Accepted Manuscript

Understanding the Determinants of Household Cooking Fuel Choice in

Afghanistan: A Multinomial Logit Estimation

Uttam Paudel, Umesh Khatri, Krishna Prasad Pant

PII: S0360-5442(18)30905-8

DOI: 10.1016/j.energy.2018.05.085

Reference: EGY 12920

To appear in: Energy

Received Date: 25 September 2017

Revised Date: 20 March 2018

Accepted Date: 11 May 2018

Please cite this article as: Uttam Paudel, Umesh Khatri, Krishna Prasad Pant, Understanding the
Determinants of Household Cooking Fuel Choice in Afghanistan: A Multinomial Logit Estimation,
Energy (2018), doi: 10.1016/j.energy.2018.05.085

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
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.
 ACCEPTED MANUSCRIPT

Understanding the Determinants of Household Cooking Fuel Choice in

Afghanistan: A Multinomial Logit Estimation

Uttam Paudela*, Umesh Khatrib and Krishna Prasad Pantc

aPhD Scholar, Tribhuvan University, Kathmandu, Nepal

Researcher, USAID, Kathmandu, Nepal
b2

cVisiting Faculty, Central Department of Economics, Tribhuvan University, Kathmandu, Nepal

*Corresponding author: uuupaudel22@gmail.com, Phone: +977-9861898926

Abstract

Cooking energy demand in Afghanistan has mostly fulfilled by traditional energy sources despite
availability of health and environment friendly clean energy options internationally.
Understanding the determinants of household cooking fuel choice either to prefer clean or
traditional energy sources is therefore important to identify policy options for clean energy
promotion. For this, a nationally representative household information using Afghanistan
Demographic and Household Survey 2015 is analyzed using multinomial logistic regression
model. The results show that residence in urban areas, availability of electricity, higher household
wealth, high education, married status and separate cooking place positively affect choice of
liquefied petroleum-gas as compared to traditional fuels; whereas large family size and aged
household head indeed have positive linkages on probability of choosing animal dung as cooking
fuel. Robustness of the results shows that wealthy households are more likely to use liquefied
petroleum gas and fuel wood compared to poor households, therefore confirm priority for wealth
generation. For motivating the rural households to use clean-cooking-fuel, government and other
cohorts should put effort on increasing affordability of rural households to clean energy sources
through provision and employment opportunities, improving access to electricity and creating
awareness about the health and environmental benefits of clean energy.

Key Words: Cooking Fuel; Clean Energy; Afghanistan; Multinomial-Logit; Choice Probability.

JEL Code: Q42

1. Introduction

A decline of petroleum oil consumption in residential sector, for example two third of total oil
consumption replaced by natural gas from 2000 to 2015, suggests quality and convenience of fuel
usually escort a tradeoff between quality and utility (benefit)[1]. Household demands for diesel
and liquefied petroleum gas (LPG) have increased while that for kerosene and gasoline
declined[1]. But, households in developing countries still depend heavily on fuel wood and other
solid fuel-based energy sources for cooking fuel. This is either due to lack of consistent sources of
modern energy or to the lavishness of cheap fuel wood nearby. In developing countries, more than
2.5 billion people use solid fuels, for example fuel wood, charcoal, agricultural waste and animal
dung to meet their energy needs for cooking [2–4].

To provide universal access to clean energy facilities, International Energy Association (IEA) has
established a global energy scenario through the year 2030 called ‘‘Universal Modern Energy
 ACCEPTED MANUSCRIPT

Access Case’’ [1]. There would be 100% entrance of clean cooking conveniences but, a notable
issue is that roughly 80% of these people reside in rural areas and some of them under conflict
states [5] and they find difficult to switch to modern energy options. According to the Global
Energy Assessment, there is significant role of strong regulation, standards, investment and
policies to support the sustainable energy transition and to encourage rapid transition from
traditional biomass to clean fuels for the improvement of energy access in developing countries
[6]. Some studies explain energy ladder or inter fuel substitution to identify and analyze potential
determinants with a gradient of household cooking energy choice starting from solid fuels
(firewood and charcoal) to liquid fuels (gas and oil) and finally, electricity [7–10] . The sustainable
energy access and efficient use of energy can contribute to reducing energy poverty and improving
energy security, reducing energy related environmental and health impacts [11–13].

Evidences since 1990’s demonstrate economics of fuel and stove type, access conditions to fuels,
technical characteristics of cook stoves and cooking practices, cultural preferences and health
impacts as major determinants of cooking fuel choice in the developed world [14,15]. However,
factors affecting choices among modern and traditional fuels like electricity, LPG, kerosene,
firewood, animal dung cake, crop residues, charcoal and coal are not clearly identified. Some
studies report that the choice relies predominantly on household wealth, education and gender of
household head, affordable price, fuel availability, cost of installation, stove price and efficiency,
cooking and consumption habits, dependability of supply, household preferences and tastes, place
of cooking and socio-cultural factors in developing world [16–19]. Some of these studies claim
that energy transition is happening towards clean energy with continuously balancing stable
economies [16,18]. However, the ground reality of unstable economy of Afghanistan has
descriptively recorded availability and affordability of fuel as predictor in the choice of cooking
fuel in the rural areas [20], which might not be representative to explore the current need of energy
transition and determinants of household cooking energy (a large part of national energy use)
choice among rural and urban Afghanistan.

Afghanistan has unhinged power supply due to destructed and disintegrated power generation and
high dependency on imported petroleum products and limited use of domestic oil, natural gas and
coal [20]. The cooking covers the highest proportion of total energy consumption in Afghanistan
and the most commonly used cooking fuels are firewood, LPG, straw/grass and animal dung [20],
among which firewood is the major source of rural subsistence and low income people in the
country [21]. The residents outside cities obtain firewood from the agricultural areas and those
cities generally purchase firewood from markets, while LPG is imported from neighboring
countries. It is obvious that LPG is commonly used by high income and middle income families,
whereas firewood, straw/grass and animal dung are mostly used by low income families [20].

This study is conducted on a humble need to contribute on understanding factors driving household
cooking energy choice in Afghanistan where people are under uncertainty mainly by conflicts.
This study is to understand determinants (non-price) of household cooking energy in both rural
and urban Afghanistan; and explain probability choice of households cooking fuel in the presence
of uncertainty or switching towards modern fuels for cooking. The research questions include why
households do choose a particular type of cooking fuel, what factors motivate these choices, what
household characteristics make more probable to use traditional dirty energy sources than clean
energy for cooking.

Based on data from Afghanistan Demographic Health Survey (AfDHS) [22], these questions are
analyzed through application of the rational consumer choice theory and random utility model [23]
2
 ACCEPTED MANUSCRIPT

confronting with more recent household survey data in both rural and urban areas. So this study
seeks the socio-economic as well as demographic factors determining energy choice of Afghani
households including wealth index (developed based on quality and quantity of selected consumer
goods such as television, bicycle, car, source of drinking water, toilet facilities, and flooring
materials that household owned using principal component analysis). Wealth quintiles made from
the household score in dividing the distribution into five equal categories of the population. A
central thesis of this paper is that cooking fuel choices are affected by a set of household
demographic and infrastructure variables such as age, marital status, education level, age and sex
of the household head, household size, household wealth and cooking place.

2. Materials and Methods

To understand the household choice for cooking fuel and factors affecting people’s decision for
using the cooking fuels in Afghanistan, household level data of AfDHS conducted during June 15,
2015, to February 23, 2016 are used. The AfDHS funded by United States Aid (USAID) is the
first sample survey of its kind implemented by the joint effort of the Central Statistical
Organization and the Ministry of Public Health, with the objective of providing reliable, and
updated accurate data for the country and policymakers in monitoring and designing programs for
improving household cooking energy and health services in Afghanistan. The AfDHS followed a
stratified two-stage sample design (cluster and systematic sampling) covering a total of 24,395
rural and urban households spread over all 34 provinces across the country.

The AfDHS collected data using three separate questionnaires: Household Questionnaire,
Woman’s Questionnaire, and the Man’s Questionnaire. Data from the Household Questionnaire
are used in this study which provide basic information on the demographic socio-economic and
economic characteristics of head and members of households and general household
characteristics such as types of dwelling, types of fuels used for cooking, place of cooking, wealth
index, source of water and other health related issues. Study variables are drawn from the single
dataset of household questionnaire splitting into many datasets with different number of cases.
Therefore, the relevant variables that can influence household cooking fuel choice are identified.

2.1 Methodology

An extensive use of both logistic regression and multinomial logistic is found from literatures for
analyzing the determinants prompting the cooking energy decision [24–29]. The multinomial logit
model, in particular, is apposite when the household has a variety of options with highly
distinguished characteristics. Most of the households in our study used LPG and bio-gas; followed
by wood, straw/grass, animal dung and others as the major source of cooking fuel. Particularly this
study has developed a truncated view being concentrated only on the analysis of household energy
probability choice for first four highly used main fuels; LPG- 33%, wood- 23.2%, straw/grass-
17% and animal dung-19% covering 92% of 24,395 households and ignoring other minor fuels
used by very few households following a study based on Nigeria [27] (less than 7% each, such as
charcoal- 0.7%, coal- 0.5%, electricity- 0.2% and remaining others- 6.5%) for the development of
policy prescriptions. Though evidence from developed and some developing countries claim better
electricity excess accelerates the economic growth of a nation, with 72% of sample households
having electrification, only 0.2% of Afghani households use electricity as a main cooking fuel, and
might be mainly used for lighting. Ergo, unfortunately exclusion of electricity and other fuels in
the analysis became mandatory to concoct the fact.

3
 ACCEPTED MANUSCRIPT

Demand theory assumes that quantity demanded of a commodity is inversely affected by its price
and available substitutes and positively affected by population. In case of fuel wood and animal
dung fuel demand the substitutes can be LPG, electricity, kerosene, cooking gas and coal. Based
on this theory, household size, availability of electricity, sex of household head, age of household
head, education level of household head, current marital status, region, place of food cooked and
wealth index are selected as independent variables to explain household cooking fuel choice. The
expected sign including definition for each independent variable that may affect household fuel
choice are summarized in Table 1.

Table 1 Variables used and hypothesis for describing adoption of clean energy

Based on the multiple choice of cooking fuel as the dependent variable, based on correlation
results, a multinomial logistic (MNL) regression is used to determine the factors influencing
household choice of cooking fuel in both the rural and urban areas of Afghanistan. A simultaneous
estimation of all the logits under MNL sets logical association between the factors and run the data
more efficiently [30]. It is expected that the odds of one alternative choice that is chosen over a
second one should be independent or irrespective to an un-chosen third alternative [31].

2.2 Modeling
A particular cooking fuel 𝑖 is chosen from the set of alternatives by a household n to maximize
household utility. The fuel chosen among households might be different, depending on the degree
of availability, wealth, demographic and other socioeconomic characteristics. Random Utility
Model (RUM) explains that individual utility of a choice among alternatives sums explanatory
component and an error component 𝑒, being independent of the deterministic part and addressing
a predetermined distribution [4]. In other words, choices among alternatives is the function of the
probability that the utility associated with a particular option i is higher than other options.

While modeling, utility equation becomes,

𝑈𝑖𝑛 = 𝑉(𝑋𝑖𝑛;𝛽) + 𝐸𝑖𝑛…………………………………….(1)

Here, 𝑈𝑖𝑛 is the utility of fuel options, (𝑖= 1,………..𝑗𝑛), for decision maker (𝑛= 1……………𝑁). 𝑈𝑛
is a vector of utilities for 𝑛 individual households. 𝑋𝑖𝑛 is a vector of regressing variables explaining
alternative 𝑖 and decision maker 𝑛. 𝑋 is a matrix of regressing variables describing all alternatives
and for decision-maker 𝑛. 𝛽 is vector of known parameters; 𝑉 is systematic utility, a function of
the regressing variables and known parameters 𝛽 , 𝐸𝑖𝑛 is a random disturbance and 𝐸𝑛 is the
vector of random disturbances.

Mathematically, RUM assumes utility maximization then decision-maker 𝑛 chooses 𝑖 if and only
if 𝑈𝑖𝑛 ≥ 𝑈𝑗𝑛 for all 𝑗 belongs to 𝑗𝑛…………………………………………(2)

Where, 𝐶𝑛 is the set of 𝑗𝑛 alternative faced by 𝑛.

The choice probability equation can be set as,

𝑃(𝑖│𝑋𝑛: 𝛽, 𝛾𝜀) = 𝑃𝑟𝑜𝑏 (𝑈𝑖𝑛 ≥ 𝑈𝑗𝑛 , 𝑉𝑗 𝜀 𝐶𝑛)…………(3)

Where, (𝛾𝜀) are unknown parameters of random disturbance𝐸𝑛 ~𝐷(𝛾𝜀).

4
 ACCEPTED MANUSCRIPT

According to McFadden, if the (𝑁) error terms (𝜀𝑖𝑛), (𝑗 = 1,…,𝑁) are independently and
identically distributed with Weibull distribution [32] 𝐹(𝐸𝑖𝑗) = 𝑒𝑥𝑝[𝑒𝑥𝑝( ‒ 𝐸𝑖𝑗)], then the
probability that household (𝑛) chooses option (𝑖) can be expressed as a function in function of
household characteristics. Now further the probability that individual household (𝑛) chooses
option (𝑖) can be expressed as a function in function of household characteristics(𝑋) as

𝑒𝑥𝑝𝑉(𝑋𝑖𝑛;𝛽)
𝑃(𝑖│𝑋𝑛: 𝛽) = ……………………………( 4)
𝑒𝑥𝑝𝑉(𝑗𝑛;𝛽)

Equation 4 explains the multinomial logit. The common practice in MNL is to choose one category
or choice as the reference for comparison. For the choice of options, we obtain the following
estimates of the probabilities:
1
𝑃(1│𝑋𝑛: 𝛽) = ………………………( 5)
1 + 𝑒𝑥𝑝𝑉(𝑗𝑛;𝛽)
For case of m, equation 4 takes the form,

𝑒𝑥𝑝𝑉(𝑋𝑖𝑛;𝛽)
𝑃(𝑚│𝑋𝑛: 𝛽) = ………………………...(6)
1 + 𝑒𝑥𝑝𝑉(𝑗𝑛;𝛽)

Because of mutually exclusive choices the sum of probability equation (5) and (6) is 1, with being
highly non-linear in nature. The risk ratio or relative risk ratio is deduced by dividing equation (6)
by (5) and is known as logarithm of the ratio of the probability that never goes beyond linear form.
The expression is,
𝑃(𝑚|𝑋𝑛: 𝛽)
𝑙𝑛 = (𝑋𝑖𝑛;𝛽)………………………………(7)
𝑃(1|𝑋𝑛: 𝛽)

Finally, equation (7) clearly seems like MNL with likelihood relationship between regressand and
regressor. Now, MNL model exactly matched in simple version without regional effect is,

𝑌
= 𝛼0 + 𝛼1𝑋1 + 𝛼2𝑋2 + 𝛼3𝑋3 + 𝛼4𝑋4 + 𝛼5𝑋5 + 𝛼6𝑋6 + 𝛼7𝑋7 + 𝛼8𝑋8 + 𝛼9𝑋9 + 𝑒𝑖
……………………………………………. ……………….(8)
Again, multinomial logit model with regional effect is modeled as,

𝑌
∗= 𝛽0 + 𝛽1𝑋1 + 𝛽2𝑋2 + 𝛽3𝑋3 + 𝛽4𝑋4 + 𝛽5𝑋5 + 𝛽6𝑋6 + 𝛽7𝑋7 + 𝛽8𝑋8 + 𝛽9𝑋9 + 𝛽10
𝑋10 + 𝑒𝑘…………………………………………….. (9)
Here,
𝑌= Household fuel choice in equation (8) 𝑌 ∗= Household fuel choice in equation (9)
𝑋1= Place of living 𝑋7= Electricity
𝑋2= Age 𝑋8 = Sex
𝑋3= Education 𝑋9= Household size
𝑋4=Wealth Index 𝑋10= Region
X5=Cooking Place αs = Coefficients of equation (8)

5
 ACCEPTED MANUSCRIPT

𝑋6=Marital Status 𝑒𝑖= Error term in equation (8)

𝛽'𝑠= coefficients of equation (9) 𝑒𝑘= Error term in equation (9)

Finally, robustness test of this study taking wealth index only as explanatory variable is carried
out by a separate regression to match with over emphasis of other past evidence explaining the
wealth status dominantly determines the household cooking fuel choice rather than other factors,
so that policy maker could prioritize the immediate need in the country.
𝑌 ∗∗= 𝜆0 + 𝜆1𝑍 + 𝑒𝑙……………………………………………. ….(10)
Here,𝑌 ∗∗ = Household fuel choice in equation (10) 𝜆𝑠 = Coefficients of
equation (10)
𝑒𝑙 = Error term in equation (10) 𝑍 = Wealth Index

3. Results

The results of empirical analysis of the study begin with the descriptive analysis of the study units
followed by MNL regression results with association details and robustness of the final outcomes
in the study. Table 2 depicts, as per AfDHS 2015, around three folds more rural household are
surveyed out of total sample households 24,395 using different cooking energy fuel, of which 30
missing cases are discarded.

3.1 Descriptive Results

In terms of household energy choice in Afghanistan, AfDHS 2015 data show that more than 92%
of the sample urban households has access to electricity but 0.4 percent of them use electricity for
cooking. The highest proportion (32.6%) of households use LPG or biogas as the main cooking
fuel which is the same as a study in Pakistan [33]. Likewise, wood is used by 23.2%, animal
dungs by 19.2% and straw/shrubs by 16.9% (in weighted average) as their main energy source for
cooking. Other fuels are used by very small proportion of samples. Two third of total sample
households depends on traditional energy and remaining households on clean and other, suggests
clean energy seems still to penetrate to large chunk of population comprising rural areas of
Afghanistan.

Table 2 Distribution of Cooking Energy Choice and Households in Afghanistan

Descriptive statistics of independent variables which are hypothesized to influence the cooking
fuel choice among the Afghani households are presented in table 3. Variables like sex of
household head, access to electricity and cooking places are dummies. The average age of
household head is 44 years with 3.4 average years of education. The association between the
education level of household head and energy choice is illustrated in figure 1. The average
education level of a household head using LPG for cooking is 5.32; using electricity 6.62; using
firewood, as low as 3.33; and using dung cakes only 2.11. Figure 1 illustrates that the level of
education of the household head is strongly related to a switch to the use of clean energy for
cooking.

Table 3 Summary Statistic of Independent Variables used in Multinomial Logit Regression

6
 ACCEPTED MANUSCRIPT

Figure 1 Average Education Level of Household Head and Wealth Index with Different Fuel
Consumption

A clear link between choice of cooking fuel and wealth quintiles is shown, as analyzed by
calculating the proportion of households that use a particular source of energy for wealth quintiles.
The proportion of households using clean fuel among the richest quintile is three folds higher to
the households in the poorest quintile (Figure 1), indicating that households with high wealth are
likely to switch from traditional fuel to clean fuel. Besides, a serious concern about near perfect
dependence among some of the independent variables, especially between wealth and the
education variables are diagnosed with the correlation matrix which shows no sign of serious
multi-collinearity issues among the regressors reported in table 4.

Table 4 Correlation Matrix

3.2 Factors Affecting the Households Fuel Choice

The MNL regression model was fitted with Afghani household choice for three types of fuels:
LPG, wood and straw/grass, controlling for animal dung having relatively high health cost [34],
with and without regional effect separately. Robustness of the MNL regression result was analysed
in order to frame a conformation picture of the choice process for these specific fuels in
Afghanistan.

The result presented as in table 5 revealed that household size has negative significant effect on
choice of LPG at 1% level of significance indicating as bigger the household size increases the
probability choice of dung cake and wood rather than modern fuels. This is obvious in developing
countries because more household members need larger amount of cooking fuel and can manage
traditional fuels such as animal dung and wood at free of cost using family laborer of very low
opportunity cost. Being parallel to the other previous studies [4,25,26,35], households across urban
area prefer LPG as a main household cooking fuel. Similarly, households with access of electricity
are more likely to use LPG and less likely to use wood and straw as main energy for cooking,
compared to the animal dung.

Increase in age of household head leads to probability choice of animal dung for cooking rather
than other fuels. The results also reveal that cooking place at outdoor has a negative significant
effect of choice for LPG, wood and straw as main cooking energy, while the likelihood of
preference is towards LPG only for households cooking at separate building, compared to cooking
inside the house. Similarly, households having separate buildings for cooking also prefer to use
animal dung as main cooking energy. This result is almost similar with study of Kenya [27].
Similarly, significance of wealth index explains that higher the wealth of household, preference
toward the LPG, wood and straw as cooking fuel is higher compared to animal dung. Household
properties and wealth affect household probability choice for a particular source of energy [4,13].
The education level of head of the household is also a significant determinant of energy choice.
Education level of the household head is more likely associated with the likelihood of choosing
LPG and wood compared to animal dung.

Married head of the household seems more likely preferring in choice of straw rather than animal
dungs as an energy for cooking. Some variables such as sex, widowed and divorced household
head and some other categories seem insignificant in the choice of particular fuel. The model

7
 ACCEPTED MANUSCRIPT

reveals that 28.68 percent variations in the fuel choice is explained by the explanatory variables
fitted. This is quite a reasonable level for the models fitted in cross sectional data.

Table 5 Multinomial Logit Estimation of Household Fuel Choice for Cooking Without
Regional Effect.

The result presented in table 6 depicts the same scenario of relationship between the choice of fuel
for cooking and its influencing factors including regional effect using the MNL, controlling for
animal dung with provinces as the fixed factors. At one percent significant level, compared to rural
area, wood is more preferable to animal dung in urban areas of Afghanistan whereas animal dung
is more preferable being irrespective to the cooking place. And other demographic variables show
same effect explained from the model without the fixed effects,

Further, the results with the fixed factors mainly show how probability choice of cooking fuels are
regionally distributed at 1% level of significance. Setting North East Afghanistan as the reference
region, households in other provinces are less likely to use LPG as the main cooking fuel. At West
Afghanistan households are less likely to use LPG and wood but more likely to use straw over
animal dungs, while households at North West, Central, East and South parts of Afghanistan are
less likely to choose LPG and Wood but more likely to choose straw over animal dung as the main
cooking energy. As a final point, animal dung is more preferable to use as main cooking fuel at
South West Afghani households. Finally, the model is fit at 40.02.

Table 6 Multinomial Logit Estimation of Household Fuel Choice for Cooking with
Regional Effect

3.3 Robustness Test of Results

To test the robustness of result in the study, we reorganized Wealth Index into richer, middle and
poorer categories. Table 7 at 1% level of significance shows that households belonging to both
richer and middle wealth index are more likely to choose clean cooking fuels as much LPG, wood
and then straw over animal dungs as main cooking fuels, compared to the poorer households. This
results further confirms that household’s movement from poorer to middle and to richer wealth
index develops a situation of using clean fuels over traditional cooking fuels. This MNL model is
fit at 26.98 meaning that wealth index covers significant proportion of the household fuel choice.

Table 7 Multinomial Logit Estimation of Household Fuel Choice for Cooking with Wealth
Index

4. Discussion

The results without involving fixed factors of regional effect conclude that household size, place
of living (urban), electricity access, age, cooking place in separate building, education of
household head, household wealth and married household head affect significantly on energy
choice move towards clean energy over animal dung. Similarly, with involving regional effect the
result concludes that straw/grass are more popular as main cooking fuel for households at West
Afghanistan North West, Central, East and South parts of Afghanistan more likely to choose over
animal dung. In contrast, animal dung is more preferable at South West Afghani households. The
robustness of the results for the confirmation of findings showed that rich and middle wealth

8
 ACCEPTED MANUSCRIPT

holding households are more likely to use cleaner fuel over animal dung compared to poor
households.

As other studies concluded with larger household size can be attributed to the cheapness traditional
fuels possibly because of large labor inputs to collect and therefore less scarce of traditional fuels
at home [10,12,14,24,36], this paper confirms similar results in Afghanistan. Our study also
supports the evidences obtained in literatures explaining the traditional energy (fuel wood, animal
dung, and crop waste) is dominant among poor households, while higher wealth households use
electricity and LPG, assuming that energy transition occurs from traditional fuels to the modern
with increasing socioeconomic status of households, either through a rise in wealth or availability
of clean energy [25,35], but a study found that the household income does not significantly affect
choice of cooking fuel used per household [3]. Robustness of the result also confirms that the first
priority should be given to the generation of household wealth, because wealth enables household
to switch towards clean fuels through increasing affordability among poor households [35].
Obviously, cost of energy source and household budget constrictions definitely play a significant
role in limiting the choice of energy attainable by households situating in normal political disorder
[11,24].

However, Afghani households with access to electricity prefer animal dungs for cooking which is
a contrast with recent study in Bhutan [4], though both are in the list of developing countries. This
may be due to less cost effective and low multiple use of electricity. Though, electricity considers
as a measure of level of development of societies and living standards, price and its availability
plays vital role on its choice [36,37], but our model excluded electricity because very small share
of households was found using electricity for cooking fuel.. It is believed that higher education
advances knowledge of fuel attributes, taste, and preference for clean fuels [28], our study also
lightly supports this argument though the betterment from well education environment is hard in
Afghanistan.

This paper has explored an unreached area urgently relevant for policy makers to be addressed the
social, cultural and organizational hurdles that determine the taking on of modern convenient
energy against other traditional energy types in conflict ongoing area. This study is therefore
crucial for policy makers in politically unstable developing countries to understand the
fundamental non-price drivers of household energy choices in order to be able to design and deliver
polices that yield optimal energy transition outcomes. Understanding the core drivers of household
energy choice being irrespective with uncertain political situation and limited wealth therefore
becomes an important policy consideration for prioritizing energy caring environment and health
in long term. Beyond energy transition and following health perspective, a study [38] argued that
clean energy adoption reduces the prevalence of chronic bronchitis, asthma and acute respiratory
incidence.

Besides providing vital policy implications, this empirical study came across some limitations. (1)
Owing to the exclusion of information in data set of price, occupation of household head,
technological parameters of fuels and appliances and availability of fuel, we could not explore the
further informative policy in depth as other studies [8,29,24,35] explained. (2) This study has not
included other minor cooking fuels in the analysis because of very small share of of households
are using them. (3) This study has has not considered association of fuel selection pattern in
household with comprehensive consequences on health and environment. Despite these pitfalls,
through the policy perspective, results of this study strongly suggest for motivating the households
in rural Afghani areas to use clean cooking fuel which will be health and environment friendly as
9
 ACCEPTED MANUSCRIPT

well as socially acceptable, government and other cohorts should put effort on increasing regular
sources of wealth through provisions or employment opportunities, regionally promoting higher
level of education opportunity, providing provision to build a separate cooking place and
improving access of modern cooking fuels to large number household specially in south west
Afghani rural areas of the country.

4. Conclusion

This paper developed a framework for analyzing household cooking fuel use pattern and,
association of household cooking fuel choices and their determinants using the multinomial logit
model on Afghanistan Demographic Health Survey 2015 database consisting of 21,835 sample
households. This paper specially investigates the probability that a household with certain
characteristics will choose alternative cooking fuels (LPG, wood and shrubs/grass) energy types
relative to animal dung fuel with and without fixed effects of regional economic differences in
cooking energy choice. This study concludes that decreasing poverty and increasing educational
level significantly increases the probability of switching from dung fuel to higher ladder alternative
fuel types such as gas, wood and shrubs. Whilst we find that the reluctance to switch from biomass
to cleaner energy types are higher and more significant in South west region of Afghanistan where
the topography of the area and remoteness of the towns make the supply of basic services like gas
difficult, relative to the capital region. Again, the study also finds that even though households
with electric power supply do not typically use electricity for cooking, electric connected
households are more likely to use gas or biogas for cooking suggesting either unreliable power
supply or inability of households to afford electricity for cooking or a mix of both. Thus, energy
policy should focus on wealth generation activities, endowment of electricity and other reliable,
and affordable sources of cooking energy for household use. This new evidence provides some
useful insight into the current household energy consumption situation and empirical basis for
energy policy in Afghanistan.

Economic enablement of the citizenry will make modern energy affordable and help people to
prefer modern fuels over traditional fuels reducing their dependence on animal dung fuel for
cooking. However, the need of empowerment seems to be more exigent in the south west region.
In this regard, locally available and reliable government modern energy distribution centers will
be vital in improving accessibility of basic energy services easier than the existing. All in all, there
needs to be a comprehensive national strategy in providing the right incentives for households to
switch sustainably from animal dungs to modern energy, specifically, by addressing the
socioeconomic and demographic determinants influencing energy choices.

Acknowledgement
We acknowledge all the experts for their potential suggestions at the course of research work.

Author Biographies
Uttam Paudel is a doctoral student at Tribhuvan University. His research interests include
Energy, Health and Environmental Economics. Address: Bairochan Marg, Kathmandu
Metropolitan city- 3, Kathmandu Nepal. E-mail: uuupaudel22@gmail.com

Umesh Khatri is a Researcher at USAID. His research interests include Energy and
Technological Economics. Address: Godawari-12, Lalitpur, Nepal. E-mail:
khatriumesh81@gmail.com

10
 ACCEPTED MANUSCRIPT

Krishna Prasad Pant is a visiting faculty in the Central Department of Economics, Tribhuvan
University. His research interest includes agricultural economics, environmental economics,
alternative energy and climate change. Address: Devnagar, Nagarjun-2, Kathmandu Nepal.
Email: kppant@yahoo.com

Funding Sources
This research did not receive any specific grant from any funding agency in the public,
commercial, or not-for-profit sectors.

References
[1] IEA. Key World Energy Statistics 2010. 2010. doi:10.1787/9789264039537-en.
[2] Gea. Global Energy Assessment - Toward a Sustainable Future: Summary for Policmakers
2012:3–30.
[3] Karimu A. Energy for Sustainable Development Cooking fuel preferences among
Ghanaian Households : An empirical analysis. Energy Sustain Dev 2015;27:10–7.
doi:10.1016/j.esd.2015.04.003.
[4] Bahadur D, Behera B, Ali A. Household energy choice and consumption intensity :
Empirical evidence from Bhutan. Renew Sustain Energy Rev 2016;53:993–1009.
doi:10.1016/j.rser.2015.09.019.
[5] Foell W, Pachauri S, Spreng D, Zerriffi H. Household cooking fuels and technologies in
developing economies. Energy Policy 2011;39:7487–96. doi:10.1016/j.enpol.2011.08.016.
[6] GEA Writing Team. Global Energy Assessment. Glob. Energy Assesssment - Towar. a
Sustain. Futur., 2012. doi:http://dx.doi.org/10.1017/CBO9780511793677.
[7] Guta DD. Application of an almost ideal demand system ( AIDS ) to Ethiopian rural
residential energy use : Panel data evidence. Energy Policy 2012;50:528–39.
doi:10.1016/j.enpol.2012.07.055.
[8] Ekholm T, Krey V, Pachauri S, Riahi K. Determinants of household energy consumption
in India. Energy Policy 2010;38:5696–707. doi:10.1016/j.enpol.2010.05.017.
[9] Tei J, Adu G. An empirical analysis of household energy choice in Ghana. Renew Sustain
Energy Rev 2015;51:1402–11. doi:10.1016/j.rser.2015.07.050.
[10] Alem Y, Beyene AD, Köhlin G, Mekonnen A. Modeling household cooking fuel choice :
A panel multinomial logit approach. Energy Econ 2016;59:129–37.
doi:10.1016/j.eneco.2016.06.025.
[11] Anozie AN, Bakare AR, Sonibare JA, Oyebisi TO. Evaluation of cooking energy cost ,
efficiency , impact on air pollution and policy in Nigeria 2007;32:1283–90.
doi:10.1016/j.energy.2006.07.004.
[12] Bonjour S, Adair-rohani H, Wolf J, Bruce NG, Mehta S, Prüss-ustün A, et al. Solid Fuel
Use for Household Cooking : Country and Regional Estimates for 1980 – 2010
2013;784:784–90.
[13] Khandker SR, Barnes DF, Samad HA. Are the energy poor also income poor? Evidence
from India. Energy Policy 2012;47:1–12. doi:10.1016/j.enpol.2012.02.028.
[14] Caraher M, Dixon P, Lang T, Carr‐Hill R. The state of cooking in England: the
relationship of cooking skills to food choice. Br Food J 1999;101:590–609.
doi:10.1108/00070709910288289.
[15] Masera OR, Saatkamp BD, Kammen DM. From linear fuel switching to multiple cooking
strategies: A critique and alternative to the energy ladder model. World Dev
2000;28:2083–103. doi:10.1016/S0305-750X(00)00076-0.
11
 ACCEPTED MANUSCRIPT

[16] Joon V, Chandra A, Bhattacharya M. Household energy consumption pattern and socio-
cultural dimensions associated with it: A case study of rural Haryana, India. Biomass and
Bioenergy 2009;33:1509–12. doi:10.1016/j.biombioe.2009.07.016.
[17] Owusu Boadi K, Kuitunen M. Factors affecting the choice of cooking fuel, cooking place
and respiratory health in the Accra metropolitan area, Ghana. J Biosoc Sci 2006;38:403–
12. doi:10.1017/S0021932005026635.
[18] Pokharel S, Chandrashekar M. Analysis of cooking energy in developing countries. Nat
Resour Forum 1995;19:331–7. doi:10.1111/j.1477-8947.1995.tb00625.x.
[19] Adegbulugbe A, Akinbami J. Urban household energy use patterns in Nigeria. Nat Resour
Forum 1995;19:125–32.
[20] Mohammad A, Shrestha P, Kumar S. Urban residential energy use in Kandahar ,
Afghanistan. Cities 2013;32:135–42. doi:10.1016/j.cities.2013.04.002.
[21] Paper W. www.econstor.eu 2010.
[22] Survey H. Afghanistan. 2015.
[23] Walker J, Ben-Akiva M. Generalized random utility model. Math Soc Sci 2002;43:303–
43. doi:10.1016/S0165-4896(02)00023-9.
[24] Rahut DB, Das S, De Groote H, Behera B. Determinants of household energy use in
Bhutan. Energy 2014;69:661–72. doi:10.1016/j.energy.2014.03.062.
[25] Pandey VL, Chaubal A. Comprehending household cooking energy choice in rural India.
Biomass and Bioenergy 2011;35:4724–31. doi:10.1016/j.biombioe.2011.09.020.
[26] Takama T, Tsephel S, Johnson FX. Evaluating the relative strength of product-speci fi c
factors in fuel switching and stove choice decisions in Ethiopia . A discrete choice model
of household preferences for clean cooking alternatives. Energy Econ 2015;34:1763–73.
doi:10.1016/j.eneco.2012.07.001.
[27] Fraser GCG. Agrekon : Agricultural Economics Research , Policy and Practice in
Southern Africa Multinomial logit analysis of household cooking fuel choice in rural
Kenya : The case of Kisumu district Multinomial logit analysis of household cooking fuel
choice in rura n.d.:37–41. doi:10.1080/03031853.2006.9523731.
[28] Adeyemi PA, Adereleye A. Determinants of Household Choice of Cooking Energy in
Ondo State , Nigeria 2016;7:131–42.
[29] Sehjpal R, Ramji A, Soni A, Kumar A. Going beyond incomes : Dimensions of cooking
energy transitions in rural India. Energy 2014;68:470–7.
doi:10.1016/j.energy.2014.01.071.
[30] Long JS. Regression models for categorical and limited dependent variables. 1997.
[31] Pleskac TJ. Decision and Choice: Luce’s Choice Axiom. Michigan Univ 2012:1–18.
doi:10.1016/B978-0-08-097086-8.43031-X.
[32] McFadden DL. Quantal Choice Analaysis: A Survey. Ann Econ Soc Meas 1976;5:363–
90.
[33] Jan I, Khan H, Hayat S. Determinants of Rural Household Energy Choices : An Example
from Pakistan 2012;21:635–41.
[34] Pant KP. Cheaper fuel and higher health costs among the poor in rural Nepal. Ambio
2012;41:271–83. doi:10.1007/s13280-011-0189-6.
[35] Hou B, Tang X, Ma C, Liu L, Wei Y. Cooking fuel choice in rural China : results from
microdata. J Clean Prod 2016:1–10. doi:10.1016/j.jclepro.2016.05.031.
[36] Policy E, Sathaye JA, Berkeley L. Urban household energy use in India : efficiency and
policy implications 2015;4215. doi:10.1016/S0301-4215(98)00008-1.
[37] Özcan M, Gülay E, Üçdo Ş. Economic and demographic determinants of household
energy use in Turkey 2013;60:550–7. doi:10.1016/j.enpol.2013.05.046.
[38] Pant KP. Economics of indoor air pollution and respiratory health: A case study of the
12
 ACCEPTED MANUSCRIPT

effects of improved stove and biogas adoptions on health costs in rural Nepal. Respirology
2013;18:20–1.

13
 ACCEPTED MANUSCRIPT

Figure 1 Average Education Level of Household Head and Wealth Index with
Different Fuel Consumption

1
 ACCEPTED MANUSCRIPT

Highlights

 Two-thirds Afghan households use traditional fuels for cooking and 0.2% electricity
 Poverty reduction and education increase probability of switching to clean fuels
 Urban couples with electricity access and separate kitchen room adopt clean fuel
 Households with aged head and large family size are less likely to adopt clean fuel
 Afghanistan government needs incentive strategy to promote adoption of clean energy
 ACCEPTED MANUSCRIPT

Table 1 Variables used and hypothesis for describing adoption of clean energy
Variable’s Name Description Expected Sign
Place of living Household situated place (1 if urban 0 otherwise) +
Age Age of the household head -
Sex Sex of household head (1 if male 0 otherwise) +
Education Education of household head (years of schooling) +
Wealth Index Wealth index of household +
Cooking Place Household kitchen setting internal, external or -
separate building
Marital status Current Marital status of household head +
Electricity Electricity access in household (1 if access 0 +
otherwise)

Household size Number of family members _

Region Provinces of Afghanistan ±

Table 2 Distribution of Cooking Energy Choice and Households in Afghanistan

Cooking Energy Choice (Percent) Household

Distribution
Area Electrici LPG/Natu Woo Anima Crop Shrub Other Number Perce
ty ral d l Dung residues s/Gra s of nt
gas/Bioga ss Househol
s/Other ds
gas
Urban 0.4 83.1 9.3 2.8 0.9 2.7 0.8 6,387 26.21
Rural 0.1 15.1 28.0 24.8 8.0 21.8 2.2 17,978 73.79
Source: Central Statistics Organization 2015

Table 3 Summary Statistic of Independent Variables used in Multinomial Logit

Regression
Variables Means Standard Deviation
Place of living (1 if urban, 0 otherwise) 0.26 0.44
Age (years) 44.15 14.16
Education (years of schooling) 3.38 4.87
Wealth Index 2.91 1.36
Electricity (1 if excess, 0 otherwise) 0.71 0.44
Household size (number) 8.35 3.82

1
 ACCEPTED MANUSCRIPT

Table 4 Correlation Matrix

Variables Househol Region Place of Electri. Sex Age C. C. Wealth Edu. Marital
d size living Fuel Place Index Status
Hh size 1.0000
Region 0.0854 1.0000
Place of -0.0238 0.0086 1.0000
Living
Electri. 0.0024 -0.0325 0.2220 1.000
Sex 0.0806 0.0007 -0.0325 -0.0044 1.0000
Age 0.3030 0.0048 0.0194 0.0230 -0.0257 1.0000
C. fuel 0.0641 -0.0223 -0.5423 -0.2370 0.0232 0.0159 1.000
C. Place -0.0766 -0.2097 0.0041 -0.0080 -0.0160 0.0233 0.059 1.000
Wealth -0.0041 0.1675 0.5781 0.1821 -0.0187 -0.044 -0.583 -0.126 1.000
Index
Education -0.0301 -0.0379 0.1854 0.1068 0.0651 -0.169 -0.219 -0.076 0.281 1.000
Marital -0015 0.0122 -0.0170 -0.0038 -0.4665 0.224 -0.002 -0.008 -0.007 -0.103 1.000
Status
Note: In the table, Hh size= household size, Electri.=electricity, C. Fuels=cooking fuels,
C. Place=cooking place and Edu =education.

2
 ACCEPTED MANUSCRIPT

Table 5 Multinomial Logit Estimation of Household Fuel Choice for Cooking

Without Regional Effect.
Explanatory Variables LPG Wood Straw/Shrubs
Household Size -0.035*** 0.0545*** 0.006
(0.008) (0.006) (0.006)
Place of living 1.248*** 0.120 0.184**
(0.086) (0.081) (0.086)
Electricity 0.453*** -0.256*** -0.671***
(0.069) (0.048) (0.048)
Sex -0.270 0.288 0.055
(0.251) (0.200) (0.218)
Age 0.003 00008 -0.004**
(0.002) (0.002) (0.002)
Cooking Place
In a separate building 0.242*** -0.031 -0.251***

(0.070) (0.055) (0.060)

Outdoors -0.692*** -0.851*** -0.155***
(0.081) (0.060) (0.056)
Other cooking areas -1.037*** -1.387*** -0.664***
(0.213) (0.161) (0.137)
Wealth Index
Poorer 1.632*** 1.820*** 0.546***
(0.124) (0.059) (0.054)
Middle 2.870*** 2.629*** 0.988***
(0.120) (0.065) (0.062)
Richer 4.971*** 3.675*** 1.148***
(0.129) (0.089) (0.097)
Richest 7.413*** 4.324*** 1.317***
(0.293) (0.284) (0.334)
Education 0.0529*** 0.028*** -0.013**
(0.006) (0.005) (0.006)
Marital Status
Married 0.091 -0.016 0.564***
(0.221) (0.175) (0.207)
Widowed -0.027 0.310 0.181
(0.299) (0.228) (0.267)
Divorced 0.902 0.483 -0.439
(0.785) (0.620) (0.858)
Constant -3.382*** -1.780*** -0.441
Observations 21,759
Pseudo R2 0.287
Note: In the model, dependent variable is types of fuel used, animal dung is taken as base
category. Explanatory variables includes poorest, unmarried household head and cooking
inside the home in the model as reference category. Standard errors in parentheses, ***
p<0.01, ** p<0.05, * p<0.1

3
 ACCEPTED MANUSCRIPT

Table 6 Multinomial Logit Estimation of Household Fuel Choice for Cooking with
Regional Effect

Explanatory Variables LPG Wood Straw/Shrubs

Household size -0.039*** 0.032*** -0.0001
(0.009) (0.007) (0.007)
Place of Living 1.230*** 0.223*** 0.264***
(0.088) (0.085) (0.088)
Electricity 0.318*** -0.314*** -0.663***
(0.071) (0.051) (0.048)
Sex -0.242 0.265 0.026
(0.254) (0.212) (0.220)
Age 0.004 0.001 -0.004**
(0.002) (0.002) (0.002)
Cooking place
In a separate building 0.037 -0.260*** -0.306***
(0.073) (0.058) (0.062)
Outdoors -0.560*** -0.348*** -0.039
(0.086) (0.069) (0.059)
Other -0.780*** -0.824*** -0.538***
(0.218) (0.181) (0.141)
Wealth Index
Poorer 1.626*** 1.669*** 0.466***
(0.125) (0.065) (0.056)
Middle 2.922*** 2.547*** 0.855***
(0.122) (0.072) (0.064)
Richer 5.058*** 3.663*** 1.018***
(0.131) (0.096) (0.099)
Richest 7.565*** 4.300*** 1.191***
(0.295) (0.286) (0.335)
Education 0.054*** 0.029*** -0.017***
(0.007) (0.006) (0.006)
Marital Status
Married 0.152 0.032 0.582***
(0.226) (0.185) (0.208)
Widowed 0.0357 0.304 0.216
(0.307) (0.242) (0.269)
Divorced 1.025 0.476 -0.318
(0.822) (0.700) (0.855)
Region
West Afghanistan -0.506*** -3.071*** 0.345***
(0.105) (0.108) (0.082)
North West Afghanistan -1.182*** -3.133*** 0.288***
(0.098) (0.093) (0.079)
Central Afghanistan -0.537*** -0.675*** 0.914***
(0.092) (0.069) (0.081)
East Afghanistan -0.866*** -0.902*** 0.968***
4
 ACCEPTED MANUSCRIPT

(0.157) (0.127) (0.136)

South East Afghanistan -0.439*** -0.331*** 0.566***
(0.108) (0.085) (0.103)
South West Afghanistan -0.810*** -0.963*** -0.007
(0.123) (0.092) (0.109)
Constant -2.787*** -0.635** -0.776**
(0.362) (0.289) (0.311)
Observations 21,759
Pseudo R2 0.345
Notes: In the model, dependent variable is types of fuel, animal dung is taken as base
category. Explanatory variables includes poorest, unmarried household head and cooking
inside the home and North East Afghanistan in the model as reference category Standard
errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Table 7 Multinomial Logit Estimation of Household Fuel Choice for Cooking with
Wealth Index
Explanatory Variables LGP Wood Straw/grass

Wealth Index
Middle 2.034*** 1.636*** 0.816***
(0.082) (0.052) (0.057)
Richer Category 5.452*** 2.809*** 0.978***
(0.089) (0.076) (0.089)
Constant -2.296*** -0.307*** -0.330***
(0.055) (0.026) (0.026)
Observations 21,835
Pseudo R2 0.193
Note: In the model, dependent variable is types of fuel, animal dung is taken as base
category. Explanatory variables includes poor as reference category. Standard errors in
parentheses. *** p<0.01, ** p<0.05, * p<0.1