confidence interval, respectively. Interestingly, ceteris paribus, the first panel of Figure 4 traces the adverse impacts of increases in the use of labor input in the electricity sector on human development. In the short run, there is an adverse and statistically significant, impact on human development, which can be confirmed from the first panel of Figure 4. The long-run effect is not statistically significant. The second panel of Figure 4 confirms the positive impact of a 10% decrease in GEL on human development. This effect is statistically significant in the short run but not in the long run.

Figure 5 traces the path of human development following a 10% increase (decrease) in growth in capital input. In other words, Figure 5 explores the impacts of a 10% increase/decrease in capital input in year t = 10, ceteris paribus, on human development. The effects are not statistically significant.

Figure 6 examines the effects of a 10% increase (decrease) in the index of globalization (LX₁), ceteris paribus in year t = 10, on subsequent human development. The first panel shows that an increase in the globalization index, ceteris paribus, by 10% raises human development in India. The short-run effect is positive and statistically significant, while the effect in the long run is also positive but marginally significant. The effect of globalization is re-confirmed from the second panel of Figure 6 when we consider a 10% decrease in globalization in year t = 10.

4.2. Causality from Frequency Domain Analysis

In this section, we examine the causal relation between electricity efficiency (EFF) and human development(X₂) using the frequency domain causality test. For obtaining the frequency domain causality test results, we apply the Breitung–Candelon Spectral Causality approach [22]. Figure 7 describes the findings from the Breitung–Candelon Spectral Causality. The first panel of Figure 7 confirms that energy efficiency causes human development. The second panel of Figure 7 confirms that there is no reverse causality running from human development to energy efficiency.

We present

Table 8. Results from the Spectral Causality.

Direction of Causality	Very Long Term	Long Term	Medium Term	Short Term	Very Short Term
	ω = 0.05	ω = 0.85	ω = 1.65	ω = 2.45	ω = 3.14
EFF => X2	2.8490 (0.2406)	6.1981 (0.0451) **	2.3144 (0.3144)	0.9292 (0.6284)	1.5618 (0.4580)
X2 => EFF	0.3577 (0.8362)	0.6555 (0.7206)	3.6247 (0.1633)	2.1802 (0.3362)	1.9703 (0.3734)

** implies significant at less than 5% level, in parenthesis, we have presented the p-values with respected to the Wald test statistic.

to document the frequency domain causality as follows:

From the frequency domain causality analysis, we confirm that the direction of causality flows from the efficiency in the electricity production to per capita GDP in the long term. Thus, our results reveal that the electricity sector drives human development—measured by the per capita GDP (X₂). Consequently, if efforts to expand energy access leads to a decrease in EFF and increases in GEK and GEL—such changes in the electricity sector can seriously compromise the per capita GDP (X₂). This effect on X₂ can thereby compromise energy affordability—or purchasing power to buy an adequate energy bundle. This will, in turn, lower human development. This chain of events can be called the vicious cycle of energy access. On the contrary, if improving energy access is achieved by increases in EFF and decreases in GEK and GEL, we are in a virtuous cycle: as improved energy access improves EFF as well as lowers GEK and GEL—the per capita GDP (X₂) rises. Such increases in X₂ leads to improvements in human development—ceteris paribus. The impact of globalization seems to have a positive and highly significant effect on per capita GDP.

5. Conclusions

In order to explain the observation that access to energy lowers human development in South Asia, we take India as the case study. In India, electricity is the most important source of energy and also a special means to fight energy poverty for improving human development. Our departure from the current literature is twofold: first, instead of focusing upon the total energy output, which determines energy access by enlarging the total energy pool, we focus upon the energy efficiency in the electricity sector of India. Secondly, using the per capita GDP as a proxy for human development, we found that energy efficiency in the electricity sector plays a pivotal role in driving economic development in India. If improvements in access to energy are attained using increases in capital input (GEK) and increases in labor input (GEL) with decreases in energy efficiency in the electricity sector, then improved access to energy can trigger decreases in human development.

Even if we do not equate per capita GDP with human development, we can deduct from the research that per capita GDP is a significant driver of human development. If energy efficiency in the electricity industry impacts per capita GDP, then the consequences of energy access depend on changes in energy efficiency in the electricity sector. We discovered two major drivers of per capita GDP from the electricity industry using the conventional methodology as:

• There is a long-term relationship between per capita GDP vis-à-vis energy efficiency in the electricity sector, growth in labor and capital inputs in the electricity sectors.
• Increases (decreases) in energy efficiency in the generation of electricity increases (decreases) per capita GDP while increases (decreases) in the growth in labor and capital inputs, ceteris paribus, decrease (increase) per capita GDP.

The above two observations suggest whether access to energy improves, or compromises, human development will be predicated upon the inner dynamics in the electricity sector. If access to energy is achieved by raising energy efficiency in the electricity sector—along with decreases in the growth in labor and capital inputs (due to increased productivity of these factors), ceteris paribus—access to energy is bound to increase human development. On the other hand, ceteris paribus, if access to energy is accompanied by a lowering of energy efficiency and increases in the growth of labor and capital inputs—human development will decline with improved energy access. We thus highlight the inner dynamics of the electricity sector as a major determinant of the effects of access to energy upon human development. This is because the relationship between energy access and human development will be shaped by the balance between energy efficiency and growth in input uses. The main weakness of our analysis is the mutual dependence of the chosen variables, which can significantly compromise the ARDL models. The frequency domain causality ensures that there is no mutual dependence. The future research will explore time-varying causality.