Journal Pre-proof

Energy poverty and education: Fresh evidence from a panel of

developing countries

Nicholas Apergis, Michael Polemis, Simeoni-Eleni Soursou

PII: S0140-9883(21)00323-6
DOI: [Link]
Reference: ENEECO 105430

To appear in: Energy Economics

Received date: 19 January 2021

Revised date: 4 July 2021
Accepted date: 4 July 2021

Please cite this article as: N. Apergis, M. Polemis and S.-E. Soursou, Energy poverty
and education: Fresh evidence from a panel of developing countries, Energy Economics
(2018), [Link]

This is a PDF file of an article that has undergone enhancements after acceptance, such
as the addition of a cover page and metadata, and formatting for readability, but it is
not yet the definitive version of record. This version will undergo additional copyediting,
typesetting and review before it is published in its final form, but we are providing this
version to give early visibility of the article. Please note that, during the production
process, errors may be discovered which could affect the content, and all legal disclaimers
that apply to the journal pertain.

© 2018 © 2021 Published by Elsevier B.V.

 Journal Pre-proof

Energy Poverty and Education: Fresh Evidence from a Panel of Developing

Countries

Nicholas Apergis1 napergis@[Link], Michael Polemis2 mpolemis@[Link], and Simeoni-

Eleni Soursou3 simonisoursou@[Link]

1
University of Texas at El Paso
2
University of Piraeus, Hellenic Competition Commission and University of Patras
3
University of Piraeus

of
ro
Abstract

The goal of this study is to empirically assess the impact of education on energy poverty
-p
through the lens of the human capital theory. To this end, a sample of 30 developing
re
economies, spanning the period 2001-2016 was used, while GMM estimators were employed
lP

to effectively deal with cross-sectional dependence and endogeneity. The empirical results

clearly document that education generates a negative impact on energy poverty. The findings
na

remain robust under different approximations of energy poverty, incurring significant

ur

implications for policymakers and government officials.

Jo

Keywords: Energy poverty; Education; Energy transition; Emerging economies; Human

capital

1. Introduction

Energy deprivation and poverty alleviation are at the top of scientific interest over the

last years. The lack of access to electrification and more sophisticated energy forms,

especially in emerging economies, has given birth to various studies. The so-called energy

poverty issue is currently gaining momentum with scientific attention around the dimensions,

the spillovers, and the potential responses against the energy poverty phenomenon. Energy
 Journal Pre-proof

poverty constitutes a widespread problem that has hit hard many households all over the

world in the aftermath of the global financial crisis. To give an example of its penetration, it

is argued that, more than 34 million people across Europe in 2018 were unable to afford

proper indoor thermal comfort. Moreover, nearly 2.4 billion people across the globe live

without electricity or have limited access to unreliable electricity grids (Sovacool et al. 2012).

Energy poverty can thus be regarded as a global phenomenon that raises serious economic,

political, and social concerns (Thomson et al, 2016).

of
Under Sustainable Development Goal 7 (SDG7), energy is a major challenge to

ro
achieve economic development (Ahmed and Gasparatos, 2020). In the spirit of the SDG7, the

‘access to affordable, reliable, sustainable, and modern energy’ should be an imperative goal
-p
(Nations, 2020). However, IEA in the latest World Energy Outlook 2020 outlines that the
re
ongoing uncertainties due to COVID-19 tantalize clean energy transition and vulnerable
lP

populations. The ongoing outbreak of the COVID-19 pandemic stresses further the

implications of energy poverty for human wellbeing and climate change as it can delay
na

significant energy investments and expected growth. Thus, the trade-off between energy
ur

poverty, prosperity, and livelihood might expand as access to energy has a pivotal role in

economic flourishing and overall social welfare (Bilegsaikhan, 2016).

Jo

Many scholars have drawn attention to the incidence of energy poverty across specific

countries with relatively low economic growth (Olang, 2018; Khan, 2019). Boardman (1991)

was the first who coined the term “energy poverty” suggesting that if the energy expenditure

of a household exceeds the threshold of 10% of its annual income then, it can be considered

as “energy-poor”. Subsequent studies have attempted to link energy poverty with limitations

in fuel access and the security of affordable and adequate heat in developed countries (Qurat-

ul-Ann and Mirza, 2020). As it is argued in Kahouli, (2020), fuel poverty implies thermal

discomfort and is associated with households’ specific characteristics. Moore (2012), also
 Journal Pre-proof

uses the fuel poverty framework and proposes that affordability and accessibility play a

crucial role, given that households spend a significant amount of income to cover basic

energy needs. Based on this study, a household is typically considered energy-poor, if the

costs of energy exceed 10% -15% of its total income. Amin et al., (2020) investigate the

various ways in which energy poverty affects the development of South Asian countries,

while Adusah-Poku and Takeuchi (2019) explore the changes in energy poverty status in

Ghana among different time periods. Energy poverty concerns mostly emerging economies,

while in the context of industrialized ones is assessed through the prism of fuel poverty

of
(Adusah-Poku and Takeuchi, 2019), or the conservation of energy resources for future

ro
generations (Ogwumike and Ozughalu, 2016; Bilegsaikhan, 2016).
-p
The need to mitigate climate change fosters sustainable development, which is
re
reflected in pivotal socio-economic aspects, such as GDP, employment, FDI, energy
lP

consumption, intensity, education, national income, etc. Lessening energy deprivation not

only enhances social welfare (Bilegsaikhan, 2016) and economic growth, but also the
na

attainment of higher literacy rates. As referred to Energy, Poverty and Development

ur

hypothesis (Karekezi, 2012), school attendance is linked to adequate and affordable energy;

energy is a prerequisite for studying at home, distance learning methods, and the use of
Jo

modern equipment in-situ. Improvements in energy-efficient buildings and access to clean

energy enhance the reduction of educational costs, such as heating and cooling providing

more students the opportunity to continue studying.

Based on the above, we argue that while a few studies examine the educational

aspects of energy poverty, several questions remain unanswered. First, on what theoretical

grounds is based the transmission channel between education and energy poverty. Second,

the absence of a conceptual and methodological framework on this issue has been regarded as

one of the main crucial points hindering the ability of policymakers to tackle energy poverty.
 Journal Pre-proof

Third, concerning the drivers of energy poverty, much of the existing literature sheds light on

household characteristics (e.g., dwelling size, members, income level, energy prices,

electricity consumption) ignoring social and behavioral determinants including the role of

human capital accumulation (education) on improving the access to electricity.

Even though there is a clear link between energy poverty and education especially in

low-income countries, the literature contribution to the topic remains at a premature stage. It

is highlighted that the impact of energy poverty on various economic factors has been

extensively studied during recent years (e.g., growth, sustainable development). However, it

of
is acknowledged that social and behavioral characteristics, such as literacy, vocational

ro
training, trust, and ethnic diversity can be important determinants of energy poverty
-p
(Awaworyi Churchill and Smyth, 2020). The related strand of literature is still in its infancy.
re
The main motivation of this paper compared to other related studies is to demonstrate
lP

the mechanism in which education impacts energy poverty by lending support on a solid

theoretical background. Providing that the severe outcomes of energy poverty on literacy
na

mainly can be seen under the multiple dimensional approaches of energy poverty (Ogwumike
ur

and Ozughalu, 2016; Mendoza et al., 2019; Zhang, 2019), the goal of this study is to

investigate the link between education and energy poverty and thus to bridge the gap in the
Jo

current literature.

This is important for policymakers and government officials especially in emerging

economies for at least two reasons. First, most of the developing economies especially the

low and lower-middle-income countries (i.e., Benin, Kora, Vietnam, Nigeria, Togo,

Bangladesh, etc) have limited access to electricity and other “clean” fuels such as natural gas,

LPG, ethanol, and methanol. We argue that the problem of energy poverty in these countries

is more severe than in high-income emerging countries (Chile, Uruguay among others), while

impacts less the developed economies. The limited access to electricity negatively affects the
 Journal Pre-proof

level of education generating multiplied effects on entertainment, health conditions, comfort,

or even productivity of people (Donev et al., 2018). Better access to power grid networks

provides families with improved education as children could study more easily after dark,

while on the other hand simplifies household tasks accelerating the productivity of home

businesses (see Donev et al., 2018). Second, in many of the emerging countries, efficient

access to education is hindered by political, economic, social, or even religious reasons and

beliefs. This outcome might impose serious threats on access to electricity (electrification

rate) and thus increase energy poverty since there is a direct link between human capital and

of
economic growth (Azariadis and Drazen, 1990; Mankiw et al., 1992; Barro, 2001). As a

ro
result, the more (less) educated people in a country the less (more) severe the energy poverty
-p
since the electrification rate can be indicative of the economic development of a country
re
(Acharya and Sadath, 2019; Donev et al., 2018).
lP

Our study contributes to the literature on many fronts. To begin with, it is the first

study, to the best of our knowledge, focusing solely on the impact of education on energy
na

poverty. For this reason, we utilize a panel data set consisting of 30 emerging economies over
ur

the period 1990-2019 based on the current World Bank Country Classification. The reason

for limiting our analysis to the emerging economies can be attributed to the fact that these
Jo

countries (e.g., South Asia, Sub-Saharan, and East Asia regions) are characterized by

significant weaknesses in their energy systems (e.g., limited access to the energy grid

network, inadequate energy services, energy dependence on solid fuels) contributing to

energy poverty at the national levels. The empirical framework uses pooled econometric

analysis to uncover the main drivers of energy poverty in the sample countries.
 Journal Pre-proof

Second, the theoretical underpinnings of this study rely on the intersection of the

human capital and ‘energy transition’ theories.1 Along this line of reasoning, investing in

education accumulates human capital stimulating economic growth. However, limited access

to “clean” energy resources (renewables, electricity, natural gas) broadly considered as one

aspect of energy poverty, hinders the level of economic growth. Consequently, energy

poverty could be attributed either to the non-availability of “clean” energy inputs compared

to traditional biomass fuels, or non-affordability and illiteracy (Acharya and Sadath, 2019). In

simple words, a better education level, triggers human capital as a share of GDP, facilitating

of
the transmission to “clean” energy fuel sources through appropriate pricing and technological

ro
improvement policies and thus mitigating energy poverty.
-p
The empirical findings based on GMM techniques that control for endogeneity and
re
small sample bias indicate that there is an inversely statistically significant related
lP

relationship between education and energy poverty in all the estimated models. This outcome

coincides with other related studies. We argue that a 10% increase (decrease) in the
na

percentage of primary-school-age children proxied by the adjusted net enrolment rate,

ur

decreases (increases) energy poverty by about 4.2% on average. The other education

indicator, namely the government expenditure on total education over the GDP seems to have
Jo

a slightly larger effect since a 10% increase in total education expenditure accelerates access

to electricity by 4.34% on average. The rest of the covariates (electric power consumption,

government expenditure, economic growth, carbon emissions, and energy intensity) are

statistically significant and plausible signed. Moreover, our findings survive robustness

checks under alternative definitions of energy poverty. The empirical findings are extremely

important for governments and policymakers, bringing to the fore the important role that

education can play in alleviating energy poverty. In a micro-level framework, the

1
The energy transition theories explain how societies switch from traditional to more advanced fuels that are
cleaner and efficient (Grubler, 2012).
 Journal Pre-proof

implementation of suitable policies to boost the household’s energy transition toward more

efficient and clean energy sources paves the way for lessening energy poverty that has hit

emerging economies.

The remainder of the paper unfolds as follows. Section 2 discusses the related

literature presenting the key findings of the previous studies. To ground the concept of energy

poverty, Section 3 develops a theoretical framework for describing the basic interactions

between education and energy poverty. Section 4 illustrates the data and the sample variables,

of
while Section 5 describes the research design. Section 6 discusses the results of the empirical

ro
analysis along with the necessary robustness tests. Finally, Section 7 concludes the paper by

providing some useful policy implications and avenues for future research.
-p
re

2. Literature review
lP

This section provides a conceptual discussion of how education can potentially

na

influence energy poverty. The main potential mechanisms via which education may influence

energy poverty are also discussed. In 2010, over 50 developing countries have tracked
ur

encouraging results in the achievement of basic education. However, energy poverty still
Jo

tantalizes the global education goal. The energy use is necessary for studying at home,

distance learning methods, and the use of modern equipment in-situ. Furthermore,

improvements in energy efficiency and access to clean energy reduce educational costs.

Nevertheless, the literature that contributes to the topic remains at a premature stage.

The incidence of educational deprivation as an unpleasant outcome of energy poverty can be

mainly seen under the multiple socioeconomic dimensions of energy poverty. Notably,

through the approach of Multidimensional Energy Poverty Index (MEPI), researchers assess

the effects of energy deprivation on education and especially about houses’ living conditions.
 Journal Pre-proof

Ogwumike and Ozughalu, (2016) use the headcount ratio and construct a MEPI to assess the

impact of energy poverty on sustainable development in Nigeria. Their results suggest that

72% of Nigerian households lack access to modern energy, whereas over 54% lack access to

grid energy. The educational level of the household head is negatively related to energy

poverty status. By contrast, age, the household’s size, the region, as well as general poverty,

are positively associated with energy poverty.

Geographical characteristics are also examined by Oum (2019) who finds a

of
significant reduction in the energy poverty rate associated with an increase in electricity

ro
connections in rural areas in Laos. Specifically, Oum (2019) studies the impact of energy

poverty on education and health in the case of Lao and finds that energy poverty negatively
-p
impacts the average years of schooling for energy-poor households, whereas demographic
re
and geographic households’ characteristics also play an important role in school attendance.
lP

The findings of this study support that the energy-poor households have a higher illiteracy

level in comparison to that of non-energy poor, whereas energy poverty negatively influences
na

the years of schooling at the 1% significance level. Additionally, the education years are

positively related to “female or older and literate household heads”. The MEPI approach is
ur

used by Mendoza et al. (2019) in the case of the Philippines, who conclude that 9 out of 10
Jo

households lack education and communication-related devices, whereas Ahmed and

Gasparatos (2020) assesses that the energy poverty in different rural areas in Ghana is defined

through the MEPI approach; his findings document that the majority of groups “deprive in

terms of cooking fuel and lighting”, which is reflected on the strong usage of traditional fuels,

like biomass and the absence of electrification.

Differences across the regions using the MEPI can be found in a study by Adusah-

Poku, and Takeuchi (2019). The authors suggest that there is an important increase in both

electricity access and the use of modern energy, though there is a significant distinction
 Journal Pre-proof

between rural and urban households’ energy poverty conditions. Similarly, demographic and

geographic effects ascertain the level of energy poverty in Northern and Southern China

(Zhang, 2019), while Liu et al. (2020) assess the development of rural centralized residence

policies and the choice of cooking fuels in the case of Sichuan Province in China. Their

findings illustrate that the increase in clean energy through the RCR does not augment the

households’ expenditure. Education has a positive effect, indicating that household heads

with higher education tend to choose cleaner cooking fuels.

of
Applying the integration method in their panel data analysis, Amin et al., (2020) and

ro
Liu et al. (2020) also investigate the energy poverty impact on economic growth for South

Asian countries. Their studies analyze the relationship between energy poverty, education,
-p
inflation, income, employment, and economic growth. Their results sustain that in the long
re
run, an increase in education by 1% fosters economic growth by 6.41%. The estimates also
lP

indicate that energy poverty has a negative impact on economic growth, whereas education

influences positively economic growth. Abbas et al. (2020) employ both Tobit and OLS
na

models for seven different Asian countries and they highlight that education has a negative
ur

relation with the socio-economic determinants of MEPI (Abbas, 2020). Finally, Rahut et al.

(2019), following previous studies (Heltberg, 2004; Baiyegunh, 2004; Liu, 2020) sustain that
Jo

higher education results in higher uses of cleaner energy sources.

The existing studies on the education-energy poverty nexus though limited in number

(see Table 1), unravel a negative and statistically significant direct channel between

educational attainment and energy poverty (Koomson and Danquah, 2021; Crentsil et al.,

2019; Sharma et al, 2019; Sharma, 2016; Tewathia, 2014). Specifically, in the earlier study of

Tewathia (2014) the main determinants of energy poverty expressed by the average monthly

household electricity consumption are examined over a survey dataset of 395 Delhi

households. The econometric analysis relying on simple OLS estimations reveals that
 Journal Pre-proof

household income, electrical appliances, family size, dwelling size, time spent out by the

family members, and higher education level are the most important drivers of energy poverty.

The education variables that are used in this study are binary variables taking values zero and

one reflecting the education level of the head of the household (primary, graduate, and post-

graduate level). The education estimates come with a negative sign and only graduate and

post-graduate dummy variables are statistically significant at the 5% level. The empirical

results strongly indicate that higher education level appears to have a negative relationship

with household electricity consumption. The magnitude of the coefficients ranges from -

of
0.281 to -0.306, implying that a 10% increase (decrease) in the education level incurs a

ro
decrease (increase) of the energy poverty level by about 3% approximately. The empirical
-p
findings do not drastically change if the sample is split based on the weather periods
re
(moderate weather, summer, and winter season).

Sharma (2016) argues that in India the increased expenditures on education are
lP

associated with the reduction in energy poverty, mostly in households having lower income.
na

Similarly, Crentsil et al. (2019) indicate that energy poverty in Ghana decreases as the

educational level of the household head improves. It is also argued that the energy poverty for
ur

households with heads who had no formal schooling is higher than for households whose
Jo

heads had attained post-secondary schooling.

Acharya and Sadath (2019), explore the impact of education and economic

development on energy poverty in India between two distinct periods (2004-2005 & 2011-

2012). The authors calculate a Multidimensional Energy Poverty Index by collecting data

obtained from the 25 Indian states and Union territories across the country. The empirical

findings indicate that education exerts a greater impact on reducing energy poverty compared

with income.
 Journal Pre-proof

The most related study to ours is the work of Sharma et al, (2019). This study uses

survey data collected from 1,000 households in Mumbai, equally divided into four income

groups. The sample includes several socio-economic variables, such as the household

expenses on electricity services, the consumption expenditure, the dwelling size, the

education level, and the awareness of the household headcount on energy and environment.

The results indicate that energy poverty proxied by monthly expenditure on electrical energy

by the household mainly depends on the consumption expenditure of the household.

However, this outcome is differentiated across the four different income groups. This study

of
uses two alternative education indicators, namely the number of years of education of the

ro
household head and a dummy variable taking the value of one should the household head is
-p
aware of the benefits of energy conservation, and zero, otherwise. The regression results for
re
the whole sample (all households) unravel that the education level of households does not

impose any statistically significant effect on energy poverty. In other words, a higher
lP

education level of a household does not translate into a reduction of their energy consumption
na

level. This indicates that even the educated households, particularly, despite their high

awareness, maybe either negligent on energy conservation or, due to their higher income,
ur

may not be much bothered about their energy expenses. In contrast, the alternative education
Jo

variable expressing the awareness of the households to the energy conservation measures is

positively and statistically significant correlated with energy poverty, particularly in

households having high and very low incomes.

Lastly, in a subsequent study, Koomson and Danquah, (2021) by utilizing survey data

from Ghana, investigate the impact of financial inclusion on energy poverty. The latter is

proxied by a dummy variable equal to one if the household’s energy deprivation score

exceeds 0.33 and zero otherwise, while the former dummy variable takes the value of one if

the household financial deprivation score is less than 0.5 and zero otherwise. This study uses
 Journal Pre-proof

also multidimensional energy poverty indicators (modern cooking fuel, indoor pollution,

electricity access, and household appliance ownership) to assess the impact of education

along with several other characteristics (gender, marital and employment status, age,

household size, etc) on energy poverty. The impact of education on energy poverty is

examined through the inclusion of a dichotomous variable, taking the value of one if the

household head is educated and zero otherwise. Based on the main OLS findings, the

education level is negatively and statistically significant correlated with the energy poverty

indicator, ranging in its magnitude from -0.148 to -0.185. In other words, a 10% increase in

of
the education level of the household, mitigates energy poverty even by 1.85%. The study

ro
claims that the improvements in education, due to financial inclusion increase the probability
-p
of higher occupational incomes which makes it easier for households to afford energy
re
services, limiting thus energy poverty.
lP
na

Table 1: Selected papers on the key determinants of the education-energy poverty nexus
ur

Dependent Covariates/sign
variable
Study Sample/Co Energy Education Income/eco Ener Carbo Govern Other
Jo

untry Poverty nomic gy n ment variable

Indicator growth intens dioxid expendit s
ity e ure
emissi
ons
Tewat Survey/New Average 1) Binary Monthly - - - 1) Stock
hia Delhi monthly variable average of the
(2014) (India) household equals 1 if household electrica
electricity the income / l
consumptio household positive applianc
n head is es
educated up /positive
till school 2) Use
level/negativ of
e electrica
2) Binary l
variable applianc
equals 1 if es
the /positive
household 3)
 Journal Pre-proof

head is Family
graduate/neg size
ative /positive
3) Binary 4)
variable Dwellin
equals 1 if g size
the /positive
household 5)
head is post- Average
graduate / time
negative spent
outside
by a
family
member
/negativ
e

of
6)
Binary
variable

ro
equals 1
if the
-p househo
ld is
located
in
re
South-
Delhi
/positive
lP

Achar Survey Weighted 1) Literary / Gross - - - -

ya, /India states energy negative District
and and Union poverty 2) Domestic
Sadath territories index Percentage Product /
na

(2019) (lighting, of the negative

cooking, population
and above the
additional 12th
ur

measures) standard
education/ne
gative
Jo

3)
Constructed
education
index/negati
ve
Sharm Survey Monthly 1) Number of Total - - - Dwellin
a et al /India expenditure years of monthly g size /
(2019) on electrical education of expenditure positive
energy by the HH of the
the head/negativ household /
household e positive
head 2) Binary
variable
equals 1 if
the
household
head is aware
of the
benefits of
energy
 Journal Pre-proof

conservation
/ positive
Crents Survey Multidimen 1) No - - - - 1) Urban
il et al /Ghana sional education / househol
(2019) energy positive d/
poverty 2) positive
index Incomplete 2) Rural
primary/posit househol
ive d/
3) Complete positive
primary / 3) Male
positive head of
4) househol
Incomplete d/
secondary / positive
positive 4)
5) Complete Female

of
secondary / head of
positive househol
6) Higher d/

ro
education / positive
positive 5) Age /
-p positive
Koom Survey 1) Energy Binary - - - - 1)
son /Ghana deprivation variable Financia
and 2) equals 1 if l
re
Danqu Multidimen household inclusio
ah sional head is n/
(2021) energy educated/neg negative
lP

poverty ative 2)
index Househo
ld size /
positive
na

3)
Employ
ment
status
ur

/negativ
e
4) Age /
Jo

positive
5)
Female
headed /
negative
6)
Married
/
negative

3. Theoretical framework

As widely documented by the literature, there exists a clear link between education

and economic development (see among others Hanushek, and Wößmann, 2010; Cohen and
 Journal Pre-proof

Soto, 2007; Kalaitzidakis et al, 2001). A country's economy becomes more productive as the

proportion of the educated labor force increases. This happens since educated workers can

more efficiently carry out tasks that require literacy and critical thinking. Therefore, investing

in education and thus accumulating human capital is one of the main determinants of

economic growth (Schultz, 1960; Becker, 1994; Krueger and Lindahl, 2001). Education can

be also regarded as one of the most powerful instruments for reducing poverty and inequality,

while it sets the foundation for sustainable development (Psacharopoulos and Patrinos, 2004;

Gillette and Patrinos, 2014). Education drives economic growth, resulting in efficiency gains

of
which in turn benefits poor and low-income households (Koomson and Danquah, 2021).

ro
These financial improvements may increase households’ access to resources including energy
-p
services, mitigating energy poverty.
re
Based on the above, a higher education level would allow people to be more energy
lP

aware by switching to an efficient energy fuel mix (Rahut, 2019; Ogwumike and Ozughalu,

2016; Hills, 1994; Leach, 1992), given that energy inefficiency is a significant source of
na

energy poverty (Koomson and Danquah, 2021; Mattioli et al., 2017). From the demand side
ur

perspective, a society with a significant proportion of highly educated people would

efficiently manage their energy use, partly due to the awareness of a possible shortage of
Jo

fossil fuel soon. Moreover, most educated professionals are willing to spend more on

purchasing star-rated energy-efficient products to reduce energy demand. This would

subsequently reduce their energy bills since these green-certified products are often more

expensive than the non-rated ones. Consequently, a higher education ‘consumers’ profile’

accelerates the transition from traditional fuels like biomass fuels (fuel woods), kerosene,

crop waste, and animal dung to more efficient modern commercial energy sources (i.e.,

renewables, electricity, petroleum products, natural gas) mitigating energy poverty. It is

noteworthy that the lack of modern energy sources may also lead to health hazards such as
 Journal Pre-proof

malnutrition, cardiovascular diseases, hindering the overall human development process

(Sharma et al, 2019; WTO, 2014).

To conclude, this study is developed on the theoretical underpinnings of human

capital and energy transition theories to provide fresh evidence of the prevalence and extent

of energy poverty in emerging economies.

4. Data and variables

Energy access is utilized widely as a proxy to measure energy poverty (Culver, 2017),

of
while many studies lend support to more sophisticated indicators also capturing the various

ro
aspects of sustainable development. To better describe the dimensions of energy poverty

there are many indicators used, such as the IEA’s Energy Development Index (EDI), the
-p
Energy Poverty Index (EPI), the Compound Energy Poverty Indicator (CEPI), and the
re
Multidimensional Energy Poverty Index (MEPI).2
lP

The relevant indicators employ micro-level variables (e.g., access to electricity by

na

households, household income, household expenditure on energy, housing costs, type of

household, age of the property, area of residence, etc) usually obtained by conducting
ur

primary research (surveys) to encapsulate the different aspects of energy poverty.

Jo

Specifically, the IEA’s indicator incorporates the weighted average of per capita energy

consumption and the percentage of commercial energy in total final energy consumption. The

EPI assesses the inequalities in different geographic regions concerning regional energy

prices, energy poverty, and income level. Moreover, the CEPI approach proposed by Maxim

et al. (2016) is more extensive compared to the previous indicator. However, the MEPI

analysis developed in Nussbaumer et al (2012), constitutes the most widely used approach to

measure energy poverty, considering specific socioeconomic drivers that are absent from

2
See the discussion developed in Mendoza et al, (2019).
 Journal Pre-proof

other energy poverty indicators. The relevant index is based on microdata (e.g., cooking,

lighting, household appliances, communication, education, entertainment, etc) drawn from

household surveys and calculates the share of people identified as energy poor (headcount

ratio) and the average intensity of deprivation of the energy-poor household. In its original

version, an individual is identified as energy poor if the combination of the deprivations faced

exceeds a pre-defined threshold (usually 50%). Despite its simplicity and ease of its

application, the related indicator is not free from limitations such as the arbitrary selection of

the threshold and the excessive sensitivity to energy prices. To overcome these shortcomings

of
other objective and subjective indicators of energy poverty have been used by the recent

ro
empirical literature (see among others Churchill and Smyth, 2021; Churchill et al, 2020;
-p
Thomson et al, 2016).
re
The first indicator is the Low-Income/High-Cost Index (LIHC) which was developed
lP

in Hills (2012). The relevant indicator is based on both the high costs of energy and the low-

income status of households. According to the LIHC indicator, a household is defined as

na

energy-poor when income falls below a certain (relative) poverty line and when its energy
ur

costs are higher than an energy expenditure threshold (Hills 2012). The main advantage of

this indicator is that it corrects the 10% “energy poverty line” threshold indicator by
Jo

considering not only the expenditure on energy but also an income threshold (Churchill and

Smyth, 2021; Romero et al, 2018; Hills 2012). However, this approach has several limitations

mostly related to its complexity and non-transparent nature. Moreover, there is difficult to

find out those households that can come out of energy poverty by way of reducing their

energy costs, while its doubly relative character makes it very difficult to isolate causes and

effects when analyzing time series (see Romero et al, 2018).

The second indicator of energy poverty based on the related study of Churchill et al,

(2021), is a subjective measure that reflects the “feeling of material deprivation felt by
 Journal Pre-proof

households who are unable to adequately heat their homes” (Churchill et al., 2020). The

relevant indicator is important since it allows respondents to express how they feel (Price et

al., 2012), helping the researchers to better embed the social dimension of energy poverty

(see also Rodriguez-Alvarez et al., 2019).

The third indicator comprises a composite measure of energy poverty based on the

multidimensional poverty index and the MEPI (see also Alkire and Santos, 2010). The

relevant dummy variable takes the value of one if respondents were energy-poor based on at

of
least one of the objective and subjective measures of energy poverty and zero otherwise.

ro
Lastly, another commonly used measure of fuel poverty considers households as fuel
-p
poor if the proportion of their income spent on energy exceeds 10% of their income on

energy. Several related studies have used this approach (see among others Churchill et al,
re
2020; Romero et al, 2018) that provides several advantages. Specifically, this indicator is
lP

simple to calculate and due to its ease of communication is relatively versatile from a

pragmatic point of view (see Romero et al, 2018). However, the lack of any reference to the
na

household income, in tandem with its excessive sensitivity to energy prices and the arbitrary
ur

selection of the threshold at 10% mitigate to some extent its positive aspects of this indicator.
Jo

All these related studies measure energy poverty either at a country level or at a

“union” level (i.e., EU-28, OECD, etc) making use of micro-level indicators. However, this

approach can be linked with the use of macro-level indicators like the ones reported in our

study since the adopted energy poverty indicators (i.e., access to electricity, access to clean

fuels and technologies for cooking, GDP per capita, etc) are measuring quite similar aspects

of energy poverty but at a larger scale (i.e., macro perspective). This constitutes another

novelty of the present study.

 Journal Pre-proof

Annual data from 30 countries (see the list in the Appendix), spanning the period

1990-2019, on the following variables were obtained from the World Bank database: 1)

energy poverty, measured as access to electricity (total) defined as the percentage of the

population, or as access to electricity (rural) as a percentage of the population, or as access to

electricity (urban) as a percentage of the population, or as access to clean fuels and

technologies for cooking as a percentage of the population, 2) electric power consumption

measured as kWh per capita, 3) total carbon dioxide emissions from electricity and heat

production per capita measured as the percentage of total fuel consumption per capita, 4)

of
GDP per capita, 5) school enrolment measured as the percentage of secondary school-age

ro
children, and 6) education expenditure. The last two variables are the proxies for the primary
-p
independent variable, which is the education measure. Table 2 provides certain descriptive
re
statistics.
lP

Table 2. Descriptive statistics

na

Variable Mean SD Min Max

Access to electricity (total) 73.91 29.34 7.10 100.00
Access to electricity (rural) 63.05 30.51 0.13 100.00
ur

Access to electricity (urban) 68.99 31.68 0.00 100.00

Access to clean fuels and technologies for cooking 55.97 34.98 1.01 99.10
Electric power consumption per capita 1,479.44 1,784.03 29.81 10,496.51
Jo

CO2 emissions per capita 32.25 16.49 0.28 71.83

Secondary school enrolment rate 66.32 17.74 29.81 89.63
Education expenditure (%) 4.19 1.78 1.15 10.68
GDP per capita 9,377.07 7,317.15 929.88 39,857.09
SD = standard deviation.

5. Empirical specification

The model specification is indicated below:

p1 p2

ENPOVit = αi + Σγ1it EDUCi(t-1) + Σγ2it Xit + vit (1)

i=0 i=0
 Journal Pre-proof

where ENPOV denotes energy poverty, EDUC is the proxy for the education variable, X is a

vector of control variables provided in the relevant literature and discussed in the data section

above, and v is the error term. The model includes N=30 countries (indexed by i), observed

over T=30 periods (years) (indexed by t), and allows for country-specific effects (αi) for the

ith individual unit. In a panel country framework, the disturbances vi,t are uncorrelated. They

are assumed to be independently distributed across countries with a zero mean.

However, in the presence of cross-section error dependence, conventional panel

of
estimation methods could lead to inconsistent estimates and incorrect inferences. The analysis

ro
implements GMM system estimates by Arellano and Bover (1995) and Blundell and Bond

(1998) while following Roodman (2009) for eliminating any small sample bias. Finally, it
-p
also employs the Windmeijer (2005) finite sample correction for standard errors.
re
The selection of the sample variables is based on theoretical grounds or dictated by
lP

the empirical literature. Specifically, this study uses traditional measures of energy poverty

namely alternative indicators for access to electricity and clean fuels which are in alignment
na

with similar studies (Amin et al., 2020; Ogwumike and Ozughalu, 2016; Ozughalu and
ur

Ogwumike 2019). We also include two education variables in the reduced-form equations
Jo

that have been used by earlier studies (Koomson and Danquah, 2021; Sharma et al., 2019;

Tewathia, 2014). In this study, we use three alternative GDP proxies to account for the

economic growth effect on energy poverty. However, as is previously documented, most of

the related studies use household survey data to analyze the key determinants of energy

poverty (see Table 1). Therefore, they use household expenditures to quantify the income

effect on energy poverty (Tewathia, 2014; Sharma et al., 2019). We also use GDP as a proxy

of economic growth. This approach is also followed in the literature (see for example

Acharya, and Sadath, 2019).

 Journal Pre-proof

Following earlier works (see Phoumin, and Kimura, 2019; Romero et al., 2018), we

have used electricity consumption as one of the key determinants of energy poverty.

Moreover, similarly to Martinez and Poveda (2013), where final energy consumption per

GDP was used to account for the impact of energy intensity on energy poverty, this study

employs the primary energy intensity level.

The relevant literature also considers the role of carbon emissions (CO2) as a

significant driver of energy poverty; for instance, Kanti (2017) and Ouedraogo (2017) report

that poor households make use of unclean (“dirty”) cooking fuels to survive, with the

of
combustion of such fuels results in large emission of hazardous indoor air pollutants, such as

ro
carbon monoxide, nitrogen oxides, and particulate matter, which severely affect the health of
-p
household members. Jayendran et al. (2018) also provide evidence that many households in
re
emerging economies are heavily dependent on inefficient fuel (e.g., kerosene) for lighting.

Indoor combustion without proper ventilation poses significant health risks, while certain by-
lP

products also contribute to climate-altering carbon emissions.

na

6. Results and discussion

ur

6.1. Preliminary unit root and cross-sectional dependence testing

Jo

The first step of the empirical analysis examines the unit root properties in the data

through advanced panel unit root tests. Panel unit root tests of the first-generation can lead to

spurious results if significant degrees of positive residual cross-section dependence exist and

are ignored. Consequently, the implementation of second-generation panel unit root tests is

desirable only when it has been established that the panel is subject to a significant degree of

residual cross-section dependence. Therefore, before selecting the appropriate panel unit root

test, it is crucial to provide some evidence on the degree of residual cross-sectional

dependence.
 Journal Pre-proof

The cross-sectional dependence (CD) statistic by Pesaran (2004) is based on a simple

average of all pair-wise correlation coefficients of the OLS residuals obtained from standard

augmented Dickey-Fuller regressions for each variable in the panel. Under the null

hypothesis of cross-sectional independence, the CD test statistic follows asymptotically a

two-tailed standard normal distribution. The results, reported in Table 3, uniformly reject the

null hypothesis of cross-section independence, providing evidence of cross-sectional

dependence in the data, given the statistical significance of the CD statistics.

of
Table 3. Cross-section dependence tests

ro
Variables CD test [p-values]
Access to electricity (total) 7.86 [0.00]
-p
Access to electricity (rural) 8.19 [0.00]
Access to electricity (urban) 7.04 [0.00]
re
Access to clean fuels & technologies for cooking 6.93 [0.00]
Electric power consumption 7.57 [0.00]
lP

CO2 emissions 7.14 [0.00]

Secondary school enrolment 7.11 [0.00]
Education expenditure 6.94 [0.00]
na

GDP per capita 6.98 [0.00]

Next, a second-generation panel unit root test is employed to determine the degree of
ur

integration of the respective variables. The Pesaran (2007) panel unit root test does not
Jo

require the estimation of factor loading to eliminate cross-sectional dependence. Specifically,

the usual ADF regression is augmented to include the lagged cross-sectional mean and its

first difference to capture the cross-sectional dependence that arises through a single-factor

model. The null hypothesis is a unit root for the Pesaran (2007) test. The results of this test

are reported in Table 4 and support the presence of a unit root across all panel variables.

Table 4: Panel unit root tests

H0: Contains a unit root
CIPS
Variables Levels 1st
Differences
Access to electricity (total) -1.162 -5.759***
 Journal Pre-proof

Access to electricity (rural) -1.144 -5.809***

Access to electricity (urban) -1.138 -5.774***
Access to clean fuels and technologies -1.152 -5.884***
for cooking
Electric power consumption -1.045 -5.898***
CO2 emissions -1.122 -6.924***
Secondary school enrolment -1.153 -5.996***
Education expenditure -1.059 -6.704***
GDP per capita
Note: ***: denotes statistical significance at a=1%.

6.2. Main findings

of
The empirical results are reported in Table 5 with columns indicating certain

ro
specifications. In particular: Column (1) displays the estimates concerning the net enrolment
-p
measure of the education variable, while Column (2) shows the estimates in relevance to the
re
education expenditure measure of the educational variable. These first two columns measure

GDP in PPP prices, while Columns (3) and (4) repeat the previous results (Columns (1) and
lP

(2)), but this time GDP is measured as per unit of energy use. Columns (5) and (6) provide
na

the estimates when GDP is measured as GDP per capita. Finally, Columns (7) and (8) show

the estimates in the cases of the adjusted net national income and adjusted net national
ur

income per capita, respectively.

Jo

The results clearly document the positive and statistically significant impact of the

two measures of education on access to electricity (the proxy of energy poverty). The results

remain consistently similar across all eight specifications. With respect to the economic

interpretation of the findings, the results indicate that the immediate impact of school

enrolment on access to electricity is that 1% increase in the school enrolment leads to a

contemporaneous increase in the percentage of electricity access by 0.429, 0.438, 0.430, and

0.452, respectively, as well as a lagged increase by 0.226, 0.248, 0.239, and 0.274,

respectively. Moreover, the findings illustrate that a 1% increase in total education

 Journal Pre-proof

expenditure leads to a contemporaneous increase in the percentage of electricity access by

0.446, 0.460, 0.453, and 0.450, respectively, as well as a lagged increase by 0.229, 0.252,

0.255, and 0.263, respectively. In terms of the remaining determinants of energy poverty, the

estimates are in accordance with theoretical expectations, with all, but one, the remaining

drivers exerting a positive and statistically significant effect on access to electricity. In

contrast, carbon emissions exert a negative impact on access to electricity, indicating that

higher carbon emissions clearly illustrate that households have reduced access to electricity

(clean energy) and mostly use unclean forms of heating that generate high levels of carbon

of
emissions.

ro
The relevant diagnostics are reported at the bottom of Table 5. In particular, the
-p
findings report the LM test for the appropriateness of the random effects modeling approach.
re
The null hypothesis of no random effects is rejected, indicating that a random effects model is
lP

more suitable. Moreover, for the validity of the instruments, the results need to reject the null

hypothesis of difference-in-Hansen tests of the exogeneity of instruments. The difference-in-

na

Hansen test fails to reject the respective null. Thus, the test supports the validity of the
ur

instruments used, while difference-in-Hansen tests imply the exogeneity of the instruments

employed. The table also reports the Hansen test for over-identifying restrictions. In the
Jo

estimation process, instruments were generated from one lag for levels and difference in the

regressors. As the number of instruments remained lower than the number of observations, it

did not cause any identification problem, as reflected in the Hansen test.

6.3. Sensitivity/robustness checks

For robustness purposes, we also consider the three alternative definitions of energy

poverty. The new results are presented in Tables 6 through 8, where energy poverty is

measured as access to electricity (rural), access to electricity (urban), and access to clean
 Journal Pre-proof

fuels and technologies for cooking, respectively. Focusing on the link between energy

poverty and education, the new findings provide robust support to the baseline results.

Table 5: Dynamic panel estimations (energy poverty: access to electricity, total)

Variables (1) (2) (3) (4) (5) (6) (7) (8)
ΔAccess to 0.697* 0.714* 0.709* 0.668* 0.661* 0.693* 0.665* 0.675*
electricity(- ** ** ** ** ** ** ** **

of
1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔAdjusted 0.429* 0.438* 0.430* 0.452*
net ** ** ** **

ro
enrolment [0.00] [0.00] [0.00] [0.00]
rate -p
ΔAdjusted 0.226* 0.248* 0.239* 0.274*
net * * * *
enrolment [0.03] [0.02] [0.02] [0.02]
re
rate(-1)
ΔEducation 0.446* 0.460* 0.453* 0.450*
lP

expenditure ** ** ** **
[0.00] [0.00] [0.00] [0.00]
ΔEducation 0.229* 0.252* 0.255* 0.263*
na

expenditure * * * *
(-1) [0.03] [0.03] [0.02] [0.02]
ΔElectric 0.125* 0.137* 0.144* 0.150*
ur

power ** ** ** **
consumptio [0.00] [0.00] [0.00] [0.00]
n
Jo

ΔElectric 0.081* 0.086* 0.094* 0.080*

power * * * *
consumptio [0.03] [0.03] [0.02] [0.03]
n(-1)
ΔCO2 - - - - - - - -
emissions 0.514* 0.520* 0.529* 0.511* 0.536* 0.532* 0.553* 0.567*
** ** ** ** ** ** ** **
[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔCO2 - - - - - - - -
emissions(- 0.224* 0.229* 0.237* 0.228* 0.239* 0.236* 0.265* 0.274*
1) * * * * * * * **
[0.03] [0.03] [0.03] [0.03] [0.02] [0.02] [0.02] [0.01]
ΔGDP per 0.302* 0.316*
capita ** **
[0.00] [0.00]
ΔGDP per 0.117* 0.148*
 Journal Pre-proof

capita(-1) * *
[0.03] [0.02]
Diagnostics
R2-adjusted 0.61 0.65 0.69 0.66 0.64 0.68 0.56 0.63
AR(1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
AR(2) [0.33] [0.42] [0.27] [0.30] [0.37] [0.31] [0.32] [0.38]
Hansen test [0.42] [0.50] [0.42] [0.44] [0.51] [0.46] [0.31] [0.42]
Difference [0.70] [0.77] [0.60] [0.67] [0.76] [0.62] [0.63] [0.62]
Hansen test
Country YES YES YES YES YES YES YES YES
fixed
effects
LM test [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
No. of 9 8 10 9 9 11 8 11

of
instruments
No. of 900 900 900 900 900 900 900 900
observation

ro
s
-p
Notes: Figures in brackets denote p-values. LM stands for the Lagrange Multiplier test for random effects
(Breusch and Pagan, 1980). AR(1) is the first-order test for residual autocorrelation. AR(2) is the test for
re
autocorrelation of order 2. Hansen is the test for the overidentification check for the validity of instruments. The
difference-in-Hansen test checks the exogeneity of the instruments. The number of lags was determined through
the Akaike criterion. All estimations were performed with time dummies (time fixed effects). **: p<0.05; ***
lP

p<0.01.
na

Table 6: Dynamic panel estimations (energy poverty: access to electricity, rural)

ur

Variables (1) (2) (3) (4) (5) (6) (7) (8)

ΔAccess to 0.612* 0.625* 0.621* 0.632* 0.636* 0.640* 0.635* 0.640*
Jo

electricity(- ** ** ** ** ** ** ** **
1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔAdjusted 0.389* 0.394* 0.398* 0.411*
net ** ** ** **
enrolment [0.00] [0.00] [0.00] [0.00]
rate
ΔAdjusted 0.197* 0.210* 0.218* 0.231*
net * * * *
enrolment [0.03] [0.03] [0.03] [0.03]
rate(-1)
ΔEducation 0.419* 0.424* 0.430* 0.435*
expenditure ** ** ** **
[0.00] [0.00] [0.00] [0.00]
ΔEducation 0.211* 0.219* 0.227* 0.231*
expenditure * * * *
(-1) [0.03] [0.03] [0.03] [0.03]
 Journal Pre-proof

ΔElectric 0.109* 0.115* 0.121* 0.129*

power ** ** ** **
consumptio [0.00] [0.00] [0.00] [0.00]
n
ΔElectric 0.065* 0.070* 0.075* 0.076*
power * * * *
consumptio [0.04] [0.04] [0.02] [0.02]
n(-1)
ΔCO2 - - - - - - - -
emissions 0.485* 0.493* 0.499* 0.507* 0.516* 0.513* 0.520* 0.527*
** ** ** ** ** ** ** **
[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔCO2 - - - - - - - -
emissions(- 0.210* 0.215* 0.217* 0.220* 0.224* 0.229* 0.225* 0.232*

of
1) * * * * * * * *
[0.03] [0.03] [0.03] [0.03] [0.03] [0.03] [0.03] [0.02]
ΔGDP per 0.277* 0.283*

ro
capita ** **
[0.00] -p [0.00]
ΔGDP per 0.101* 0.109*
capita(-1) * *
[0.04] [0.03]
re
Diagnostics
R2-adjusted 0.53 0.55 0.51 0.56 0.53 0.56 0.52 0.57
lP

AR(1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]

AR(2) [0.30] [0.37] [0.25] [0.26] [0.28] [0.30] [0.27] [0.33]
Hansen test [0.35] [0.44] [0.36] [0.34] [0.38] [0.36] [0.33] [0.39]
na

Difference [0.65] [0.72] [0.54] [0.61] [0.68] [0.57] [0.56] [0.60]

Hansen test
Country YES YES YES YES YES YES YES YES
ur

fixed
effects
LM test [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
Jo

No. of 8 8 9 8 9 9 10 10
instruments
No. of 900 900 900 900 900 900 900 900
observation
s
Notes: Figures in brackets denote p-values. LM stands for the Lagrange Multiplier test for random effects
(Breusch and Pagan, 1980). AR(1) is the first-order test for residual autocorrelation. AR(2) is the test for
autocorrelation of order 2. Hansen is the test for the overidentification check for the validity of instruments. The
difference-in-Hansen test checks the exogeneity of the instruments. The number of lags was determined through
the Akaike criterion. All estimations were performed with time dummies (time fixed effects). **: p<0.05; ***
p<0.01.

Table 7: Dynamic panel estimations (energy poverty: access to electricity, urban)

 Journal Pre-proof

Variables (1) (2) (3) (4) (5) (6) (7) (8)

ΔAccess to 0.729* 0.742* 0.749* 0.748* 0.768* 0.774* 0.775* 0.789*
electricity(- ** ** ** ** ** ** ** **
1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔAdjusted 0.447* 0.455* 0.462* 0.469*
net ** ** ** **
enrolment [0.00] [0.00] [0.00] [0.00]
rate
ΔAdjusted 0.240* 0.249* 0.260* 0.269*
net * * * *
enrolment [0.02] [0.02] [0.02] [0.02]
rate(-1)
ΔEducation 0.461* 0.468* 0.479* 0.492*
expenditure ** ** ** **

of
[0.00] [0.00] [0.00] [0.00]
ΔEducation 0.240* 0.254* 0.259* 0.270*
expenditure * * * *

ro
(-1) [0.02] [0.02] [0.02] [0.02]
ΔElectric 0.138* 0.149* -p 0.160* 0.166*
power ** ** ** **
consumptio [0.00] [0.00] [0.00] [0.00]
n
re
ΔElectric 0.098* 0.109* 0.119* 0.129*
power * * * *
lP

consumptio [0.03] [0.02] [0.02] [0.02]

n(-1)
ΔCO2 - - - - - - - -
na

emissions 0.538* 0.543* 0.549* 0.542* 0.556* 0.553* 0.558* 0.578*

** ** ** ** ** ** ** **
[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔCO2 - - - - - - - -
ur

emissions(- 0.240* 0.246* 0.251* 0.254* 0.259* 0.255* 0.269* 0.282*

1) * * * * * * * **
Jo

[0.02] [0.02] [0.02] [0.02] [0.02] [0.02] [0.02] [0.00]

ΔGDP per 0.337* 0.358*
capita ** **
[0.00] [0.00]
ΔGDP per 0.146* 0.174*
capita(-1) * *
[0.02] [0.02]
Diagnostics
R2-adjusted 0.64 0.68 0.70 0.69 0.68 0.71 0.60 0.68
AR(1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
AR(2) [0.36] [0.49] [0.35] [0.38] [0.44] [0.37] [0.39] [0.44]
Hansen test [0.46] [0.55] [0.48] [0.49] [0.56] [0.50] [0.39] [0.48]
Difference [0.74] [0.81] [0.69] [0.73] [0.79] [0.68] [0.67] [0.70]
Hansen test
Country YES YES YES YES YES YES YES YES
fixed
 Journal Pre-proof

effects
LM test [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
No. of 10 10 11 9 11 11 9 11
instruments
No. of 900 900 900 900 900 900 900 900
observation
s

Notes: Figures in brackets denote p-values. LM stands for the Lagrange Multiplier test for random effects
(Breusch and Pagan, 1980). AR(1) is the first-order test for residual autocorrelation. AR(2) is the test for
autocorrelation of order 2. Hansen is the test for the overidentification check for the validity of instruments. The
difference-in-Hansen test checks the exogeneity of the instruments. The number of lags was determined through
the Akaike criterion. All estimations were performed with time dummies (time fixed effects). **: p<0.05; ***

of
p<0.01.

ro
-p
Table 8: Dynamic panel estimations (energy poverty: access to clean fuels and technologies
for cooking)
re
Variables (1) (2) (3) (4) (5) (6) (7) (8)
ΔAccess to 0.656* 0.701* 0.694* 0.659* 0.654* 0.683* 0.652* 0.664*
lP

electricity(- ** ** ** ** ** ** ** **
1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔAdjusted 0.412* 0.417* 0.422* 0.431*
na

net ** ** ** **
enrolment [0.00] [0.00] [0.00] [0.00]
rate
ur

ΔAdjusted 0.210* 0.217* 0.219* 0.225*

net * * * *
enrolment [0.03] [0.03] [0.03] [0.02]
Jo

rate(-1)
ΔEducation 0.430* 0.435* 0.433* 0.438*
expenditure ** ** ** **
[0.00] [0.00] [0.00] [0.00]
ΔEducation 0.214* 0.218* 0.224* 0.230*
expenditure * * * *
(-1) [0.03] [0.03] [0.03] [0.02]
ΔElectric 0.114* 0.117* 0.121* 0.126*
power ** ** ** **
consumptio [0.00] [0.00] [0.00] [0.00]
n
ΔElectric 0.070* 0.074* 0.075* 0.077*
power * * * *
consumptio [0.03] [0.03] [0.03] [0.03]
n(-1)
ΔCO2 - - - - - - - -
 Journal Pre-proof

emissions 0.492* 0.498* 0.506* 0.509* 0.514* 0.518* 0.513* 0.525*

** ** ** ** ** ** ** **
[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
ΔCO2 - - - - - - - -
emissions(- 0.215* 0.217* 0.218* 0.223* 0.229* 0.230* 0.235* 0.245*
1) * * * * * * * *
[0.03] [0.03] [0.03] [0.03] [0.03] [0.03] [0.02] [0.02]
ΔGDP per 0.280* 0.293*
capita ** **
[0.00] [0.00]
ΔGDP per 0.104* 0.115*
capita(-1) * *
[0.03] [0.03]
Diagnostics

of
R2-adjusted 0.58 0.55 0.58 0.56 0.57 0.55 0.53 0.58
AR(1) [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
AR(2) [0.30] [0.36] [0.24] [0.25] [0.32] [0.27] [0.29] [0.33]

ro
Hansen test [0.38] [0.44] [0.37] [0.40] [0.43] [0.44] [0.27] [0.29]
Difference [0.62] [0.71] [0.54] [0.60]
-p [0.71] [0.54] [0.56] [0.58]
Hansen test
Country YES YES YES YES YES YES YES YES
fixed
re
effects
LM test [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
lP

No. of 8 8 9 10 8 10 11 13
instruments
No. of 900 900 900 900 900 900 900 900
na

observation
s
Notes: Figures in brackets denote p-values. LM stands for the Lagrange Multiplier test for random effects
(Breusch and Pagan, 1980). AR(1) is the first-order test for residual autocorrelation. AR(2) is the test for
ur

autocorrelation of order 2. Hansen is the test for the overidentification check for the validity of instruments. The
difference-in-Hansen test checks the exogeneity of the instruments. The number of lags was determined through
the Akaike criterion. All estimations were performed with time dummies (time fixed effects). **: p<0.05; ***
Jo

p<0.01.

7. Conclusion

This study aimed to contribute to the growing literature on the behavioral aspects of

energy poverty in developing countries. Building upon the well-established human capital

theory, the analysis investigated the impact of education on access to electricity on a balanced

panel dataset of 30 emerging economies over the period 2001-2016. Moreover, it further

explored the potential channels through which education influenced energy poverty.
 Journal Pre-proof

The empirical findings provided fresh insights into the education-energy poverty

nexus, supporting the presence of a negative and statistically significant relationship between

education and energy poverty. This result gave sufficient ground to the validity of the energy

transition theories, indicating that households with a low level of educational attainment,

have restricted access to clean energy forms, such as electricity, and mostly use unclean

forms of energy (e.g., biomass fuels and fossil fuels) that generate high levels of carbon

emissions. This study also uncovered positive and statistically significant relationships of

energy poverty with other covariates, such as economic growth, energy intensity, and

of
electricity consumption. Moreover, it unraveled a negative and statistically significant effect

ro
of environmental pollution on energy poverty. -p
Since the energy policy of the developing economies has been considered insufficient
re
to tackle energy poverty issues, focusing mainly on short-term goals, rather than long-lasting
lP

solutions, this study could be important for policymakers and government officials in their

effort to mitigate energy poverty and achieve sustainable growth level for the developing
na

countries. The positive impact of education on the access to electricity stipulates the
ur

governments to implement various programs, such as training and education of householders

toward energy-saving and use of clean energy sources (electricity, natural gas, renewables).
Jo

In line with this, energy awareness, focusing on the training of a consumer in

understanding how much his actions and use of appliances contribute to total consumption,

can turn out to be of paramount importance in achieving substantial, and sustained energy

consumption reduction. An energy consumer that can know in real-time the energy

consumption that her behavior is causing and compare it to that of other similar users, is more

likely to initiate efficiency mitigating energy poverty. This is related to the use of smart

metering systems. A smart meter is usually an electronic device that records consumption of

electric energy in intervals of an hour or less and communicates that information at least daily
 Journal Pre-proof

back to the utility for monitoring and billing. Smart meters enable two-way communication

between the meter and the central system. Unlike home energy monitors, smart meters can

gather data for remote reporting. Specifically, smart metering devices offer the utility market

the benefits of improved reliability and accuracy, ease of calibration, security, and advanced

billing features, and they offer energy consumers the opportunity to better manage energy

usage and reduce costs. Moreover, smart meters allow consumers to access real-time

information on energy usage across distributed energy resources. In this way, they enable

consumers to make data-driven decisions to streamline energy consumption and save on their

of
annual utility bills.

ro
Finally, this study offers possible avenues for future research. A possible extension
-p
would be to increase the number of developing countries participating in the analysis (given
re
data availability) and then to split the sample to isolate the emerging economies based solely
lP

on their income level (e.g., high income, upper middle income, low-middle income

countries). The reason for doing this can be justified by the fact that different regulatory and
na

institutional regimes exist between these economies. For example, regulation in middle-
ur

income (emerging) countries is driven, to a large extent, by political goals, thus, distorting the

growth path of these economies, along with the energy poverty level. Consequently, these
Jo

countries are truly in need of a regulation system that balances accountability, transparency,

and consistency. As a result, there might be different relationships between the emerging

economies classified by their income level. In this way, several useful findings will emerge

providing important policy implications.

Acknowledgments

We would like to thank both the Editor and the Guest Editor for giving us the opportunity to

revise our study. Our gratitude also goes to two referees whose constructive comments and
 Journal Pre-proof

suggestions enhanced the merit of our work. Any remaining errors are the authors’ sole

responsibility.

Credit author statement

Nicholas Apergis: Conceptualization, Methodology, Software, Data curation, Writing -
Original draft preparation.
Michael Polemis: Methodology, Visualization, Validation, Formal Analysis, Writing -
Original draft preparation.
Simeoni-Eleni Soursou: Conceptualization, Data curation, Validation, Writing - Original
draft preparation.

of
Appendix.

Argentina, Bangladesh, Benin, Botswana, Brazil, Cambodia, Chile, China, Colombia, Costa

ro
Rica, Egypt, Ecuador, Guatemala, India, Kenya, Korea, Mexico, Morocco, Namibia,
-p
Nicaragua, Niger, Peru, Philippines, Senegal, South Africa, Tanzania, Togo, Tunisia,
Uruguay, Vietnam.
re
lP

Supplementary data
Supplementary material
na

References
ur

Abbas, K.L. 2020. Do socioeconomic factors determine household multidimensional energy

Jo

poverty? Empirical evidence from South Asia. Energy Policy, 146.

Acharya, R., Sadath, A. 2019. Energy poverty and economic development: Household-level

evidence from India, Energy and Buildings, 183, 785-791.

Adusah-Poku, F., Takeuchi, K. 2019. Energy poverty in Ghana: any progress so far?

Renewable and Sustainable Energy Reviews, 112, 853-864.

 Journal Pre-proof

Ahmed, A., Gasparatos, A. 2020. Multi-dimensional energy poverty patterns around

industrial crop projects in Ghana: Enhancing the energy poverty alleviation potential of

rural development strategies, Energy Policy, 137, 111123

Alkire, S., and Santos, M.E, 2010. Acute Multidimensional Poverty: A New Index for

Developing Countries. Oxford Poverty & Human Development Initiative (OPHI)

Working Paper No. 38, United Nations Development Programme Human Development

Report Office Background Paper No. 2010/11, Available at SSRN:

of
[Link] or doi:10.2139/ssrn.1815243

ro
Amin, A., Liu, Y., Yu, J., Chandio, A., Rasool, S. 2020. How does energy poverty affect
-p
economic development? A panel data analysis of South Asian countries. Environmental

Science and Pollution Research, 27(25), 31623-31635.

re

Arellano, M., Bover, O. 1995. Another look at the instrumental variable estimation of error-
lP

components models. Journal of Econometrics 68, 29-51.

na

Azariadis C, Drazen A .1990. Threshold externalities in economic development. Quarterly

Journal of Economics 105: 501-526.

ur

Barro., R.J. 2001. Human capital and growth. American Economic Review, 91: 12-17.
Jo

Baiyegunh, L.H. 2004. Rural household fuel energy transition: evidence from Giwa LGA

Kaduna State, Nigeria. Energy for Sustainable Development, 20, 30-35.

Becker, G.S., 1994. Human Capital, University of Chicago Press Economics Books, 3,

9780226041209.

Blundell, R., Bond, S. 1998. Initial conditions and moment restrictions in dynamic panel data

models. Journal of Econometrics 87, 115-143.

 Journal Pre-proof

Boardman, B. 1991. In Fuel Poverty: From Cold Homes to Affordable Warmth. Belhaven

Press, London.

Breusch, T.S., Pagan, A.R. 1980. The Lagrange Multiplier test and its applications to model

specification in econometrics. Review of Economic Studies 47, 239-253.

Mendoza, C.B., Cayonte, D.D., Leabres, M.S., Manaligod, R.L. (2019). Understanding

multidimensional energy poverty in the Philippines, Energy Policy, 133, 110886

Churchill, S.A, Smyth, R. 2020. Ethnic diversity, energy poverty, and the mediating role of

of
trust: Evidence from household panel data for Australia. Energy Economics, 86,

ro
104663. -p
Churchill, S. A., Smyth, R., and Farrell, L. (2020). Fuel poverty and subjective wellbeing.
re
Energy Economics, 86, 104650.
lP

Churchill, S. A., and Smyth, R. (2021). Energy poverty and health: Panel data evidence from

Australia. Energy Economics, 105219.

na

Cohen, D. and Soto, M. 2007. Growth and human capital: good data, good results, Journal of
ur

Economic Growth, 12 (1), 51-76.

Jo

Crentsil, A.O. Asuman, D., A.P. Fenny. 2019. Assessing the determinants and drivers of

multidimensional energy poverty in Ghana. Energy Policy, 133, 110884.

Donev, J., Hanania, J., Stenhouse, K. 2018. Energy Education - Access to electricity

[Online]. Available: [Link]

[Accessed: January 12, 2021].

Gillette H., Patrinos, H.A (ed.), 2014. Indigenous Peoples, Poverty, and Development,

Cambridge Books, Cambridge University Press, number 9781107415140.

 Journal Pre-proof

Grubler, A. 2012. Energy transitions research: Insights and cautionary tales. Energy Policy

50, 8–16.

Hanushek, E.A. and Wößmann, L. 2010. Education and economic growth, in Peterson, P.,

Baker, E. and McGaw, B. (Eds), International Encyclopedia of Education, 2, Oxford,

245-252.

Heltberg, R. 2004. Fuel switching: evidence from eight developing countries. Energy

Economics, 26, 869-887.

of
Hills, P. 1994. Household energy transition in Hong Kong. Energy, 19, 517-528.

ro
Hills, J., 2012. Getting the Measure of Fuel Poverty: Final Report of the Fuel Poverty
-p
Review.
re
Centre for the Analysis of Social Exclusion, LSE, London.
lP

Ifa, A., Guetat, I. 2018. Does public expenditure on education promote Tunisian and
na

Moroccan GDP per capita? ARDL approach. The Journal of Finance and Data Science,

4, 234-246.
ur

Jayendran, V., Solanki, S., Werner, K., Gautam, Y.N. 2018. Addressing energy poverty in
Jo

India: a systems perspective on the role of localization, affordability, and saturation in

implementing solar technologies. Energy Resources and Social Sciences, 40, 205-210.

Kalaitzidakis, P., Mamuneas, T.P., Savvides, A. and Stengos, T. 2001. Measures of human

capital and nonlinearities in economic growth, Journal of Economic Growth, 6, 229-

254.

Kanti, A. 2017. 1.3 million deaths every year in India due to indoor air pollution. Business

World.
 Journal Pre-proof

Karekezi, S.M. 2012. Energy, poverty, and development. In: Global Energy Assessment

Writing Team Karekezi, S.M. (Ed.), Global Energy Assessment: Toward a Sustainable

Future. Cambridge: Cambridge University Press, 151-190.

Khan, H. Z. 2019. Socio-economic and environmental factors influenced pro-poor growth

process: new development triangle. Environmental Science and Pollution Research, 26

29157–29172.

Koomson, I., Danquah, M. 2021. Financial inclusion and energy poverty: Empirical evidence

of
from Ghana, Energy Economics, 94. 105085.

ro
Krueger, A.B. and Lindahl, M. 2001. Education for growth: why and for whom?, Journal of
-p
Economic Literature, 39 (4), 1101-1136.
re
Leach, G. 1992. The energy transition. Energy Policy, 20, 116-123.
lP

Liu, Z.W. 2020. Does centralized residence promote the use of cleaner cooking fuels?

Evidence from rural China. Energy Economics, 104895.

na

Mankiw., N.G., Romer, D, Weil, D.N. 1992. A contribution to the empirics of economic
ur

growth. Quarterly Journal of Economics 107, 407-437.

Jo

Martinez, C., and Poveda. A.C. 2013. Energy, Development, and Economic Growth in

Colombia, Lecture Notes in Energy, 16, 311-330

Mattioli, G., Lucas, K., Marsden, G. 2017. Transport poverty and fuel poverty in the UK:

from analogy to comparison. Transport Policy 59, 93–105.

Maxim, A., Mihai, C., Apostoaie, C.M., Popescu, C., Istrate, C., and Bostan, I. (2016).

Implications and Measurement of Energy Poverty across the European Union,

Sustainability, 8(5), 1-20.

 Journal Pre-proof

Mendoza, C., Darcy, Cayonte, M., Leabres, L., Rose A., 2019. Understanding

multidimensional energy poverty in the Philippines. Energy Policy, 133, 110886.

Nussbaumer, P. B. (2012). Measuring energy poverty: Focusing on what matters. Renewable

and Sustainable Energy Reviews, 16 (1), 231-243.

Ozughalu, U.M., Ogwumike, F.O. 2019. Extreme Energy Poverty Incidence and

Determinants in Nigeria: A Multidimensional Approach. Social Indicators Research

142, 997–1014.

of
Ogwumike, F.O., Ozughalu, U.M. 2016. Analysis of energy poverty and its implications for

ro
sustainable development in Nigeria. Environment and Development Economics, 21(3),
-p
273-290.
re
Ouedraogo, N.S. 2017. Africa energy future: alternative scenarios and their implications for
lP

sustainable development strategies. Energy Policy, 106, 457-471.

Oum, S. 2019. Energy poverty in the Lao PDR and its impacts on education and health.
na

Energy Policy, 132, 247-253.

ur

Pesaran, M.H. 2007. A simple panel unit root test in the presence of cross‐section
Jo

dependence. Journal of Applied Econometrics 22, 265-312.

Pesaran, M.H. 2004. General diagnostic tests for cross section dependence in panels.

Cambridge Working Papers in Economics, No. 435 and CESifo Working Paper, No.

1229.

Phoumin, H., Kimura, F., 2019. Cambodia's energy poverty and its effects on social

wellbeing: Empirical evidence and policy implications, Energy Policy, 132, 283-289.

Price, C.W., Brazier, K., Wang,W., 2012. Objective and subjective measures of fuel poverty.

Energy Policy 49, 33–39.

 Journal Pre-proof

Psacharopoulos, G., Patrinos, H.A. 2004. Returns to investment in education: a further

update, Education Economics, 12(2), 111-134.

Rahut, D.A. 2019. Wealth, education and cooking-fuel choices among rural households in

Pakistan. Energy Strategy Reviews, 24, 236-243.

Rodriguez-Alvarez, A., Orea, L., Jamasb, T., 2019. Fuel poverty and well-being: a consumer

theory and stochastic frontier approach. Energy Policy 131, 22–32.

Romero, J.C., Linares, P., López, X. 2018. The policy implications of energy poverty

of
indicators, Energy Policy,.115, 98-108.

ro
Roodman, D. 2009. A note on the theme of too many instruments. Oxford Bulletin of
-p
Economics and Statistics 71, 135-158.
re
Sharma, D. 2016. Nexus between financial inclusion and economic growth: evidence from
lP

the emerging Indian economy. Journal of Financial Economic Policy, 8 (1), 13-36.
na

Sharma, S., Han, P., Sharma, V. 2019. Socio-economic determinants of energy poverty

amongst Indian households: a case study of Mumbai. Energy Policy, 32, 1184-1190.
ur

Schultz, T.W. 1960. Capital Formation by Education, Journal of Political Economy, 68, 571-
Jo

571.

Sovacool, B.K., Cooper, C., Bazilian, M., Johnson, K., Zoppo, D., Clarke, S., Raza, H.A.

2012. What moves and works: broadening the consideration of energy poverty. Energy

Policy, 42, 715-719.

Tewathia, N. 2014. Determinants of the household electricity consumption: a case study of

Delhi. International Journal of Energy Economics and Policy, 4, 337-348.

 Journal Pre-proof

Thomson, H., Snell, C.J., Liddell, C. 2016. Fuel poverty in the European Union: a concept in

need of definition? People, Place and Policy Online, 5-24.

WHO, 2014. World Health Statistics 2014. World Health Organization, Geneva, Switzerland.

Windmeijer, F. 2005. A finite sample correction for the variance of linear efficient two-step

GMM estimators. Journal of Econometrics 126, 25-51.

United Nations, 2020. Sustainable development goals. Retrieved from

[Link]

of
Zhang, D.L. 2019. A multidimensional measure of energy poverty in China and its impacts

ro
on health: An empirical study based on the China family panel studies. Energy Policy,
-p
131, 72-81.
re
lP

Highlights
 We examine the effect of education on energy poverty in 30 developing countries

na

The study uses GMM methods

 The findings illustrate that education mitigates energy poverty
 Important channels from education attainment to energy poverty are identified
ur

 Useful policy implications to alleviate energy poverty are provided

Jo