Content uploaded by Christopher O Olopade
Author content
All content in this area was uploaded by Christopher O Olopade on Jan 25, 2018
Content may be subject to copyright.
Sustained usage of bioethanol cookstoves shown in an urban Nigerian
city via new SUMs algorithm
Amanda Northcross
a,b,
⁎,MattShupler
a
, Donee Alexander
e,1
, John Olamijulo
c
, Temitope Ibigbami
c
,
Godson Ana
d
, Oladosu Ojengbede
d
, Christopher O. Olopade
e
a
Department of Environmental and Occupational Health, School of Public Health and Health Services, The George Washington University, USA
b
Global Alliance for CleanCook Stoves, USA
c
Healthy Life for All Foundation, Ibadan, Nigeria
d
University of Ibadan, Ibadan, Nigeria
e
Center for Global Health, University of Chicago, USA
abstractarticle info
Article history:
Received 18 February 2016
Revised 4 April 2016
Accepted 28 May 2016
Available online xxxx
An unbiased assessment of cooking patterns during a cookstove intervention can provide strong evidence for
sustained usage of a cookstove among the target population. A bioethanol cookstove was used as an intervention
within a randomized controlled trial being conducted in Ibadan, Nigeria to assess the ability of a clean stove to
improve birth outcomes. Sustained usage of the intervention was quantified using a newly developed method
of analyzing cooking patterns based on time integrated temperature data from Stove Use Monitors (SUMs)
installed on householdcookstoves. The method accountsfor household levelvariations in ambient temperatures.
We report a significant decline of traditional kerosene stove usage, 84% of women in the Bioethanol arm giving
away their kerosene stove before the conclusion of the study (56% within the first month ofenrollment), suggest-
ing the bioethanol stove replaced the kerosene stove. This is the first study to objectively evaluate a liquid-to-
liquid fuel substitution.
© 2016 Published by Elsevier Inc. on behalf of International Energy Initiative.
Keywords:
Household air pollution
Kerosene
Bioethanol
Cookstove
Introduction
Household air pollution(HAP) is the number one environmentalrisk
factor for death and disability worldwide (Lim et al., 2013), attributing
to over 4 million deaths annually (Smith et al., 2014). HAP exposures
vary greatly between rural and urban areas, especially in low- and
middle-income countries (LMICs) (Martin et al., 2013). While residents
of rural communitiesin LMIC continue to rely on biomass for their daily
cooking needs, those living in urban areas in several developing coun-
tries of Africa, Asia, and Latin America use kerosene frequently as a sub-
stitute (Lam et al., 2012).
The use of kerosene fuel for cookingis a public health concern as ker-
osene cookstoves emit particulate matter (PM), carbon monoxide (CO),
volatile organic compounds (VOC), nitric oxides (NO
x
) and sulfur diox-
ide (SO
2
)(Lam et al., 2012.). Studies have reported that households
using kerosene cookstoves are exposed to kitchen PM concentrations
ranging from 300 to 750 μg/m
3
(Habib et al., 2008; Zhang et al., 2000).
While not as high as traditional biomass combustion, these PM concen-
trations greatly exceed current World Health Organization (WHO)
guidelines.
Lam et al. (2012) summarized previous epidemiological studies of
kerosene used for cooking or lighting, which provided evidence that
kerosene emissions may impair lung function and increase risk of asth-
ma. A hospital-based case–control study conducted among women in
Nepal found that use of a kerosene fueled stove was significantly associ-
ated with 3.36 times the odds of developing tuberculosis (Pokhrel et al.,
2010). Recently,WHO released new health-based air quality guidelines
for household fuel combustion, which discourages the use of kerosene
until further research into its health impacts is conducted.
Ethanol is a clean-burning fuel, comparable to liquefied petroleum
gas (LPG). In one study, it was found to be cleaner-burning than kero-
sene under certain conditions and according to certain measures
(Rajvanshi, 2006). It is similar to LPG in terms of combustion efficiency
and particle emissions. Ethanol may be a viable option as a liquid
Energy for Sustainable Development 35 (2016) 35–40
Abbreviations: HAP, household air poll ution; SUMs, stove use monitors; RCT,
randomized controlled trial; LMICs, low- and middle-income countries; PM, particulate
matter; CO, carbon monoxide; NO
x
, nitrous oxides; SO
2
, sulfur dioxide; LPG, liquefiedpe-
troleum gas;PHCs, primary health centers; IHV, initial home visit; EM, expectation maxi-
mization; UCL, upper confidence limit; SD, standard deviation; IQR, inter quartile range;
ICC, intraclass correlati on coefficient; VOC, volatile organic compounds; WHO, World
Health Organization.
⁎Corresponding author at: Department of Environmental and Occupational Health,
School of Public Health and Health Services, The George Washington University, USA.
Tel.: +1 202 994 3970.
E-mail address: northcross@email.gwu.edu (A. Northcross).
1
DA is currently at the Global Alliance for CleanCook stoves.
http://dx.doi.org/10.1016/j.esd.2016.05.003
0973-0826/© 2016 Published by Elsevier Inc. on behalf of International Energy Initiative.
Contents lists available at ScienceDirect
Energy for Sustainable Development
cooking fuel in Nigeria because it can be produced locally and in a re-
newable manner (Obueh, 2006). Given the health damaging nature of
kerosene, a randomized controlled trial (RCT) aimed at quantifying im-
provements in pregnancy outcomes through reductions in exposures to
HAP from cookstoves was conducted. An ethanol fueled stove named
the CleanCook (Dometic Group, Durban South Africa) was chosen as
the intervention stove in the trial. The CleanCook surpasses WHO
benchmarks for PM
2.5
and International Organization for Standardiza-
tion (ISO) International Workshop Agreement (IWA) Tier 4 standards
for emissions (Berkeley Air Monitoring Group 2012). The CleanCook
received the best rating possible, which is matched only by LPG stoves,
induction stoves, electricity, biogas, and solar-powered stoves.
While a high performingstove is crucial for HAP reduction, its usage
is just as important for improving health outcomes (Johnson and
Chiang, 2015). Efforts to implement improved cooking technologies
(ICTs) have been met with significant challenges. The translation of
high energy efficiency and smoke removal standards from cookstoves
in laboratory testing has not led to consistent, reproducible perfor-
mance in the household. Frequently, when ICTs are used within a
home, ‘stove stacking’results. Stove stacking occurs when individuals
continue to utilize their traditional stove in conjunction with the new
cooking technology they have received. Thus, the potential health
benefits of the ICT are hampered because individuals are still exposed
to levels of HAP above WHO guidelines from continued use of their
traditional cookstove (Johnson and Chiang, 2015).
In order to effectively reduce exposure to HAP and achieve the
greatest health benefit, complete displacement of traditional stoves
with clean cooking technologies must be achieved (Johnson and
Chiang, 2015). Cooking patterns must be closely monitored within the
target population to systematically evaluate if use of a newly introduced
cookstove is consistently maintained, resulting in significant disuse of
the traditional cookstove.
This paper presents an analysis of cooking patterns via stove use
monitor (SUM) data from the CleanCook stoves disseminated in the
RCT in Nigeria. It is quantitatively demonstrated that, in an urban set-
ting, the transition from a kerosene cookstove to an ethanol cookstove
can be achieved with minimal occurrence of stove stacking.
Methods
Study overview
The RCT was conducted in Ibadan, Nigeria, a metropolis of over 3
million people located in Southwest Nigeria. Pregnant women less
than 18 weeks gestational age, who cooked primarily with kerosene
and/or biomass, were recruited from one of five local, primary health
centers (PHCs). Participants were randomized into control and inter-
vention groups. Participants in the control group continued to cook
with their traditional stove. The intervention group participants were
given a bioethanol cookstove called the CleanCook, valued at $60, and
free bioethanol fuel until the delivery of the baby. SUMs were placed
on all cookstoves used in participant homes. Cooking patterns and
stove preferences were monitored throughout their pregnancy using a
combination of the SUMs and interview-administered questionnaires
data regarding cooking habits and daily activities. The data were collect-
ed every two to three weeks during subsequent home visits. At the con-
clusion of their participation in the study, participants are given the
option to purchase bioethanol fuel subsidized to match the current
cost of kerosene.
Temperature readings via SUMs
Thermochron iButtons 1921G (Maxim Integrated Products,
Sunnyvale, CA) were used to monitor the temperature of each
stove and are described in detail elsewhere as SUMs (Ruiz-Mercado
et al., 2012). The SUMs record temperatures to the nearest 0.5 °C and
were programmed to monitor either every 3, 10 or 13 min based on
the length of time between field visits. The SUMs were placed 10 cm
from the center of the kerosene cookstove burner and 14 cm directly
in between the double burner of the CleanCook stove. These distances
were determined pre-trial by defining an optimum length away from
the stove burner that provides sufficient resolution of temperature
fluctuation while not causing the SUMs to overheat and rupture.
Inclusion criteria
While SUMs remained on each cookstove for the entire study dura-
tion, each cookstove did not have SUMs data available for the complete
study duration (detail on SUMs field performance is provided in SI). A
reliability analysis was conducted to determine how many days of
SUMs cookstove monitoring were necessary to be representative of
cookstove use during the entire period of the intervention (Ruiz-
Mercado, 2012). The analysis took into consideration both overall and
monthly days of SUMs data during a participant's enrollment in the
intervention to account for potential variations in cooking occurring at
different months during the pregnancy.
In the reliability analysis, ‘study months’were defined as 30-day pe-
riods, beginning with a participant's entry into the study (established as
the day a participant received her initial home visit), and ending with
the birth of her baby. Because study participants were recruited and
randomized at no later than 18 weeks of gestational age, approximately
five study months of SUMs data was collected from the cookstove(s) of
each participant. Participants with at least eight days of stove monitor-
ing with SUMs in study months one through four, on at least a kerosene
or CleanCook stove, prior to October 1, 2014, were included. Higher var-
iability (also reported in an Indianintervention (Pillarisetti et al., 2014))
coupled with less days of data due to the delivery of the child resulted in
the exclusion of study month five from the analysis presented. More
details about how sufficiency was assessed are provided in SI.
Converting temperature readings to stove usage
All data management and statistical analysis was conducted in
RStudio, version 0.98.507 (R Core Team, 2014). A stove was determined
as in-use when the SUMS temperature was above a threshold tempera-
ture. A unique threshold was determined for each home to account for
ambient temperature variations among the households.
The temperature distribution from each SUM followed a bimodal
distribution with the two peaks occurring at the mean ambient temper-
ature and the mean cooking temperature (Fig. 1).
The ambienttemperature curve was assumed to benormally distrib-
uted due to large number of data points (average per stove = 10,070
data points) for each cookstove. Using the Expectation Maximization
(EM) algorithm via the Mixtools package, version 1.0.2 (Benaglia et al.,
2009), in RStudio, the average and standard deviation of the ambient
temperature was obtained. The 99.9% upper confidence limit (UCL) of
the mean ambient temperature was estimated as the cutoff between a
stove being in and out of use for a particular SUM, creating a unique
threshold temperature for each cookstove (Fig. 1).
The mixed EM algorithm requires a minimum number of data points
with-in both modes of the expected bi-modal distribution. Secondary
stoves (kerosene stoves in the intervention arm) were not used enough
by study participantsfor the mixed EM algorithm to converge and iden-
tify two modes. For this reason, only theprimary cookstove (CleanCook
stove in intervention arm and kerosene cookstove in control arm) were
used to define a temperature threshold for ‘stove in-use’versus ‘stove
nonuse’. This cutoff was applied to all SUMs in the home, regardless of
the type of stove. Additionally, the mixed EM algorithm is only effective
on stove types that heat and cool rapidly such that a distinct dichotomy
of ambient and cooking temperatures is present.
For participants that owned and used more than one kerosene stove,
the kerosene cookstove with the higher usage was deemed the primary
36 A. Northcross et al. / Energy for Sustainable Development 35 (2016) 35–40
cookstove for that household and only one threshold temperature was
used. Higher usage was quantitatively determined by the cookstove
that had a higher cooking temperature mixing proportion as deter-
mined by the EM algorithm. This was the kerosene cookstove that had
a higher proportion of temperature data corresponding to the curve of
cooking temperatures as compared with ambient temperatures. The
variability of threshold temperatures of all kerosene stoves in the con-
trol arm was modeled using random effects (accounting for multiple
kerosene stoves within a control arm home) to confirm that the main
source of variance in ambient temperatures was between homes and
not within homes.
Metrics of cookstove usage
Stove usage was evaluated by length of a cooking event, duration of
cooking per day and number of cooking events per day. Cooking event
lengths were calculated as the number of consecutive temperature
readings above the threshold temperature multiplied by the SUMs log-
ging interval. Cooking events were calculated as the number of discrete
times that a SUM recorded at least one temperature above the temper-
ature cutoff during each 24-h monitoring day. Each event is separated
by a minimum of 10 min from a previous cooking event. SUMs data
from two kerosene stoves belonging to the same participant were
aggregated to effectively evaluate each participant's total daily stove
usage.
Quantification of sustained cookstove usage
A‘stove-day’is defined as a particular day with any amount of stove
usage (Ruiz-Mercado, 2012). Any amount of stove usage on a SUMs
monitoring-day is defined as having at least one cooking event regis-
tered to a SUM, regardless of the length of that single cooking event.
The number of stove-days per study month was counted for eachpartic-
ipant to assess the prevalence of stove stacking within the intervention
arm.
Pre-stove dissemination SUMs
Kerosene stove usage among intervention arm participants prior to
stove-dissemination was compared to their CleanCook stove usage dur-
ing the intervention to observe potential changes in cooking patterns
furnished by introduction of the bioethanol stove. SUMs data from the
same time period for control arm participants was leveraged to assess
the reliability of cooking patterns during this time as a reflection of
cooking patterns during the intervention (Pillarisetti et al., 2014).
Statistical analysis
Mixed effects linear regression was conducted to account for varia-
tions both between household and within a participant's household to
determine differences in average cooking length between the primary
cookstoves. Mixed effects Poisson regression was used to compare the
rates of cooking events/day between primary cookstoves. Mixed effects
logistic regression was used to assess differences in the proportion of
days during the intervention that the primary cookstoves were used
at least once daily by study participants. The same analyses were
conducted to compare traditional kerosene stove usage pre-stove
dissemination to ethanol stove usage during the intervention, among
intervention arm participants. In all regression models involving
cooking length estimation, only similar SUMs sampling intervals
(three, ten, thirteen) were compared, and unless otherwise noted, all
cooking length regression was done with 13 min monitoring, as this
was the most prevalent sampling interval during our intervention (see
SI for more information). In analyses of cooking events per day, all
SUMs data was aggregated. Study month was controlled for (when
applicable).
Results
Study population
The study population included in the analysis consists of 50 partici-
pants (25 intervention; 25 control) with an average of 71 SUMs
monitoring-days (SD = 8, range = 47–88) during enrollment months
one through four. The 50 participants had a total of 3564 SUMs
monitoring-days from their study entry through either the last day of
their fourth enrollment month or through October 1, 2014, whichever
came first. These 50 pregnant Nigerian women used kerosene cook-
stoves exclusively prior to randomization. Descriptive statistics are
provided in Table 1.Table 1 shows that there were no significant differ-
ences in measured baseline characteristics, including participant-
estimated average monthly amount of money spent on kerosene fuel
and estimated monthly amount of kerosene fuel used.
Table 1 shows that there were no significantdifferences in measured
baseline characteristics, including participant-estimated average
monthly amount of money spent on kerosene fuel and estimated
monthly amount of kerosene fuel used.
Variation in ambient temperatures
The individual household threshold temperature used to determine
if a stove wasin use ranged by nine degrees Celsius(range =30.6–39.1,
IQR = 33.4–35.0, mean = 34.2, SD = 1.5) across the entire study
population. Within a home very little variation in temperature was
observed. An intraclass correlation coefficient (ICC) of 80% was found
in the random effects model of temperature thresholds from all
kerosene cookstoves within a home.
Consistency of stove usage among participants
Fig. 2 illustrates that the percent of ‘stove days’of the CleanCook
stove consistently remained at approximately 90% throughout months
1–4 of participants' duration in the intervention, similar to the percent
of stove days for kerosene stoves among control arm participants.
Based on logistic regression analysis, there was no significant difference
(p = 0.9) between thepercentage of stove dayswith kerosene stove use
among control arm participants and the percentage of stove days with
CleanCook stove use, controlling for each study month. Using two
Fig. 1. Distribution of temperatures recorded to SUM placed on CleanCo ok stove from
participant HAP12-184). Reflective of 106 SUMs monitoring-days and 12,238
temperatures collected over the course of the intervention.
37A. Northcross et al. / Energy for Sustainable Development 35 (2016) 35–40
weeks of pre-stove dissemination SUMs data, no significant difference
(p = 0.2) was found in percent stove days of kerosene usage between
both study arms.
In the intervention arm, average percent of stove days for the kero-
sene stove was 85% prior to receiving the CleanCook stove. After thedis-
semination of the CleanCook, the average percent stove days for
kerosene cookstoves dropped to approximately 25% in the first month
and at or below 5% for subsequent months.
Among the 25 intervention arm participants, 21 (84%) gave their
stove(s) away during the course of the intervention and used the
CleanCook stove exclusively for the remainder of the study period. Of
these 21 participants, 14 gave their stove(s) away within the first
month of receiving the CleanCook stove, four participants during the
fifth month of the study, close to child birth, and three in study months
2–4.
As it is difficult to pinpoint the exact date the participants gave their
stoves away, the day of stove removal from the household was estimat-
ed as the last day SUMs data was downloaded from the removed kero-
sene stove. Once the kerosene stove was given away (or both kerosene
stoves given away, if the participant had two), every subsequent study
month was assumed to have zero stove-days for that kerosene stove.
Stove usage analysis
Controlling for study month, CleanCook users cooked an average of
17 min less per day than kerosene users; however, the difference was
not significant (p = 0.15). Kerosene stoves were in use for an average
of 131 min per day (95% CI = [114, 151]), while average duration of
CleanCook stove use was 114 min per day (95% CI = [99, 130]). No sta-
tistical differences were found in the number of cooking events per day
between the CleanCook and kerosene groups, after controlling for study
month; p = 0.5. Over the entire study period, the average number of
cooking events per day was 1.84 (95% CI = [1.65, 2.19]) for kerosene
stove users and 2.05 (95% CI = [1.78, 2.36]) for CleanCook stove users.
Cooking lengths were log transformed to meet the assumption of
normality. The average length of a cooking event on kerosene stoves
was 56 min (95% CI = [52, 61]), with the CleanCook stove being used
an average of 45 min (95% CI = [42, 49]). The eleven minute difference
was statistically significant; p = 0.001. The average cooking length in
the second and third study months were not statistically different
from that of the first study month (p = 0.9 & p = 0.2, respectively)
within both study arms. However, in the fourth month of the study,
there was a significantly higher increase in average length of a cooking
event among the control arm participants compared to the intervention
group (p
interaction
= 0.003). The average cooking length on kerosene
stoves in study month four was 62 min (95% CI = [58, 67]) and
46 min (95% CI = [43, 50]) on CleanCook stoves.
Pre-stove dissemination stove usage analyses
Twenty-three of the 25 intervention participants had at least seven
pre-stove dissemination SUMs monitoring-days on their kerosene
cookstove. The 23 kerosene cookstoves had between 12 and 14 days
of monitoring, combining to a total of 303 pre-CleanCook stove dissem-
ination SUMs monitoring-days in the analysis. When comparing the
number cooking events per day on kerosene stoves used by interven-
tion arm participants, prior to receiving the CleanCook stove, to the
number of cooking events per day for CleanCook, there was no statisti-
cal difference (p = 0.2). Kerosene cookstoves were used an average of
1.97 times per day (95% CI = [1.68, 2.32]) by intervention arm partici-
pants, pre-stove dissemination, compared to an average of 2.05 events
per day (95% CI = [1.78, 2.36]) on CleanCook stoves.
No cooking event length analysis was completed due to the low ker-
osene cookstove usage among intervention arm participants. Seventeen
of the 25 control arm participants had at least 7 days of SUMs monitor-
ing pre-randomization (range = 11–14), for a total of 219 SUMs
monitoring-days. The average number of cooking events per day was
1.83 (95% CI = [1.49, 2.24]) pre-stove dissemination compared to 1.84
events per day (95% CI = [1.65, 2.19]) during the trial. There was no sta-
tistical difference (p = 0.9). Cooking events per day compared between
control and intervention groups pre-stove dissemination were not
statistically significantly different (p = 0.4).
Discussion
Sustained use of the CleanCook bioethanol cookstoves
We report sustained use during pregnancy of an ethanol fueled
cookstove intervention within a cohort of Nigerian women living in an
urban setting, with access to kerosene fuel. Unlike other studies which
have included a cookstove intervention, we report minimal stove stack-
ing. The average percent stove-days of the kerosene stoves in the inter-
vention arm decreased to 5% by the second month of the subjects'
participation in the study. The high usage of the bioethanol stove sug-
gests that the stove met the needs of the participants to complete
cooking tasks. Comparing the percent of stove days, as well as the
number of cooking events the amount of stove usage was not statistical-
ly different between pre-and post-stove dissemination periods (Fig. 2),
suggesting that the bioethanol stove did not alter cooking patterns. The
bioethanol stove replaced the kerosene stove as shown by 85% of the
participants in the intervention arm removing their kerosene cookstove
from their household and adopting the CleanCook stove exclusively for
all of their cooking tasks.
The conclusion that the bioethanol stove replaced the cooking tasks
previously completed on a kerosene stove is also supported by the lack
of statistical difference of the percent stove-days, average duration of
cooking per day and average number of cooking events per day when
comparing the intervention and control study groups. Table 1 shows
no significant difference in participant-estimated average monthly
amount of money spent on kerosene fuel and estimated monthly
amount of kerosene fuel used, highlighting the initial similarity of
cooking fuel use.
Sustained use of the CleanCook stove may be attributed to similari-
ties between the keroseneand CleanCook stove.Both stoves utilize a liq-
uid fuel, require one fueling event prior to stove usage (not continuous
feeding), allow for modulation of temperatures, have similar size
Table 1
Characteristics of Nigerian women in HAP study.
Characteristic Intervention
(n = 25)
Control
(n = 25)
Age (years) (mean (SD)) 27 (5) 28 (5)
Number of children prior to pregnancy (count (%))
0–1 9 (36) 7 (28)
≥2 16 (64) 18 (72)
Education level (count (%))
Junior secondary or lower 8 (32) 9 (36)
Senior secondary or higher 17 (68) 16 (64)
Occupation (count (%))
a
Trader 12 (48) 19 (73)
Tailor 5 (20) 3 (12)
Other 9 (36) 4 (15)
PHC recruited from (count (%))
Agbongbon 15 (60) 15 (60)
Oranyan 10 (40) 10 (40)
Participant estimated monthly kerosene usage at
baseline (liters) (mean (SD))
14 (9) 17 (10)
Participant estimated monthly kerosene expenditure
at baseline (Naira) (mean (SD))
1950 (1195) 2204 (1400)
Fisher's exact test used for categorical data; t-test used for age; Wilcoxon test used for all
other continuous data .
All p-values N0.1.
a
Some participants have more than one occupation.
38 A. Northcross et al. / Energy for Sustainable Development 35 (2016) 35–40
burners, are portable and do not require electricity to operate. These
similarities may have limited changes to the behavioral rhythm of
cooking when the CleanCook stove was introduced. Behaviors of food
preparation and meal cooking may be modified when improved bio-
mass stoves require changes in the timing of feeding or processing of
fuel compared to the traditional biomass stove, possibly accounting
for the higher levels of stove stacking reported in previous studies
(Pillarisetti et al., 2014; Bailis et al., 2007).
Potential impacts on exposure
A decrease in daily cooking time may lead to reductions in HAP ex-
posures, and possibly increase the time that women have for other
daily activities. The average of 17 min less time spent cooking per day
on Cleancook stoves versus kerosene cookstoves over the four months
studied may be due to the bioethanol stove having two burners com-
pared to the single burner on the kerosene stoves (three participants
did have two kerosene stoves). Cumulatively, there is almost two
hours per week less cooking time among bioethanol stove users com-
pared to the control group. The duration of daily cooking may be a pos-
sible proxy for personal exposures to household air pollution and more
health-relevant statistic and a better measure of potential risk other
proxy measures such as cooking fuel type, or kitchen concentrations
(Pillarisetti et al., 2014).
SUMs algorithm to assess liquid fueled cookstove usage
To our knowledge, the algorithm presented is the first to account for
household level variation in ambient temperatures within a study set-
ting by establishing a uniquethreshold temperature for each participant
with data collected from a single SUMs. The newly developed algorithm
used in this analysis is cost-effective as it only requires temperature
measurements captured from the SUMs deployed on cookstoves; no
additional ambient temperature measurements or SUMs are needed.
Developing within home temperature thresholds also reduces the
need to correct for the differences in measured temperatures between
SUMSs. Previously, deviations from local ambient temperature data
(Pillarisetti et al., 2014) and time–temperature slope thresholds (Ruiz-
Mercado et al., 2013) have been used in interventions in India and
Guatemala, respectively. The individual threshold temperatures
covered a range of 9 °C in this population. Use of a single ambient
temperature to identify a single threshold temperature for the entire
study population would result with an inaccurate estimate of cooking
length, as possible number of cooking events.
The SUMs algorithm presented is applicable to the liquid fueled
cookstoves used within the study, as the temperature rises and declines
rapidly when the cookstove is turned on and off, respectively. The
algorithm can be used with any stove with similar heating and cooling
properties such as LPG. The only limitation is the requirement for a
significant amount of ambient data to be collected. The EM algorithm
requires the ambient temperature to have a normal distribution there
for this algorithm may not work for short period of SUMs temperature
data collection.
Study limitations
Providing participants with fuel during the length of the study may
have resulted in higher usage rates compared to a natural experiment
where women are solely given the stove and required to purchase
their own fuel (Ruiz-Mercado et al., 2011). However, 76% of the partic-
ipants have continued to purchase ethanol fuel and use the CleanCook
stove since the conclusion of their participation in the study. This
suggests that financial incentives, while potentially a factor in theinitial
uptake of the CleanCook stove, participants are satisfied with the
CleanCook stove.
There may be important seasonal trends in cooking patterns in
Nigeria due to characteristics such as religion and occupation of the par-
ticipants. The analysis in this paper only focused on cooking patterns
over the course of pregnancy, and not on calendar months, due to the
study's ongoing recruitment. The sample size limited the power for de-
tecting seasonal changes in cooking patterns. However, randomization
evenly distributed measured population characteristics between the
two study arms, theoretically minimizing potential residual bias intro-
duced by seasonal cooking differences.
Author contributions
ALN contributed to the study design, training of field team for data
collection, design of data collection instruments, statistical analysis
and writing of the manuscript. MS conducted statistical analysis and
contributed to writing of the manuscript. DA contributed to coordina-
tion of data collection, and writing of the manuscript. JO and TI
Fig. 2. (Left) The average percent stove days (95% CI) for each study month among intervention arm participants. (Right) The average percent stove days (95% CI) for kerosene stoves
among control arm participants. Pre-stove dissemination data only includes 40 participants having at least seven SUMs monitoring days. Pre-stove dissemination percent stove days
point estimates are representative of approximately two weeks of data.
39A. Northcross et al. / Energy for Sustainable Development 35 (2016) 35–40
contributed to data collection, coordination and manuscript editing. GA
and OO contributed to study design, field work supervision and edito-
rials. COO contributed to study design, manuscript writing and editing
and had oversight over study design and administration.
Funding
Partial funding was provided by the Global Alliance for CleanCook
Stoves, (UNF-12-378) The George Washington University Milken Insti-
tute School of Public Health, Project Gaia, USA and the Shell Foundation.
Acknowledgments
The authors thank the participants in Ibadan, Nigeria for the gen-
erosity of their time and cooperation. We thank the Healthy Life for
All field staff, as well as a host of study collaborators and colleagues
at the University of Ibadan, University of Chicago, and The George
Washington University.
Appendix A. Supplementary data
Additional tables; methodology detailing sufficiency of SUMs
monitoring-days; intraclass correlation coefficients from linear mixed
models; accounting for different SUMs sampling intervals; sensitivity
analysis of Expectation Maximization method. Supplementary data as-
sociated with this article can be found in the online version, at http://
dx.doi.org/10.1016/j.esd.2016.05.003.
References
Bailis R, Berrueta V, Chengappa C, Dutta K, Edwards R, Masera O, et al. Performance test-
ing for monitoring improved biomass stove interventions: experiences of the House-
hold Energy and Health Project1. Energy Sustain Dev 2007;11(2):57–70.
Benaglia T, Chauveau D, Hunter D, Young D. Mixtools: an r package for analyzing finite
mixture models. J Stat Softw 2009;32(6):1–29.
Habib G, Venkataraman C,Bond TC, Schauer JJ. Chemical, microphysical and optical prop-
erties of primary particles from the combustion of biomass fuels. Environ Sci Technol
2008;42(23):8829–34.
Johnson MA, Chiang RA. Quantitative gu idance for stove usage and performance to
achieve health and environmental targets. Environ Health Perspect 2015;123(8):
820–6.
Lam NL, Smith KR, Gauthier A, Bates MN. Kerosene: a review of household uses and their
hazards in low- and middle-income countries; 2012 [http://www.tandfonline.com/
doi/abs/10.1080/10937404.2012.710134 accessed Apr 21, 2015].
Lim SS, Vos T, Flaxman AD, DanaeiG, Shibuya K, Adair-Rohani H, et al. A comparative risk
assessment of burden of diseaseand injury attributableto 67 risk factors and risk fac-
tor clusters in 21 regions, 1990–2010: a systematic analysis for the Global Burden of
Disease Study 2010. Lancet 2013;380(9859):2224–60.
Martin II WJ, Glass RI, Araj H, Balbus J, Collins FS, Curtis S,et al. Household air pollution in
low- and middle-income countries: health risks and research priorities. PLoS Med
2013;10(6), e1001455.
Obueh J. Methanol stoves for indoor air pollution reduction in Delta State, Nigeria —ad-
dressing the needs of people for clean energy. Boil Point 2006;52:27–9.
Pillarisetti A, Vaswani M, Jack D, Balakrishnan K, Bates MN, Arora NK, et al. Patterns of
stove usage after introduction of an advanced cookstove: the long-term application
of household sensors. Environ Sci Technol 2014;48(24):14525–33.
Pokhrel AK, Bates MN, Verma SC, Joshi HS, SreeramareddyCT, Smith KR. Tuberculosis and
indoor biomass and kerosene use in Nepal: a case–control study. Environ Health
Perspect 2010;118(4):558–64.
RCoreTeam.R: a language and environment for statistical computing.; Vienna, Austria;
2014.
Rajvanshi AK. Ethanol fuel for rural households. Maharashtra India: Nimbkar Agric. Res.
Inst.;2006.
Ruiz-Mercado I. The stove adoption process: quantification using sto ve use monitors
(SUMs) in households cooking with fuelwood; 2012.
Ruiz-Mercado I, Masera O, Zamora H, Smith KR. Adoption and sustained use of improved
cookstoves. Energy Policy 2011;39(12):7557–66.
Ruiz-Mercado I, Canuz E, Smith KR. Temperatur e dataloggers as stove use monitors
(SUMs): field methods and signal analysis. Biomass Bioenergy 2012;47:459–68.
Ruiz-Mercado I, Canuz E, Walker JL, Smith KR. Quantitative metrics of stove adoption
using stove use monitors (SUMs). Biomass Bioenergy 2013;57:136–48.
Smith KR, Bruce N, BalakrishnanK, Adair-Rohani H, BalmesJ, Chafe Z, et al. Millions dead:
how do we know and what does it mean? Methods used in the comparative risk as-
sessment of household air pollution. Annu Rev Public Health 2014;35(1):185–206.
Zhang J, Smith KR, Ma Y, Ye S, Jiang F, Qi W, et al. Greenhouse gases and other airborne
pollutants from household stoves in China: a database for emission factors. Atmos
Environ 2000;34(26):4537–49.
40 A. Northcross et al. / Energy for Sustainable Development 35 (2016) 35–40
