Energy Economics 65 (2017) 115–126

Contents lists available at ScienceDirect

Energy Economics
journal homepage: www.elsevier.com/locate/eneco

Income and energy use in Bangladesh: A household level analysis

Syed Abul Hasan a, * , Pallab Mozumder b, c
a
Crawford School of Public Policy, Australian National University, Canberra, Australia
b
Dept. of Earth and Env., Florida International University, FL, USA
c
Dept. of Economics, Florida International University, FL, USA

A R T I C L E I N F O A B S T R A C T

Article history: We examine how energy use at the household level moves with income in Bangladesh. Using the 2010
Received 5 February 2017 wave of Bangladesh Household Income and Expenditure Survey data, our analyses indicate a U-shaped rela-
Received in revised form 28 April 2017 tionship of both electricity use and other types of energy use (combined) with household expenditure. The
Accepted 2 May 2017 findings imply that as income grows households increase their energy expenditure less than proportion-
Available online 11 May 2017
ally, up to a threshold level of income. Beyond the threshold income, energy use increases as a proportion
of income, particularly for electricity use. We identify the threshold (turning point) for both electricity and
JEL classification: other types of energy use. Based on the current level of expenditure and its growth, reaching the turning
O13
point would require 17years for the former category but only 7years for the latter group.
D12
© 2017 Elsevier B.V. All rights reserved.
Q40
Q56

Keywords:
Household energy expenditure
Energy Engel curve
Turning point

1. Introduction
Bangladesh is making steady economic progress in the midst of
acute energy shortage. Per capita annual energy consumption in the
country is 371 kW h which is one of the lowest in the world (World
Bank, 2016). Electricity is the major source of energy for most eco-
nomic activities in Bangladesh. While around 78% of the population
now has access to electricity, highly intermittent power supply makes
their life quite miserable (Mujeri et al., 2014). A large portion of
households, on the other hand, do not have access to electricity while
other sources of energy are not adequate to meet the rising demand,
resulting in a high rate of energy poverty in the country. A recent
study finds that 58% of rural households in Bangladesh are energy
poor compared to the income poverty of 45% (Barnes et al., 2011).
The energy resources and infrastructure in Bangladesh are not
only inadequate but also poorly managed. In particular, electricity
generation plants have been unable to meet demand over the past
decade. Corruption, high system losses, low plant efficiency, erratic
* Corresponding author at: 3.27 JG Crawford Building, Crawford School of Public
Policy, Australian National University, Canberra, Australia.
E-mail address: syed.hasan@anu.edu.au (S.A. Hasan).
power supply, electricity theft, blackouts, and poor maintenance are
major issues in the power sector in Bangladesh (World Bank, 2017).
Unavailability of modern energy leads to a high share of energy con-
sumption from traditional sources such as fuel wood, animal waste
and crop residues. Fortunately, the country has small reserves of oil
and coal and relatively large reserves of natural gas, to enhance the
energy security of the country.
Bangladesh has joined in the group of lower–middle income
countries in 2015 and currently has a per capita GDP of US$1384
(Finance Division, 2016). Energy consumption rises sharply between
per capita income US$1000 and US$10,000, the range in which
Bangladesh would belong to at least for a decade with the current
growth scenario (Mujeri et al., 2014). The country has also set up
an agenda to promote itself to a middle-income economy by 2021,
which would require an annual GDP growth rate of 7.5–8.0% for the
remaining period. Such growth is beyond historical rates (around
6.3% since 2010) and will need substantial improvement in the phys-
ical infrastructure of the country including the energy sector (World
Bank, 2016). High economic growth is also expected to substantially
increase the residential use of energy (Burke and Csereklyei, 2016).
At the same time, Bangladesh is aiming to ensure access to electricity
for 96 % of its citizens together with an uninterrupted power supply
to the industries (Planning Commission, 2015).

http://dx.doi.org/10.1016/j.eneco.2017.05.006
0140-9883/© 2017 Elsevier B.V. All rights reserved.
116 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126
Anticipating this rapid increase in the future energy demand,
the Government plans to increase its electricity generation capac-
ity from 14,000 MW in 2015 to 23,000 MW by 2020. With this
goal, the current energy policy of the country largely focuses on
expanding energy generation capacity from all available sources,
which combines both large projects based on traditional sources
(fossil fuel, hydro and nuclear energy) as well as expanding off grid
small projects based on renewable sources (wind, solar and biogas)
(Planning Commission, 2015).
While focusing on energy supply is essential, a better under-
standing of its demand can provide critical insights for forecasting
its future consumption. In particular, an effective knowledge of the
changes in energy demand with respect to income, illustrated by the
Energy Engel Curve (EEC), may provide useful insights into many
aspects of consumer behavior. For example, the EEC may provide
important policy implications which can be useful for a country to
develop an effective energy and power management strategy. Most
importantly, EECs may provide forecasts of future energy demand
as a result of changing demographic scenario (e.g., average house-
hold size) in an economy. Such forecasts can be useful in predicting
changes in demand to match with the energy generation capacity.
EECs are also useful to analyze the share of burden borne by the low-
income people resulting from a tax imposed on energy (Almås, 2012;
Banks et al., 1997; Blundell et al., 1998; Deaton and Muellbauer,
1980; Hasan, 2016).
Most of the studies on Bangladesh that model the relationship
between energy expenditure and income are based on aggregate
data and do not cover the behavioral dynamics at the household
level. For instance, Mozumder and Marathe (2007) examine the
causal relationship between per capita electricity consumption and
per capita GDP in Bangladesh over time. They find a unidirectional
causality from per capita GDP to per capita electricity consumption.
Interestingly, studies find a similar unidirectional causality for the
consumption of natural gas and for the aggregate energy use in the
country (e.g., Das et al., 2013; Paul and Uddin, 2011). In contrast,
Ruhul et al. (2008) find no causal relation between GDP and energy
use for Bangladesh. Among the few studies that rely on house-
hold survey data to investigate the dynamics of energy use with
income, Barnes et al. (2011) estimate an energy poverty line (the
critical minimum amount of energy needed to sustain life) for rural
Bangladesh.
Among studies estimating the EEC (or some variants of it) for
other developing countries, Navajas (2009) explores the contribution
of income and other household characteristics to explain house-
hold consumption of natural gas and liquefied petroleum (but not
electricity) in Argentina. The study estimates Engel curves for these
two types of energy use and explores welfare implications of tar-
iff reforms in natural gas. Related to this strand of literature, a
number of studies investigate the price and income elasticities of
electricity consumption. For instance, to evaluate energy policies in
India, Gundimeda and Köhlin (2008) estimate the price and income
elasticities for different types of fuel (electricity, kerosene, lique-
fied petroleum gas and fuelwood) for different income groups in
urban and rural areas in the country. Golley and Meng (2012) find
that richer households in China consume more energy compared to
their poor counterparts, both directly and indirectly through con-
suming energy intensive goods and services. In a recent paper, Youn
and Jin (2016) study the sensitivity of price increases on household
electricity use in Seoul, Korea and find that progressive pricing has
substantial curbing effects on household electricity use.
From developed country perspectives, Ironmonger et al. (1984)
have estimated Engel curves for disaggregated energy use in
Australia. Using longitudinal household expenditure survey data,
they estimate Engel curves for electricity, natural gas, gasoline and
other fuels for different demographic groups. They find that while
aggregate energy is a necessity, gasoline is a luxury good in Australia.
More recently, Meier et al. (2013) estimate Engel curves for elec-
tricity and gas expenditures using British household panel survey
data and find that income elasticities for energy spending are U-
shaped. They show that income elasticities for both gas and elec-
tricity expenses are less than one (implying energy services are a
necessity) although they rise in the long-run.
Energy consumption, especially when it is clean, may have
improved outcomes on health, education and welfare of household
members (Barnes et al., 2011; Pachauri and Jiang, 2008). Thus raising
household energy consumption is critical, especially in developing
countries with rapidly growing income like Bangladesh. For instance,
a large portion of households in Bangladesh do not have access to
electricity while other sources of energy are not adequate to meet
the rising demand, resulting in a high rate of energy poverty in the
country (Barnes et al., 2011). As a result, the type of relationship
between energy use and income may provide key insights for policy
formulation. However, the issue has not been widely explored in
empirical Engel curve literature.
Against this background, we make an attempt to analyze how
the use of energy at the household level varies with income in
Bangladesh. For our analysis we disaggregate energy use into two
major categories, electricity and all other energy combined (other
energy hereafter). Using the 2010 round of Bangladesh Household
Income and Expenditure Survey (HIES) data, we estimate EECs for
both categories. Our investigation employs a wide range of empiri-
cal techniques for estimation, including quantile, nonparametric and
semiparametric regression methods. Our study finds that, as we
move towards higher income households, expenditure shares reduce
initially for both types of energy but exhibit an opposite pattern
thereafter. We identify the threshold level (turning point) for both
categories and, relying on the current expenditure and its growth,
the number of years required to reach the turning point. Due to
a close association between income and expenditure, the findings
imply a U-shaped relation between energy use and income. Our find-
ings contribute to the literature on energy demand by estimating
disaggregated energy Engel curves for Bangladesh and discussing the
policy implications for energy management and planning in similar
countries.
We organize the rest of the article as follows. In Section 2 we
discuss the empirical strategy and identification issues. A brief dis-
cussion on the data is presented in Section 3. Estimated results are
presented and discussed in Section 4. Section 5 concludes.
2. Empirical strategy and identification
This study follows Hasan (2016) in using similar methods and
data. The latter study is a methodological exercise to indicate that
the empirical specification of food Engel curve has important con-
sequences on food consumption in response to income shocks. This
study uses the same technique and data to understand how the pat-
tern of energy use changes with income, a relatively unexploited area
in the field of Energy Economics.
We use household survey data to estimate Engel curves for elec-
tricity and other energy use. An important consideration in using
household surveys is that the data on expenditure are more reliable
than income, which is particularly true for low-income and agricul-
tural economies (Deaton, 1997; Engel and Kneip, 1996). Thus studies
commonly use expenditure instead of income in modeling Engel
curves. Furthermore, to capture a range of functional forms, depen-
dent variables in such models are usually defined as the share of a
good in total expenditure (budget/expenditure share Engel curves)
(Lewbel, 2008). Since Bangladesh is a low-income country that relies
highly on agriculture, we estimate budget share Engel curves in our
analysis. However, some lumpy non-consumption expenditures are
dropped from our sample and we use expenditure and consumption
 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126
interchangeably. Thus a primary dependent variable in our analysis
is the money spent on electricity use as a share of total expendi-
ture at the household level. We employ a similar approach for other
energy use. Following the convention, we model energy shares on
the logarithm of total expenditure.
Engel curves depend on a number of economic and non-economic
factors (Assimakopoulos and Domenikos, 1991). Demographic char-
acteristics play a particularly important role in modeling Engel
curves as expenditures can be subject to an economy of scale in
household size and number of children. However, when demo-
graphic characteristics enter linearly in an expenditure share Engel
curve, they impose strong restrictions — if the relationship of a
good is linear with (log of) total expenditure, all goods need to be
related with it in the same way (Blundell et al., 1998). To ensure the
functional flexibility over expenditure categories, we employ total
expenditure that is rescaled with an equivalence scale (ES).1
The functional form of expenditure in models of Engel curve is an
empirical issue. Semiparametric (SP) models impose no restrictions
on the functional form of the variables that enter nonparametri-
cally. Thus we employ the following SP specification to examine the
functional form of total expenditure in our models
wih = F(log yh ) + Vh k + uih , (1)
where for each energy type i and household h, w is the energy share
(with separate models for the use of electricity and other energy), F
is an unknown function, ȳ is total expenditure, V is a vector of demo-
graphic and other relevant control variables that enter into the model
linearly and u is the error term [u ∼ NID(0, s 2 )]. The vector V in our
model includes variables like household size, number of adults, age
of household head and its’ square and education of household head
and his/her spouse. Our models additionally include indicators for
urbanization and district fixed effects (sixty-four) to capture regional
heterogeneity. SP estimates from model (1) can be used to partial out
the effect of the variables entering linearly into the model, allowing
to infer the effect of total expenditure on energy shares.
Unfortunately, Engel curve models may suffer from endogeneity
when energy shares are regressed on total expenditure (Summers,
1957). Specifically, a simultaneous decision about total expendi-
ture together with allocation on each category can make the for-
mer endogenous. The usual solution is to employ an Instrumental
Variable (IV) technique or a Control Function (CF) approach. We
employ the latter, which runs regressions of the endogenous vari-
ables on a set of instruments (1st stage), and add the residuals
as additional covariates in the original model (2nd stage). Run-
ning nonparametric regressions of the endogenous variable(s) on the
instrument(s), saving residuals and including it in the SP model can
also control for endogeneity in such models (Newey et al., 1999).
While the CF approach can provide results similar to the IV tech-
nique, one advantage of using the former approach is that it provides
more precise estimates when an endogenous variable is non-linear
(Wooldridge, 2010). The significance of the residuals in the CF
approach may additionally indicate the presence of endogeneity in
such models.
The main challenge in an estimation that relies on an IV or a CF
approach is to find appropriate instrument(s). We employ income,
which is recognized as a good instrument for expenditure and has been
1
The (equivalized) total expenditure is generated by dividing total expenditure
with an ES for Bangladesh. We specifically use the ES identified in Hasan (2016),
which rely on the same data and follows a semiparametric technique to identify the
scale. The employed functional form in the ES allows for a monotonically increasing
economy of scale in household size and proportion of children. See Hasan (2016), for
details.
117
Fig. 1. Movement of energy share with total expenditure.
employed in a number of important Engel curve studies (e.g., Banks
et al., 1997; Bhalotra and Attfield, 1998; Blundell et al., 2007, 1998).
The reason is that income and expenditure of households are highly
correlated (strong instrument) while, in a two stage budgeting system,
income has no direct effect on expenditure shares (valid instrument).2
There is evidence of non-linearity in the use of energy with regard
to total expenditure/income (e.g., Meier et al., 2013). In our SP anal-
ysis we conduct formal tests to find the degree of polynomial in total
expenditure that better fits the model. In doing so, for the sake of
parsimony, our strategy is to start with the null hypothesis that the
simplest (linear) model can fit the SP model. In case we reject the null
hypothesis, next we test the null hypothesis of a quadratic model and
then cubic model and so on.
As we shall see in the Results and discussion section, there is
a U-shaped relationship between energy (electricity/other energy)
share and total expenditure as depicted in Fig. 1. The figure shows
that when total expenditure (C) reaches at the threshold level (C∗ ),
energy (Z) becomes a luxury (expenditure elasticity, nd > 1), while it
is necessary (nd < 1) at values lower than the threshold. The thresh-
old level of total expenditure (C∗ ), which is the turning point, can be
identified by estimating the Energy Engel curve. Once we have some
idea about the functional form of total expenditure from our semi-
parametric model, we can replace the nonparametric part in Eq. (1)
with a parametric specification to estimate a fully parametric Engel
curve for each type of energy in our analysis.
3. Data
We use the 2010 round of Bangladesh Household Income and
Expenditure Survey (HIES) data. The HIES is a cross-sectional sur-
vey repeated with an interval of 5 years and collects household and
individual level data on income, consumption and other important
socioeconomic and demographic variables. The survey relies on a
two-stage stratified random sampling approach for the selection of
households. The total number of surveyed households in 2010 round
of HIES is 12,240 (Bangladesh Bureau of Statistics, 2012). Of those,
we exclude a total of 407 observations which contain missing data
on either income or energy expenditure.
The HIES collects detailed data on energy use, separately for elec-
tricity, kerosene, gas, coal, firewood, cow dung, jute stick, candle,
matchbox and agricultural residue like rice bran, sugarcane bagasse
and tree leaves. As the use of electricity becomes more impor-
tant with growing income, we analyze the use of electricity in our
2
In a two stage budgeting system, given income, household first decide total
expenditure and then given total expenditure, decides expenditure for each category.
The two stage budgeting system is commonly assumed in empirical Engel curve
studies. See Liviatan (1961), Blundell (1988), Baker et al. (1989), Blundell et al. (1998),
and Blundell et al. (2007), for detail.
118 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126

Table 1 Table 3
Summary statistics of household energy use (Taka/month). Mean household energy share (%) by quintiles of total expenditure.

Mean SD Min Max p5 p25 p50 p75 p95 N Quintile All

Electricity 258 243 20 8000 70 130 200 300 650 6515 1 2 3 4 5

Other energy 489 287 4 6280 180 325 450 575 987 11,800
Electricity 2.94 2.44 2.26 2.18 1.92 2.35
Note: Monthly average exchange rate at July 2010 (middle of the survey period): Other energy 7.10 5.64 5.00 4.21 3.06 5.00
1 US$ ≈ 69.44 Taka (the local currency).

Means and standard deviations of key independent variables in

investigation. The use of other types of energy is quite low and, as a our model are presented in Table 4. A high variability of household
result, we combine all of them (other energy hereafter) to conduct income compared to household expenditure is evident which can be
a separate analysis. Table 1 presents the means, standard deviations due to the measurement error in income. This justifies our reliance
and some selected quantiles of money spent on energy use, sepa- on expenditure, rather than income, for analyzing the use of electric-
rately for electricity and other energy. The table demonstrates that ity and other energy.
nearly half of the selected households do not have any expenditure
on electricity.
The zero expenditure of electricity for some households is due 4. Results and discussion
to the lack of access to electricity connections resulting from an
inadequate infrastructure of electricity generation and transmission To examine the association of expenditure shares of electricity
in Bangladesh. Not having an electricity connection can affect the and other energy with total expenditure over the quantiles of the
pattern of a household’s energy use (Heltberg, 2004). Spending on dependent variables, we first perform quantile regressions. The plot
energy use for those households, are the outcome of a constrained of coefficients in Fig. 2 (a) indicates a negative association of (log
optimization. However, an ambitious plan and huge programs for of) total expenditure with electricity share. However, the association
electricity generation have enabled the country to expand the access becomes much stronger as we move along to higher shares of elec-
to electricity from 45% in 2008 to over 74% in 2015 with a tar- tricity expenditure, indicating a nonlinear relation between the two.
geted rate of 96% in 2021 (Planning Commission, 2015). This rising We also observe a similar pattern for other energy share, although
trend of access to electricity would imply that a large proportion of the slope of the plot is much steeper than that of electricity (Fig. 2
households in Bangladesh will start consuming electricity in the near (b)).
future. Since our objective is to understand future energy expen- Assuming other important variables to be uncorrelated with total
diture patterns, it appears reasonable to conduct the analysis on expenditure allows us to employ a nonparametric (NP) technique to
households with electricity connections. examine the relation of energy shares with (log of) total expenditure.
Another reason for excluding households without electricity con- The NP regression results again indicate a nonlinear relationship
nection from our sample is the concern about potential reverse of total expenditure with the expenditure shares of both types of
causality between electricity use and income. In particular, while
income of a household explains whether it has an electricity con-
nection or not, connectivity to a power source may also significantly Table 4
affect income (Bridge et al., 2016). Thus, our final sample includes Summary statistics of independent variables.
a total of 6515 households that have electricity connections at their Mean SD
residence.
Finances
Summary statistics for the dependent variables in our sample is Household expenditure (Taka/month) 11,961 8572
presented at Table 2, showing that the mean expenditure on elec- Household income (Taka/month) 16,172 32,297
tricity use and other energy as a ratio of total expenditure is 2.4% Expenditure per capita (Taka/month) 2730 1986
and 5.0%, respectively. Households at the bottom 5 percentiles have Income per capita (Taka/month) 3789 7243
a share of electricity use of 0.8% while families at the top 5 per-
Demographics
centiles have a share of 4.8% indicating a good deal of variation in Family size 4.64 1.92
this spectrum. A similar pattern is also observed in the use of other Adult member 2.74 1.28
energy. Interestingly, households without any electricity connection, Child member 1.90 1.32
who are excluded from our analysis sample, also have a similar Age of household head 46.05 13.44

expenditure share for other energy.

Education
Table 3 indicates that mean proportion of both electricity and Household head with no education 0.41 0.49
other energy use reduce with the increase in total expenditure. Household head with primary education 0.15 0.36
However, while the mean expenditure share of other energy is Household head with secondary education 0.28 0.45
reduced to half as we move from the bottom to the top quintile of Household head with higher secondary education 0.12 0.32
Household head with graduate degree 0.04 0.19
total expenditure, it is only reduced by one-third for electricity. This Spouse of head with no education 0.42 0.49
indicates a faster increase in the use of electricity compared to the Spouse of head with primary education 0.18 0.38
use of other energy. Spouse of head with secondary education 0.33 0.47
Spouse of head with higher secondary education 0.06 0.24
Spouse of head with graduate degree 0.01 0.11

Other
Table 2 Urban 0.54 0.50
Household energy use as share (%) of total expenditure. N 6515

Mean SD Min Max p5 p25 p50 p75 p95 N Note: Education, employment and other categories a dummy variables and associated
means indicate their sample proportions. We considered the senior most male mem-
Electricity 2.35 1.49 0.17 27.33 0.79 1.40 2.05 2.88 4.78 6515
ber as the head of the household. Monthly average exchange rate at July 2010 (middle
Other energy 5.00 2.83 0.00 34.66 1.44 3.09 4.52 6.46 10.08 6515
of the survey period): 1 US$≈ 69.44 Taka (the local currency).
 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126 119

a.Electricity

0.00 -0.01
Total expenditure
-0.01 -0.01
-0.02

.05 .15 .25 .35 .45 .55 .65 .75 .85 .95
Quantile
b.Other energy
-0.01
-0.02
Total expenditure
-0.03 -0.04

.05 .15 .25 .35 .45 .55 .65 .75 .85 .95
Quantile

Fig. 2. Coefficient plot of energy shares by quantiles (OLS estimates in dashed lines).

energy use, electricity and other energy (Fig. 3). Moreover, the share
of other energy appears to decrease at a faster rate than the share of
electricity.
Since other independent variables may be correlated with total
expenditure, a SP model would be more appropriate to investigate
the relationship of energy use with (log of) total expenditure com-
pared to a model that is NP in nature. We employ total income, as
described in Section 2, to address the endogeneity in total expen-
diture. The SP model estimates, presented in Table 5, indicate that
key independent variables have significant impact on the use of both
electricity and other energy. As we expect, education of households
(head and spouse) as well as urban residency appear to have a posi-
tive association with the energy expenditure shares, particularly for
electricity. Among other variables, significance of residuals from the
first stage indicates that total expenditure may be endogenous in our
model for electricity use.
Employing the estimates in Table 5, we calculate the partialed out
(of other independent variables) value of energy shares. Scatter plots
of these values in Fig. 4, together with their NP fit, again indicate
nonlinear relationships between energy shares (both electricity and
other energy) and total expenditure. Formal tests of the functional
forms, outlined in Hardle and Mammen (1993), reject models that
are linear in (log of) total expenditure but cannot reject a quadratic
specification for both types of energy use.3 As we cannot reject the
3
All tests in our analysis are conducted at the 5% level of significance. See Appendix,
Table A.1 for the p-values from Hardle and Mammen (1993) test.
120 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126

a. Electricity

.3
Actual data
Local linear fit
95 % CI

Expenditure share
.2
.1
0

6 8 10 12
Total expenditure
b. Other energy
.4

Actual data
Local linear fit
95 % CI
.3
Expenditure share
.2
.1
0

6 8 10 12
Total expenditure
Fig. 3. Nonparametric fit of energy shares.

null hypothesis of a quadratic fit for both types of energy in our
analysis, we do not test the null hypothesis of a cubic fit.
Since our test of SP models indicates quadratic specifications for
both types of energy use, we fit models that include both (log of) total
expenditure and its square. Endogeneity in such models may occur
due to both the linear and the quadratic term. Thus, in the first stage,
we regress total expenditure on total income and its square and col-
lect the residuals. We repeat the same process with squared (log of)
total expenditure. Next, in the second stage, along with other covari-
ates, we include residuals from previous regressions in the original
model (2nd stage).
For the case of electricity share, the included residuals from two
first stage regressions are not significant individually (in the second
stage as two additional explanatory variables) but highly significant
jointly. As a result, following the convention of the CF approach, we
only include the set of residuals which are generated in the first
stage by regressing total expenditure on the instruments and other
covariates. The significance of the residuals in the model indicates
that while total expenditure can be endogenous in the model for
electricity share, residuals generated through regressing the linear
term are enough to control for such endogeneity. On the other hand,
both sets of residuals from the first stage regressions are individu-
ally and jointly significant in our model for other energy share. Thus
they indicate that controlling for the endogeneity of total expendi-
ture requires using residuals generated through regressing the linear
as well as the quadratic term in our model for other energy share.
Our parametric model estimates indicate negative marginal
effects of total expenditure on the shares of both types of energy use
(Table 6). In the model for electricity, total expenditure is only sig-
nificant at the 10% level when all the control variables are dropped
 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126 121

Table 5 expenditure shares are typically large even within income groups
Semiparametric estimates from energy share model. (Cronin et al., 2017; Pizer and Sexton, 2015).
Electricity Other energy The individual coefficient of total expenditure and its square are
Family size −0.0008 ∗∗∗
−0.0003 presented in Table 7 for both types of energy use in our analy-
(0.0002) (0.0002) sis. For electricity use, we observe a negative coefficient for total
Number of adults −0.0014∗∗∗ −0.0046∗∗∗ expenditure and a positive coefficient for its square. This implies a
(0.0003) (0.0004) U-shaped relationship of electricity share with total expenditure. The
Household head’s age 0.0001 0.0001
coefficients for the share of other energy indicate a similar pattern.
(0.0001) (0.0001)
Square of household head’s age 0.0000 −0.0000 Thus, as total expenditure increases households initially increase
(0.0000) (0.0000) their expenditure of energy less than proportionally. However, after
Household head has primary education 0.0002 −0.0016 reaching a threshold level of total expenditure, energy expenditure
(0.0005) (0.0010) increases more than proportionally as total expenditure continues
Household head has secondary education 0.0012∗∗ −0.0024∗∗∗
(0.0005) (0.0008)
to rise. Our results thus indicate a quadratic EEC for both electric-
Household head has higher secondary education 0.0032∗∗∗ −0.0018* ity and other energy use in Bangladesh and are largely similar to
(0.0009) (0.0011) earlier studies. For example, Banks et al. (1997) employ a nonlin-
Household head has graduate degree 0.0023∗∗ −0.0009 ear model in expenditure and find a quadratic Engel curve for fuel.
(0.0012) (0.0018)
Barnes et al. (2011) observe a similar pattern for energy use (all
Household head has other education 0.0031* 0.0014
(0.0018) (0.0048) combined) in Bangladesh. A close association between total expendi-
Spouse of head has primary education 0.0018∗∗∗ 0.0001 ture and income of households also indicates a U-shaped relationship
(0.0006) (0.0009) of energy shares with income.4
Spouse of head has secondary education 0.0033∗∗∗ 0.0015* The observed pattern of energy demand is plausible for low-
(0.0005) (0.0009)
Spouse of head has higher secondary education 0.0052∗∗∗ 0.0034∗∗
income countries where people may have unmet demands for nec-
(0.0011) (0.0015) essary goods and services (e.g., food, housing and education) at low
Spouse of head has graduate degree 0.0064∗∗∗ 0.0061∗∗ incomes. As their incomes rise, they spend their additional earn-
(0.0017) (0.0031) ings on these necessary goods and services more than proportionally,
Spouse of head has other education 0.0131* 0.0094
resulting in a reduction in the share of other types of expenditure
(0.0067) (0.0070)
Urban 0.0054∗∗∗ −0.0056∗∗∗
(Hasan, 2016). When income increases enough, people may increase
(0.0005) (0.0012) spending more than proportionally on categories like energy. This
Residual −0.0064∗∗∗ −0.0014 may result in a U-shaped relationship between energy share and
(0.0011) (0.0017) total expenditure. Furthermore, for electricity, this effect can be
Adjusted R2 0.17 0.17
stronger as people at higher incomes usually buy electronic items
N 6515 6515
(e.g., phone, TV and computer) and appliances (e.g., refrigerator, air
Note: We reported marginal effects from semiparametric models with (64) district conditioner and fan) that heavily rely on electricity.
fixed effects; clustered standard errors are reported in parentheses.
∗
p < 0.10.
Our results are robust to a number of modifications in model
∗∗
p < 0.05. specification. We obtain similar results as we repeat the analysis
∗∗∗ with either total (unequivalized) or per capita value of total expen-
p < 0.01.
diture (Tables 7, A.2 and A.3). We also find similar results when we
employ expenditure that is equivalized by either the OECD or the
square root of family size (SRFS) scale (Tables A.4 and A.5). Earlier
(Column 1). However, as we add controls variables (Column 2) and studies indicate that there are often economies of scale in family
district fixed effects (Column 3), the coefficient becomes highly sig- size as well as number of children in the household (Hasan, 2016).
nificant. In all cases residuals from the first stage remain highly Furthermore, demographic structures like family size is expected to
significant, again validating our suspicion about the exogeneity of change with time and the level of development. In such circum-
total expenditure in our model. While the pattern is similar for other stances, analyses with equivalized expenditure appear to be more
energy, a higher marginal value of the coefficient of total expenditure appropriate for analyzing energy use. As a result, for the rest of
reflects the fact that at higher levels of income households reduce this analysis, we emphasize on models that employ equivalized
their expenditure share of other energy at a faster rate compared to expenditure.
that of electricity. Finally, a model that is cubic in total expenditure indicates a
Estimates of other important variables from the preferred model slightly higher (in absolute value) marginal effect of total expendi-
(Columns 3 and 6, Table 6) indicate effects that are largely similar to ture on the share of both types of energy use (Table A.6). However,
that of the semiparametric models. For example, household head’s Bayesian information criterion (BIC) and the preference for parsi-
age and head/spouse’s education have positive impact on energy mony ranks quadratic model over the cubic one for both electricity
shares. This can be due to the fact that household member’s edu- and other energy (see Tables 6 and A.6). As a result, in combina-
cation is usually associated with their socioeconomic status, which tion with our findings in the SP analysis, we have chosen quadratic
again can be related with the ownership of energy intensive items specifications for modeling both energy groups in our analysis.
like electronics and appliances. Among other important variables, Our conclusions also do not change qualitatively with the use of
urban households consume more of electricity and less of other the CF approach compared to the conventional IV technique. Since
energy than rural households. Interestingly, the two quantities are with the former approach both sets of residuals from the first stage
roughly similar indicating that urban households substitute electric- are significant in the second stage of the model for other energy,
ity for other energy. Coefficient of family size is significant in Table 6 the coefficients for both the endogenous variables are exactly the
but negative for electricity use and positive for other energy use. same regardless of the use of the CF approach or the IV technique.
Since we have already employed total expenditure that is equival-
ized, this implies that smaller households consume less electricity
but more of other energy. Thus our results indicate differences in the 4
More appropriately, we mean permanent income which refers to the long-run
expenditure pattern of the two types of energy. Finally, the explana- income of households. The advantage of using total expenditure as a proxy for
tory power of our models appears reasonable as variations in energy permanent income is provided in Meyer and Sullivan (2012).
122 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126

a. Electricity

.3
Partiled out value
Local linear fit
95 % CI

.2
Expenditure share
.1
0
-.1

6 8 10 12
Total expenditure
b. Other energy
.4

Partiled out value

Local linear fit
95 % CI
.3
Expenditure share
.2
.1
0

6 8 10 12
Total expenditure
Fig. 4. Nonparametric fit of partialed out energy shares.

On the other hand, IV estimates for both the endogenous variables
were similar but higher in absolute value in the model for electricity
share. However, employing the CF approach indicates that residuals
from the first stage regression of (the endogenous) squared total
expenditure are not significant and we drop them (while keeping the
residuals from the other first stage regression) to have more precise
estimates compared to the usual IV estimates.5
To make our analysis useful for policy formulation, Fig. 5 plots the
predicted expenditure shares of energy use and their quadratic fit
against total expenditure (in levels). It indicates that the expenditure
share of electricity exhibits a U-shaped pattern. On the other hand,
5
Results are available from the authors upon request.
we observe a rapid reduction in the share of other energy which does
not rise much with total expenditure and more closely resembles
a flipped J-shaped pattern within the range of total expenditure in
our data. Although, other energy takes a U-shaped pattern when we
consider a broader spectrum of total expenditure that is beyond the
range in our sample.
Next, we employ the earlier model estimates to find the mean
expenditure (with 95% confidence intervals) at which the shares
of energy use start to increase (Table 8). Based on the current
total expenditure and the growth rate of the economy (as a proxy
to total expenditure growth), we also report the number of years
required to reach at the turning points. Our analysis indicates that,
for Bangladesh, it would require 17 additional years for electricity
expenditure to be at a minimum (turning point) as a proportion of
 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126 123

Table 6
OLS estimates from quadratic energy share model with equivalised expenditure.

Electricity Other energy

(1) (2) (3) (4) (5) (6)

∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗
Total expenditure −0.0016* −0.0055 −0.0044 −0.0279 −0.0198 −0.0406∗∗∗
(0.0008) (0.0010) (0.0008) (0.0018) (0.0030) (0.0031)
Age of household head 0.0001∗∗∗ 0.0001∗∗∗ 0.0001 0.0001∗∗∗
(0.0000) (0.0000) (0.0000) (0.0000)
Family size −0.0010∗∗∗ −0.0013∗∗∗ −0.0015∗∗∗ 0.0010∗∗
(0.0002) (0.0002) (0.0004) (0.0004)
Number of adults −0.0007∗∗∗ −0.0004 −0.0030∗∗∗ −0.0057∗∗∗
(0.0003) (0.0003) (0.0006) (0.0005)
Household head has primary education −0.0000 −0.0001 −0.0006 −0.0003
(0.0006) (0.0006) (0.0013) (0.0011)
Household head has secondary education 0.0005 0.0005 −0.0032∗∗∗ −0.0003
(0.0006) (0.0006) (0.0009) (0.0008)
Household head has higher secondary education 0.0024∗∗∗ 0.0021∗∗ −0.0035∗∗ 0.0006
(0.0009) (0.0009) (0.0014) (0.0012)
Household head has graduate degree 0.0015 0.0011 −0.0034* 0.0012
(0.0014) (0.0013) (0.0021) (0.0019)
Household head has other education 0.0034 0.0025 −0.0009 0.0031
(0.0021) (0.0018) (0.0058) (0.0045)
Spouse of head has primary education 0.0013∗∗ 0.0013∗∗ −0.0011 0.0021∗∗
(0.0006) (0.0006) (0.0010) (0.0010)
Spouse of head has secondary education 0.0023∗∗∗ 0.0023∗∗∗ −0.0007 0.0037∗∗∗
(0.0006) (0.0005) (0.0010) (0.0010)
Spouse of head has higher secondary education 0.0041∗∗∗ 0.0036∗∗∗ 0.0002 0.0041∗∗∗
(0.0012) (0.0011) (0.0017) (0.0015)
Spouse of head has graduate degree 0.0054∗∗∗ 0.0040∗∗ 0.0029 0.0012
(0.0020) (0.0018) (0.0033) (0.0032)
Spouse of head has other education 0.0094 0.0089 0.0013 0.0086
(0.0067) (0.0070) (0.0066) (0.0069)
Urban 0.0053∗∗∗ 0.0050∗∗∗ −0.0062*** −0.0050∗∗∗
(0.0007) (0.0007) (0.0013) (0.0013)
Residual 1 −0.0095∗∗∗ −0.0054∗∗∗ −0.0066∗∗∗ 0.1483∗∗ −0.1284 0.4752∗∗∗
(0.0011) (0.0013) (0.0011) (0.0689) (0.0976) (0.1005)
Residual 2 −0.0088∗∗ 0.0071 −0.0274∗∗∗
(0.0040) (0.0056) (0.0058)
Constant 0.2252∗∗∗ 0.1795∗∗∗ 0.1834∗∗∗ 1.2534∗∗∗ 0.0748 2.6826∗∗∗
(0.0551) (0.0516) (0.0556) (0.2837) (0.4129) (0.4289)
District fixed effects (64) No No Yes No No Yes
Adjusted R2 0.07 0.14 0.20 0.22 0.27 0.35
BIC −36,715 −37,177 −37,689 −29,541 −29,902 −30,694
N 6515 6515 6515 6515 6515 6515

Note: We reported marginal effects from OLS estimates; bootstrapped clustered standard errors are reported in parentheses.
∗
p < 0.10.
∗∗
p < 0.05.
∗∗∗
p < 0.01.

total expenditure, which would increase thereafter. The estimated
time is much lower (7 years) for other types of energy to reach at the
turning point.
5. Conclusion
This study investigates the pattern of household energy expendi-
ture in Bangladesh. We employ the 2010 round of Household Income
and Expenditure Survey data and select households who have elec-
tricity connection at their residences. Based on a wide range of
empirical specifications we find that the expenditure share of both
electricity and other energy changes in a nonlinear U-shaped fash-
ion as total expenditure rises. This indicates that at higher incomes
expenditure on energy as a proportion of income reduces initially.
However, after reaching a threshold level of income, energy expen-
diture as a proportion of income starts to increase. We also provide

Table 7
OLS estimates of total expenditure coefficients.

Equivalised Household Per capita

Electricity Other Electricity Other Electricity Other

energy energy energy

Linear term −0.0333∗∗∗ −0.5846∗∗∗ −0.0504∗∗∗ −0.2817∗∗∗ −0.0607∗∗∗ −0.3129∗∗∗

(0.0129) (0.0999) (0.0109) (0.0517) (0.0114) (0.0715)
Quadratic term 0.0017∗∗ 0.0322∗∗∗ 0.0025∗∗∗ 0.0135∗∗∗ 0.0029∗∗∗ 0.0180∗∗∗
(0.0008) (0.0058) (0.0006) (0.0027) (0.0006) (0.0045)
Adjusted R2 0.20* 0.35 0.22 0.36 0.22 0.34
N 6515 6515 6515 6515 6515 6515

Note: We report OLS estimates with CF approach which include (64) district fixed effects; bootstrapped clustered standard errors are reported in parentheses.
∗
p < 0.10.
∗∗
p < 0.05.
∗∗∗
p < 0.01.
124 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126

a. Electricity

.05
Predicted (means)
Quadratic fit
95% CI of prediction

.04
Expenditure share
.03
.02
.01

0 5000 10000 15000 20000

Total expenditure (Taka/month)
b. Other energy
Predicted (means)
.12

Quadratic fit
95% CI of prediction
.1
Expenditure share
.08
.06
.04
.02

0 5000 10000 15000 20000

Total expenditure (Taka/month)
Fig. 5. Predicted energy share and fitted line.

a socio-demographic snapshot of energy expenditure as Bangladesh
makes transition from a low- to middle-income country. Our analy-
ses predict that electricity expenditure of urban, educated and larger
households is higher compared to their counterparts in Bangladesh
while the opposite is true for other types of energy.
The findings in our analysis are important for ensuring access to
energy, particularly to electricity, in Bangladesh. For example, identi-
fication of the turning points can give us some idea about the income
level after which the demand for electricity and other energy would
increase rapidly in the country. Thus the turning point estimates can
be specifically useful for future planning in expanding the access
to electricity and other energy sources in Bangladesh. Such plan-
ning is crucial as increasing electricity supply from some targeted
sources, like nuclear and hydro based plants as well as importing
from neighboring countries, may take a long time to materialize
(Planning Commission, 2015).
Bangladesh also aims to reduce the use of certain types of
energy like fuel wood, animal waste and crop residues that emits
green house gases and leads to adverse health outcomes (Planning
Commission, 2015). Thus the Government may consider incentiviz-
ing energy uses from sources that are benign to environment and
human health. Such instruments may also be used to facilitate access
to affordable electricity in promoting economic growth and poverty
reduction. However, such incentives need to be well targeted with
minimum price distortions as the growth prospect of the country can
be lower otherwise (World Bank, 2010). Findings from this study can
be specifically useful to indicate the share of the burden borne by
low-income people resulting from a tax/subsidy imposed on energy
and related welfare implication in the country.
This study may also supplement the estimation of global energy
demand. World energy consumption is expected to rise by 48%
between 2012 and 2040. Since a rise in global energy demand would
 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126 125

Table 8
Turning points of energy shares (in Taka).

Electricity Other energy

Lower Upper Lower Upper

With respect to bound Mean bound bound Mean bound

Total expenditure (equivalised) 5448 16,650 50,892 7546 8770 10,191

Years required [16.77] [7.30]

Implied expenditure levels for households with:

1 adult & 0 child 5448 16,650 50,892 7546 8770 10,191
1 adult & 1 child 6119 18,702 57,162 8476 9850 11,447
1 adult & 2 children 6765 20,677 63,198 9371 10,890 12,656
1 adult & 3 children 7390 22,587 69,038 10,237 11,897 13,825
1 adult & 4 children 7997 24,443 74,709 11,078 12,874 14,961
2 adults & 0 child 9098 27,809 84,998 12,604 14,647 17,021
2 adults & 1 child 9664 29,540 90,287 13,388 15,558 18,081
2 adults & 2 children 10,219 31,235 95,470 14,156 16,451 19,119
2 adults & 3 children 10,764 32,899 100,555 14,910 17,328 20,137
2 adults & 4 children 11,298 34,534 105,552 15,651 18,189 21,138
3 adults & 0 child 12,282 37,540 114,740 17,014 19,772 22,978
3 adults & 1 child 12,793 39,103 119,517 17,722 20,595 23,934
3 adults & 2 children 13,298 40,644 124,228 18,421 21,407 24,878
3 adults & 3 children 13,795 42,165 128,877 19,110 22,208 25,808
3 adults & 4 children 14,287 43,667 133,467 19,791 22,999 26,728
4 adults & 0 child 15,196 46,446 141,962 21,050 24,463 28,429
4 adults & 1 child 15,671 47,899 146,402 21,709 25,228 29,318
4 adults & 2 children 16,141 49,336 150,796 22,360 25,985 30,198
4 adults & 3 children 16,607 50,759 155,145 23,005 26,735 31,069
4 adults & 4 children 17,068 52,168 159,452 23,644 27,477 31,931

Notes: 1. Turning points and upper and lower limits (95% CI) are based on estimates in Table 6.
2. Values correspond to the exponentials of the estimates, as total expenditure enters into the model in logarithmic form.
3. Numbers inside the square brackets report the average years required to reach the turning point, given the country’s recent growth of GDP per capita (≈ 7%/year) and assuming
no change in the distribution of total expenditure over time.

also be accompanied by a significant switch towards modern energy,
the growth is estimated to be much higher for electricity (69%).
On the other hand, economic growth is a key determinant of rising
energy demand and the fastest rates of energy growth are pro-
jected for emerging countries where combined GDP would increase
annually by 4.2% from 2012 to 2040, compared to a global rate
3.3% per year (USEIA, 2016). Thus, significant growth in the elec-
tricity demand is expected to take place in the developing nations
where rising living standards would increase the use of home appli-
ances and electronic devices as well the use of commercial services,
including hospitals, schools, office buildings and shopping malls
(IEA, 2016; USEIA, 2016). Our study provides a case study of future
energy demand for Bangladesh in the likely case of a changing demo-
graphic structure with varying household size. Together with other
country level estimates, this may contribute in providing a more
comprehensive picture of future energy demand at the global level.
Acknowledgments
We thank Co-Editor Perry Sadorsky, two anonymous reviewers,
Jane Golley, Shahe Emran, Syed Ahsan and participants of the 50th
Annual Conference of the Canadian Economics Association (CEA) for
providing useful comments.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://
dx.doi.org/10.1016/j.eneco.2017.05.006.
References
Almås, I., 2012. International income inequality: measuring PPP bias by estimating
Engel curves for food. Am. Econ. Rev. 102 (2), 1093–1117.
Assimakopoulos, V., Domenikos, H., 1991. Consumption preferences structure of Greek
households. Energy Econ. 13 (3), 163–167.
Baker, P., Blundell, R., Micklewright, J., 1989. Modelling household energy expendi-
tures using micro-data. Econ. J. 99 (397), 720–738.
Bangladesh Bureau of Statistics, 2012. Report of the household income and expendi-
ture survey - 2010. Final Report. Bangladesh Bureau of Statistics, (April).
Banks, J., Blundell, R., Lewbel, A., 1997. Quadratic Engel curves and consumer demand.
Rev. Econ. Stat. 79 (4), 527–539.
Barnes, D.F., Khandker, S.R., Samad, H.A., 2011. Energy poverty in rural Bangladesh.
Energy Policy 39 (2), 894–904.
Bhalotra, S., Attfield, C., 1998. Intrahousehold resource allocation in rural Pakistan: a
semiparametric analysis. J. Appl. Econ. 13 (5), 463–480.
Blundell, R., 1988. Consumer behaviour: theory and empirical evidence—a survey.
Econ. J. 98 (389), 16–65.
Blundell, R., Chen, X., Kristensen, D., 2007. Semi-nonparametric IV estimation of
shape-invariant Engel curves. Econometrica 75 (6), 1613–1669.
Blundell, R., Duncan, A., Pendakur, K., 1998. Semiparametric estimation and consumer
demand. J. Appl. Econ. 13 (5), 435–461.
Bridge, B.A., Adhikari, D., Fontenla, M., 2016. Household-level effects of electricity on
income. Energy Econ. 58, 222–228.
Burke, P.J., Csereklyei, Z., 2016. Understanding the energy-GDP elasticity: a sectoral
approach. Energy Econ. 58, 199–210.
Cronin, J.A., Fullerton, D., Sexton, S.E., 2017. Vertical and horizontal redistributions
from a carbon tax and rebate. NBER Working paper 23250. National Bureau of
Economic Research.
Das, A., McFarlane, A.A., Chowdhury, M., 2013. The dynamics of natural gas consump-
tion and GDP in Bangladesh. Renew. Sust. Energ. Rev. 22, 269–274.
Deaton, A., 1997. The Analysis of Household Surveys: A Microeconometric Approach
to Development Policy. World Bank, Washington D.C, USA.
Deaton, A., Muellbauer, J., 1980. Economics and Consumer Behaviour. 1st edition,
Cambridge University Press, Cambridge, USA.
Engel, J., Kneip, A., 1996. Recent approaches to estimating Engel curves. J. Econ. 63 (2),
187–212.
Finance Division, 2016. Bangladesh economic review. Annual Review. Ministry of
Finance, Bangladesh Government, (June. Available from: http://www.mof.
gov.bd/en/index.php?option=com_content&amp;view=article&amp;id=367&
amp;Itemid=1 [Accessed: 29 January 2017]).
Golley, J., Meng, X., 2012. Income inequality and carbon dioxide emissions: the case of
Chinese urban households. Energy Econ. 34 (6), 1864–1872.
Gundimeda, H., Köhlin, G., 2008. Fuel demand elasticities for energy and environ-
mental policies: Indian sample survey evidence. Energy Econ. 30 (2), 517–546.
Hardle, W., Mammen, E., 1993. Comparing nonparametric versus parametric regres-
sion fits. Ann. Stat. 21 (4), 1926–1947.
Hasan, S.A., 2016. Engel curves and equivalence scales for Bangladesh. J. Asia Pac. Econ.
21 (2), 301–315.
Heltberg, R., 2004. Fuel switching: evidence from eight developing countries. Energy
Econ. 26 (5), 869–887.
126 S. Hasan, P. Mozumder / Energy Economics 65 (2017) 115–126
IEA World Energy Outlook, 2016. Annual publication, The International Energy
Agency (IEA), November 2016. Available from: http://www.iea.org/bookshop/
720-World_Energy_Outlook_2016 [Accessed: 29 January 2017].
Ironmonger, D., Manning, I., Van Hoa, T., 1984. Longitudinal working models: esti-
mates of household energy consumption in Australia. Energy Econ. 6 (1), 41–46.
Lewbel, A., 2008. Engel curves. In: Durlauf, S.N., Blume, L., Blume, E.L. (Eds.), The New
Palgrave Dictionary of Economics. Palgrave Macmillan Basingstoke, NY, pp. 1–7.
chapter 1.
Liviatan, N., 1961. Errors in variables and Engel curve analysis. Econometrica 29 (3),
336–362.
Meier, H., Jamasb, T., Orea, L., 2013. Necessity or luxury good? Household energy
spending and income in Britain 1991–2007. Energy J. 34 (4), 109–129.
Meyer, B.D., Sullivan, J.X., 2012. Consumption and income poverty in the United States.
In: Jefferson, P.N. (Ed.), The Oxford handbook of the economics of poverty. Oxford
University Press, New York, pp. 49–74.
Mozumder, P., Marathe, A., 2007. Causality relationship between electricity consump-
tion and GDP in Bangladesh. Energy Policy 35 (1), 395–402.
Mujeri, M.K., Chowdhury, T.T., Shahana, S., 2014. Energy Sector in Bangladesh: An
Agenda for Reforms. GSI Report, Available from: https://www.iisd.org/sites/
default/files/publications/energy_sector_bangladesh.pdf [Accessed: 25 January
2017].
Navajas, F.H., 2009. Engel curves, household characteristics and low-user tariff
schemes in natural gas. Energy Econ. 31 (1), 162–168.
Newey, W.K., Powell, J.L., Vella, F., 1999. Nonparametric estimation of triangular
simultaneous equations models. Econometrica 67 (3), 565–603.
Pachauri, S., Jiang, L., 2008. The household energy transition in India and China. Energy
Policy 36 (11), 4022–4035.
Paul, B.P., Uddin, G., 2011. Energy and output dynamics in Bangladesh. Energy Econ.
33 (3), 480–487.
Pizer, W.A., Sexton, S., 2017. Distributional Impacts of Energy Taxes. Nber working
paper 23318, National Bureau of Economic Research.
Commission, Planning, 2015. Seventh Five Year Plan (FY2016 - FY2020): Accelerating
Growth, Empowering Citizens. Planning Commission, Bangladesh Government.
December.
Ruhul, S., Rafiq, S., Hassan, A.K., 2008. Causality and dynamics of energy consumption
and output: evidence from non-OECD Asian countries. J. Econ. Dev. 33 (2), 1–26.
Summers, R., 1957. A Note on Least Squares Bias in Household Expenditure
Analysis. Discussion Paper 29, Cowles Foundation for Research in Economics,
Yale University.
USEIA, 2016. International Energy Outlook 2016. Annual publication, U.S. Energy Infor-
mation Administration, May 2016. Available from: www.eia.gov/forecasts/ieo/
pdf/0484(2016).pdf [Accessed: 29 January 2017].
Wooldridge, J.M., 2010. Econometric Analysis of Cross Section and Panel Data. 2
edition, MIT Press, Cambridge MA., pp. 02142–11209.
World Bank, 2010. Addressing the Electricity Access Gap. Background Paper for
the World Bank Group Energy Sector Strategy, World Bank, Washington,
D.C. Available from: http://siteresources.worldbank.org/EXTESC/Resources/
Addressing_the_Electricity_Access_Gap.pdf [Accessed: 29 January 2017].
World Bank, 2016. Bangladesh - Country Snapshot. Country Report, World Bank,
Washington, D.C.. Available from: http://documents.worldbank.org/curated/
en/654391476782032287/Bangladesh-Country-snapshot [Accessed: 25 January
2017].
World Bank, 2017. Bangladesh: Ensuring a Reliable and Quality Energy Supply.
Media Report, Available from: http://www.worldbank.org/en/results/2016/10/
07/bangladesh-ensuring-a-reliable-and-quality-energy-supply [Accessed: 25
January 2017].
Youn, H., Jin, H.J., 2016. The effects of progressive pricing on household electricity use.
J. Policy Model 38 (6), 1078–1088.