A Recipe for Success? Randomized Free Distribution of Improved Cooking Stoves in Senegal

Abstract

Today more than 2.7 billion people rely on biomass as their primary cooking fuel, with profound implications for the environment and people’s well-being. Wood provision is often time-consuming and the emitted smoke has severe health effects – both burdens that afflict women in particular. The dissemination of Improved Cooking Stoves (ICS) is frequently considered an eff ective remedy for these problems. This paper evaluates the take-up of ICS and their impacts through a randomized controlled trial in rural Senegal. Although distributed for free, the ICS are used by almost 100 % of households. Furthermore, we find substantial effects on firewood consumption, eye infections, and respiratory disease symptoms. These findings substantiate the increasing efforts of the international community to improve access to improved cooking stoves and call for a more direct promotion of these stoves.

Full text

Journal
of
Health
Economics
42
(2015)
44–63
Contents
lists
available
at
ScienceDirect
Journal
of
Health
Economics
journa
l
h
om
epa
ge:
www.elsevier.com/locate/econbase
The
intensive
margin
of
technology
adoption
–
Experimental
evidence
on
improved
cooking
stoves
in
rural
Senegal夽
Gunther
Benscha,
Jörg
Petersa,b,∗
aRheinisch-Westfälisches
Institut
für
Wirtschaftsforschung
(RWI),
Essen,
Germany
bAMERU,
University
of
the
Witwatersrand,
Johannesburg,
South
Africa
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
30
July
2014
Received
in
revised
form
12
March
2015
Accepted
12
March
2015
Available
online
20
March
2015
JEL
classification:
C93
I12
O12
O13
Q53
Keywords:
Household
air
pollution
Energy
access
Technology
adoption
Development
economics
Biomass
fuel
a
b
s
t
r
a
c
t
Today,
almost
3
billion
people
in
developing
countries
rely
on
biomass
as
primary
cooking
fuel,
with
profound
negative
implications
for
their
well-being.
Improved
biomass
cooking
stoves
are
alleged
to
counteract
these
adverse
effects.
This
paper
evaluates
take-up
and
impacts
of
low-cost
improved
stoves
through
a
randomized
controlled
trial.
The
randomized
stove
is
primarily
designed
to
curb
firewood
consumption,
but
not
smoke
emissions.
Nonetheless,
we
find
considerable
effects
not
only
on
firewood
consumption,
but
also
on
smoke
exposure
and,
consequently,
smoke-related
disease
symptoms.
The
reduced
smoke
exposure
results
from
behavioural
changes
in
terms
of
increased
outside
cooking
and
a
reduction
in
cooking
time.
We
conclude
that
in
order
to
assess
the
effectiveness
of
a
technology-oriented
intervention,
it
is
critical
to
not
only
account
for
the
incidence
of
technology
adoption
–
the
extensive
margin
–
but
also
for
the
way
the
new
technology
is
used
–
the
intensive
margin.
©
2015
Elsevier
B.V.
All
rights
reserved.
1.
Introduction
In
developing
countries,
almost
3
billion
people
rely
on
tradi-
tional
biomass-based
fuels
for
their
daily
cooking
purposes.
In
rural
sub-Saharan
Africa,
virtually
all
households
cook
with
biomass,
夽We
thank
Mark
Andor,
Manuel
Frondel,
Rachel
Griffith,
Michael
Grimm,
Subhrendu
Pattanayak,
Fiona
Ross,
Christoph
M.
Schmidt,
Maximiliane
Sievert,
and
in
particular
Colin
Vance
for
helpful
comments.
Participants
of
the
Nordic
Conference
in
Development
Economics,
Gothenburg/Sweden
in
June
2012,
the
Cen-
tre
for
the
Studies
of
African
Economies
conference
in
Oxford/United
Kingdom
in
March
2013
and
research
seminars
at
University
of
Göttingen/Germany
and
Witwa-
tersrand
University
Johannesburg/South
Africa
provided
valuable
input.
Financial
support
from
the
German
Federal
Ministry
for
Economic
Cooperation
and
Develop-
ment
(BMZ)
through
the
Independent
Evaluation
Unit
of
Deutsche
Gesellschaft
für
Internationale
Zusammenarbeit
(GIZ)
is
gratefully
acknowledged.
Peters
gratefully
acknowledges
the
support
of
a
special
grant
(Sondertatbestand)
from
the
German
Federal
Ministry
for
Economic
Affairs
and
Energy
and
the
Ministry
of
Innovation,
Science,
and
Research
of
the
State
of
North
Rhine-Westphalia.
∗Corresponding
author
at:
RWI,
Hohenzollernstrasse
1–3,
45128
Essen,
Germany.
Tel.:
+49
0201
8149
247;
fax:
+49
0201
8149
200.
E-mail
address:
peters@rwi-essen.de
(J.
Peters).
mostly
firewood.
The
collection
of
and
cooking
with
firewood
is
associated
with
various
negative
effects
on
the
living
conditions
of
the
poor.
According
to
the
World
Health
Organization
(WHO),
the
emitted
smoke
is
the
leading
environmental
cause
of
death
and
is
responsible
for
4.3
million
premature
deaths
every
year
–
more
deaths
than
are
caused
by
malaria
or
tuberculosis
(WHO,
2014;
Martin
et
al.,
2011).
Medical
research
throughout
the
last
decades
found
links
between
air
pollution
induced
by
open
fires
and
vari-
ous
illnesses
including
pneumonia,
chronic
obstructive
pulmonary
disease
(COPD),
and
eye
infections,
but
also
stunted
growth
of
chil-
dren,
tuberculosis,
and
cardiovascular
diseases
(see
Armstrong
and
Campbell,
1991;
Campbell
et
al.,
1989;
Dherani
et
al.,
2008;
Kan
et
al.,
2011;
McCracken
et
al.,
2011;
Pandey,
1984a,b;
Pandey
et
al.,
1989).
Furthermore,
biomass
usage
for
cooking
is
a
major
source
of
climate-relevant
emissions
(Shindell
et
al.,
2012).
Improved
biomass
cooking
stoves
(ICSs)
are
often
believed
to
be
a
game
changer
for
cooking
in
developing
countries.
It
is
in
this
context
that
the
United
Nations
set
out
the
Sustainable
Energy
for
All
initiative
with
the
ambitious
goal
of
globally
universal
adoption
of
clean
cooking
stoves
and
electricity
by
2030.
There
is,
however,
a
wide
range
of
ICSs
with
different
levels
of
sophistication
that
have
http://dx.doi.org/10.1016/j.jhealeco.2015.03.006
0167-6296/©
2015
Elsevier
B.V.
All
rights
reserved.

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
45
strong
implications
for
smoke
emissions
and
thus
cleanliness.
It
is
hence
still
a
matter
of
ongoing
debate
under
which
conditions
ICSs
can
be
considered
as
clean,
also
compared
to
modern
fuels
like
electricity
and
gas.1
This
paper
presents
findings
from
a
Randomized
Controlled
Trial
(RCT)
among
253
households
in
twelve
villages
in
Senegal
to
ana-
lyze
behavioural
responses
and
impacts
following
the
introduction
of
an
ICS.
The
ICS,
which
was
assigned
free
of
charge,
is
a
low-
cost
and
maintenance-free
portable
clay-metal
stove.
It
is
produced
in
a
fairly
standardized
way
by
local
manufacturers
(potters
and
whitesmiths)
in
their
workshops
and
is
marketed
at
a
retail
price
of
around
10
US$.
The
stove
has
an
expected
life
span
of
one
to
three
years
before
it
deteriorates
and
has
to
be
replaced.
It
has
already
been
widely
used
in
large
governmental
dissemination
pro-
grammes
in
urban
and
rural
Africa.
As
such,
this
is
the
first
study
to
assess
a
type
of
ICS
whose
design
is
geared
towards
fuel
savings,
ease
of
use,
affordability
and,
hence,
large-scale
applicability,
but
one
that
lacks
specific
health-conducive
technical
features
such
as
a
cleaner
burning
process
or
a
chimney.
Without
further
changes
in
cooking
behaviour,
the
reduction
in
particulate
matter
emissions
that
the
randomized
ICS
can
technically
achieve
would
probably
be
insufficient
to
affect
the
health
of
users.
This
is
due
to
the
non-linear
particulate
exposure–response
relation
found
in
medical
research
suggests
that
large
reductions
in
smoke
exposure
are
required
in
order
to
ensure
positive
health
effects
(see,
for
example,
Ezzati
and
Kammen,
2001;
Pope
et
al.,
2011;
Burnett
et
al.,
2014).
The
main
impact
indicators
of
this
study
are
firewood
consump-
tion,
time
use,
respiratory
disease
symptoms
and
eye
infections.
They
are
supplemented
by
various
indicators
along
the
results
chain
of
the
intervention
with
regard
to
cooking
behaviour.
Effects
on
these
indicators
were
assessed
12
months
after
randomiza-
tion
following
a
baseline
study
in
November
2009.
The
behavioural
changes
we
look
at
–
firewood
usage
patterns
and
smoke
exposure
–
can
be
expected
to
materialize
already
in
the
first
few
months
after
ICS
adoption.
The
changes
in
these
indicators
we
observe
after
one
year
of
ICS
ownership
therefore
reflect
impacts
to
be
expected
in
the
long
run
–
as
long
as
people
continue
to
use
the
ICS
and
replace
it
by
a
new
one
once
it
is
not
functional
anymore.
The
third
wave
of
interviews
in
March
2013
is
used
to
track
the
longer-
term
usage
behaviour
and
the
stove’s
durability
at
the
end
of
what
technically
is
the
life
span
of
the
ICS.
A
couple
of
factors
contribute
to
a
high
external
validity
of
this
RCT
for
the
African
context:
the
study
was
implemented
in
an
unob-
trusive
way
in
order
to
ensure
that
we
observe
real-world
cooking
behaviour.
It
was
designed
and
conducted
in
cooperation
with
the
ICS
dissemination
programme
of
the
Government
of
Senegal,
so
that
an
upscaling
of
the
intervention
under
real-world
conditions
would
be
possible.
Furthermore,
the
dominating
cooking
fuel
in
our
study
area
is
firewood,
which
is
also
the
case
in
most
other
African
countries
(Bonjour
et
al.,
2013).
Firewood
scarcity
in
our
study
region
and,
consequently,
the
incentive
to
use
more
effi-
cient
stoves
is
pronounced
and
comparable
to
other
dry
areas
in
non-equatorial
Africa.2
We
find
that
the
ICSs
are
taken
up
by
virtually
all
households
and
intensively
used,
even
after
three
and
a
half
years.
For
the
most
part,
people
only
give
up
using
the
stove
when
it
is
not
functional
anymore
and
not
because
they
lose
interest
in
using
it.
We
further-
more
observe
substantial
effects
on
firewood
consumption,
which
1See
World
Bank
(2011)
for
a
more
detailed
discussion
of
different
types
of
improved
cooking
stoves
and
Martin
et
al.
(2011)
for
a
recent
overview
on
the
improved
stoves
and
air
pollution
policy
debate.
2External
validity
and
potential
challenges
to
it
are
discussed
further
in
Section
3.5
and
Appendix
D.
confirm
savings
rates
determined
in
lab
tests.
In
addition,
we
find
a
decrease
in
early
indicators
for
respiratory
diseases
and
eye
infec-
tions.
These
effects
on
people’s
health
status
cannot
be
explained
only
by
the
take-up
of
the
new
ICS
and
the
firewood
savings,
but
rather
by
an
additional
reduction
in
smoke
exposure
due
to
more
outside
cooking
and
a
reduced
cooking
time
that
is
enabled
by
the
new
stove.
Our
findings
add
to
the
existing
body
of
evidence
on
ICS
impacts,
which
so
far
is
mainly
represented
by
two
RCTs:
the
RESPIRE
study
in
Guatemala
(see,
for
example,
Smith-Sivertsen
et
al.,
2004,
2009;
Díaz
et
al.,
2007;
Smith
et
al.,
2011)
and
a
study
conducted
by
J-Pal
in
India
(Hanna
et
al.,
2012).3Both
studies
used
stationary
chim-
ney
ICSs
that
are
installed
in
the
user’s
kitchen,
with
the
difference
that
the
RESPIRE
stoves
are
of
higher
quality,
thus
more
expen-
sive
(100–150
US$),
and
require
less
maintenance
than
those
used
in
the
Hanna
et
al.
(2012)
study.
A
more
detailed
comparison
of
technical
features
of
the
ICSs
used
in
the
different
studies
is
pro-
vided
in
Appendix
A.
While
the
RESPIRE
study
detects
a
substantial
reduction
in
household
air
pollution
and
a
reduction
in
the
risk
of
respiratory
disease
symptoms
and
eye
problems,
Hanna
et
al.
observe
reductions
in
smoke
inhalation
only
in
the
first
year
but
not
over
a
four
year
time
horizon.
This
is
mainly
driven
by
mainte-
nance
being
more
and
more
neglected
over
time,
which
leads
to
a
weak
performance
and
low
usage
rates
after
some
years.
Against
this
background,
our
paper
is
the
first
to
add
evidence
on
how
people
use
an
adapted
and
simple
ICS
in
an
unsupervised
setup
that
is
deemed
to
represent
a
more
realistic
study
environment
than
the
highly
controlled
medical
trials
conducted
for
RESPIRE.
Our
study
contributes
to
the
literature
by
providing
compelling
evidence
that
such
a
simpler
and
cheaper
ICS
can
actually
also
trigger
substantial
impacts
–
if
cooking
behaviour
also
changes.
Conceptually,
these
results
confirm
the
findings
of
Hanna
et
al.:
Looking
at
the
technical
features
of
an
ICS
is
not
enough,
since
the
real-world
behaviour
of
users
strongly
co-determines
the
results.
Unlike
Hanna
et
al.,
though,
we
find
that
behavioural
adaptations
to
a
simple
ICS
may
trigger
sizable
positive
health
effects.
These
differences
in
findings
of
the
two
studies
show
the
poten-
tials
of
disseminating
ICS
that
are
adapted
to
the
target
population
and
that
facilitate
cleaner
cooking.
The
stove
used
in
the
Hanna
et
al.
study
requires
regular
maintenance,
for
which
people
in
turn
need
to
be
trained
(which
not
all
of
them
were),
while
the
stove
random-
ized
for
our
study
is
maintenance-free.
Furthermore,
our
portable
stove
is
well
adapted
to
the
local
cooking
habits,
whereas
the
stove
distributed
in
Hanna
et
al.
interferes
more
with
local
cooking
habits
by
requiring
people
to
cook
inside,
which
they
are
not
accustomed
to.
In
this
sense,
the
stove
in
our
study
increases
the
number
of
choice
variables
for
the
users,
while
the
one
used
in
Hanna
et
al.
decreases
it.
In
this
broader
behavioural
context,
our
study
adds
to
a
nascent
strand
in
the
health
economics
literature
studying
adop-
tion
behaviour
of
households
for
health
relevant
technologies
and
goods
such
as
bednets
(Cohen
and
Dupas,
2010;
Tarozzi
et
al.,
2014),
point-of-use
drinking
water
disinfectants
(Luby
et
al.,
2008;
Kremer
et
al.,
2009),
deworming
drugs
(Kremer
and
Miguel,
2007),
condoms
(e.g.
Kamali
et
al.,
2003),
or
a
range
of
such
technolo-
gies
(Wendland
et
al.,
2015).
More
specifically,
it
demonstrates
3In
addition
to
these
two
studies,
further
evidence
with
mixed
results
exists
for
China
(Mueller
et
al.,
2013;
Yu,
2011),
Mexico
(Masera
et
al.,
2007)
and
urban
Senegal
(Bensch
and
Peters,
2013).
Burwen
and
Levine
(2012)
conducted
an
RCT
in
Ghana
using
a
very
simple
mud
stove.
As
a
major
difference
to
the
present
study
as
well
as
the
RESPIRE
and
the
J-Pal
study,
tests
in
a
controlled
field
lab
setting
already
find
that
the
stove
does
not
perform
better
than
the
traditional
ones.
The
poor
performance
is
also
reflected
in
low
usage
rates
after
a
few
months.

46
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
that
the
analysis
of
technology
adoption
and
related
promotion
programmes
should
encompass
both
a
technical
and
an
economic
perspective,
not
only
an
assessment
of
the
mechanical
perfor-
mance.
This
is
in
line
with
the
concept
of
intensive
and
extensive
margins
of
behaviour
that
has
recently
been
brought
into
the
debate
on
public
health
interventions
(see
Dupas,
2011):
It
is
not
only
the
mere
technology
adoption
that
counts
(extensive
margin).
Rather,
the
full
effect
can
only
be
determined
if
the
way
the
new
technology
is
used
is
accounted
for
as
well,
the
intensive
margin.
The
remainder
of
the
paper
is
organized
as
follows:
Section
2
reviews
the
country
and
intervention
background
and
outlines
the
research
design
including
the
identification
strategy.
Section
3
presents
the
study
results
for
all
our
impact
indicators,
and
Section
4
concludes.
2.
Programme
background
and
methodological
approach
2.1.
Improved
stove
dissemination
and
cooking
fuels
in
Senegal
Despite
its
seeming
superiority
to
traditional
biomass
cooking,
the
ICS
technology
has
not
made
significant
inroads
into
African
households.
There
may
be
various
reasons
for
this,
which
are
comprehensively
discussed
in
Rehfuess
et
al.
(2014)
and
Lewis
and
Pattanayak
(2012).
One
explanation
relevant
for
the
rural
setting
is
that
firewood
can
typically
be
collected
for
free
so
that
most
of
the
benefits
of
ICS
usage
are
not
monetary
ones.
This
makes
it
more
difficult
for
households
to
finance
the
investment
given
liquidity
and
credit
constraints.
On
the
supply
side,
the
stove
design
may
fail
to
meet
user
needs
in
preparing
local
dishes
with
available
fuels
and
cooking
utensils.
Earlier
programmes
in
various
African
countries
relied
on
subsidies
for
ICS
production
or
distributed
them
for
free.
Most
of
these
programmes
did
not
succeed,
however,
in
triggering
sustainable
ICS
usage.
Based
on
such
experience,
development
practitioners
frequently
argue
that
people
do
not
appreciate
and
use
ICS
that
they
receive
as
a
gift
and,
consequently,
reject
the
option
of
distributing
ICSs
for
free
(Barnes
et
al.,
1994;
Martin
et
al.,
2011).
This
is
also
the
spirit
underlying
the
ICS
dissemination
pro-
gramme
Foyer
Amélioré
au
Sénégal
(FASEN),
which
is
implemented
by
the
Senegalese
Ministry
of
Energy
in
cooperation
with
Deutsche
Gesellschaft
für
Internationale
Zusammenarbeit
(GIZ).4In
contrast
to
earlier
ICS
interventions,
FASEN
focuses
on
establishing
a
sus-
tainable
and
autonomous
market
for
ICSs
by
testing
performance,
training
producers
and
distributors,
and
supporting
communica-
tion
and
promotional
campaigns.
Similar
to
other
countries,
FASEN
so
far
concentrated
its
ICS
dissemination
on
charcoal
ICSs
in
urban
areas.
The
main
ICS
type
disseminated
by
FASEN
since
2006,
the
Jam-
baar,
is
also
used
in
the
present
RCT.
It
is
a
portable
single-pot
stove
with
a
fired
clay
combustion
centre
enclosed
by
a
metal
casing.
Owing
to
basic
design
improvements
of
the
Jambaar
compared
to
traditional
stoves,
the
woodfuel
burns
more
efficiently
and
the
heat
is
better
conserved
and
focused
towards
the
cooking
pot.
Both
charcoal
and
firewood
models
exist.
We
chose
the
firewood
Jambaar
for
our
experiment
as
firewood
is
the
dominant
fuel
in
rural
Senegal
with
89%
of
rural
households
using
it
as
their
primary
cooking
fuel
(ANSD,
2006).
In
rural
areas
ICSs
have
not
been
available
so
far.
Stove
types
used
here
are
either
three-stone
stoves
available
at
zero
cost
or
traditional
metal
stoves
and
open
fire
grills
that
can
be
bought
for
between
500
and
2500
CFA
Francs,
4GIZ
provides
technical
assistance
on
behalf
of
the
German
Federal
Ministry
for
Development
and
Economic
Cooperation
(BMZ)
and
is
one
of
the
largest
bilateral
development
agencies
in
the
world.
which
is
equivalent
to
1–5
US$
(see
Appendix
B
for
pictures
of
the
ICS
and
other
stove
types
used
in
the
study
region).
The
GIZ
programme
intends
to
expand
its
activities
to
rural
areas
and
expects
the
price
of
the
Jambaar
for
the
rural
market
to
be
around
4000–5000
CFA
Francs
(8–11
US$),
which
is
well
below
the
prices
of
the
more
sophisticated
ICS
technologies
widely
disseminated
in
Latin
America
or
Asia.
Cooking
fuels
are
an
issue
of
major
importance
in
the
daily
life
of
Senegalese
households.
Households
have
the
custom
to
cook
inside,
which
leads
to
a
higher
exposure
to
smoke
emissions
than
outside
cooking.
WHO
(2009)
holds
household
air
pollution
induced
by
solid
fuel
usage
for
cooking
accountable
for
6300
pre-
mature
deaths
every
year
in
Senegal
alone.
Apart
from
agricultural
land
clearance,
wood
usage
for
cooking
purposes
is
moreover
the
most
important
driving
force
of
ongoing
deforestation
in
the
mostly
arid
and
Sahelian
country
(see
WEC/FAO,
1999;
Tappan
et
al.,
2004;
FAO,
2005a,b).
A
constant
population
growth
of
2.6%
per
year
puts
further
pressure
on
fuelwood
resources.
As
a
consequence,
house-
holds
face
an
increasing
scarcity
of
fuelwoods:
firewood
collection
is
becoming
increasingly
time-consuming,
while
fuelwood
prices
are
rising.
This
circumstance
applies
particularly
to
the
Bassin
Arachidier,
the
study
area
of
this
evaluation,
situated
some
200
km
southeast
of
Dakar.
2.2.
Impact
indicators
The
first
impact
indicator
of
our
study
is
the
household
consump-
tion
of
firewood.
This
indicator
aggregates
each
dish
cooked
in
a
typical
week,
with
a
dish
being
one
component
of
a
meal
that
is
pre-
pared
on
a
separate
stove,
for
example
rice
and
sauce.
We
thereby
account
for
the
fact
that
several
stoves
may
be
used
simultaneously
for
the
preparation
of
a
single
meal.
The
rationale
for
this
indicator
is
that
a
reduction
in
firewood
consumption
not
only
has
immedi-
ate
implications
for
wood
scarcity
and
deforestation
pressures,
but
is
also
a
strong
intermediate
indicator
for
other
ultimately
relevant
impacts
such
as
health
and
time
use.
Impacts
on
health
and
time
use
are
examined
directly.
We
investigate
the
indicator
time
spent
by
household
members
on
fire-
wood
collection
and
cooking
and
the
prevalence
of
diseases
that
are
potentially
related
to
firewood
usage.
For
this
purpose,
we
look
at
symptoms
that
are
likely
to
be
affected
in
the
short-term
after
smoke
emissions
are
reduced;
these
are
captured
by
the
indicators
household
member
with
symptoms
of
respiratory
diseases
and
house-
hold
member
with
eye
problems.
We
examine
this
indicator
both
on
the
household
level
and
the
household
member
level.
For
respira-
tory
diseases,
these
symptoms
are
cough,
asthma,
or
difficulty
in
breathing.
They
indicate
acute
respiratory
infections
and
chronic
obstructive
pulmonary
diseases,
which
are
the
leading
causes
of
mortality
and
diseases
induced
by
exposure
to
air
pollution
from
solid
fuels
(Ezzati
and
Kammen,
2002).
Exposure
to
particles
could
be
detected
as
a
causal
agent
of
these
and
other
serious
respiratory
diseases
such
as
lung
cancer
or
pneumonia
(see
Duflo
et
al.,
2008b;
Pattanayak
and
Pfaff,
2009).
Respiratory
diseases
and
eye
problems
are
elicited
on
a
self-
reporting
basis:
respondents
are
asked
to
give
information
on
those
household
members
who
exhibited
the
symptoms
of
interest
in
the
six
months
preceding
the
interviews.
While
such
self-reported
health
indicators
are
sometimes
viewed
with
concern
because
of
potential
measurement
errors,
the
literature
supports
their
appli-
cation
by
highlighting
the
correlation
with
actual
illnesses
(see
Idler
and
Benyamini,
1997;
Miilunpalo
et
al.,
1997;
Peabody
et
al.,
2006;
Butrick
et
al.,
2010).
In
particular,
if
specific
symptoms
are
asked
about
precisely
as
was
done
in
this
study,
respondents
can
be
expected
to
report
accurately.
A
deterioration
in
recall
accuracy
of
reported
morbidity
as
found
in
Das
et
al.
(2012)
and
Kjellsson

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
47
et
al.
(2014)
is
a
concern
in
this
study
but
would
only
reduce
the
precision
of
our
health
estimates
and
not
induce
any
bias.
To
record
firewood
consumption
and
cooking
time,
the
person
responsible
for
cooking
is
asked
to
specify
the
number
of
people
cooked
for
and
the
types
of
stoves
used
for
every
meal
throughout
a
typical
day.
For
each
stove
application,
we
then
record
the
cooking
duration
and
the
cooking
fuel
type.
In
case
of
firewood,
the
cooking
person
is
additionally
asked
to
pile
up
the
amount
of
firewood
used
for
the
respective
stove
application,
which
is
then
weighed
with
scales.
In
combination
with
information
on
the
frequency
with
which
the
respective
stoves
are
used
throughout
a
typical
week,
this
data
serves
to
determine
the
weekly
household
consumption
of
firewood.
Enumerators
crosschecked
stove
usage
as
part
of
the
interviews
by
verifying
which
stove
was
currently
in
use
or
had
been
used
recently.
The
indicator
time
spent
by
household
members
on
cooking
aggregates
the
self-reported
cooking
duration
for
all
meals
of
a
typical
day,
whereas
the
time
spent
by
household
members
on
firewood
collection
aggregates
the
spells
in
which
household
members
are
occupied
with
gathering
firewood
in
the
course
of
a
week.
Technically
achievable
savings
rates
for
the
Jambaar
(referred
to
as
ICS
in
the
following)
have
already
been
determined
in
con-
trolled
cooking
tests
(CCTs),
where
a
cooking
person
prepares
the
same
meal
on
both
a
traditional
stove
and
an
improved
stove
in
order
to
compare
the
woodfuel
consumption
of
both
stove
types.
However,
the
effective
savings
in
real-life
households
might
devi-
ate
from
such
laboratory
field
tests
for
various
reasons
summarized
by
Bensch
and
Peters
(2013).5The
deficiencies
of
CCTs
can
be
over-
come
by
evaluating
the
woodfuel
consumption
based
on
a
survey
among
a
larger
sample
of
households
in
which
the
diversity
and
dynamics
of
real-life
cooking
practises
are
captured.
This
is
what
is
done
in
the
present
paper.
2.3.
Identification
strategy
We
employ
two
approaches
to
estimate
the
impact
of
ICS
usage
in
this
experimental
setup.
The
intention-to-treat
effect
(ITT)
is
obtained
by
simply
comparing
mean
values
of
impact
indicators
for
the
treatment
and
control
group,
without
accounting
for
non-
compliance
from
households
that
were
assigned
to
the
treatment
group
but
for
some
reason
do
not
use
the
ICS.
In
our
case,
the
ITT
serves
to
estimate
the
effect
of
providing
the
ICS
for
free
to
house-
holds
who
do
not
yet
own
one.
The
average
treatment
effect
on
the
treated
(ATT),
by
contrast,
accounts
for
non-usage
in
the
treat-
ment
group
and
potential
take-up
in
the
control
group
and
thereby
serves
to
estimate
the
impact
of
effective
ICS
usage.
For
this
pur-
pose,
instrumental
variable
(IV)
estimations
are
applied
with
the
random
assignment
into
the
treatment
group
as
an
instrument
for
the
effective
usage
of
the
ICS.
In
our
case,
ITT
and
ATT
are
very
simi-
lar
given
the
high
compliance
rate
in
the
treatment
group
and
given
that
only
one
household
in
the
control
group
acquired
an
ICS
from
another
source.
Although
RCTs
allow
for
a
simple
comparison
of
the
impact
indicators
at
the
time
of
the
follow-up,
the
precision
of
the
estimates
can
be
increased
by
controlling
for
other
household
char-
acteristics
that
have
been
collected
in
a
baseline
survey.
We
there-
fore
implement
both
the
ITT
and
ATT
approach
with
and
without
5For
example,
the
tests
frequently
concentrate
on
the
main
meal
only
and
they
cannot
account
for
the
fact
that
households
might
prepare
more
hot
meals
because
cooking
becomes
cheaper
due
to
the
higher
efficiency
of
the
ICS
(or
less
exhausting
in
terms
of
firewood
collection)
–
a
phenomenon,
which
is
referred
to
as
the
rebound
effect
in
the
energy
economics
literature
(see
Frondel
et
al.,
2008;
Herring
et
al.,
2009).
controlling
for
baseline
household
characteristics
such
as
educa-
tion
and
income
using
Ordinary
Least
Squares
(OLS)
regression.
In
order
to
shed
more
light
on
how
reductions
in
firewood
con-
sumption
are
induced
by
ICS
usage,
we
also
do
an
OLS
regression
on
the
individual
dish
level,
additionally
controlling
for
a
set
of
poten-
tial
dish-
and
meal-specific
confounders
such
as
the
number
of
people
cooked
for.
This
dish-level
regression
has
to
be
interpreted
with
some
care,
since
–
in
spite
of
the
random
ICS
assignment
–
the
households
that
received
a
new
stove
can
still
choose
whether
to
use
the
ICS
or
a
traditional
stove
for
the
respective
meal.
This
choice
might
then
be
driven
by
unobservable
factors,
which
would
distort
the
savings
estimates
if
the
unobservables
are
also
correlated
with
firewood
consumption.
Finally,
we
employ
probit
regressions
on
the
health
status
of
households
and
of
individual
household
members.
In
principle,
these
estimations
might
as
well
suffer
from
some
endogeneity
induced
by
intra-household
bargaining
processes:
healthier
and
more
powerful
women
might
bargain
themselves
out
of
cooking
with
the
dirtier
stove
and
into
cooking
with
the
cleaner
ICS
(see
Pitt
et
al.,
2006).
This
potentially
leads
to
a
spurious
correlation
between
ICS
ownership
and
improvements
in
the
health
status.
In
our
context,
though,
this
is
very
unlikely,
since
the
assignment
to
the
cooking
duty
does
not
seem
to
be
a
result
of
short-term
nego-
tiations,
but
it
is
rather
determined
by
cultural
norms
with
one
or
two
women
per
household
being
continuously
responsible
for
cooking.
Even
if
post-randomization
selection
processes
occurred,
they
would
be
uncovered
by
the
health
indicators
we
use,
because
we
observe
both
the
people
responsible
for
cooking
and
those
who
are
not.
2.4.
RCT
design
and
implementation
The
study
design
followed
the
guidelines
on
the
implementa-
tion
of
RCTs
provided
in
Duflo
et
al.
(2008a).
The
first
decision
that
had
to
be
taken
was
the
level
on
which
to
randomize
the
treatment
–
the
village
or
the
individual
household.
In
the
present
case,
it
is
sensible
to
randomize
on
the
household
level,
since
the
decision
about
whether
to
adopt
an
ICS
is
taken
in
the
household
and
not
on
a
regional
level.
Furthermore,
our
impact
indicators
are
mea-
sured
on
household
level
(or
below).
One
reason
to
randomize
on
the
village
level
instead
of
the
household
level
would
be
to
account
for
spillover
effects.
These
are
expected
to
be
negligible,
since
the
ICSs
are
only
used
by
the
households
themselves
and
the
penetra-
tion
rate
per
village
envisaged
in
this
RCT
is
too
low
to
affect,
for
example,
local
firewood
supply.
The
next
decision
regards
the
sample
size,
both
in
terms
of
households
and
villages.
We
determined
the
sample
size
based
on
a
power
calculation
focusing
on
the
indicator
firewood
savings.
We
approximated
the
relevant
parameters
ex-ante
using
the
data
col-
lected
for
the
quasi-experimental
study
presented
in
Bensch
and
Peters
(2013).
Taking
into
account
these
parameters
and
the
prob-
ability
of
being
assigned
to
the
treatment
group,
we
obtained
a
required
sample
size
of
250
households
spread
across
12
villages
(see
Appendix
C).
We
selected
villages
that
are
far
away
from
GIZ-
supported
ICS
producers
in
order
to
avoid
treatment
contamination
that
might
occur
if
households
randomly
assigned
to
the
control
group
obtain
an
ICS
independently.6Furthermore,
we
selected
the
6Two
further
channels
exist
through
which
the
treatment
may
be
contaminated.
First,
treatment
households
may
share
their
stoves
with
control
households.
This
did
not
occur.
Second,
the
two
household
groups
may
exchange
about
determinants
of
respiratory
health,
for
example.
Yet,
the
treatment
did
not
involve
any
awareness
raising
and
cooking
is
also
a
rather
private
issue,
as
stated
in
open
interviews,
that
seems
less
of
a
talking
point
in
women’s
conversations.
As
a
consequence,
only

48
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
Base
an
Nove
line
sur
vey
d lotter
y
e
mber 2009
Stove
a
and user
i
few
after t
h
a
llocat
ion
i
nstructions
day
s
h
e lott
er
y
Interm
e
vis
i
1, 2 and
7
after
all
o
e
diary
i
ts
7
month
s
o
cation
Follow-up
Survey
November
2010
ICS
u
s
tra
cking
s
March
2
s
age
s
ur
vey
2
013
Fig.
1.
Steps
in
RCT
implementation.
12
villages
from
the
target
region
of
a
planned
GIZ
rural
elec-
trification
intervention
so
that
we
could
introduce
the
study
as
preparatory
field
work
related
to
the
electrification
project
and,
thereby,
reduce
attention
paid
to
the
randomization.
In
November
2009,
we
conducted
the
baseline
survey
among
253
randomly
sampled
households
(see
Fig.
1
for
the
timeline
of
the
RCT).
Information
was
gathered
using
a
structured
question-
naire
covering
the
socio-economic
dimensions
that
characterize
the
relevant
living
conditions
of
the
households.
Since
the
study
also
served
as
a
baseline
in
the
context
of
the
envisaged
electrifi-
cation
intervention
–
a
solar
home
system
dissemination
project
–
a
particular
focus
of
the
questionnaire
was
on
energy
sources
(including
electricity)
and
energy
services
(including
cooking).
Consequently,
the
cooking-related
parts
of
the
interviews
did
not
draw
particular
attention.
This
is
important
to
avoid
auspices
biases
and
Hawthorne
effects
(see
Appendix
D).
We
complemen-
tarily
gathered
qualitative
information
in
focus
group
discussions
and
semi-structured
interviews
with
key
informants
such
as
women’s
groups,
stove
and
charcoal
producers,
teachers,
regional
administrators,
and
village
chiefs.
The
random
assignment
was
put
into
practice
through
a
lot-
tery
directly
following
the
baseline
interviews.
We
presented
the
prizes
of
this
lottery,
an
ICS
or
a
5
kg
bag
of
rice,
as
recompense
for
participation
in
the
baseline
study.
Participants
were
therefore
not
aware
of
being
part
of
an
experiment.
The
connotation
of
the
ICS
receivers
as
the
treatment
group
and
the
bag
of
rice
receivers
as
a
control
group
was
not
communicated
to
the
participants.7In
order
to
increase
trust
in
the
fairness
of
the
lottery,
we
conducted
it
in
each
of
the
villages
directly
after
completing
the
interviews
and
informed
the
households
immediately
about
which
recompense
they
would
get.
Hence,
we
applied
simple
stratified
randomization
with
the
villages
as
the
stratification
criterion.
Of
the
253
house-
holds
interviewed
for
the
baseline,
98
received
an
ICS
and
155
a
bag
of
rice.
The
rice
and
ICSs
were
distributed
within
three
days
of
the
baseline
interview.
The
households
that
were
drawn
to
get
an
ICS
received
a
brief
15-min
introduction
on
how
to
use
the
stove.8
The
ICS
and
rice
bag
distribution
as
well
as
the
instruction
were
done
by
field
workers
who
were
involved
in
the
preparation
of
the
electrification
project
and
who
were
visiting
the
village
anyhow.
No
specific
village
gathering
was
organized.
The
ICS
was
presented
minor
contamination
effects
are
conceivable
that,
furthermore,
would
rather
lead
to
an
underestimation
of
effects.
7The
average
rice
consumption
per
capita
in
Senegal
is
84
kg
per
year
(GAIN,
2011).
Hence,
the
bag
of
rice
received
by
the
control
group
corresponds
to
0.5%
of
annual
rice
consumption
for
the
average
household
size
in
our
sample
and
will
most
likely
not
affect
any
of
our
impact
indicators
that
were
measured
one
year
after
the
distribution.
8Because
many
other
ICS
types
require
more
extensive
maintenance
and
more
usage
instructions,
one
might
think
of
these
instructions
as
a
treatment
in
its
own
right,
which
might
be
introduced
as
a
random
second
treatment
arm.
For
our
ICS,
this
is
however
not
the
case.
Given
the
simplicity
in
the
use
of
the
ICS
and
given
that
it
is
virtually
maintenance
free,
additionally
randomizing
the
instruction
within
the
treatment
group
in
our
case
would
not
make
a
difference.
as
a
fuel-saving
device,
which
requires
a
few
precautions.
House-
holds
were,
for
example,
informed
that,
in
contrast
to
open
fires
for
which
people
typically
use
large
branches
or
even
trunks,
the
firewood
has
to
be
chopped
first
in
order
to
fit
the
relatively
small
fuel
feed
entrance
of
the
ICS.
In
line
with
what
real-world
users
are
told
about
this
type
of
ICS,
households
were
briefly
informed
about
the
convenience
co-benefits
of
fuel
savings,
which
are
a
quicker
cooking
process,
less
smoke
and
a
cleaner
kitchen
(if
cooking
is
done
indoors).
No
information
about
potential
repercussions
on
the
health
status
was
provided.
The
complete
instructions
on
the
functioning
and
proper
usage
of
the
ICS
and
related
information
provided
are
presented
in
Appendix
E.
Between
the
baseline
and
the
follow-up
phase,
local
community
workers
conducted
three
preparatory
visits
in
the
survey
villages
for
the
planned
electrification
project.
It
is
worth
highlighting
that
the
electrification
intervention
was
not
implemented
in
any
of
the
sampled
villages
before
the
end
of
this
study.
Furthermore,
electricity
is
virtually
never
used
for
cooking
in
rural
Africa,
in
par-
ticular
not
in
the
case
of
solar
home
systems
whose
capacity
is
not
sufficient
for
cooking
purposes.
Once
in
the
field,
the
commu-
nity
workers
additionally
checked
if
ICS
households
were
using
the
ICS
and
whether
they
had
encountered
technical
problems
(which
were
in
any
case
very
rare).
Again,
no
further
treatment
in
terms
of
awareness
raising
or
usage
encouragement
was
undertaken.
While
a
few
of
the
households
were
not
yet
making
frequent
use
of
their
new
stove
one
month
after
ICS
allocation,
by
the
time
of
the
second
visit
virtually
all
ICS
households
cooked
regularly
on
the
ICS.9For
the
follow-up
phase
at
the
end
of
2010,
the
same
structured
ques-
tionnaire
was
used
as
in
the
baseline
phase.
Attrition
was
very
low:
only
four
households
either
could
not
be
located
or
had
moved
out
of
the
village,
three
in
the
control
and
one
in
the
treatment
group.
None
of
the
households
refused
to
participate
in
the
follow-up
survey.
We
excluded
two
groups
of
households
from
the
analysis:
four
households
with
affiliated
Quran
schools,
where
usually
between
50
and
150
students
live
and
eat
and
which
are
therefore
not
com-
parable
to
family
households,
as
well
as
households
that
prior
to
the
study
had
already
received
improved
stoves
other
than
the
ICS
used
in
the
RCT
from
urban
relatives.
These
six
treatment
and
ten
control
households
cannot
be
expected
to
have
bought
another
ICS
in
a
non-RCT
world
and
therefore
do
not
represent
the
population
of
interest.
They
were
originally
included
in
the
randomization
only
because
they
were
a
priori
not
clearly
discernible
and
since
we
conducted
the
randomization
on-site
and
directly
after
the
survey.
No
further
restrictions
were
made
on
who
to
include
in
the
sample
9It
is
not
likely
that
the
delayed
take-up
was
triggered
by
the
visits
or
in
any
way
related
to
them.
Instead,
the
visits
revealed
that,
first,
a
few
housewives
travelled
outside
the
village
and
therefore
had
not
used
the
ICS
so
far.
Second,
some
women
needed
to
adapt
to
the
quicker
cooking
with
the
ICS,
which
at
the
beginning
created
a
feeling
of
insecurity.
Third,
some
households
were
reluctant
at
the
beginning
as
they
wanted
to
preserve
their
ICS
and
used
it
only
sparsely.
Fourth,
a
few
polygamous
households
needed
some
time
to
decide
on
who
would
use
the
ICS
and
when.

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
49
Table
1
Baseline
characteristics
of
randomly
assigned
ICS
owners
and
non-owners.
Treatment
Mean
(sd)
Control
Mean
(sd)
p-value
for
test
on
difference
in
means
(2)
−
(1)
(1)
(2)
(3)
Socio-economic
characteristics
Household
size
12.88
(5.55)
12.94
(5.82)
0.94
Family
structure
(%) 0.96
Extended
family
77.8
74.8
Nuclear
family
15.6
18.0
Couple
or
monoparental
family
6.6
7.2
Household
head
is
of
Wolof
ethnicity
(%)
52.2
50.4
0.92
Father
with
more
than
one
wife
(1
=
yes)
33.7
30.2
0.58
Father’s
education
level
(%) 0.88
None
12.5 9.8
Alphabetization
77.3
77.4
Primary
5.7
7.5
At
least
secondary
4.5
5.3
Main
wife’s
education
level
(%)
0.84
None
41.6
39.6
Alphabetization
51.7
53.9
Primary
6.7
5.8
At
least
secondary 0.0 0.7
Telecommunication
expenditures
(CFAF)
4250
(3830)
5090
(8640)
0.43
Ownership
of
bank
account
(1
=
yes)
0.08
0.06
0.55
Household
receives
remittances
(1
=
yes)
0.42
0.45
0.65
Thatched
roof
(1
=
yes)
0.67
0.67
0.97
Wall
material
of
house
is
stone
or
brick
(1
=
yes)
0.49
0.51
0.75
Flooring
material
is
soil
(1
=
yes)
0.36
0.27
0.16
Land
is
completely
owned
by
household
(1
=
yes) 0.94
0.93
0.62
Ownership
of
sheep
(1
=
yes)
0.62
0.63
0.87
Number
of
mobile
phones
owned
1.86
(1.31)
2.04
(2.15)
0.48
Main
wife
is
member
of
an
association
(1
=
yes)
0.71
0.73
0.71
Father’s
primary
activity
(%)
0.57
Subsistence
farming
79.3
83.4
Services
and
manufacturing 16.1
14.3
Retirement
4.6
2.3
Cooking-related
characteristics
Most
utilized
stove
type
(%)
0.33
Open
fire
(three-stones
or
Os†)
72.2
72.7
Traditional
metal
wood
stove
24.4
26.6
LPG
stove
3.3
0.7
Stove
usage
in
times
per
week 21.01
(4.10)
21.43
(4.98)
0.50
Firewood
consumption
per
dish
(kg)
Three-stones
4.85
(2.14)
5.14
(2.57)
0.37
Os†4.84
(2.32)
4.84
(2.84)
1.00
Per
capita
and
dish
firewood
consumption
of
three-stones
and
Os,
main
dishes
lunch
and
dinner
(kg)
0.50
(0.32)
0.47
(0.26)
0.52
Firewood
provision
(%)
0.95
Only
collected
76.1
76.3
Only
bought
1.2
0.7
Both
collected
and
bought
22.7
23.0
Number
of
observations
90
139
Note:
sd
–
standard
deviation;
p-values
are
determined
by
means
of
t-
and
chi-square
tests
†The
Os
is
a
stove
in
which
an
open
fire
burns
between
three
metal
feet.
or
the
analysis.
Altogether,
the
sample
used
for
the
subsequent
impact
analysis
in
Sections
3.2–3.4
comprised
229
households.
As
a
robustness
check
shows,
not
discarding
these
two
groups
of
households
and,
hence,
performing
the
analysis
with
all
249
households
for
which
baseline
and
follow-up
data
is
available
does
not
change
any
of
our
findings,
neither
when
applying
ITT
nor
ATT.
In
March
2013,
approximately
three
and
a
half
years
after
the
randomization,
an
ICS
usage
tracking
survey
among
the
households
that
had
received
an
ICS
was
conducted
by
enumerators
famil-
iar
with
the
ICS.
All
but
one
of
the
90
ICS
households
included
in
the
impact
analysis
could
be
retrieved
for
this
interview
wave.
In
addition
to
asking
the
households
simple
usage
questions,
the
enumerators
recorded
their
own
assessment
on
the
condition
of
the
ICS.
The
results
of
this
usage
tracking
survey
are
presented
in
Section
3.5.
3.
Results
3.1.
Socio-economic
conditions
and
cooking
behaviour
The
primary
purpose
of
this
section
is
to
scrutinize
the
balancing
of
the
two
randomized
groups,
since
we
abstained
from
explic-
itly
balancing
them
through
re-randomization
before
assigning
the
ICSs.
The
second
purpose
is
to
illustrate
the
socio-economic
envi-
ronment
in
which
the
RCT
was
implemented.
Table
1
documents
the
baseline
socio-economic
and
cooking-related
characteristics
of
the
229
households
before
stove
distribution.
On
average,
house-
holds
consist
of
13
members;
household
size
varies
in
a
range
between
2
and
42
persons
per
household.
Larger
households
are
more
common:
78%
are
extended
families
and
16%
nuclear
families
(two
parents
plus
children).
Four
in
five
households
are
subsistence

50
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
Fig.
2.
Distribution
of
non-farm
income
at
baseline.
Fig.
3.
Distribution
of
farm
income
at
baseline.
farmers,
the
majority
of
them
living
in
houses
with
thatched
roofs.
As
can
be
seen
from
the
p-values
in
the
right-hand
column,
two-
sided
tests
of
equality
of
the
values
for
the
two
compared
groups
do
not
reveal
statistically
significant
differences.
The
groups
are
balanced
in
the
relevant
observable
characteristics.
In
addition,
Figs.
2
and
3
show
the
distribution
in
non-agricultural
and
agricul-
tural
income:
the
treatment
and
the
control
group
strongly
overlap.
Accordingly,
a
two-sample
Kolmogorov–Smirnov
test
cannot
reject
the
null
of
identical
distributions
at
the
10
percent
level.10
Regarding
the
baseline
stove
usage
patterns
reflected
in
the
table,
two
stove
types
dominate
rural
kitchens
in
Senegal:
open
fires
(three-stone
stoves
or
Os
in
which
the
open
fire
burns
between
metal
feet)
and
traditional
metal
stoves,
the
Malagasy
and
Cire.
LPG
stoves
are
rarely
used
in
rural
Senegal;
in
our
sample
only
three
households
mainly
use
LPG
for
cooking.
90%
of
dishes
are
prepared
with
firewood.
Around
15%
of
all
meals
are
prepared
with
more
than
one
stove,
primarily
to
prepare
rice
on
one
stove
and
a
sauce
on
a
second
one.
On
average,
each
household
prepares
21
hot
dishes
10 We
additionally
ran
a
probit
regression
to
check
the
correlation
between
ICS
allocation
and
the
joint
set
of
cooking-related
as
well
as
socio-economic
character-
istics
and
village
dummies.
As
part
of
this
estimation,
we
performed
a
Likelihood
Ratio
chi-square
test
with
the
null
hypothesis
that
all
of
the
regression
coefficients
are
simultaneously
equal
to
zero.
The
p-value
of
0.98
validates
the
findings
from
the
univariate
comparisons
of
no
correlation.
All
tests
have
as
well
been
carried
out
with
the
original
sample
of
253
baseline
households,
for
which
statistically
significant
differences
cannot
be
observed
either.
Table
2
Utilization
rates
of
different
stove
types
at
follow-up.
Treatment
Control
Open
fire
19.5%
70.5%
Traditional
metal
wood
stove
7.7%
24.1%
ICS
69.1%
0.7%†
LPG
stove 3.7% 4.7%
Average
total
number
of
stove
applications
per
household
and
week
25.3
22.6
Note:
The
shares
represent
the
ratio
between
the
number
of
times
the
respective
stove
type
is
used
and
the
total
number
of
stove
applications
per
household
and
week.
†ICS
usage
among
the
control
group
is
due
to
the
fact
that
one
household
which
was
not
randomly
assigned
to
receive
an
ICS
acquired
one
individually
after
the
randomization.
per
week
using
one
of
its
stoves.
As
sometimes
more
than
one
stove
is
used
for
one
meal,
the
range
of
weekly
stove
applications
is
between
14
and
49.
The
follow-up
data
on
stove
usage
shows
no
changes
in
the
control
group:
the
most
often
used
stove
types
are
three-stone
stoves
(53%),
traditional
metal
stoves
(25%)
or
Os
(20%).
Accord-
ingly,
the
savings
potentials
of
ICS
usage
are
relatively
high
with
73%
of
households
mainly
using
open
fire
stoves
in
the
absence
of
an
ICS.
For
the
treatment
group,
the
follow-up
data
shows
that
the
ICSs
have
achieved
broad
acceptance
among
users.
There
are
only
two
non-compliers:
one
ICS
was
completely
broken
in
an
accident
and
one
household
did
not
use
the
new
stove.
Otherwise,
as
many
as
95%
of
the
distributed
ICSs
are
used
at
least
seven
times
per
week;
for
85%
of
treatment
households
the
ICS
became
the
predominantly
used
stove.
The
proportion
of
individual
dishes
prepared
with
the
different
stove
types
also
mirrors
this
usage
pattern
(see
Table
2).
As
such,
our
set-up
mimics
the
most
likely
scenario
where
treat-
ment
households
have
one
ICS
at
their
disposal
and
continue
to
use
less
efficient
traditional
stoves,
because
one
stove
is
not
sufficient
to
prepare
the
required
amount
of
food
or
because
the
ICS
is
too
small
for
the
pot
sizes
used
in
a
few
large
households.
The
table
also
shows
that
treatment
households
increased
the
number
of
dishes
prepared.
This
is
probably
not
due
to
rebound
effects
(see
footnote
**),
since
the
total
number
of
hot
meals
cooked
does
not
increase
in
the
treatment
group
and
households
reported
that
the
quantity
and
type
of
food
prepared
has
not
changed
since
receiving
the
ICS.
Instead,
the
increase
simply
reflects
the
fact
that
ICS
households
have
an
additional
stove
at
their
disposal
such
that
the
different
components
of
one
meal
that
were
formerly
prepared
on
a
single
stove
are
now
prepared
on
two
stoves.
3.2.
Firewood
consumption
ITT
and
ATT
estimates
for
the
household
consumption
of
fire-
wood
indicator
are
calculated
both
with
and
without
the
baseline
household-level
control
variables
taken
from
Table
1:
in
addition
to
income,
telecommunication
expenditures
are
used
as
a
proxy
for
living
standards.
Bank
account
ownership
is
used
as
a
proxy
for
the
household’s
access
to
credits
and
ability
to
pay.
Housing
condi-
tions
as
a
wealth
indicator
are
captured
by
whether
the
flooring
material
in
the
household
is
soil
and
whether
the
wall
material
is
stone
or
brick.
As
another
wealth
metric,
we
include
a
dummy
indi-
cating
sheep
ownership.
The
results
do
not
change
if
other
wealth
and
socio-economic
indicators
shown
in
the
table
are
included.
As
suggested
in
Bruhn
and
McKenzie
(2009),
we
additionally
include
village
dummies
in
order
to
account
for
the
stratified
randomiza-
tion.
According
to
our
findings
presented
in
columns
(1)
and
(2)
of
Table
3,
firewood
savings
are
substantial,
with
around
27
kg
being
saved
per
week
in
every
household
after
introduction
of
the
ICS.
These
are
ITT
results.
ATT
estimates
differ
only
marginally
being

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
51
Table
3
Effect
of
ICS
usage
on
firewood
consumption
per
week
and
per
dish.
Estimator:
Coefficient
(Standard
Error
in
parentheses)
Ordinary
Least
Squares,
ITT
Dependent
variable:
Firewood
consumption
per
week
in
kg
Firewood
weight
per
dish
in
kg
Variable
(1)
(2)
(3)
(4)
Dish
variables†
Dish
is
cooked
on
open
fire
Ref.
Ref.
Dish
is
cooked
on
ICS
−1.99*** (0.24)
−2.04*** (0.24)
Dish
is
cooked
on
traditional
metal
stove
−0.03
(0.53)
Ref.
Main
dish 1.00*** (0.33)
Short
cooking
(<30
min)
−0.94*** (0.19)
Meal
variables†
Number
of
people
the
meal
is
cooked
for
(in
terms
of
the
logarithm
of
adult
equivalents)
1.96*** (0.57)
Lunch
Ref.
Breakfast
−1.66*** (0.18)
Dinner
−0.32*** (0.10)
Multiple
stoves
−0.14
(0.33)
Household
variables
Household
with
ICS
−26.78*** (6.33)
−26.96*** (6.10)
Average
number
of
people
cooked
for
(in
terms
of
the
logarithm
of
adult
equivalents)
42.79*** (14.36)
Father
has
formal
education
2.63
(8.27)
0.01
(0.31)
Mother
has
formal
education 5.95
(5.39) 0.23
(0.20)
Household
income
(in
logarithmic
terms)
1.25
(2.48)
0.03
(0.07)
Telecommunication
expenditures
(in
logarithmic
terms) 0.56
(0.97)
0.03
(0.03)
Bank
account
ownership
1.66
(16.85)
0.46
(1.18)
Flooring
material
is
soil
−17.63** (7.09)
−0.60** (0.25)
Wall
material
of
house
is
stone
or
brick
2.55
(9.83)
0.42
(0.36)
Ownership
of
sheep
−7.31
(7.80)
−0.32
(0.29)
Association
membership
of
the
mother −7.02
(7.93) −0.71** (0.29)
Village
dummies
Included
Included
Included
Included
Constant
−13.60
(42.09)
82.56*** (8.83)
−0.77
(1.64)
3.90*** (0.36)
Mean
of
treatment
group
60.80
(3.92)
60.69
(3.24)
2.28
(0.16)
2.25
(0.12)
Mean
of
control
group
87.58
(4.68)
87.65
(5.03)
4.27
(0.17)
4.29
(0.21)
Savings
rate
(%)
30.6
30.8
46.7
47.5
Number
of
observations 228
228
627
633
Adjusted
R-squared
0.25
0.13
0.43
0.18
F-test
4.83*** 3.71*** 16.18*** 9.44***
Note:
Computations
on
household
level
(columns
1
and
2)
are
performed
with
heteroskedasticity
corrected
standard
errors
accounting
for
heterogeneity
in
treatment
responses;
standard
errors
for
the
dish-level
estimations
(columns
3
and
4)
are
clustered
by
household.
†For
an
explanation
of
the
dish-
and
meal-level
control
variables,
see
Bensch
and
Peters
(2013).
*Significance
level
of
10%.
** Significance
level
of
5%.
*** Significance
level
of
1%.
slightly
higher.
As
these
observations
hold
in
the
same
way
for
the
other
impact
indicators,
we
will
only
present
the
more
conserva-
tive
ITT
estimates
in
the
following
(ATT
estimates
can
be
taken
from
Table
F1
in
Appendix
F).
Inserting
in
the
regression
the
values
1
and
0
for
the
binary
treatment
variable
and
average
values
for
the
covariates
gives
us
the
absolute
ICS
consumption
values
shown
at
the
bottom
of
the
table.
This
implies
that
30%
of
the
households’
total
firewood
consumption
is
saved.
This
is
clearly
less
than
the
40–50%
found
in
CCTs.
As
noted
above,
rebound
effects
as
one
potential
driver
for
the
difference
to
CCT
results
do
not
seem
to
play
a
role.
Another
likely
reason
is
the
fact
that
treatment
households
do
not
switch
completely
to
ICS
usage
and
still
prepare
parts
of
their
meals
on
traditional
stoves.
In
order
to
assess
the
savings
potentials
in
case
they
would
fully
switch
to
ICS
usage,
we
additionally
compare
the
firewood
con-
sumption
for
dishes
prepared
on
an
ICS
in
the
treatment
group
to
dishes
prepared
on
traditional
stove
types
in
the
control
group.
Even
though
the
analysis
of
firewood
savings
on
the
dish
level
may
be
endogenous,
it
provides
an
upper
bound
estimate
of
savings
potentials
where
households
had
access
to
several
ICSs
to
poten-
tially
abandon
traditional
stoves
completely.
These
estimations
furthermore
provide
insights
into
how
the
savings
materialize,
since
they
make
it
possible
to
examine
the
influence
of
dish-
and
meal-specific
factors.
Table
3
shows
in
columns
(3)
and
(4)
the
results
for
the
OLS
regression
that
controls
for
household
charac-
teristics
and
characteristics
specific
to
the
stove
application.
The
results
reveal
the
differential
effects
of
various
dish-
and
meal-
specific
variables
whose
coefficient
signs
are
as
expected
and
reflect
consistent
firewood
consumption
figures
across
dish
types.
The
R2of
0.43
for
the
estimation
including
control
variables
in
col-
umn
3
indicates
that
a
good
part
of
the
variation
in
the
dependent
variable
can
be
explained
by
observable
factors.
The
statistically
highly
significant
ICS
coefficient
would
imply
an
average
ICS
sav-
ings
rate
of
47%.
It
is,
thus,
in
the
range
of
the
CCT
results.
An
unbiased
alternative
to
come
up
with
a
firewood
savings
estimate
for
the
case
of
adopting
ICS
for
the
entire
range
of
stove
applications
is
to
perform
a
slightly
adapted
version
of
the
IV
estimation
in
the
calculation
of
the
ATT
for
total
firewood
consumption.
We
now
instrument
a
new
treatment
variable,
ICS
usage
intensity,
by
the
random
assignment.
Usage
intensity
is
coded
as
a
continuous
variable
obtained
by
dividing
the
number
of
dishes
prepared
on
an
ICS
by
the
total
number
of
dishes
prepared

52
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
Table
4
Effect
of
ICS
usage
on
time
expenditures.
Treatment
Control
Difference
in
means
(2)
−
(1)
Regression-adjusted
difference
in
means
Mean
(se)
Mean
(se)
Mean
(se)
p-Value
(H0:
Diff
=
0)
Mean
(se)
p-Value
(H0:
Diff
=
0)
(1)
(2)
(3)
(4)
(5)
(6)
Duration
of
firewood
collection
per
week
(min)
719((75.1)
867((69.4)
148(103.3)
0.15
136(102.0)
0.19
Number
of
observations†86
134
Cooking
duration
per
day
(min)
25133381(21.8)
0.00*** 75(20.7)
0.00***
Number
of
observations 90
139
Note:
All
values
derived
from
ITT
estimations
with
heteroskedasticity
corrected
standard
errors
(in
parentheses)
and
including
village
dummies;
se
–
standard
error.
†For
the
firewood
collection
indicator,
the
nine
missing
observations
(5
control
and
4
treatment)
are
due
to
households
that
were
not
able
to
specify
the
firewood
collection
time
spells.
*** Significance
level
of
1%.
in
the
respective
household.
It
thus
ranges
from
0%
to
100%.
The
resulting
Wald
estimator
yields
an
average
rate
of
43.8–45.0%
(with
and
without
controls).
This
unbiased
IV
estimate
still
suffers
from
the
fact
that
treatment
households
increased
the
number
of
stove
applications
over
which
they
spread
the
food
preparation.
Nevertheless,
we
can
conclude
that
if
all
meals
in
a
household
were
cooked
on
an
ICS,
the
savings
rate
obtained
in
columns
(1)
and
(2)
of
Table
3
could
well
be
around
40%.
The
results
on
firewood
consumption
turn
out
to
be
robust
to
outliers
just
like
the
results
for
the
time
and
health
indicators
assessed
in
the
following
two
sections.
The
two
robustness
checks
we
applied
were,
first,
to
estimate
the
median
of
the
dependent
variables
by
using
quantile
regression
techniques
and,
second,
to
exclude
outliers
defined
as
values
more
than
two
standard
devia-
tions
away
from
the
mean.
The
results
can
be
taken
from
Table
F2
in
Appendix
F.
3.3.
Time
use
As
many
as
96%
of
all
households
collect
at
least
part
of
the
fire-
wood
they
use
for
cooking.
A
reduction
in
firewood
consumption
is
likely
to
lead
to
households
spending
less
time
on
firewood
collec-
tion.
In
fact,
the
reduction
in
the
aggregate
time
spent
by
household
members
on
firewood
collection
is
approximately
two
and
a
half
hours
per
week,
which
corresponds
to
16–17%
(Table
4).
The
reduc-
tion,
though,
is
statistically
only
borderline
significant
(p-values
of
between
0.15
and
0.19
for
ITT
with
and
without
controls),
a
finding
that
does
not
seem
to
be
fully
consistent
with
the
reduction
in
total
firewood
consumption
of
around
30%
found
in
the
previous
section.
Still,
it
is
not
surprising
that
time
savings
are
less
pronounced
than
savings
in
firewood.
One
reason
for
the
lower
savings
is
that
ICS-
using
households
might
just
collect
less
wood
during
one
excursion
instead
of
reducing
the
number
of
excursions.
The
lack
of
statisti-
cal
significance
of
the
difference
might
be
due
to
inaccuracies
in
the
time
usage
variable,
which
increases
the
standard
error
and,
thus,
reduces
power.
The
inaccuracies
are
induced,
for
example,
by
the
fact
that
31%
of
households
collect
the
wood
on
their
own
land
while
farming,
which
makes
it
difficult
to
disentangle
time
spent
on
the
task
of
collection
from
time
spent
on
ordinary
field
work.
Also,
some
households
use
a
variety
of
wood
supply
strategies
depending
on
different
factors,
most
notably
the
season:
some,
for
example,
do
not
collect
the
firewood
every
week
but
instead
hold
a
stock
that
is
typically
replenished
before
the
rainy
season.
ICS
households
might
moreover
save
time
because
cook-
ing
is
facilitated
and
quicker.
In
qualitative
interviews
women
repeatedly
pointed
out
that
the
ICS
allows
them
to
regulate
the
temperature
more
easily,
which,
in
turn,
makes
it
easier
to
do
other
things
while
cooking.
The
cooking
duration
of
all
three
meals
throughout
a
typical
day
decreases
significantly
by
more
than
75
min
(Table
4),
where
preparation
of
an
individual
meal
on
an
ICS
takes
around
one
and
a
half
hours.
These
savings
far
exceed
the
time
that
households
additionally
invest
in
cutting
the
firewood
into
smaller
pieces,
which
takes
not
more
than
15
min/day.
Due
to
a
lack
of
local
job
and
business
opportunities,
a
shift
of
time
towards
income-generating
activities
cannot
be
observed.
The
cooking
women
do
not
seem
to
sleep
more
either,
since
their
time
awake
differs
by
mere
5
min
between
the
two
compared
groups.
Qualitative
discussions
rather
suggest
that
the
facilitation
of
the
cooking
task
helps
them
to
execute
household
duties
in
a
less
hurried
way
and
to
take
more
rest
during
the
day.
3.4.
Health
The
negative
effect
of
firewood
usage
on
people’s
health
may
be
alleviated
by
ICS
usage
via
two
channels.
First,
the
reductions
in
firewood
consumption
found
in
Section
3.2
can
be
expected
to
reduce
harmful
smoke
emissions,
although
it
is
–
as
discussed
in
the
introduction
–
unclear
whether
simple
ICSs
like
those
used
in
this
RCT
reduce
smoke
emission
sufficiently
to
induce
positive
health
effects.
Second,
exposure
to
the
emitted
smoke
might
be
reduced,
either
via
reductions
in
the
cooking
duration
(as
found
in
Section
3.3)
or
if
cooking
behaviour
changes
because
of
the
new
stove.
In
general,
smoke
exposure
is
very
high
in
rural
Senegal,
with
around
two-thirds
of
the
household
members
responsible
for
cooking
staying
next
to
the
stove
most
of
the
time
they
are
cooking.
Furthermore,
the
vast
majority
of
households
cook
inside,
predominantly
in
a
separate
kitchen.
While
in
the
control
group
the
proportion
of
outside
cookers
stays
stable,
in
the
treatment
group
it
doubles
from
11%
to
23%
between
baseline
and
follow-up.
The
main
reason
for
this
can
be
traced
to
the
fact
that
the
ICS
better
shields
the
fuel
from
wind
than
three-stone
stoves;
also,
from
the
households’
perspective,
wind
and
dust
are
indeed
the
main
draw-
back.
In
addition,
the
ICS
requires
less
supervision,
allowing
the
cook
to
dedicate
more
of
her
attention
to
other
tasks
away
from
the
smoke
source.
Virtually
all
persons
responsible
for
cooking
are
women,
on
average
two
per
household
with
no
difference
between
treatment
and
control.
We
examine
whether
chronic
symptoms
of
respiratory
diseases
and
eye
infections
prevail
among
the
women
respon-
sible
for
cooking
and,
as
placebo
outcomes,
among
the
women
not
responsible
for
cooking
and
male
household
members.
We
first
look
at
two
dummy
variables:
at
least
one
household
mem-
ber
with
symptoms
of
respiratory
diseases
and
at
least
one
household
member
with
eye
problems
take
the
value
one
if
at
least
one
house-
hold
member
of
the
respective
group
reports
having
suffered
from
these
symptoms
at
some
point
in
the
last
six
months
before
the
interview.
The
results
are
displayed
in
Table
5
and
indicate
the
share
of
households
for
which
these
variables
take
the
value
one.
The
gender-differentiated
data
provides
for
striking
indications
of
health
effects:
for
women
responsible
for
cooking,
9.0%
of
treated

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
53
Table
5
Effect
of
ICS
usage
on
health
status.
Treatment
Control
Difference
in
means
(2)-(1)
Regression-adjusted
difference
in
means
Mean
Mean
Mean
p-Value
(H0:
Diff
=
0)
Mean
p-Value
(H0:
Diff
=
0)
(1)
(2)
(3)
(4)
(5)
(6)
Household
level
analysis
Respiratory
system
disease
(%) †
Any
woman
responsible
for
cooking
9.0
17.7
8.7
0.07*
Any
male 7.9 6.5 −1.4
0.69
Any
woman
not
responsible
for
cooking
3.5
6.2
2.7
0.38
Eye
problems
(%) †
Any
woman
responsible
for
cooking
4.5
14.0
9.5
0.02**
Any
male
4.5
7.3
2.8
0.40
Any
woman
not
responsible
for
cooking
8.1
7.1
−1.0
0.78
Number
of
observations‡86–90 127–139
Individual
level
analysis
Cook
in
household
shows
symptoms
of
.
.
.
(%)
A
respiratory
system
disease
4.7
11.8
7.1
0.01** 6.9
0.01***
Eye
problems 2.9 9.8
6.9
0.01** 5.7
0.00***
Non-cooking
person
in
house
hold
shows
symptoms
of.
.
.
(%)
A
respiratory
system
disease
1.7
2.0
0.3
0.63
0.5
0.46
Eye
problems
1.9
2.0
0.1
0.87
0.2
0.73
Number
of
observations
778
1199
Note:
Standard
errors
for
the
household
level
estimations
are
heteroskedasticity
corrected,
those
for
individual
household
member
level
estimations
are
clustered
by
household,
all
estimations
include
village
dummies
in
order
to
account
for
the
stratified
randomization.
†ITT
with
inclusion
of
controls
is
not
shown
in
this
table,
since
for
some
control
variables
(bank
account
ownership,
flooring
material,
village
dummies)
failure
is
perfectly
predicted
in
the
estimated
probit
regressions.
‡Differences
in
the
number
of
observations
are
due
to
a
few
missing
values
and
some
households
without
any
woman
not
responsible
for
cooking.
§The
values
in
this
analysis
are
marginal
means
and
marginal
effects
derived
from
estimations
that
can
be
found
in
regression
form
in
Appendix
F,
Table
F3.
They
are
conventionally
calculated
at
the
mean
of
the
other
independent
variables
taking
into
account
the
particularities
of
calculating
margins
for
interaction
terms
in
non-linear
models
and
conditioning
on
household
members
who
are
cooks.
*Significance
level
of
10%.
** Significance
level
of
5%.
*** Significance
level
of
1%.
households
report
at
least
one
of
them
suffering
from
respiratory
disease
symptoms.
The
corresponding
value
for
the
control
group
of
17.7%
is
almost
twice
as
large
–
with
this
difference
being
statis-
tically
significant.
If
we
look
at
the
same
proportion
for
male
household
members,
who
usually
do
not
spend
time
around
the
cooking
spot,
treatment
and
control
group
households
do
not
differ
significantly
from
each
other,
nor
do
we
find
a
difference
for
women
not
responsible
for
cooking.
The
same
pattern
is
observable
for
eye
infections:
14.0%
of
households
report
that
at
least
one
woman
responsible
for
cook-
ing
suffers
from
eye
problems
in
the
control
group
compared
with
4.5%
in
the
treatment
group.
The
difference
is
significantly
different
from
zero.
No
such
statistically
significant
difference
is
observed
for
men
and
women
not
responsible
for
cooking.
With
respect
to
the
potential
bargaining
into
or
out
of
cooking
selection
processes
out-
lined
in
Section
2.3,
one
would
expect
changes
in
prevalence
rates
in
the
group
of
women
who
are
not
responsible
for
cooking
if
that
bargaining
process
was
strong.
However,
this
is
not
the
case.
The
bottom
of
Table
5
refers
to
results
derived
from
ITT
pro-
bit
regressions
for
the
same
disease
symptoms
on
the
level
of
individual
household
members.11 We
now
look
at
the
dummy
variables
household
member
with
symptoms
of
respiratory
diseases
and
household
member
with
eye
problems,
which
take
the
value
one
if
the
respective
household
member
reports
having
suffered
from
these
symptoms
at
some
point
in
the
last
six
months
before
the
interview.
We
find
prevalence
rates
of
between
3%
and
12%
11 We
abstain
from
showing
ATT
estimations
here,
since
the
specification
requires
interacting
the
treatment
status
with
the
dummy
variable
that
indicates
the
cooking
responsibility.
We
would
thus
need
to
instrument
the
ICS
uptake
and
the
interaction
term,
respectively.
Using
the
random
assignment
as
instrument
for
both
ICS
uptake
and
also
in
the
interaction
term
(which
is
a
controversial
procedure)
does
not
deliver
any
result
in
our
case,
since
the
estimations
do
not
converge.
on
individual
level.12 The
results
confirm
the
findings
of
the
household
level
estimations.
In
the
group
of
household
members
responsible
for
cooking
the
prevalence
rates
for
both
respiratory
disease
symptoms
and
eye
infections
go
down
by
almost
seven
percentage
points.
Significance
levels
are
even
more
pronounced
with
p-values
of
0.01
for
both
estimations
with
and
without
control
variables
respectively
reflecting
the
more
accurate
definition
of
the
indicator
and
the
larger
sample
size.
The
estimations
as
well
corroborate
that
the
treatment
has
no
effect
at
all
on
the
group
of
household
members
not
responsible
for
cooking.
Altogether,
while
the
reduction
in
smoke
due
to
fuel
savings
might
be
too
modest
to
trigger
perceivable
health
effects
by
itself,
it
is
likely
that
the
combination
with
the
change
in
cooking
behaviour
enabled
by
the
ICS
explains
the
observed
improvements
in
health
indicators:
the
ICS
facilitates
outdoor
cooking,
the
cooking
duration
is
reduced,
and
the
cooking
and
combustion
process
requires
less
supervision.
3.5.
Impact
sustainability
and
upscaling
the
intervention
Hitherto
we
have
found
quite
strong
and
robust
evidence
for
high
take-up
and
impacts
of
ICS
usage
after
one
year
that
are,
given
the
experimental
set-up,
internally
valid.
Internal
validity,
though,
is
only
a
necessary
condition
for
high
policy
relevance.
The
decisive
questions
in
a
next
step
are,
first,
whether
these
usage
rates
and
impacts
persist
over
time,
second,
whether
the
intervention
yields
12 Comparable
data
on
respiratory
system
diseases
and
eye
problems
for
Sub-
Saharan-African
countries
is
very
sparse
(van
Gemert
et
al.,
2011).
Studies
with
indicator
definitions
that
come
closest
to
ours
show
comparable
levels
in
these
health
problems
(ANSD
and
ICF
International,
2012;
Adeloye
et
al.,
2013).

54
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
Fig.
4.
Decline
in
the
percentage
of
ICS
users
among
randomized
households.
benefits
that
outweigh
the
costs
and
if
so,
third,
whether
it
can
be
upscaled.
In
order
to
assess
the
sustainability
of
the
observed
impacts
we
conducted
an
ICS
usage
tracking
survey
three
and
a
half
years
after
the
random
assignment.
This
enables
us
to
examine
the
durability
of
the
randomized
ICS
under
day-to-day
rural
cooking
conditions
and
the
usage
behaviour
over
the
full
life-span
of
the
ICS.
In
this
sec-
ond
follow-up
round,
we
did
not
collect
information
on
impacts,
because
a
majority
of
the
stoves
would
have
already
exceeded
their
useable
lifetimes.
For
statistical
power
reasons,
this
reduced
sample
size
would
have
made
an
examination
of
impacts
difficult.
Considering
an
expected
life
span
of
one
to
three
years,
the
propor-
tion
of
49%
of
treatment
households
still
using
the
randomized
ICS
can,
nevertheless,
be
considered
surprisingly
high.
In
the
enumer-
ators’
appraisal,
half
of
these
ICS
were
still
in
good
condition.
The
proportion
of
dishes
prepared
with
an
ICS
among
ICS
users
declined
only
slightly
from
70%
in
2010
to
62%.
As
can
be
seen
in
Fig.
4,
those
treatment
households
who
do
not
use
the
ICS
anymore
(51%)
only
slowly
ceased
to
use
their
ICS.
All
of
them
have
done
so
because
the
stove
has
deteriorated
and
90%
of
them
still
used
their
ICS
two
years
after
randomization.13
Against
this
background
of
persisting
usage
behaviour
we
con-
duct
a
simple
cost–benefit
analysis.
The
costs
of
the
ICS
are
represented
by
the
market
price
of
around
10
US$.
For
a
conser-
vative
estimate
of
the
benefits,
to
begin
with,
we
only
account
for
reductions
in
firewood
consumption.
We
take
the
average
price
of
0.02
US$/kg
of
firewood
paid
by
firewood-purchasing
house-
holds
at
the
time
of
the
follow-up
survey
as
an
upper
bound
of
the
shadow
price
for
collected
firewood.
Valuing
the
firewood
that
ICS
users
save
compared
to
traditional
stove
users
shows
that
the
savings
amount
to
2.03
US$
per
month.
Even
with
a
lower
shadow
price
for
collected
firewood,
it
is
obvious
already
at
this
stage
that
the
benefits
of
ICSs
outweigh
the
costs
by
far
over
its
life
span.
If
health
benefits
and
the
reduction
in
cooking
duration
were
taken
into
account,
the
benefits
would
be
even
greater.
Similarly,
bene-
fits
would
turn
out
to
be
larger
when
social
costs
were
additionally
included,
i.e.
forest
degradation,
village
air
pollution,
and
carbon
emissions.
As
a
consequence,
upscaling
the
intervention
seems
to
be
economically
sensible.
However,
some
challenges
for
external
validity
of
the
RCT
need
to
be
considered
when
transferring
the
results
to
an
upscaled
intervention
or
to
other
regions.
In
Appendix
D,
we
discuss
the
aspects
raised
by
Duflo
et
al.
(2008a):
general
equilibrium
effects,
13 Within
the
complete
investigation
period
of
three
and
a
half
years,
the
ICS
was
destroyed
in
two
cases,
once
because
of
heavy
rainfall
and
once
because
the
kitchen
wall
collapsed.
In
four
cases,
the
ICS
was
stolen.
Hawthorne
and
John
Henry
effects
as
well
as
possible
limitations
to
generalizations
beyond
our
specific
intervention
and
sample.
Over-
all,
the
external
validity
of
this
RCT
is
quite
high.
In
particular,
the
fact
that
our
field
experiment
was
implemented
in
an
unobtrusive
way
enables
us
to
transfer
the
findings
to
a
non-experimental
set-
up.
In
terms
of
transferability
of
the
high
take-up
rates,
the
severe
firewood
scarcity
in
our
study
area
may
have
increased
the
incen-
tives
for
households
to
effectively
use
the
ICS.
Take-up
in
more
biomass-abundant
regions
could
hence
be
lower.
Another
driver
of
high
usage
rates
is
the
fact
that
we
are
working
with
a
type
of
ICS
that
is
adapted
to
the
rural
conditions
not
only
in
our
study
area
but
also
beyond.
In
sum,
if
permanent
access
to
ICS
is
ensured
and
provided
that
the
ICS
is
slightly
modified
in
response
to
potentially
different
cook-
ing
habits
elsewhere
(e.g.
pot
sizes
or
cooking
fuel),
our
findings
are
transferable
to
different
populations
in
(Western)
Africa.
4.
Discussion
and
conclusion
In
this
paper
we
evaluated
take-up
behaviour
and
impacts
of
improved
cooking
stoves
(ICSs)
in
rural
Senegal
by
means
of
a
ran-
domized
controlled
trial
(RCT).
ICSs
are
widely
seen
as
an
option
for
developing
countries
to
combat
the
devastating
effects
of
wood-
fuel
usage
for
cooking
purposes
on
people’s
health,
work
load
as
well
as
the
environment.
The
first
finding
is
that
ICS
take-up
was
close
to
100%
among
the
randomly
assigned
households
and
that
people
only
cease
to
use
the
ICS
if
it
deteriorates.
This
sustain-
ably
high
take-up
rate
comes
as
a
surprise,
since
it
is
often
argued
among
development
practitioners
that
people
would
not
use
ICSs
for
which
they
have
not
paid.
It
also
constitutes
a
major
difference
to
the
findings
in
Hanna
et
al.
(2012).
Major
reasons
for
this
are
probably
differences
in
how
convenient
and
advantageous
the
ICS
technology
is
from
the
household
perspective
and
to
which
degree
the
ICS
has
a
better
performance
than
the
existing
stove
portfo-
lio.
First,
the
ICS
used
in
our
study
is
maybe
closer
to
the
regular
cooking
habits
of
the
target
population.
It
is
easier
to
use,
does
not
require
any
particular
maintenance
and
due
to
its
portability
households
can
decide
themselves
where
to
cook.
Second,
wood
scarcity
is
probably
higher
in
our
study
area
thereby
increasing
the
relevance
of
an
ICS.
Third,
more
than
a
fourth
of
the
households
in
the
study
in
India
already
also
used
cleaner
fuels
like
electricity
and
gas
before
the
randomization
so
that
the
randomized
ICS
did
not
necessarily
represent
an
improvement
for
them.
The
firewood
savings
were
found
to
be
statistically
significant
and
substantial.
They
amount
to
around
30%
per
week
in
the
most
likely
scenario
where
households
have
one
ICS
and
continue
to
use
traditional
stoves
complementarily.
If
these
complementar-
ily
used
traditional
stoves
were
also
replaced
by
ICSs,
the
savings
could
increase
further
up
to
around
40%.
Such
a
reduction
in
fire-
wood
consumption
is
an
important
impact
in
an
arid
country
like
Senegal,
where
forests
are
permanently
under
pressure
and
fire-
wood
provision
is
a
daily
hardship
for
rural
women.
Moreover,
the
CO2that
is
sequestered
in
both
dead
wood
and
green
wood
is
set
free
with
obvious
implications
for
climate
change
processes.
Deforestation
and
forest
degradation
are
in
fact
a
relevant
source
of
global
CO2emissions.
IPCC
(2013)
estimates
that
net
land-use
change,
mainly
deforestation,
is
responsible
for
about
10%
of
the
total
anthropogenic
CO2emissions.
To
the
extent
woodfuel
usage
contributes
to
these
processes,
dissemination
of
ICS
as
used
in
this
study
can
help
to
reduce
such
losses
of
carbon
sinks.
We
also
observe
a
reduction
in
firewood
collection
time,
but
this
is
only
borderline
significant.
Furthermore,
we
find
that
cooking
duration
is
decreased
significantly
by
over
20%.
In
addition,
the
cooking
process
is
facilitated
so
that
the
time
the
cook
needs
to
be

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
55
in
direct
proximity
to
the
cooking
spot
is
reduced.
Together
with
an
increase
in
outdoor
cooking,
this
leads
to
an
evident
reduction
in
exposure
to
harmful
smoke.
Consequently,
we
also
find
a
clear
indication
of
a
decrease
in
respiratory
disease
symptoms
and
eye
problems,
with
a
drop
of
around
9
percentage
points
each
for
the
women
responsible
for
cooking.
Our
self-reported
health
outcomes
might
of
course
feed
criti-
cism
that
objective
indicators
such
as
individual
particulate
matter
exposure
as
measured
in
the
RESPIRE
study
deliver
more
accurate
information.
Apart
from
the
high
costs
of
executing
such
a
sur-
vey,
there
is
also
a
trade-off
between
the
increased
accuracy
and
a
Hawthorne
effect.
Study
participants
can
be
expected
to
behave
differently
if
they
are
asked
to
wear
exposure
monitoring
tools
for
24
h,
for
example.
Hence,
self-reported
and
objective
measure-
ments
can
rather
be
seen
as
complements.
In
addition,
one
might
suspect
an
auspices
or
courtesy
bias
in
our
data
where
respon-
dents
express
their
gratitude
for
having
received
the
ICS
or
expect
additional
benefits
from
a
satisfied
implementing
agency.
In
their
stove
study
in
Ghana,
Burwen
and
Levine
(2012)
suspect
that
this
effect
biases
their
results,
since
the
positive
effects
on
self-reported
health
they
observe
are
not
plausible
given
that
smoke
exposure
is
not
reduced.
However,
this
bias
is
not
likely
in
the
present
case,
since
participating
households
were
not
aware
of
the
study’s
focus
on
ICSs.
Even
if
some
households
noticed
the
role
the
ICS
played
in
this
study,
they
were
unlikely
to
relate
its
usage
to
health
outcomes.
The
fact
that
we
did
not
observe
any
health
effect
among
house-
hold
members
not
responsible
for
cooking
strongly
underpins
this
view.
Hence,
different
from
the
Burwen
and
Levine
(2012)
study,
placebo
outcome
indicators
corroborated
our
findings.
Finally,
the
magnitude
of
observed
savings
is
in
the
range
of
what
is
expected
based
on
laboratory
tests
and,
thus,
does
not
feed
the
suspicion
of
biased
responses.
Altogether,
the
substantial
and
statistically
significant
impacts
on
different
levels
of
indicators
including
positive
external
effects
such
as
reduced
deforestation
and
household
air
pollution
substan-
tiate
the
efforts
that
the
international
community
dedicates
to
the
dissemination
of
ICSs.
The
findings
on
the
health
level
fit
into
the
concept
of
intensive
and
extensive
margins
of
behaviour
that
has
a
longer
tradition
in
agricultural
economics
(Feder
et
al.,
1985)
and
has
recently
been
brought
into
the
debate
on
public
health-relevant
behaviour
in
developing
countries
(see
Dupas,
2011).
The
present
analysis
suggests
that
not
only
the
extensive
margin
of
cooking
should
be
addressed
by
disseminating
cleaner
stoves,
but
also
the
intensive
margin
by,
for
instance,
raising
awareness
of
the
need
to
reduce
smoke
exposure.
This
behavioural
dimension
should
also
be
taken
into
account
by
the
Global
Alliance
for
Clean
Cookstoves
and
the
United
Nations
in
outlining
future
policies
to
increase
access
to
improved
or
clean
cooking
stoves.
Even
ICSs
that
still
emit
con-
siderable
amounts
of
smoke
might
trigger
positive
health
effects
if
they
also
induce
exposure-relevant
behavioural
changes.
The
almost
universal
take-up
among
randomly
assigned
ICS
owners
suggests
that
if
they
have
an
easy
opportunity
to
obtain
an
ICS
that
is
adapted
to
local
cooking
habits
people
also
use
it.
A
sim-
ple
back-of-the-envelope
cost–benefit
calculation
further
made
it
clear
that
investing
in
an
ICS
would
be
a
profitable
investment
from
the
point
of
view
of
the
individual
households.
The
inter-
play
of
cash
and
credit
constraints,
the
lack
of
information,
and
the
fact
that
in
many
cases
the
women
responsible
for
cooking
do
not
manage
the
household
budget,
all
this
however
raises
doubts
about
whether
households
would
be
able
and
willing
to
pay
the
market
price
for
ICSs,
even
if
the
stoves
were
readily
available
on
the
market.
The
experience
from
long-standing
pilot
dissemination
activities
in
neighbouring
rural
areas
in
Senegal
seems
to
support
the
presumption
that
the
majority
of
rural
households
would
prob-
ably
stick
to
the
cheaper
traditional
three-stone
or
metal
stoves.14
As
the
strategy
of
promoting
the
creation
of
sustainable
ICS
mar-
kets
has
already
proven
to
be
difficult
in
urban
areas,
where
fuels
are
purchased
and
ICS
benefits
are
clearly
monetary
ones,
it
can
be
expected
to
require
even
more
efforts
and
resources
in
rural
areas.
In
combination,
the
high
take-up
and
the
positive
external
effects
of
ICS
usage
observed
in
this
study
would
suggest
that
more
direct
options
of
ICS
promotion
should
be
reconsidered.
This
could
mean,
for
example,
directly
subsidizing
the
production
of
ICSs
in
rural
areas
so
that
end-user
prices
can
compete
with
traditional
stoves.
If
the
findings
can
be
confirmed
in
other
rural
areas,
it
might
even
be
an
option
to
distribute
ICSs
directly
to
the
households,
either
for
free
or
at
a
very
low,
symbolic
price.
While
this
would
be
in
contrast
to
the
strategies
pursued
by
most
ICS
dissemination
programmes,
and
many
practitioners
are
opposed
to
a
free
distribu-
tion
policy,
the
empirical
literature
provides
evidence
from
other
field
experiments
that
supports
the
idea.
Paying
a
positive
price
does
not
necessarily
lead
to
higher
usage
rates
of
health-relevant
goods
(Cohen
and
Dupas,
2010;
Tarozzi
et
al.,
2014),
charging
cost-
sharing
prices
substantially
reduce
take-up
(Kremer
and
Miguel,
2007)
and
there
is
only
weak
evidence
yet
that
price
serves
to
allo-
cate
the
health-relevant
goods
to
those
with
the
most
need
(Okeke
et
al.,
2013).
Any
ICS
promotion
policy
has
to
be
designed
in
close
coop-
eration
with
local
stakeholders,
putting
particular
effort
into
the
choice
of
technically
and
culturally
appropriate
ICS
models.
Insti-
tutions
have
to
be
created
to
sustain
the
distribution
of
direct
subsidies
for
the
ICSs,
thereby
avoiding
the
flash-in-the-pan
effect
that
has
been
observed
in
unsuccessful
earlier
ICS
subsidization
programmes.
As
these
recommendations
can
only
be
an
interim
conclusion,
further
research
on
the
take-up
behaviour
and
on
the
impacts
of
ICS
usage
has
to
follow
up
in
other
regions
and
potentially
other
seasons
as
well.
The
indication
of
positive
health
effects
of
the
simpler
ICS
used
in
this
RCT
calls
for
taking
into
account
cooking
behaviour
in
these
studies.
As
evidenced
by
the
lower
take-up
of
ICSs
in
the
Hanna
et
al.
(2012)
study
in
India,
the
results
may
vary
in
different
environments
and
if
other
ICS
types
are
used.
In
addi-
tion,
further
experimental
studies
should
examine
the
mechanisms
behind
take-up
behaviour,
such
as
the
households’
willingness-to-
pay
for
ICSs,
but
also
the
role
of
credit
constraints,
information,
and
woodfuel
scarcity.
Such
research
efforts
can
substantiate
–
or
contradict
–
the
findings
in
this
study
and
will
thereby
help
to
decide
under
which
circumstances
and
to
which
degree
sub-
sidies
might
in
fact
be
required
to
encourage
rural
people
to
obtain
ICSs.
14 See
also
Miller
and
Mobarak
(2013)
for
evidence
on
low
purchase
rates
of
ICS
in
Bangladesh.

56
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
Appendix
A.
Technical
features
of
improved
cookstove
used
in
different
studies
Study
reference
Stove
type/model
name
Combustion
chamber
type
Fuel
type
Feed
type
Chimney
Portability
Approx.
cost
(US$)
Further
stove
references
This
study
Jambaar
wood
Ceramic
Wood
Continuous
No
Yes
10
GIZ
(2011a)
Bensch
and
Peters
(2013)
Jambaar
charcoal
Ceramic
Charcoal
Batch
fed
No
Yes
9–19
GIZ
(2011b)
Burwen
and
Levine
(2012)
Council
of
Scientific
and
Industrial
Research
(CSIR)
improved
stove
Mud
Wood
Continuous
Yes
No
<10
–
Hanna
et
al.
(2012)
Appropriate
Rural
Technology
Institute
(ARTI)
improved
stove
Mud
Wood
Continuous
Yes
No
12.5
–
Masera
et
al.
(2007)
Patsari
stove
Mud/brick
Wood
Continuous
Yes
No
35
Kshirsagar
and
Kalamkar
(2014)
Miller
and
Mobarak
(2013)
Bangladesh
Council
of
Scientific
and
Industrial
Research
(BCSIR)
“efficiency”
(E)
and
“chimney”
(C)
stove
Clay
Wood
Continuous
E:
no
C:
Yes
E:
Yes
C:
No
E:
$5.8
C:
$10.9
Mobarak
et
al.
(2012)
RESPIRE
Plancha
mejorada
Brick
Wood
Continuous
Yes
No
100–150
Díaz
(2008)
Notes:
All
listed
stoves
are
direct
combustion
stoves
with
natural
draft.
Further
main
(and
more
advanced)
combustion
types
are
gasifier
and
rocket
type
direct
combustion;
forced
draft
is
an
alternative
to
natural
draft.
In
addition,
the
combustion
chamber
may
be
metallic.
Appendix
B.
Stove
types
used
in
the
survey
area
Stove
type/model
name Combustion
chamber
type Fuel
type Feed
type Chimney
Portability
Approx.
cost
(US$)
Three-stone
stoves
None
Biomass
Continuous
No
Yes
–
Os
None
Biomass
Continuous
No
Yes
1–2
Cire
khatach
Metal
Crop
residues
Batch
fed
No
Yes
3–5
Cire
wood
Metal
Wood
Continuous
No
Yes
3–5
Malagasy
stove
Metal
Charcoal
(wood)
Continuous
No
Yes
3–5
Jambaar
Wood
Ceramic
Wood
Continuous
No
Yes
10

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
57
Appendix
C.
Power
calculation
Since
information
on
our
primary
impact
variable,
firewood
consumption,
was
not
available
in
existing
data
sets
for
the
target
region
of
our
study,
we
took
data
collected
in
the
quasi-
experimental
study
presented
in
Bensch
and
Peters
(2013)
from
urban
Senegal
to
approximate
the
relevant
parameters
(prospec-
tive
power
analysis).
After
the
follow-up
survey,
we
verified
these
parameters
by
rerunning
the
analysis
with
the
actual
baseline
data
for
those
households
included
in
the
analysis
(retrospective
power
analysis).The
sample
size
n
is
given
by
the
following
formula:
n
=
D[(Z˛+
Zˇ)21
r(sd2
1+
sd2
2)
(X2−
X1)2]
Table
C1
provides
the
description,
the
values
and
the
sources
of
the
different
parameters.
The
decisive
parameter
to
be
defined
by
the
researcher
is
the
minimum
detectable
effect
size
(ES),
which
reflects
the
smallest
relative
reduction
in
woodfuel
consumption
that
we
are
able
to
detect
at
the
given
significance
level
(see
Bloom,
1995).
While
the
CCT
suggest
an
effect
size
of
40%,
we
chose
a
minimum
detectable
effect
size
of
30%
in
order
to
account
for
the
possibility
of
an
overestimated
effect
size
in
the
CCT.
We
defined
the
probability
of
being
assigned
to
the
control
group
to
be
60%
and
that
for
the
ICS
treatment
group
to
be
40%.
Taking
these
parameters
into
account,
we
obtain
a
required
sample
size
of
around
200
households,
as
is
indicated
in
the
last
row
of
the
column
for
the
prospective
analysis
in
Table
C1.
In
order
to
account
for
the
sensitivity
of
the
different
parameters
in
the
power
calculation
and
potential
attrition
or
non-compliance,
we
built
in
a
cushion
and
increased
the
number
of
households
to
be
interviewed
to
250.
With
respect
to
health
and
time
savings
impacts,
the
sample
size
required
to
measure
significant
effects
tends
to
be
substantially
higher.
The
reason
is
that
the
effect
on
respiratory
diseases,
for
example,
can
be
expected
to
be
less
pronounced.
The
implication
of
this
is
that
the
power
of
our
study
is
not
necessarily
sufficient
to
detect
all
relevant
health
and
time
savings
effects.
Appendix
D.
External
validity
External
validity
prevails
if
a
study’s
findings
can
be
transferred
from
the
study
population
to
the
policy
population.
In
other
words,
external
validity
is
concerned
with
whether
findings
obtained
from
a
small
sample
group
represent
the
wider
population
in
real
world
situations.
In
the
following,
we
discuss
how
our
RCT
design
took
into
account
the
three
dimensions
of
external
validity
as
defined
by
Duflo
et
al.
(2008a):
general
equilibrium
effects,
Hawthorne
and
John
Henry
effects
as
well
as
possible
limitations
to
generalizations
beyond
our
specific
intervention
and
beyond
our
sample.
General
equilibrium
effects
may
occur
in
the
present
case
if
widespread
ICS
usage
leads
to
a
sizable
reduction
in
firewood
demand
and,
in
turn,
to
a
reduction
in
the
costs
of
firewood
pro-
vision,
either
because
prices
decrease
or
because
firewood
is
less
scarce
and
easier
to
collect.
This
might
induce
households
to
con-
sume
more
of
the
now
cheaper
fuel.
Although
this
would
bring
welfare
benefits
such
as
more
hot
meals,
from
a
public
health
and
resource
saving
perspective
this
might
be
considered
an
adverse
second-round
effect.
Since
most
households
in
rural
Senegal
col-
lect
firewood
and
do
not
buy
it,
this
effect
can
be
expected
to
be
less
pronounced
than
for
market-based
energy
sources.
Another
major
risk
to
the
external
validity
of
RCT
results
is
if
par-
ticipants
change
their
behaviour
because
they
know
that
they
are
participating
in
an
experiment
or
are
somehow
under
observation.
While
so-called
Hawthorne
effects
(if
treatment
group
members
change
their
behaviour)
or
John
Henry
effects
(if
control
group
mem-
bers
change
their
behaviour)
can
never
be
ruled
out
completely,
we
reduced
the
risk
considerably
through
various
precautionary
measures:
first,
we
embedded
the
interviews
in
a
baseline
survey
for
an
electrification
intervention
under
preparation
in
the
studied
areas
(the
intervention
was
not
implemented
in
any
of
the
sampled
villages
before
the
end
of
this
study).
The
applied
questionnaire
covered
a
comprehensive
set
of
socio-economic
and
energy-related
dimensions
such
as
electricity
so
that
attention
was
not
focused
primarily
on
cooking-related
parts
of
the
interviews.
Second,
the
lottery
was
framed
as
a
reward
for
all
households
to
recompense
Table
C1
Table
C1
Parameters
for
power
calculation.
Description
Value
Source
Prospective
Retrospective
D
=
1
+
(m
+
1)
with
Design
effect,
accounting
for
the
loss
of
variation
in
the
data
if
clustered
instead
of
simple
random
sampling
is
used
1.59
2.25
Household
data†

Intra-cluster
correlation,
i.e.
the
proportion
of
the
overall
variance
with
respect
to
firewood
consumption
explained
by
within-village
(cluster)
variance
in
the
data
0.031
0.069
Household
data
m
Mean
number
of
interviewed
households
per
cluster
(village)
20
229/12
=
19.1
Defined
Z˛Critical
value
(Z-score)
for
a
given
level
of
confidence
˛
reflecting
the
probability
that
the
null
hypothesis
is
rejected
given
that
it
is
in
fact
true
1.96
(˛
=
5%)
1.96
(˛
=
5%)
Defined
(conventional)
ZˇZ-score
for
a
given
level
of
confidence
ˇ
reflecting
the
probability
that
the
null
hypothesis
is
rejected
given
that
it
is
in
fact
false
0.84
(ˇ
=
80%)
0.84
(ˇ
=
80%)
Defined
(conventional)
R
Ratio
of
treatment
and
control
observations
(ICS
owners
to
non-owners)
0.66
90/139
=
0.65
Lottery
outcome
defined
in
sampling
design
sd1Standard
deviation
of
firewood
consumption
of
ICS
non-owners
0.266
0.259
Household
data
sd2Standard
deviation
of
firewood
consumption
of
ICS
owners
0.186
0.181
Implicitly
defined
through
minimum
detectable
effect
size
(see
below)
X1Per
capita
firewood
consumption
of
ICS
non-owners
(in
kg)
0.384
0.411
Household
data
X2Expected
per
capita
firewood
consumption
of
ICS
owners
(in
kg)
0.269
0.288
Implicitly
defined
through
minimum
detectable
effect
size
(see
below)
ES
=
|X2–X1|/X1Minimum
detectable
effect
size
30%
30%
Defined
based
on
experiences
with
laboratory
tests
n
=
n
(ICS
owners)
+
n
(non-owners)
Result
of
power
calculation:
required
minimum
sample
sizes
for
treatment
and
control
group
192
=
76
+
116
229
=
90
+
139
†Household
data
refers
to
the
data
from
the
urban
quasi-experimental
study
(“prospective”)
and
to
the
baseline
data
from
the
present
study
(“retrospective”)
to
corroborate
the
calculations
of
the
prospective
analysis.

58
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
them
for
participation
in
the
electrification
baseline
survey,
a
sim-
ilar
procedure
as
applied
by
De
Mel
et
al.
(2008)
in
an
RCT
on
business
grants
among
micro-enterprises
in
Sri
Lanka.
Third,
all
survey
activities
were
conducted
in
an
unobtrusive
way
by
local
interviewers
and
community
workers.15
According
to
Duflo
et
al.
(2008a),
three
problems
may
hamper
a
valid
generalization
beyond
the
specific
programme
and
sample.
First,
it
may
be
that
the
particular
care
with
which
the
random-
ized
treatment
was
implemented
makes
it
difficult
to
upscale
the
intervention.
As
outlined
in
Section
2.4
and
the
instructions
given
to
participants
(see
Appendix
E)
we
keep
to
what
real-world
users
are
told
about
the
randomized
ICS.
Furthermore,
we
conducted
the
study
together
with
the
Government
of
Senegal
and
GIZ
and
thereby
mimicked
a
typical
ICS
dissemination
intervention.
Sec-
ond,
the
question
arises
as
to
whether
we
can
transfer
the
results
to
a
slightly
modified
intervention.
Here,
the
fact
that
we
dis-
tributed
the
ICS
for
free
deserves
some
attention
as
usage
behaviour
might
change
if
households
need
to
pay
for
the
ICS.
If
a
change
can
be
suspected
when
households
with
sufficient
willingness-to-
pay
self-select
into
the
treatment,
then
most
practitioners
would
expect
an
intensification
of
usage
and,
thus,
also
impacts.
Yet,
usage
intensity
is
already
high
so
that
no
substantial
increase
can
be
expected.
The
third
point
is
the
particularity
of
the
study
popu-
lation.
The
most
important
characteristics
here
are
the
fuels
used
for
cooking
and
their
availability.
Firewood
is
the
dominant
cooking
fuel
in
our
sample
as
it
is
in
major
parts
of
rural
Sub-Saharan
Africa.
98%
of
sample
households
use
firewood
as
their
primary
cooking
fuel.
The
national
average
for
rural
Senegal
is
slightly
lower
at
89%
(ANSD,
2006).
Across
Sub-Saharan
Africa
this
value
amounts
to
87%
(UNDP/WHO,
2009).16 The
reason
for
these
slightly
lower
numbers
of
firewood
usage
is
that
the
national
rural
averages
include
peri-
urban
areas,
where
charcoal
is
also
used.
Firewood
usage
patterns
in
rural
Africa
excluding
peri-urban
areas
will
be
very
much
the
same
as
in
our
sample
for
the
vast
majority
of
countries.
Firewood
availability
in
our
study
area
is
typical
for
large
parts
of
interior
Western
Africa
and
dry
savannah
regions
in
general.
All
the
households
in
our
target
area
use
firewood,
which
is
the
case
in
virtually
all
rural
areas
in
Africa.
Take-up
rates
and
consequently
impacts
might
change,
though,
in
regions
in
which
firewood
is
more
abundantly
available
(e.g.
the
southern
region
of
Senegal)
or
in
which
cleaner
fuels
are
already
available
such
as
in
urban
Africa
or
in
parts
of
rural
Asia
(see
Hanna
et
al.,
2012
for
an
example).
Appendix
E.
Experimental
design
E.1.
Design
of
the
field
experiment
The
original
design
of
the
experiment
was
drafted
in
an
incep-
tion
report
for
the
Independent
Evaluation
Unit
of
Deutsche
Gesellschaft
für
Internationale
Zusammenarbeit
(GIZ)
and
finalized
on
August
20,
2009.
It
was
concretized
during
an
in-country
prepa-
ration
mission
between
October
12
and
22,
2009
and
is
outlined
in
Section
2.4
of
this
paper
(‘RCT
design
and
implementation’).
E.2.
Selection
and
eligibility
of
participants
We
selected
twelve
villages
from
the
target
region
of
a
planned
GIZ
rural
electrification
intervention
in
Foundiougne
District
that
are
far
away
from
GIZ-supported
producers
of
improved
cooking
15 See
Zwane
et
al.
(2011)
for
an
examination
of
how
being
surveyed
might
affect
response
behaviour.
The
authors
generally
call
for
an
unobtrusive
method
of
data
collection.
16 See
Bonjour
et
al.
(2013)
for
individual
country
estimates.
Fig.
E1.
Location
of
survey
sites.
stoves
(see
Fig.
E1
and
Table
E1).
Within
the
villages,
all
households
were
eligible.
They
were
randomly
sampled
and
none
of
the
sam-
pled
households
refused
to
participate
in
the
RCT
(see
also
Fig.
E2).
E.3.
Instructions
given
to
participants
On
the
day
of
the
ICS
distribution,
households
were
reminded
via
phone
in
order
to
make
sure
the
person
responsible
for
cooking
in
the
household
was
present.
A
local
staff
member
with
several
years
of
experience
in
ICS
usage
training
had
a
meeting
with
those
women
who
had
received
an
ICS.
In
the
local
language
Wolof,
he
presented
the
ICS
as
a
fuel-saving
device
and
briefly
informed
about
convenience
co-benefits:
a
quicker
cooking
process,
less
smoke,
and
a
cleaner
kitchen.
He
verbally
informed
the
women
about
the
functioning
of
the
stove
and
the
proper
utilization.
He
explained
to
them
that
the
clay
inlay
of
the
ICS
serves
the
purpose
of
storing
the
heat
and
that
it
could
easily
break
if
the
embers
were
doused
with
water;
instead
they
were
told
to
put
the
fire
out
with
sand
on
the
ground.
Moreover,
unlike
with
open
fires
for
which
people
typically
use
entire
branches
or
even
trunks,
the
firewood
has
to
be
chopped
in
order
to
fit
the
fuel
feed
entrance
of
the
ICS.
He
advised
them
not
to
use
pot
sizes
that
are
too
big
for
the
stove
and
not
to
move
the
pot
when
it
is
placed
on
the
stove.
Households
were
also
given
a
leaflet
summarizing
these
instructions
(Fig.
E3).
This
is
all
regular
information
that
is
also
provided
by
ICS
traders
in
a
non-RCT
setup.
In
addition,
in
order
to
avoid
ICS
misuse
the
women
were
also
asked
not
to
share
the
ICS
with
other
households
or
lend
it
to
other
women.
From
a
methodological
point
of
view,
this
was
also
intended
to
avoid
treatment
contamination.
Table
E1
Table
E1
List
of
survey
sites.
Village
Rural
community
Pethie
Djilor
Keur
Mandao
Djilor
Ndoffane
Ndarry
Djilor
Keur
Omar
Djilor
Goudeme
Sidy
Djilor
Darou
Keur
Mor
Khoredi
Diossong
Thiamene
Ndiagnene
Diossong
Simong
Bambara
Nioro
A.
Tall
Ndiayene
Kad
Nioro
A.
Tall
Keur
Bacar
Santhie
Keur
S.
Diané
Keur
Maniane
Keur
S.
Diané
Nema
Bah
Toubacouta

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
59
Fig.
E2.
Participant
flow.

60
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
Fig.
E3.
Leaflet
provided
to
households
that
received
an
ICS.

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
61
Appendix
F.
Additional
estimation
results
Table
F1
Table
F1
ATT
results
for
household
level
indicators
on
firewood
consumption,
time
expenditures,
and
health.
Difference
in
means
Regression-adjusted
difference
in
means
(se)
p-Value
(H0:
Diff
=
0)
Mean
(se)
p-Value
(H0:
Diff
=
0)
(1)
(2)
(3)
(4)
Firewood
consumption
per
week
(kg) 27.74
(6.53) 0.00*** 27.64
(5.94) 0.00***
Duration
of
firewood
collection
per
week
(min) 153((102.9) 0.14 140((96.2) 0.15
Cooking
duration
per
day
(min) 84((22.1) 0.00*** 77((20.3)
0.00***
Respiratory
system
disease
(%)
Any
woman
responsible
for
cooking 9.1 0.05*9.2
0.05**
Any
male −1.0 0.77 −1.3 0.71
Any
woman
not
responsible
for
cooking 2.7 0.39 3.1
0.32
Eye
problems
(%)
Any
woman
responsible
for
cooking 9.9
0.02** 10.0
0.01**
Any
male 2.5 0.45
2.5
0.44
Any
woman
not
responsible
for
cooking
−1.5
0.70
−0.8
0.82
Note:
All
computations
are
performed
with
heteroskedasticity
corrected
standard
errors
accounting
for
heterogeneity
in
treatment
responses
and
include
village
dummies;
se
–
standard
error.
*Significance
level
of
10%.
** Sgnificance
level
of
5%.
*** Significance
level
of
1%.
Table
F2
Table
F2
Outlier
analysis
for
household
and
dish
level
indicators.
Outlier
analysis
using
median
regressions
Outlier
analysis
using
outlier
exclusion
Difference
in
means
Regression-adjusted
difference
in
means
Difference
in
means
Regression-adjusted
difference
in
means
(se)
p-Value
(H0:
Diff
=
0)
Mean
(se)
p-Value
(H0:
Diff
=
0)
(se)
p-Value
(H0:
Diff
=
0)
Mean
(se)
p-Value
(H0:
Diff
=
0)
(1)
(2)
(3)
(4)
Firewood
consumption
per
week
(kg) 26.50
(4.63) 0.00*** 26.51
(3.51) 0.00*** 19.15
(4.41)
0.00*** 18.53
(4.45)
0.00****
Firewood
weight
per
dish
(kg) 1.54
(0.19)
0.00*** 1.77
(0.13)
0.00*** 1.50
(0.12)
0.00*** 1.61
(0.13)
0.00***
Duration
of
firewood
collection
per
week
(min) 60(58)
0.30
76(66)
0.25
123(66)
0.06*117(62)
0.06*
Cooking
duration
per
day
(min)
68(21.6)
0.00*** 77((16.0)
0.00*** 40((17.1)
0.02** 40((17.2)
0.02**
Note:
Median
regressions
are
quantile
regressions
that
determine
the
median
of
the
dependent
variable
conditional
on
the
values
of
the
independent
variables.
For
outlier
exclusion,
outliers
are
defined
as
values
more
than
two
standard
deviations
away
from
the
mean.
All
values
are
computed
using
robust
standard
errors;
se
–
standard
error.
*Significance
level
of
10%.
** Significance
level
of
5%.
*** Significance
level
of
1%.
Table
F3
Table
F3
Probit
regression
on
health
status
of
household
members.
Estimator:
Coefficient
(Standard
Error
in
parentheses)
Probit,
ITT
Dependent
variable:
Household
member
with
respiratory
system
disease
Household
member
with
eye
problem
Variable
(1)
(2)
(3)
(4)
ICS
dummy
−0.11
(0.16)
−0.08
(0.16)
−0.06
(0.17)
−0.02
(0.15)
Household
member
is
responsible
for
cooking
0.85*** (0.18)
0.87*** (0.16)
0.68*** (0.16)
0.77*** (0.15)
Household
member
is
responsible
for
cooking
×
ICS
dummy
−0.41
(0.28)
−0.43
(0.28)
−0.62** (0.29)
−0.59** (0.28)
Further
household
member
variables
Household
member’s
sex
0.03
(0.17)
0.30*(0.16)
Household
member’s
age
0.01** (0.00)
0.02*** (0.00)
Household
variables
Average
number
of
people
cooked
for
(in
terms
of
the
logarithm
of
adult
equivalents)
−0.34** (0.16)
−0.17
(0.17)
Father
has
formal
education
−0.13
(0.17)
−0.26
(0.25)
Mother
has
formal
education
0.19
(0.12)
0.20
(0.14)
Household
income
(in
logarithmic
terms)
0.01
(0.04)
0.01
(0.06)
Telecommunication
expenditures
(in
logarithmic
terms)
0.04*(0.02)
0.00
(0.02)
Bank
account
ownership
−0.50
(0.44)
−0.52
(0.38)
Wall
material
of
house
is
stone
or
brick
−0.09
(0.14)
−0.30** (0.15)
Ownership
of
sheep
0.04
(0.13)
−0.08
(0.14)
Association
membership
of
the
mother
0.08
(0.12)
−0.09
(0.14)
Village
dummies
Included
Included
Included
Included
Constant
−2.02*** (0.58)
−2.07*** (0.20)
−2.41*** (0.77)
−2.14*** (0.22)
Number
of
observations
1977
1977
1977
1977
p-Value
of
interaction
term
0.143
0.119
0.031
0.037
Pseudo
R-squared
0.131
0.103
0.176
0.090
Note:
The
household
control
variable
flooring
material
is
not
included,
since
it
predicts
failure
perfectly
in
the
estimated
probit
regressions;
standard
errors
are
clustered
by
household.
*Significance
level
of
10%.
** Significance
level
of
5%.
*** Significance
level
of
1%.

62
G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
References
Adeloye,
D.,
Chan,
K.Y.,
Rudan,
I.,
Campbell,
H.,
2013.
An
estimate
of
asthma
preva-
lence
in
Africa:
a
systematic
analysis.
Croatian
Medical
Journal
54
(6),
519–531.
ANSD
(Agence
Nationale
de
la
Statistique
et
de
la
Démographie),
2006.
Résul-
tats
du
troisième
recensement
général
de
la
population
et
de
l’habitat
(2002):
Rapport
National
de
présentation,
http://www.ansd.sn/publications/rapports
enquetes
etudes/enquetes/RGPH3
RAP
NAT.pdf
(last
accessed
03.03.10).
ANSD
(Agence
Nationale
de
la
Statistique
et
de
la
Démographie),
ICF
International,
2012.
Enquête
Démographique
et
de
Santé
à
Indicateurs
Multiples
au
Sénégal
(EDS-MICS)
2010–2011.
ANSD
and
ICF
International,
Calverton,
MD,
USA.
Armstrong,
J.R.,
Campbell,
H.,
1991.
Indoor
air
pollution
exposure
and
lower
respiratory
infections
in
young
Gambian
children.
International
Journal
of
Epi-
demiology
20
(2),
424–429.
Barnes,
D.F.,
Openshaw,
K.,
Smith,
K.,
Van
der
Plas,
R.,
1994.
What
Makes
People
Cook
with
Improved
Biomass
Stoves?
A
Comparative
International
Review
of
Stove
Programs,
World
Bank
Technical
Paper
No.
242.
World
Bank.
Bensch,
G.,
Peters,
J.,
2013.
Alleviating
deforestation
pressures?
Impacts
of
improved
stove
dissemination
on
charcoal
consumption
in
urban
Senegal.
Land
Economics
89
(4),
676–698.
Bloom,
H.,
1995.
Minimum
detectable
effect
size
–
a
simple
way
to
report
the
sta-
tistical
power
of
experimental
designs.
Evaluation
Review
19
(5),
547–556.
Bonjour,
S.,
Adair-Rohani,
H.,
Wolf,
J.,
Bruce,
N.G.,
Mehta,
S.,
Prüss-Ustün,
A.,
Lahiff,
M.,
Rehfuess,
E.A.,
Mishra,
V.,
Smith,
K.R.,
2013.
Solid
fuel
use
for
household
cooking:
country
and
regional
estimates
for
1980–2010.
Environmental
Health
Perspectives
121
(7),
784–790.
Bruhn,
M.,
McKenzie,
D.,
2009.
In
pursuit
of
balance:
randomization
in
practice
in
development
field
experiments.
American
Economic
Journal:
Applied
Eco-
nomics
1
(4),
200–232.
Burnett,
R.T.,
Pope
III,
C.A.,
Ezzati,
M.,
Olives,
C.,
Lim,
S.S.,
Mehta,
S.,
Shin,
H.H.,
Singh,
G.,
Hubbell,
B.,
Brauer,
M.,
Anderson,
H.R.,
Smith,
K.R.,
Balmes,
J.R.,
Bruce,
N.G.,
Kan,
H.,
Laden,
F.,
Prüss-Ustün,
A.,
Turner,
M.C.,
Gapstur,
S.M.,
Diver,
W.R.,
Cohen,
A.,
2014.
An
integrated
risk
function
for
estimating
the
global
burden
of
disease
attributable
to
ambient
fine
particulate
matter
exposure.
Environmental
Health
Perspectives
122
(4),
397–403.
Burwen,
J.,
Levine,
D.E.,
2012.
A
rapid
assessment
randomized-controlled
trial
of
improved
cookstoves
in
rural
Ghana.
Energy
for
Sustainable
Development
16
(3),
328–338.
Butrick,
E.,
Peabody,
J.,
Solon,
O.,
DeSalvo,
K.,
Quimbo,
S.,
2010.
A
compari-
son
of
objective
biomarkers
with
a
subjective
health
status
measure
among
children
in
the
Philippines.
Asia-Pacific
Journal
of
Public
Health
24
(4),
565–576.
Campbell,
H.,
Armstrong,
J.R.,
Byass,
P.,
1989.
Indoor
air
pollution
in
developing
countries
and
acute
respiratory
infection
in
children.
Lancet
1,
1012.
Cohen,
J.,
Dupas,
P.,
2010.
Free
distribution
or
cost-sharing?
Evidence
from
a
ran-
domized
Malaria
prevention
experiment.
Quarterly
Journal
of
Economics
125
(1),
1–45.
Das,
J.,
Hammer,
J.,
Sánchez-Paramo,
C.,
2012.
The
impact
of
recall
periods
on
reported
morbidity
and
health
seeking
behaviour.
Journal
of
Development
Eco-
nomics
98
(1),
76–88.
De
Mel,
S.,
McKenzie,
D.,
Woodruff,
C.,
2008.
Returns
to
capital
in
micro-enterprises:
evidence
from
a
field-experiment.
Quarterly
Journal
of
Economics
123
(4),
1329–1372.
Dherani,
M.,
Pope,
D.,
Mascarenhas,
M.,
Smith,
K.R.,
Weber,
M.,
2008.
Indoor
air
pollution
from
unprocessed
solid
fuel
use
and
pneumonia
risk
in
children
aged
under
5
years:
a
systematic
review
and
meta-analysis.
Bulletin
of
the
World
Health
Organization
86
(5),
390–398.
Díaz,
E.,
Smith-Sivertsen,
T.,
Pope,
D.,
Lie,
R.,
Diaz,
A.,
McCracken,
J.,
Arana,
B.,
Smith,
K.,
Bruce,
N.,
2007.
Eye
discomfort,
headache
and
back
pain
among
Mayan
Guatemalan
women
taking
part
in
a
randomized
stove
intervention
trial.
Journal
of
Epidemiology
&
Community
Health
61
(1),
74–79.
Díaz,
E.,
(Ph.D.
dissertation)
2008.
Impact
of
Reducing
Indoor
Air
Pollution
on
Women’s
Health.
RESPIRE
Guatemala-Randomised
Exposure
Study
of
Pollu-
tion
Indoors
and
Respiratory
Effects.
Department
of
Public
Health
and
Primary
Health
Care,
The
University
of
Bergen.
Duflo,
E.,
Glennerster,
R.,
Kremer,
M.,
2008a.
Using
randomization
in
development
economics
research:
a
toolkit.
In:
Schultz,
P.,
Strauss,
J.
(Eds.),
Handbook
of
Development
Economics.
North
Holland,
Amsterdam,
pp.
3895–3962.
Duflo,
E.,
Greenstone,
M.,
Hanna,
R.,
2008b.
Indoor
air
pollution,
health
and
economic
well-being.
Sapiens
Journal
1
(1),
1–9.
Dupas,
P.,
2011.
Do
teenagers
respond
to
HIV
risk
information?
Evidence
from
a
field
experiment
in
Kenya.
American
Economic
Journal:
Applied
Economics
3
(1),
1–34.
Ezzati,
M.,
Kammen,
D.M.,
2001.
Indoor
air
pollution
from
biomass
combustion
and
acute
respiratory
infections
in
Kenya:
an
exposure–response
study.
The
Lancet
358,
619–624.
Ezzati,
M.,
Kammen,
D.M.,
2002.
Household
energy,
indoor
air
pollution,
and
health
in
developing
countries:
knowledge
base
for
effective
interventions.
Annual
Review
of
Environment
and
Resources
27,
233–270.
Feder,
G.,
Just,
R.E.,
Zilberman,
D.,
1985.
Adoption
of
agricultural
innovations
in
developing
countries:
a
survey.
Economic
Development
and
Cultural
Change
33
(2),
255–298.
FAO
(Food
and
Agriculture
Organization
of
the
United
Nations),
2005a.
State
of
the
World’s
Forests.
Food
and
Agriculture
Organization
of
the
United
Nations,
Rome.
FAO
(Food
and
Agriculture
Organization
of
the
United
Nations,
2005b.
Global
Forest
Resources
Assessment.
Food
and
Agriculture
Organization
of
the
United
Nations,
Rome.
Frondel,
M.,
Peters,
J.,
Vance,
C.,
2008.
Identifying
the
rebound:
evidence
from
a
German
household
panel.
Energy
Journal
29
(4),
154–163.
GAIN
(Global
Agricultural
Information
Network),
2011.
Senegal,
Grain
and
Feed
Annual.
West
Africa
Rice
Annual,
http://gain.fas.usda.gov/Recent%20GAIN%
20Publications/Grain%20and%20Feed%20Annual
Dakar
Senegal
5-6-2011.pdf
(last
accessed
08.02.12).
GIZ
(Gesellschaft
für
Internationale
Zusammenarbeit),
2011a.
Firewood
Jam-
bar
Stove,
Senegal,
https://energypedia.info/wiki/File:GIZ HERA
2011
Jambar
Bois
Senegal.pdf
(last
accessed
25.11.14).
GIZ
(Gesellschaft
für
Internationale
Zusammenarbeit),
2011b.
Charcoal
Jam-
bar
Stove,
Benin,
Kenya,
Senegal.
https://energypedia.info/images/b/b1/
GIZ HERA 2011 Jambar Charbon
Senegal.pdf (last
accessed
25.11.14).
Hanna,
R.,
Duflo,
E.,
Greenstone,
M.,
2012.
Up
in
Smoke:
The
Influence
of
Household
Behavior
on
the
Long-run
Impact
of
Improved
Cooking
Stoves,
CEEPR
WP
2012-
008.
MIT
Center
for
Energy
and
Environmental
Policy
Research.
Herring,
H.,
Sorrell,
S.,
Elliott,
D.,
2009.
Energy
Efficiency
and
Sustainable
Consump-
tion
–
The
Rebound
Effect.
Palgrave
Macmillan,
New
York.
Idler,
E.,
Benyamini,
Y.,
1997.
Self-assessed
health
and
mortality:
a
review
of
twenty-
seven
community
studies.
Journal
of
Health
and
Social
Behavior
38
(1),
21–37.
IPCC
(Intergovernmental
Panel
on
Climate
Change),
2013.
Climate
Change
2013:
The
Physical
Science
Basis.
Contribution
of
Working
Group
I
to
the
Fifth
Assess-
ment
Report
of
the
Intergovernmental
Panel
on
Climate
Change.
Cambridge
University
Press,
Cambridge/New
York.
Kamali,
A.,
Quigley,
M.,
Nakiyingi,
J.,
Kinsman,
J.,
Kengeya
Kayondo,
J.,
Gopal,
R.,
Ojwiya,
A.,
Hughes,
P.,
Carpenter,
L.M.,
Whitworth,
J.,
2003.
Syndromic
manage-
ment
of
sexually-transmitted
infections
and
behaviour
change
interventions
on
transmission
of
HIV-1
in
rural
Uganda:
a
community
randomised
trial.
The
Lancet
361
(9358),
645–652.
Kan,
X.,
Chiang,
C.Y.,
Enarson,
D.A.,
Chen,
W.,
Yang,
J.,
Chen,
G.,
2011.
Indoor
solid
fuel
use
and
tuberculosis
in
China:
a
matched
case–control
study.
BMC
Public
Health
11
(1),
1–7.
Kjellsson,
G.,
Clarke,
P.,
Gerdtham,
U.-G.,
2014.
Forgetting
to
remember
or
remem-
bering
to
forget:
A
study
of
the
recall
period
length
in
health
care
survey
questions.
Journal
of
Health
Economics
35,
34–46.
Kremer,
M.,
Miguel,
E.,
2007.
The
illusion
of
sustainability.
Quarterly
Journal
of
Economics
122
(3),
1007–1065.
Kremer,
M.,
Miguel,
E.,
Mullainathan,
S.,
Null,
C.,
Zwane,
A.P.,
2009.
Making
Water
Safe:
Price,
Persuasion,
Peers,
Promoters,
or
Product
Design?
Mimeo.
Kshirsagar,
M.P.,
Kalamkar,
V.R.,
2014.
A
comprehensive
review
on
biomass
cook-
stoves
and
a
systematic
approach
for
modern
cookstove
design.
Renewable
&
Sustainable
Energy
Reviews
30,
580–603.
Lewis,
J.J.,
Pattanayak,
S.K.,
2012.
Who
adopts
improved
fuels
and
cookstoves?
A
systematic
review.
Environmental
Health
Perspectives
120
(5),
637–645.
Luby,
S.,
Mendoza,
C.,
Keswick,
B.,
Chiller,
T.M.,
Hoekstra,
R.,
2008.
Difficulties
in
bringing
point-of-use
water
treatment
to
scale
in
rural
Guatemala.
The
Ameri-
can
Journal
of
Tropical
Medicine
and
Hygiene
78
(3),
382–387.
Martin
II,
W.J.,
Glass,
R.I.,
Balbus,
J.M.,
Collins,
F.S.,
2011.
A
major
environmental
cause
of
death.
Science
334
(6053),
180–181.
Masera,
O.,
Edwards,
R.,
Armendariz,
C.,
Berrueta,
V.,
Johnson,
M.,
Rojas
Bracho,
L.,
Riojas-Rodríguez,
H.,
Smith,
K.R.,
2007.
Impact
of
Patsari
improved
cookstoves
on
indoor
air
quality
in
Michoacán,
Mexico.
Energy
for
Sustainable
Development
11
(2),
45–56.
McCracken,
J.,
Smith,
K.,
Stone,
P.,
Díaz,
A.,
Arana,
B.,
Schwartz,
J.,
2011.
Intervention
to
lower
household
wood
smoke
exposure
in
Guatemala
reduces
ST-segment
depression
on
electrocardiograms.
Environmental
Health
Perspectives
119
(11),
1562–1568.
Miilunpalo,
S.,
Vuori,
I.,
Oja,
P.,
Pasanen,
M.,
Urponen,
H.,
1997.
Self-rated
health
status
as
a
health
measure:
the
predictive
value
of
self-reported
health
status
on
the
use
of
physician
services
and
on
mortality
in
the
working-age
population.
Journal
of
Clinical
Epidemiology
50
(5),
517–528.
Miller,
G.,
Mobarak,
M.,
2013.
Gender
Differences
in
Preferences,
Intra-household
Externalities,
and
Low
Demand
for
Improved
Cookstoves,
NBER
Working
Paper
18964.
National
Bureau
of
Economic
Research.
Mobarak,
A.M.,
Dwivedi,
P.,
Bailis,
R.,
Hildemann,
L.,
Miller,
G.,
2012.
Low
demand
for
nontraditional
cookstove
technologies.
Proceedings
of
the
National
Academy
of
Sciences
of
the
United
States
of
America
109
(27),
10815–10820,
http://dx.doi.org/10.1073/pnas.1115571109.
Mueller,
V.,
Pfaff,
A.,
Peabody,
J.,
Liu,
Y.,
Smith,
K.R.,
2013.
Improving
stove
evalu-
ation
using
survey
data:
who
received
which
intervention
matters.
Ecological
Economics
93,
301–312.
Okeke,
E.N.,
Adepiti,
C.A.,
Ajenifuja,
K.O.,
2013.
What
is
the
price
of
prevention?
New
evidence
from
a
field
experiment.
Journal
of
Health
Economics
32
(1),
207–218.
Pandey,
M.R.,
1984a.
Prevalence
of
chronic
bronchitis
in
a
rural
community
of
the
hill
region
of
Nepal.
Thorax
39,
331–336.
Pandey,
M.R.,
1984b.
Domestic
smoke
pollution
and
chronic
bronchitis
in
a
rural
community
of
hill
region
of
Nepal.
Thorax
39,
337–339.
Pandey,
M.R.,
Smith,
K.R.,
Boleij,
J.S.M.,
Wafula,
E.M.,
1989.
Indoor
air
pollution
in
developing
countries
and
acute
respiratory
infection
in
children.
The
Lancet
1,
427–429.
Pattanayak,
S.,
Pfaff,
A.,
2009.
Behavior,
environment,
and
health
in
developing
countries:
evaluation
and
valuation.
Annual
Review
of
Resource
Economics
1,
183–217.

G.
Bensch,
J.
Peters
/
Journal
of
Health
Economics
42
(2015)
44–63
63
Peabody,
J.W.,
Nordyke,
R.J.,
Tozija,
F.,
Luck,
J.,
Munoz,
J.A.,
Sunderland,
A.,
DeSalvo,
K.,
Ponce,
N.,
McCulloch,
C.,
2006.
Quality
of
care
and
its
impact
on
population
health:
a
cross-sectional
study
from
Macedonia.
Social
Science
&
Medicine
62
(9),
2216–2224.
Pitt,
M.M.,
Rosenzweig,
M.R.,
Hassan,
M.N.,
2006.
Sharing
the
Burden
of
Disease:
Gender,
the
Household
Division
of
Labor
and
the
Health
Effects
of
Indoor
Air
Pollution.
Mimeo.
Pope,
C.A.,
Burnett,
R.T.,
Turner,
M.C.,
Cohen,
A.,
Krewski,
D.,
Krewski,
D.,
Jerrett,
M.,
Gapstur,
S.M.,
Thun,
M.J.,
2011.
Lung
cancer
and
cardiovascular
disease
mortality
associated
with
ambient
air
pollution
and
cigarette
smoke:
shape
of
the
exposure–response
relationships.
Environmental
Health
Perspectives
119,
1616–1621.
Rehfuess,
E.A.,
Puzzolo,
E.,
Stanistreet,
D.,
Pope,
D.,
Bruce,
N.,
2014.
Enablers
and
barriers
to
large-scale
uptake
of
improved
solid
fuel
stoves:
a
systematic
review.
Environmental
Health
Perspectives
122
(2),
120–130.
Shindell,
D.,
Kuylenstierna,
J.C.I.,
Vignati,
E.,
van
Dingenen,
R.,
Amann,
M.,
Klimont,
Z.,
Anenberg,
S.C.,
Muller,
N.,
Janssens-Maenhout,
G.,
Raes,
F.,
Schwartz,
J.,
Falu-
vegi,
G.,
Pozzoli,
L.,
Kupiainen,
K.,
Höglund-Isaksson,
L.,
Emberson,
L.,
Streets,
D.,
Ramanathan,
V.,
Hicks,
K.,
Kim
Oanh,
N.T.,
Milly,
G.,
Williams,
M.,
Demkine,
V.,
Fowler,
D.,
2012.
Simultaneously
mitigating
near-term
climate
change
and
improving
human
health
and
food
security.
Science
335,
183–189.
Smith,
K.R.,
McCracken,
J.P.,
Weber,
M.W.,
Hubbard,
A.,
Jenny,
A.,
Thompson,
L.M.,
Balmes,
J.,
Díaz,
A.,
Arana,
B.,
Bruce,
N.,
2011.
Effect
of
reduction
in
household
air
pollution
on
childhood
pneumonia
in
Guatemala
(RESPIRE):
a
randomised
controlled
trial.
The
Lancet
378
(9804),
1717–1726.
Smith-Sivertsen,
T.,
Díaz,
E.,
Pope,
D.,
Lie,
R.T.,
Díaz,
A.,
McCracken,
J.P.,
Bakke,
P.,
Arana,
B.,
Smith,
K.R.,
Bruce,
N.,
2009.
Effect
of
reducing
indoor
air
pollution
on
women’s
respiratory
symptoms
and
lung
function:
the
RESPIRE
randomized
trial,
Guatemala.
American
Journal
of
Epidemiology
170
(2),
211–220.
Smith-Sivertsen,
T.,
Díaz,
E.,
Bruce,
N.,
Díaz,
A.,
Khalakdina,
A.,
Schei,
M.A.,
McCracken,
J.P.,
Arana,
B.,
Klein,
R.,
Thompson,
L.M.,
Smith,
K.R.,
2004.
Reducing
indoor
air
pollution
with
a
randomized
intervention
design
–
a
presentation
of
the
stove
intervention
study
in
the
Guatemalan
highlands.
Norsk
Epidemiologi
14
(2),
137–143.
Tappan,
G.,
Sall,
M.,
Wood,
E.C.,
Cushing,
M.,
2004.
Ecoregions
and
land
cover
trends
in
Senegal.
Journal
of
Arid
Environments
59
(3),
427–462.
Tarozzi,
A.,
Mahajan,
A.,
Blackburn,
B.,
Kopf,
D.,
Krisham,
L.,
Yoong,
J.,
2014.
Micro-
loans,
insecticide-treated
bednets,
and
malaria:
evidence
from
a
randomized
controlled
trial
in
Orissa,
India.
American
Economic
Review
104
(7),
1909–1941.
UNDP/WHO
(United
Nations
Development
Programme
and
World
Health
Organi-
zation),
2009.
The
Energy
Access
Situation
in
Developing
Countries
–
A
Review
Focused
on
the
Least
Developed
Countries
and
Sub-Saharan
Africa.
United
Nations
Development
Programme,
New
York.
van
Gemert,
F.,
van
der
Molen,
T.,
Jones,
R.,
Chavannes,
N.,
2011.
The
impact
of
asthma
and
COPD
in
sub-Saharan
Africa.
Primary
Care
Respiratory
Journal
20
(3),
240–248.
WEC/FAO
(World
Energy
Council
and
Food
and
Agriculture
Organization
of
the
United
Nations),
1999.
The
Challenge
of
Rural
Energy
Poverty
in
Developing
Countries.
World
Energy
Council,
London.
Wendland,
K.J.,
Pattanayak,
S.K.,
Sills,
E.O.,
2015.
National-level
differences
in
the
adoption
of
environmental
health
technologies:
a
cross-border
comparison
from
Benin
and
Togo.
Health
Policy
Planning
30
(2),
145–154.
WHO
(World
Health
Organisation),
2014.
Burden
of
Disease
from
House-
hold
Air
Pollution
for
2012,
http://www.who.int/phe/health
topics/outdoorair/
databases/FINAL
HAP
AAP
BoD
24March2014.pdf
(last
accessed
26.04.14).
WHO
(World
Health
Organization),
2009.
Country
Profile
of
Environmental
Burden
of
Disease
–
Senegal.
World
Health
Organisation,
Geneva.
World
Bank,
2011.
Household
Cookstoves,
Environment,
Health,
and
Climate
Change.
A
New
Look
on
an
Old
Problem.
World
Bank,
Washington.
Yu,
F.,
2011.
Indoor
air
pollution
and
children’s
health:
net
benefits
from
stove
and
behavioral
interventions
in
rural
China.
Environmental
and
Resource
Economics
50
(4),
495–514.
Zwane,
A.P.,
Zinman,
J.,
Van
Dusen,
E.,
Pariente,
W.,
Null,
C.,
Miguel,
E.,
Kremer,
M.,
Karlan,
D.S.,
Hornbeck,
R.,
Giné,
Y.,
Duflo,
E.,
Devoto,
F.,
Crepon,
B.,
Banerjee,
A.,
2011.
Being
surveyed
can
change
later
behavior
and
related
parameter
esti-
mates.
Proceedings
of
the
National
Academy
of
Sciences
of
the
United
States
of
America
108
(5),
1821–1826.