econstor
A Service of
zbw
Make Your Publications Visible.
Leibniz-Informationszentrum
Wirtschaft
Leibniz Information Centre
for Economics
Hasan, Mohammad Monirul; Gerber, Nicolas
Working Paper
The impacts of piped water on water quality, sanitation, hygiene and
health in rural households of north-western Bangladesh: A quasiexperimental analysis
ZEF Discussion Papers on Development Policy, No. 217
Provided in Cooperation with:
Zentrum für Entwicklungsforschung / Center for Development Research
(ZEF), University of Bonn
Suggested Citation: Hasan, Mohammad Monirul; Gerber, Nicolas (2016) : The impacts of piped
water on water quality, sanitation, hygiene and health in rural households of north-western
Bangladesh: A quasi-experimental analysis, ZEF Discussion Papers on Development Policy,
No. 217, University of Bonn, Center for Development Research (ZEF), Bonn
This Version is available at:
http://hdl.handle.net/10419/149602
Standard-Nutzungsbedingungen:
Terms of use:
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen
Zwecken und zum Privatgebrauch gespeichert und kopiert werden.
Documents in EconStor may be saved and copied for your
personal and scholarly purposes.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich
machen, vertreiben oder anderweitig nutzen.
You are not to copy documents for public or commercial
purposes, to exhibit the documents publicly, to make them
publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten,
gelten abweichend von diesen Nutzungsbedingungen die in der dort
genannten Lizenz gewährten Nutzungsrechte.
www.econstor.eu
If the documents have been made available under an Open
Content Licence (especially Creative Commons Licences), you
may exercise further usage rights as specified in the indicated
licence.
ZEF-Discussion Papers on
Development Policy No. 217
Mohammad Monirul Hasan and Nicolas Gerber
The impacts of piped water on water quality,
sanitation, hygiene and health in rural
households of north-western Bangladesh - a
quasi-experimental analysis
Bonn, July 2016
The CENTER FOR DEVELOPMENT RESEARCH (ZEF) was established in 1995 as an international,
interdisciplinary research institute at the University of Bonn. Research and teaching at ZEF
address political, economic and ecological development problems. ZEF closely cooperates
with national and international partners in research and development organizations. For
information, see: www.zef.de.
ZEF – Discussion Papers on Development Policy are intended to stimulate discussion among
researchers, practitioners and policy makers on current and emerging development issues.
Each paper has been exposed to an internal discussion within the Center for Development
Research (ZEF) and an external review. The papers mostly reflect work in progress. The
Editorial Committee of the ZEF – DISCUSSION PAPERS ON DEVELOPMENT POLICY includes
Joachim von Braun (Chair), Christian Borgemeister, and Eva Youkhana. Tobias Wünscher is
the Managing Editor of the series.
Mohammad Monirul Hasan and Nicolas Gerber, The impacts of piped water on water
quality, sanitation, hygiene and health in rural households of north-western Bangladesh - a
quasi-experimental analysis, ZEF – Discussion Papers on Development Policy No. 217,
Center for Development Research, Bonn, July 2017, pp. 46.
ISSN: 1436-9931
Published by:
Zentrum für Entwicklungsforschung (ZEF)
Center for Development Research
Walter-Flex-Straße 3
D – 53113 Bonn
Germany
Phone: +49-228-73-1861
Fax: +49-228-73-1869
E-Mail: zef@uni-bonn.de
www.zef.de
The author[s]:
Mohammad Monirul Hasan, Center for Development Research.
Contact: mhasan@uni-bonn.de
Nicolas Gerber, Center for Development Research. Contact: ngerber@uni-bonn.de
Acknowledgements
We are grateful to the German Federal Ministry for Economic Cooperation and Development
(BMZ) via the DAAD, and to the Fiat Panis Foundation for partially funding this research. We
also thank the Bill and Melinda Gates Foundation for financially supporting the fieldwork
leading to this paper through the ZEF AG-WATSAN project. We acknowledge the Institute of
Biological Science (IBSc) lab of Rajshahi University in Bangladesh for their excellent support
during the microbiological tests. Special thanks go to Dr. Akhter Ahmed of IFPRI Dhaka office
for hosting me during the field research period and BRAC-RED for letting us borrow some of
their data collection equipment. We would like to thank Dr. Zahid Hayat Mahmud of icddr,b
Dhaka office, for allowing their laboratory to be used for training the laboratory research
assistants and also Dr. Ariful Haque of Rajshahi University for arranging for their laboratory
facility to be used, especially when considering the difficult political situations.
We would like to thank Prof. Joachim von Braun for his guidance and important suggestions
and deeply acknowledge Dr. Evita Pangaribowo for her guidance in preparing the field
research and questionnaire. Special thanks go to the ZEF AG-WATSAN team for their
continuous support and comments on my works.
Abstract
We investigated the impacts of piped water on water quality, sanitation, hygiene and health
outcomes in marginalized rural households of north-western Bangladesh using a quasiexperimental analysis. A government organization – the Barindra Multipurpose
Development Authority (BMDA) – established a piped water network to provide these rural
households with improved water as they have poor access to potable water. Using
propensity score matching, the study compares a treatment and a control group of
households to identify gains in water-sanitation, hygiene and health outcomes. We found
that the BMDA piped water infrastructure had a positive impact on access to improved
water and significantly reduced the distance traveled for and time spent on collecting
drinking water. However, we found no improvement in the drinking water quality, which
was measured by the extent of fecal contamination (E. coli count per 100 ml of water) at the
point of use. The hygiene status of food utensils also did not show any improvement; food
utensils were tested positive for E. coli in both the control and treatment group. Although
access to BMDA piped water in the premises involves cost, it didn’t improve hygiene
behavior: handwashing with soap after defecation and before feeding children. The treated
households own larger water containers which implies that the intervention has had a clear
impact on the quantity of water used for household purposes. However, we did not find
evidence of health benefits, such as decreased diarrhea incidence of in under-five children,
improved child anthropometrics stunting, underweight and wasting of children due to piped
water use.
JEL classification: D12, I12, I31, O12, O18, Q15, Q25, P46
Keywords: Child diarrhea, Child growth, Piped water supply, Water-Sanitation, Hygiene,
Irrigation agriculture, Propensity Score Matching, water quality, food utensil hygiene.
Inhalt
1. Introduction......................................................................................................................................... 1
2. Sample and Data ................................................................................................................................. 5
2.1 Sample ........................................................................................................................................... 5
2.2 Conceptual framework .................................................................................................................. 7
2.3 Data ............................................................................................................................................... 9
2.3.1 Sample size selection.............................................................................................................. 9
2.3.2 Sampling procedure: cluster sampling ................................................................................. 10
2.4 Identification ............................................................................................................................... 10
2.5 Data collection............................................................................................................................. 11
3. Descriptive statistics .......................................................................................................................... 13
3.1 Household characteristics ........................................................................................................... 13
3.2 Access to improved drinking water ............................................................................................. 13
3.3 Microbiological quality of water and food utensils ..................................................................... 15
3.4 Sanitation facilities ...................................................................................................................... 17
3.5 Hygiene practices ........................................................................................................................ 18
3.6 Child diarrhea and medical expenditure ..................................................................................... 18
3.7 Water collection burden ............................................................................................................. 18
3.8 Child anthropometrics................................................................................................................. 21
4. Theory and Methods ......................................................................................................................... 22
4.1 Theory and Assumptions ............................................................................................................. 22
4.2 Propensity Score Matching (PSM) ............................................................................................... 23
5. Impact estimates ............................................................................................................................... 27
5.1 Impact on water quality, water access and cooking utensils ...................................................... 27
5.2 Impact on hygiene practices ....................................................................................................... 28
5.3 Impact on health ......................................................................................................................... 29
5.4 Impact on child growth ............................................................................................................... 29
5.5 Sensitivity analysis ....................................................................................................................... 30
6. Discussion of results and policy implications .................................................................................... 32
7. Conclusion ......................................................................................................................................... 34
References ............................................................................................................................................. 42
1. Introduction
The world is still suffering from a lack of proper sanitation and safe drinking water in the
twenty-first century. More than 0.7 billion people (9% of the world’s population) do not
have access to improved drinking water sources and about 2.4 billion people (33% of the
world’s population) do not have access to improved sanitation (United Nations, 2015). The
number of people with access to piped water increased from 2.3 billion in 1990 to 4.2 billion
in 2015 (United Nations, 2015). Although a significant improvement has been made in
terms of reducing open defecation, it is still practiced by a substantial portion of the
population (946 million, 13% of the world’s population) and could cause sanitation-related
health problems. The multipurpose characteristics of water use, especially for irrigation and
domestic purposes, leads to health issues which could be explained by the trade-offs
between water quantity and quality.
Water quality is very crucial for domestic use, especially for drinking, and therefore water
needs to be properly handled. Water quality at the source often differs from water quality
at the point of use. Access to piped water by itself is insufficient for improving child health
(e.g., decreasing diarrhea incidence) and child development (Jyotsna Jalan & Ravallion,
2003). Indeed, piped water interacts with a wide range of other determinants of child health
such as hygienic water storage, water treatment, sanitation infrastructure, medical
treatment and nutrition (Jalan and Ravallion, 2003). To maintain and develop human capital,
constant investment in water and sanitation (WATSAN) infrastructure is required, along with
investment in effecting behavioral and cultural changes. Household health expenditure is
seen as a growth-friendly investment; cost-effective and efficient health expenditure can
increase the quantity and productivity of labor through increasing life expectancy (European
Commission, 2013). While investing in WATSAN infrastructure and education do not benefit
households immediately, it brings long-term returns in human capital formation, reduces
the cost of treatment and provides positive external benefits to the society. Households can
invest in their health in several ways, such as establishing proper sources of water, setting
up sanitary latrines, educating themselves on proper hygiene practices, taking preventive
actions, purchasing medicine and health services, and even buying health insurances. In
fact, it is well established that safe hygiene practices are the single most cost-effective
means of preventing infectious diseases, but the investment in hygiene is low both in health
and water-sanitation sector (Curtis et al., 2011).
Access to water is not the only indicator of household well-being; water quality and quantity
are important indicators too. Crucially, the three indicators are interlinked. For instance,
distance to a water source affects water quality (Devoto, Duflo, Dupas, Parienté, and Pons,
2012) and the amount of work involved in water collection affects a household’s per capita
water use. Kremer et al. (2011) found that 58% of the surveyed households reported having
1
insufficient water for their daily use. Having insufficient water has negative consequences
on hygiene behavior, height-for-age and diarrhea incidence in under-five children (van der
Hoek, Feenstra, & Konradsen, 2002). Further, it creates intra-household inequalities as the
burden of water collection directly falls on women and girls, which reduces their economic
activities and opportunities, including the education of children. Devoto et al. (2012)
showed that an adequate amount of water from a piped water network allows for more
leisure time and higher productivity by reducing the burden of water collection.
Water quality affects the sanitation status of a household. Unimproved sanitation and poor
hygiene worsen water quality by allowing pathogens to contaminate water. Improved
sanitation decreases diarrhea morbidity and diarrhea mortality, including the probability of
hookworm infection. In comparison, better water quality plays a smaller role in reducing
diarrhea than sanitation and hygiene (S A Esrey, Potash, Roberts, & Shiff, 1991a). The
importance of water quality and sanitation in reducing health risk has been welldocumented (S A Esrey et al., 1991a; Fewtrell et al., 2005a; Waddington, Snilstveit, White, &
Fewtrell, 2009). The health benefits of access to improved water are less observable than
those of sanitation: they can only be realized if access to improved sanitation is ensured and
if there is sufficient water available for domestic use. (Steven A Esrey, 1996). A study
conducted in 145 low- and middle-income countries showed that in 2012, about 502
thousand diarrhea deaths were caused by inadequate drinking water, about 280 thousand
by inadequate sanitation and about 297 by inadequate hand hygiene (Prüss-Ustün et al.,
2014). Improved sanitation is associated with fewer diarrhea cases and improved height and
weight of children; height and weight of children were found to be higher in urban areas
than in rural areas (Steven A Esrey, 1996). Installing water filters and building high-quality
piped water systems with sewer connections are better at reducing diarrhea cases than
other kinds of intervention (Wolf et al., 2014). A study showed that diarrhea incidence and
cholera incidence in Bangladeshi households could be reduced by simple water filtration
(Colwell et al., 2003; Huo et al., 1996; Huq et al., 2010). Sanitation can be improved for
people in rural Bangladeshi villages by giving subsidies for building latrines. Such
intervention can also cause a beneficial spillover effect by encouraging neighboring villages
which have yet to receive subsidies to also improve on their sanitation infrastructure and
build latrines (Guiteras, Levinsohn, & Mobarak, 2015; Kaiser, 2015).
The nutrition status of under-five children is affected by the quality of water and food in a
household. Food and kitchen utensils can easily be contaminated with pathogenic bacteria
through washing and cooking. Food can be contaminated through preparing meals with
unimproved water. Preparing food with unimproved water therefore poses a serious health
risk and can cause adverse health effects, including malnutrition in children. Malnutrition
impairs the immune system and makes children more vulnerable to diarrhea (van der Hoek
et al., 2002). Diarrhea has a long-term negative impact on cognitive development in young
children (Keusch et al., 2006). Infants with poor nutritional intake are at higher risk of
2
diarrhea and malnutrition than those receiving nutritional supplementation (Javaid et al.,
1991). One way of breaking the vicious cycle of diarrhea and malnutrition is to increase the
use of safe water and improved sanitation. This reduces the transmission of pathogens,
thereby lowering diarrhea incidence and child mortality and improving nutritional status
(Steven A Esrey, 1996). Van der Hoek et al. (2002) found that larger water storage is
associated with higher diarrhea risk and child stunting prevalence.
There are a handful of studies that investigated the relationship between improved water
and health gains (S A Esrey, Potash, Roberts, & Shiff, 1991b; Fewtrell et al., 2005b; Hoque,
Juncker, Sack, Ali, & Aziz, 1996; Waddington et al., 2009; Wolf et al., 2014). The impact of
piped water on health has been documented in several studies under different conditions
(Devoto et al., 2012; Gamper-Rabindran, Khan, and Timmins, 2010; Jalan and Ravallion,
2003; Klasen, Lechtenfeld, Meier, and Rieckmann, 2012). Gamper-Rabindran et al. (2010)
showed that access to piped water reduced child mortality in Brazil by 20% from 1970 to
2000.
A randomized controlled trial experiment in urban Morocco, which highlighted the effects of
piped water in an urban setting, suggested that piped water improves neither water quality
nor health, but rather helps save time and reduces intra-household conflict (Devoto et al.,
2012). Another study conducted in an urban setting with quasi-experimental analysis
(Klasen et al., 2012) showed that piped water in urban Yemen worsened health outcomes if
water is rationed, thus highlighting the intercorrelations between water quantity, water
quality and human health. They suggested that piped water systems can only improve
health outcomes when water supply is continuous. On the other hand, a study conducted in
rural India showed that access to piped water only improved the health of well-educated
and high-income households and not poorly educated households (Jalan and Ravallion,
2003). However, the study did not investigate water quality.
The objective of this paper is to estimate the impact of piped water use on water,
sanitation, hygiene, and health outcomes in rural households living in the marginalized area
of north-western Bangladesh. The hypothesis is that the BMDA piped water service will
make a difference in health outcomes between a treated household (i.e., those with access
to BMDA piped water) and a control household (i.e., those without access to BMDA piped
water).
This paper differs from the previously mentioned papers in terms of its setting and scope.
This paper studied the health impact of using and handling piped water in a marginalized
rural setting and investigated the microbiological quality of water and kitchen utensils,
which is a unique aspect of this study. To study the health effects of piped water connection
in a water-scarce area of Bangladesh, the following variables were investigated: the level of
the fecal bacteria E. coli in drinking water and on kitchen utensils, water, sanitation and
3
hygiene infrastructure and behavior, and various health outcomes (such as diarrhea
incidence and child anthropometrics).
4
2. Sample and Data
2.1 Sample
In Bangladesh, safe drinking water is
becoming more and more scarce
because of salinity and arsenic
contamination. In the 1990s, 97% of the
population had access to safe water, but
this figure dropped to 74% in 2006
because of widespread and severe
arsenic contamination (GoB and UNDP,
2009), while the arsenic-adjusted figure
was 86% in 2009 (GoB, 2012). According
to the Joint Monitoring Programme
2015 report, 87% of the households in
Bangladesh have access to improved
drinking water, 61% have improved
sanitation facility, and only 1% defecate
openly (Unicef and WHO, 2015).
Humans mostly depend on ground water
for potable water. Groundwater is the
world’s largest ubiquitous source of Figure 1: Map of Bangladesh based on usable
high-quality fresh water (Shiklomanov pumping method.
and Rodda, 2003; Taylor, 2013). Source: Shamsudduha, Taylor, Ahmed, and Zahid (2011).
Groundwater depletion has recently
been detected in arid and semi-arid areas because of intensive abstraction of water for
irrigation purpose (Konikow, 2011; Rodell, Velicogna, and Famiglietti, 2009). The aquifer
level in the north-western part of Bangladesh is below normal caused by very high rates of
water extraction. In this part of the country, it is not easy to obtain groundwater by drilling
boreholes or setting up tube wells (Figure 1). A significant amount of money is required to
build a deep tube well for extracting groundwater. As shallow tube wells are not
recommended for drawing groundwater, deep tube wells are necessary for accessing pure
drinking water (Chen et al., 2007; Escamilla et al., 2011). People living in this area are
marginalized1 in term of access to fresh groundwater. Therefore, the BMDA, a public body,
1
The term “Marginality” is an involuntary position and condition of an individual or group at the margin of the
social, political, economic, ecological and biophysical system, that prevent them from access to resources,
assets, services, restraining freedom of choice, preventing the development of capabilities, and eventually
causing extreme poverty (Gatzweiler and Baumüller, 2014).
5
has started an initiative to supply water for irrigation and household uses using pipelines. It
has covered an extensive area in northern-western Bangladesh based on an analysis of
water needs in that area. As of 2014, 15,054 deep tube wells have been built, supplying
irrigation water to 255,256 hectares of land used for cultivating boro rice. Besides irrigation
water, the authority also supplies drinking water to many parts of its working areas. By
2014, they had established 1,100 overhead water tanks, each containing 25,000 liter of
water. The water flows from the overhead tanks to the households through a network of
pipes. The BMDA charges a household a minimal amount of money (approx. Tk. 10) for
every person using the water in a month.
The aims of the BMDA drinking water project are as follows:
1. Supply potable water to every household in rural areas throughout the year.
2. Ensure the around 500 thousand people in this area have access to arsenic-free
water.
3. Eradicate diseases caused by 1.) arsenic and 2.) shortage of potable water.
4. Improve the health of the people living in the rural villages
5. Create a reliable drinking water supply in the rural villages.
6
2.2 Conceptual framework
Figure 2: Water treatment, sanitation and hygiene barriers to transmission of
pathogens.
Source: Author’s calibration; adopted from Prüss, Kay, Fewtrell, and Bartram, (2002), and Waddington et
al. (2009).
Human health is affected by the transmission of pathogens from feces and waste water to
humans. Pathogens is transmitted through various agents such as improper sanitation and
hygiene, and unsafe drinking water sources (Figure 2). The transmission of pathogens from
feces to human can takes place through hands, flies and ground or surface water. Not
washing hands after defecation may allow pathogens to enter into human body through
various routes, such as eating, drinking, preparing food, and feeding. Pathogens can be
transmitted from ground and surface water to humans in various ways. Preparing food with
untreated surface water, drinking surface water, and ingesting water while bathing in a
pond or river can introduce pathogens into the human body, which may result in many
water borne diseases. Ground and surface water can be contaminated by sewage, flood and
chemical compounds. Piped water can be contaminated by sewage or flood water seeping
into a pipeline. Chemical compounds such as arsenic, chlorine, iron, manganese and sodium
can pollute water. These chemical compounds, along with the other industrial chemical
7
wastes, can even pollute underground water sources, which is more dangerous than the
surface water pollution in the long run.
The transmission of pathogens can be stopped by interventions such as water treatment at
source and the point-of-use (POU), and improving sanitation and hygiene (Waddington et
al., 2009). In Figure 2, the arrows (both dotted and solid) show the possible routes of
pathogen transmission; the dotted arrows indicate that the particular route of transmission
can be interrupted by interventions. The color of each dotted arrow indicates which method
acts as an effective barrier to pathogen transmission for that particular transmission route.
A blue dashed arrow denotes sanitation is an effective barrier to pathogen transmission, a
red dashed arrow denotes hygiene, and a green dashed arrow denotes water treatment.
The figure clearly shows that WATSAN interventions can stop pathogen transmission and
therefore reduce the risk of waterborne diseases.
Figure 3: Conceptual framework of Agriculture-water-sanitation nexus on human health
Source: (Tsegai, Mcbain, and Tischbein, 2013)
Agriculture plays an important role in maintaining water quality. The relationship between
agriculture, water and sanitation can be analyzed as shown in Figure 3, where potential
interventions with external drivers indicate the outcome of interest. Agriculture can affect
water quality, sanitation and hygiene, and vice versa. To alter the magnitudes of the
relationship between these three variables, various kinds of interventions are required. The
different kinds of interventions can be categorized as follows: (1) institutional intervention,
such as capacity building and closing gender gap; (2) cultural intervention, such as providing
education on hygienic behavior; (3) and financial intervention, such as providing
8
technological and financial resources. For the purpose of microeconometric analysis, if all
external drivers remain unaltered, the kinds of interventions mentioned above could
generate a positive impact on health outcomes.
2.3 Data
The selected study areas are located in the Rajshahi and Naogaon district (Figure 4), two dry
areas located in north-western Bangladesh. The BMDA has built a piped-water network in
these two areas. The survey data used in the analysis was obtained from a baseline survey
conducted for a randomized controlled trial experiment concerning food hygiene education
that took place in the following months.
Source: Banglapedia. www.bpedia.org/R_0079.php,
Accessed on October 5, 2015.
Source: www.bpedia.org/N_0048.php
Accessed on October 5, 2015.
Figure 4: Map of Rajshahi district (left) and Naogaon district (right)- the study area
2.3.1 Sample size selection
For this paper, 512 households were randomly chosen from two main clusters – those that
villages that have received BMDA intervention and those that have not. The sample size
satisfied the minimum sample size (498) calculated based on a Poisson statistical regression
power analysis. The analysis considered an effect size (ES) of 0.95 (i.e., the minimum
difference in the outcome between treated and non-treated subjects is on average 5%) and
a multicollinearity across the covariates of 0.7 (which is quite extreme) and allowed for a
probability of Type I error of 5% and a statistical power (1-Probability of Type II error) of
80%.
9
2.3.2 Sampling procedure: cluster sampling
The cluster sampling technique was used in this study. Cluster sampling is useful because it
does not require exhaustive lists of every single person in the population to be compiled.
According to World Health Organization (WHO) guidelines, the population within a cluster
should be as heterogeneous as possible and different clusters should be as homogenous as
possible among themselves. The WHO also recommends that the number of clusters in the
sampling frame should be at least five times larger than the number of randomly selected
clusters.
A random sample was chosen from a list of clusters determined by a list of villages (or
mouzas) in the districts studied. In this quasi-experiment, all clusters can be classified into
two main clusters: 1) villages whose households are connected to the piped-water network
(public intervention in 389 mouzas) and 2) villages without access to the piped-water
network (in 359 mouzas).
A useful rule of thumb is that there should be a minimum of 30 clusters. With our sample
size of 512 households, we needed to survey 16 households per cluster (mouza) from a total
of 32 mouzas; this means that 16 mouzas were to be randomly selected from the list of
villages with BMDA intervention and another 16 from the list of villages without BMDA
intervention. The random selection was done using Stata. We note that our data is well
within the WHO recommendations on the minimum ratio between clusters in the sampling
frame and the sample survey.
In each of the 32 villages, a small census survey was conducted to identify eligible
households. Only households that have at least one child younger than five years old were
included in this study. Then 16 households were randomly selected from the eligible
households in each village.
2.4 Identification
The survey was conducted among 512 households, 256 of which were living in areas with
BMDA drinking water coverage and the other 256 in areas without the coverage. It was
observed that many households living in areas with BMDA drinking water coverage did not
actually receive BMDA piped-water services because of technical problems, such as faulty
water pumps. Hence, only households that actually received piped-water services were
identified as BMDA-treated. By adopting this definition, we considered the actual receipt of
piped water when analyzing the impact of BMDA pipe-water services, rather than BMDA’s
intention to supply piped water to households. According to this definition, 186 households
were considered BMDA-treated and 326 households were not considered BMDA-treated.
10
2.5 Data collection
Household survey:
Household survey (baseline) was conducted in October 2014 in the rural villages in Rajshahi
and Naogaon district, located in north-western Bangladesh (Figure 4). Every household
received a detailed 28-page questionnaire that asked for information about a household’s
assets, income, food and non-food expenditure, investment and financial activities,
WATSAN- and hygiene-related practices, and agricultural activities. The households took
part in this survey willingly and did not receive any financial incentives in return. Each
questionnaire required approximately two hours to fill out. A total of ten field enumerators
and a supervisor were involved in collecting the information from households.
The study was approved by the ZEF ethical committee of the University of Bonn to protect
the rights of the survey respondents. All households received extensive information about
the study and had to sign a consent form prior to participating in the survey. All households
had the right to discontinue their participation at any time during the observation period.
Each household was given an identification card for follow-up.
Anthropometric survey:
On the same day as the household survey, a field enumerator took anthropometric
measurements of under-five children in the households. The height and weight of the
under-five children were measured in this survey. The measurements taken were
determined according to WHO guidelines for anthropometric measurements. The
measurement took place on the same day as the survey because households might not be
available on the following days, which might reduce the sample size of the anthropometric
data. The field enumerator also recorded the GIS information of all households, including
their latitude and longitude.
Microbiological testing of water and food utensil:
In the days following, a laboratory research assistant (LRA) visited the households and
collected water and food utensil samples from the household. The LRA collected a glass of
drinking water from the same jar the household use (the point of use). The water sample
was collected in a sterilized bottle and kept in a cool box for transporting to the laboratory.
The LRA also tapped or pressed a “food stamp” on the households’ drinking glass, spoon
and main cutting instrument. The number of food stamp samples was recorded, and the
media (food stamps) were kept in a cool box for transporting to the laboratory.
The LRAs are microbiology graduates and have been trained for this kind of assignment. Two
LRAs worked simultaneously in different areas. Each LRA covered 16 households in a mouza
and then returned to the lab on the same day. Before commencing their fieldwork, they
11
were trained at the International Centre for Diarrheal Disease Research, Bangladesh
(icddr,b) by a senior scientist.
E. coli testing procedure in the Laboratory:
The bacterium Escherichia coli O157 (E. coli) is the most commonly recommended indicator
of fecal contamination in water and food utensils. The WHO recommends there should be
no E. coli in a 100 ml drinking water sample. In the survey, E. coli was measured by filtering
100 ml of drinking water through a 0.22 μm filter paper (cellulose nitrate membrane filter;
47 mm diameter; pore size of 0.2 microns; Sartorius, Germany) using a vacuum filtration
unit. Then the filter paper was removed and placed onto a Compact Dry EC growth media
plate (Nissui Pharma, Japan) to incubate it at 37-39°C for 24 hours. After incubation, the
LRAs counted and recorded the number of E. coli colony forming units (cfu), indicated by
blue colonies, and the number of coliform colonies (red colonies) on each of the Compact
Dry EC media (Figure 5).
Food stamp XM-G agars (HyServe, Germany) were used to test for E. coli on kitchen utensils
(a glass, a spoon and a cutting knife). Food stamp sampling is a simple-to-use bacteriological
testing method for the presence of bacteria in food. A food stamp (10cm2 XMG agar) was
pressed once on each of the three specimens in the household and then kept in the cool box
for transporting to the laboratory. The food stamp was then incubated at 37-39°C for 24
hours. After incubation, the LRAs counted and recorded the number of E. coli colony
forming units (cfu), indicated by blue colonies, and the number of coliform colonies (red
colonies) on each of the XMG agar media (Figure 5).
Figure 5: E. coli colony forming units (cfus) in 100 ml water (left) and E. coli cfu in food
preparing utensils (right).
Source: Microbiological survey data, 2014.
12
3. Descriptive statistics
3.1 Household characteristics
The average age of a household head was similar in both the treated and control groups (35
years old). On average, the household head of a treated household completed slightly more
years of school than their counterpart in a control household; even then, they largely only
received primary education (Table 3). The average household size was 4.72, whereby the
number of male and female members are generally equal. About 98% of the household
heads were married at the time of survey, and the treatment and control households did
not show any significant difference in this regard. In total, 52% of the household heads were
wage earners, which comprised 42% of the household heads in the treatment group and
57% in control group (Table 3). About 57% and 48% of the households were agricultural and
non-agricultural households respectively. The percentage of non-agricultural households in
the treatment group (58%) was higher than in the control group (42%); the difference was
statistically significant at the 0.01 level. There was no significant difference between the
treatment and control groups in terms of irrigation field ownership; 63% of the households
had irrigation fields.
On average, a treated household had significantly more land than a control household (0.96
acres as compared to 0.54 acres). The same was also true for agricultural land. In our survey,
87% of the households owned livestock, and the average number of livestock was 15.3,
whereby on average a treated household owned more livestock than a control household
(Table 3). On average, a treated household had a higher household expenditure than a
control household; the difference between their expenditure was statistically significant.
Generally, food expenses constituted more than half of the total expenditure. Around 48%
of the households had access to microfinance services, which on average lasted for more
than 3.5 years. These marginalized households also reported saving almost 35% of their
annual expenditure, which is a rather high figure.
3.2 Access to improved drinking water
The quality of household drinking water is classified into improved and unimproved. In this
paper, an improved drinking-water source is defined as a piped-water network, private tube
well, community tube well, rainwater, protected springs or protected ring well; an
unimproved drinking-water source is defined river, pond, canal, or irrigation water from
shallow or deep tube well (Table 1). Our sample survey found that 99.46% of the households
in the treatment group (N=186) used improved water for drinking; in comparison, 93.87% of
the households in the control group (N=326) did so. The Fisher’s exact test yielded a p-value
13
of 0.002, which means that the difference between these two groups is statistically
significant at the 0.01 level, i.e. the two groups are different.
The households in the control group mostly obtained drinking water from community tube
wells (67%) and their private tube wells (27%), while very few households in the treatment
group (1.61% and 2.69% respectively) relied on these two sources for drinking water (Table
1). Most of the households in the treatment group got their drinking water from piped
sources (95.17%), whereas none of the households in the control group did so. The Fisher’s
exact test yielded a p-value of zero, which means the mean value of the treatment group
and the control group are different from each other.
Community tube wells are established by private households or, in some places, by
government or non-government agencies. Households do not need to pay for the water
they draw from a community tube well. Some households also collect drinking water from
private deep tube wells which are meant for irrigation. Different households had to travel
different distances to collect drinking water. Figure 6 shows the distribution of households
by the distance (in meters) between their house and the nearest drinking water source.
Control households were more likely to collect water from tube wells located far away from
their home. While 82% of the treated households were required to travel less than 50 m to
their drinking water source, only 57% of the control households could do that. On average,
households in the control group were required to travel longer distances than those in the
treatment group. The Fisher’s exact test yielded a p-value of zero for the difference
between the mean distance traveled by the treatment and that traveled by the control
groups.
Table 1: Drinking water facilities in sample households
Piped water from outside of house
Piped water from inside of house
own tube well
community tube well
Deep tube well (for irrigation)
Private Deep tube well (not piped)
Total
N
13
164
93
221
1
20
512
Total
%
2.54
32.03
18.16
43.16
0.2
3.91
100
Treatment
N
%
13
6.99
164
88.17
5
2.69
3
1.61
1
0.54
0
0
186
100
Control
N
%
0
0
0
0
88
26.99
218
66.87
0
0
20
6.13
326
100
Source: Baseline survey, 2014. Fisher’s exact = 0.000
14
Control
INSIDE
HOUSE
< 50M
50M -100M
100M -200M 200M - 500M
> 500M
0,31
0
0,92
0,54
5,83
1,61
4,3
11,29
15,64
20,25
26,69
28,49
30,37
53,76
treatment
> 1 KM
Figure 6: Percentage of households by traveling distance to collect drinking water.
Fisher's exact= 0.000
Source: Baseline survey, 2014.
3.3 Microbiological quality of water and food utensils
Figure 7 shows the percentage of households by their risk levels in terms of the E. coli cfu
count found in their drinking water. Only 25% of the households in the treatment group had
drinking water in which E. Coli was not detected, whereas this figure was 20% for the
control group. Although the distributions of cfu count in drinking water differs slightly
between the control and treatment groups, the overall difference is not statistically
significant; the Fischer exact test yielded a p-value of 0.126.
Control
12,37
NO RISK (<1)
LOW (1-10)
MEDIUM (11-100)
9,51
24,19
33,13
37,42
19,94
25,27
38,17
Treatment
HIGH(>100)
Figure 7: Percentage of households under different risk level for drinking water faecal
contamination based on E. coli cfu counts. Fisher's exact = 0.126.
Source: Microbiological survey data, 2014.
15
The contamination of kitchen utensils is closely associated with the use of contaminated
water. Nevertheless, kitchen utensils can also be contaminated through other routes, such
as handling utensils with unwashed hands and processing raw meat or fish. To identify
microbiological contamination on kitchen utensils, we tested three items (a water glass, a
spoon and a cutting knife) for E. coli by using the food stamps method. It is recommended
that no E. coli should be found on kitchen utensils. Figure 8 shows the percentage of
households with and without E. coil contamination on their food utensils. It is observed that
the E. coli counts on the kitchen utensils of the control households were higher than those
in the treated households. The mean difference between the treatment and the control
households is statistically significant at the 0.074 level.
36,81%
45,16%
39,84%
63,19%
Control (N=326)
54,84%
Treatment (N=186)
60,16%
Total (N=512)
NO E. COLI
PRESENCE OF E. COLI
Figure 8: Presence of E. coli in the food preparing utensils. Fisher's exact = 0.074
Source: Microbiological survey data, 2014.
16
Control
NO RISK (=0)
MEDIUM (1-10)
HIGH (11-100)
5,52
4,84
26,07
22,58
31,6
27,42
36,81
45,16
Treatment
VERY HIGH(>100)
Figure 9: Percentage of households under different risk level for food preparing utensils
fecal contamination based on E. coli cfu counts. Fisher's exact = 0.338.
Source: Microbiological survey data, 2014.
The microbiological quality of kitchen utensils between treatment and control group are not
statistically significant at the 0.1 level (Figure 9).
3.4 Sanitation facilities
The study areas had low sanitation coverage. Only 68% of the households had improved
sanitation, which comprised 75% of the households in the treatment group and 63% in the
control group (Table 3). Almost 17% of the households used hanging toilets, and 15% of the
households still practiced open defecations. Only less than 1% of all households used
community toilets. 21% of the households in the control group practiced open defecation,
while only 4% of the treated households did so. The difference between the treatment and
control groups was statistically significant at the 1% level (Table 2).
Table 2: Sanitation facility by treatment and control households
Total
Open defecation
Hanging toilet (fixed place)
Pucca/ bricked toilet (unsealed)
Sanitary toilet without flush (water sealed)
Sanitary toilet with flash (water sealed)
Community latrine
Total
N
77
86
159
186
2
2
512
%
15.04
16.8
31.05
36.33
0.39
0.39
100
Treatment
N
%
8
4.3
36
19.35
59
31.72
80
43.01
1
0.54
2
1.08
186
100
Control
N
%
69
21.17
50
15.34
100 30.67
106 32.52
1
0.31
0
0
326
100
Source: Baseline survey, 2014. Fisher's exact = 0.000
17
3.5 Hygiene practices
Many of the households in the study areas had inadequate hygiene. Only 68% of the
households reported regularly washing their hands with soap after using the toilet. Very few
(only 3%) households reported actually washing their hands with soap before feeding their
children (Table 3). The difference between the treatment group and the control group in
terms of handwashing practices was statistically significant. Around 76% of the treated
households reported washing their hand with soap after using the toilet; in comparison,
64% of the control households did so. Similarly, 5% of the treated households said that they
washed their hands with soap before feeding their child, while 2% of the control households
reported doing so. The difference between the two groups is statistically significant.
On average, the households in the treatment group used more soap than those in the
control group. The per capita soap consumption in the study areas was merely half a bar of
soap (approx. 50 g) per month. Besides, households rarely cleaned their water container
with soap. Only 26% households did so regularly, which consisted of 32% of the treated
households and 22% of the control households. The difference between the two groups is
statistically significant.
3.6 Child diarrhea and medical expenditure
Child diarrhea could result from unimproved water, unimproved sanitation and poor
hygiene practices and may have long-term consequences on child’s development. Diarrhea
cases in under-five children living in the study areas were recorded for a month, and it was
observed that child diarrhea was not highly prevalent in the areas. Only 13% of the
households reported diarrhea cases in their under-five children, which comprised 11% of
the treated households and 14% of the control households (Table 3). There was no
significant differences between the treated and control households in terms of diarrhea
incidence. The treated households generally invested more money in adult and child health
than the control households. On average, a treated household spent BDT 578 monthly on
treating illness in under-five children, whereas the control household spent BDT 519 for this
purpose. A treated household spent an average of BDT 4703 every year on adult healthcare
(older than five years old), while this figure was BDT 3994 for the control households. In
terms of the amount of money spent on healthcare, the difference between the two groups
were not statistically significant.
3.7 Water collection burden
It is mostly women who took up the duty of collecting water for the entire household.
Women were tasked with collecting drinking water in almost 97% of the households.
18
Although they did not have to spent much time on collecting water (only 12.77 minutes on
average), the activity is still a burden on them. Notably, the women in a treated household
spent around half as much time as those in a control household on collecting drinking water.
19
Table 3: Summery statistics
Variable
Household Characteristics
Female headed households (dummy)
Age of household head (years)
Completed years of schooling of household head
Maximum completed schooling in the household
Household size
Total number of male in the household
Total number of female in the household
female/male ratio
Household head currently married (dummy)
Household occupation: wage earning (dummy)
Household occupation: agriculture (dummy)
Household occupation: non-agriculture (dummy)
Total land (in acre)
Total agricultural land (in acre)
Total free land (in acre)
Number of Livestock
Number of cows
Number of goat
Number of poultry
Number of shared livestock
Food expenditure (BDT)
Non-food expenditure (BDT)
Total expenditure (BDT)
Per capita expenditure (BDT)
Participants of Microfinance program (dummy)
Duration of membership (years)
Household savings (BDT)
Irrigating households (dummy)
Sanitation
Access to improved sanitation (dummy)
Annual cost for maintaining a toilet (BDT)
Water
Access to improved drinking water (dummy)
Annual cost for water (BDT)
Time spend to collect drinking water in a day (minute)
Draw water with a mug from jar (dummy)
Size of the water container (liter)
Minutes to collect drinking water
100ml drinking water E.coli count (cfu)
100ml drinking water Coliform count (cfu)
E.coli count in the food utensils (cfu)
Coliform count in the food utensils (cfu)
Presence of E. coli in the 100 ml water (dummy)
Presence of Coliform in the 100 ml water (dummy)
Presence of E. coli in food preparing utensils (dummy)
Presence of Coliform in food preparing utensils (dummy)
Disease
Child diarrhea in last month (percentage) (dummy)
Annual disease cost for adult (BDT)
Monthly disease cost for children (BDT)
Hygiene
Hand wash with soap after coming from toilet (dummy)
Hand wash with soap before feeding child (dummy)
Clean water container with soap (dummy)
Total soap consumed per month (number, 1 soap =100gr.)
Per capita soap consumption per month (number)
Total
(N=512)
Treatment
(N=186)
Control
(N=326)
P-value
(treatment
=control)
1%
35.26
4.64
7.77
4.72
2.36
2.36
1.27
98%
52%
57%
48%
0.69
0.55
0.005
15.30
1.21
0.92
9.09
0.18
59692.67
39915.68
110835.40
23543.75
48%
3.91
36729.38
63%
0.5%
35.24
5.73
8.49
4.92
2.44
2.48
1.27
98%
42%
59%
58%
0.96
0.77
0.004
19.31
1.24
0.92
9.67
0.16
65786.71
49469.15
121502.20
24828.40
46%
4.52
43737.03
61%
1.2%
35.27
4.01
7.36
4.61
2.31
2.30
1.27
98%
57%
56%
42%
0.54
0.42
0.006
13.02
1.20
0.91
8.75
0.20
56215.71
34464.92
95411.55
20554.05
49%
3.58
32731.15
63%
0.45
0.98
0.00
0.00
0.05
0.24
0.06
1.00
0.81
0.00
0.47
0.00
0.01
0.01
0.38
0.13
0.78
0.93
0.41
0.52
0.00
0.00
0.00
0.00
0.54
0.10
0.17
0.62
68%
258.20
75%
334.25
63%
214.82
0.01
0.32
96%
231.61
12.77
35%
17.78
4.6
44.52
400.61
36.47
78.12
78%
97%
60%
94%
99%
631.61
8.09
37%
23.47
3.9
50.79
421.55
25.48
65.44
75%
99%
55%
94%
94%
3.39
15.46
34%
14.54
5.1
40.93
388.68
42.77
85.38
80%
97%
63%
93%
0.00
0.00
0.00
0.44
0.00
0.0
0.43
0.40
0.22
0.28
0.16
0.11
0.06
0.71
13%
4251.14
540.5
11%
4702.53
577.98
14%
3993.59
519.13
0.24
0.46
0.63
68%
3%
26%
2.31
0.51
76%
5%
32%
2.67
0.56
64%
2%
22%
2.11
0.48
0.01
0.05
0.02
0.00
0.00
Source: Baseline survey, 2014.
20
3.8 Child anthropometrics
Children in the treatment and control groups had similar height-for-age z-score and weightfor-age z-score. But children in the treatment households were better off than those in the
control households in terms of their weight-for-height z-score and BMI z-score, as evident in
the scores that are statistically significant from zero.
Table 4: Child anthropometrics by treatment and control households
Height-for-age z-score
Weight-for-age z-score
Weight-for-height z-score
BMI z-score
Stunted
Severely stunted
Underweight
Severely underweight
Wasted
Severely wasted
Mean
(N=569)
-1.57
-1.50
-0.88
-0.74
36%
10%
32%
7%
13%
2%
Treatment
(N=207)
Control
(N=362)
P-value
-1.59
-1.40
-0.72
-0.59
34%
10%
27%
7%
11%
2%
-1.56
-1.56
-0.97
-0.83
37%
10%
36%
7%
14%
2%
0.85
0.10
0.01
0.02
0.48
0.89
0.03
0.76
0.40
0.87
Source: Baseline survey, 2014.
In the study areas, 36% of the children exhibited stunted growth and 10% were severely
stunted (Table 4). Furthermore, 32% of the children were underweight; the treatment group
had a lower percentage of underweight children (27%) than the control group (36%). The
difference between these two groups was statistically significant in terms of the prevalence
of underweight children. The percentage of severely underweight rate was similar in both
groups (around 7%). Among the households surveyed, 13% of the children were found to be
wasted.
21
4. Theory and Methods
4.1 Theory and Assumptions
This paper is developed based on “Theory of Change” which explains the process of change
by outlining the causal linkages (short-, medium- and long-term outcomes) of an
intervention in a societal setting. Logical relationships are used to generate outcome
pathways. Theory of Change has been discussed much in literature, including Anderson and
Harris (2005), James (2011), and Stern et al. (2012). The steps to build a Theory of Change
include (1) defining interventions, objectives and outcomes; (2) laying out the main steps in
a causal chain, (3) Identifying the underlying assumptions, (4) adding a temporal dimension,
(5) identifying the key evaluation questions, and (6) validating and revising.
This analysis is based on a Theory of Change that assumes piped water improves health and
productivity (Figure 10). This causal link works through investment in the water, sanitation
and hygiene infrastructures which, in the short term, results in lower child diarrhea
incidence, reduction in time used for water collection, and clean kitchen utensils. The longterm outcomes are improved physical development in under-five children (i.e., lower
prevalence of stunting, underweight and wasting) and higher productivity (fewer sick days).
In any Theory of Change, there are always some assumptions to simplify the model.
22
Improved water,
sanitation and
hygiene investment
Piped water
to the rural
household
Activity
Water
available to
households
and water
collection
time saved
Water
quality
improved
Cleaned Food
utensil
Intermediate
process
Short-term
outcomes
Improved
health and
productivity
Long-term
outcomes
Assumptions
Public investment
for piped water
infrastructure by
govt. agencies
such as BMDA in
water scarce area
where alternative
sources of water
are scant.
-Household has
money to invest
-Household has
knowledge of
improved water
quality
Household has
knowledge of food
hygiene
Household is
rational and
maximizes health
and productivity
Figure 10: Theory of Change- impact pathways.
Source: Authors’ calibration
4.2 Propensity Score Matching (PSM)
Propensity score matching (PSM) was used to estimate the causal effects of piped water on
child health in a cross-sectional sample without random placement. In this study, the
placement of the treatment (piped water) was not random. The BMDA established its pipedwater network based on community needs and geographical location. So the households’
access to piped water is endogenous and estimating ordinary least square (OLS) will
generate biased results. Although both OLS and PSM require conditional independence
assumptions, PSM, unlike OLS, does not need a parametric model and therefore allows
mean impacts to be estimated without the arbitrary assumptions of functional forms and
error distributions (Jalan and Ravallion, 2003). The instrumental variable estimator (IVE)
could also have been used, but it also requires the conditional independence assumption,
23
which cannot be tested. The IVE requires an exclusion restriction, which is not satisfied by
using a single cross-sectional data set but rather requires longitudinal data (Jalan and
Ravallion, 2003). PSM confines its attention to matched sub-samples by dropping
unmatched comparison units from the analysis and therefore differs from regression
methods, which requires the use of the full sample. Impact estimation using the full sample
can lead to more biased results and is less robust for specifying the regression function
(Rubin and Thomas, 2000).
The PSM technique is increasingly used as a tool for program evaluation (Caliendo and
Kopeinig, 2008; Heinrich, Maffioli, and Vazquez, 2010). This technique matches individuals
in the treatment group with “identical” individuals in the control group based on observable
characteristics. Then, to determine treatment effects, participating households are matched
with non-participating households with similar “propensity scores” using some weights. A
propensity score is the conditional probability of being assigned to a specific treatment
given a set of observed covariates. In this paper, a probit regression model was used to
estimate propensity scores.
The treatment here is
𝑦𝑖 = 𝑏𝑚𝑑𝑎𝑢𝑠𝑒 = {
1 𝑖𝑓 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠 𝑢𝑠𝑒𝑠 𝑝𝑖𝑝𝑒𝑑 𝑤𝑎𝑡𝑒𝑟 𝑓𝑟𝑜𝑚 𝐵𝑀𝐷𝐴
0 𝑖𝑓 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑑𝑜𝑒𝑠𝑛′ 𝑡 𝑢𝑠𝑒 𝑝𝑖𝑝𝑒𝑑 𝑤𝑎𝑡𝑒𝑟 𝑓𝑟𝑜𝑚 𝐵𝑀𝐷𝐴
Here, the outcome of using piped water is denoted by y1 and the outcome of not using of
piped water (bmdause0) by y0 . The impact can be observed in the average treatment
effect on the treated (ATT), which is defined as
ATT E( y1 yo |bmdause1)E( y1|bmdause1)E( y0 |bmdause0) …….. (1)
The first term of Eq. (1) is observable, whereas the second term is non-observable because
it is impossible to consider an individual to be a recipient and non-recipient simultaneously.
A comparison group with similar observable characteristics can be created using PSM to
eliminate this problem when estimating the ATT.
The Stata command “pscore” (Becker and Ichino, 2002) was used to estimate the propensity
score. Table 3 shows the households characteristics and other covariates that were
considered. The first step was to estimate the propensity score so that it satisfied the
balancing property: this program generates five blocks of observations, ensuring that the
mean propensity score of the treatment group and the control group are the same in each
blocks. The balancing property was satisfied in the program, and this guarantees that the
treatment group and the control group had balanced (similar) covariates within the five
blocks. Stata identified the region of common support from the estimated propensity scores
of the two groups, ensuring that any combinations of observed characteristics among the
treatment households can also be found among the control households (Caliendo and
24
Kopeinig, 2008). The region of common support was determined by the program to be
[0.11256162, 0.89771079], which was the common area of the estimated propensity scores
of treatment and control group. Within the region of common support, the estimated
propensity score of the treatment group ranged from a minimum of 0.1125616 to a
maximum of 0.8977108, and the propensity scores of the control group ranged from
0.1145692 to 0.7893264 (Figure 11).
Control group
Treatment group
Figure 11: Estimated propensity score for treatment and control groups
After estimating the propensity scores, three types of matching were used to evaluate the
impact of piped water on different outcomes. The types of matching used were nearestneighbor matching, stratification matching and kernel matching. The different types of
matching have different advantages. Further, the regression-based nearest-neighbor
matching was also implemented to check the robustness of the results. The Stata command
“nnmatch” (Abadie, Drukker, Herr, and Imbens, 2004) was used for the analysis. In nearestneighbor matching, a household in the control group was chosen as a matching partner for a
household in the treatment group based on their closest propensity score. Matching “with
replacement” was used here to reduce the biasness and increase the average quality of the
matching estimator (Smith and Todd, 2005). Stratification matching works by partitioning
the common support of the propensity scores into a set of strata and calculating their
impact in each strata (P. Rosenbaum and Rubin, 1983). Kernel matching, on the other hand,
is a non-parametric matching estimator that uses the weighted average of all households in
the control group to construct the counterfactual outcome. The advantage here is the use of
more variance as a result of using more information (Caliendo & Kopeinig, 2008).
25
PSM eliminates only biasness in the treatment effect resulting from observable
heterogeneity. One may argue that unobserved heterogeneity (hidden bias) could also
impact the treatment effect, and thus matching estimators are not robust enough against
this hidden bias (Caliendo and Kopeinig, 2008). Since the magnitude of selection bias is
impossible to estimate, the sensitivity analysis proposed in “Rosenbaum Bounds” (P. R.
Rosenbaum, 2002) had to be implemented. The method shows how strongly unobserved
variables might affect the selection process and undermine the implication of a matching
analysis. If an outcome of interest is found to be non-sensitive, the results produced from
matching estimates would suffice for impact evaluation. For continuous outcome variables,
the Stata command “rbounds” (DiPrete and Gangl, 2004) was used, and for the binary
outcome variables, the Stata command “mhbounds” (Becker and Caliendo, 2007) was used,
which is based on the bounds by Mantel and Haenszel (1959).
26
5. Impact estimates
The impact of piped water access on the rural households were analyzed in terms of watersanitation quality, hygiene practices and child health. The results are based on two types of
analyzing units: a household-level analysis (Table 5) and an individual-level analysis (Table
6). The PSM for these two categories was first done using different matching techniques
(nearest-neighbor, stratification and kernel matching) and subsequently using a regression
technique based on the nearest-neighbor method.
The impact of the treatment is shown for each outcome of importance in the three
categories (water-sanitation facilities, hygiene behavior and health outcomes) in the ATT
and coefficient columns in Table 5 and Table 6. As mentioned earlier, the impact was
estimated using three different matching algorithms and subsequently a regression-based
technique as a robustness check. For most of the results discussed below, the estimates
were rather robust across the estimation techniques, as evident in the number of
statistically significant impacts.
5.1 Impact on water quality, water access and cooking utensils
In this category of variables, there are two direct determinants of health: the
microbiological quality (E. coli or general coliforms) of (1) drinking water and (2) kitchen
utensils. The results obtained using all three estimation techniques showed that using piped
water did not have any statistically significant impact on these two indicators.
Water quality was not improved due to piped water use. The E. coli counts in 100 ml of
drinking water found in the treated households were not significantly lower than those
found in the control households. The results produced by the three types of matching
techniques were almost consistent. The p-value of the difference between the drinking
water E. coli counts was not significant. The same was also true for the coliform counts in
drinking water. This means that there was no significant difference between the
microbiological quality of drinking water samples obtained from households with piped
water and those without piped water. The microbiological quality of kitchen utensils was
not significantly improved by using piped water (Table 5). The results obtained using the
regression-based matching technique were also consistent in terms of all water- and foodquality variables; the regression-based matching technique yielded insignificant p-values.
The only statistically significant variable was the access to improved water. The matchingbased methods as well as regression-based method yielded statistically significant
difference in access to improved water. Access to piped water increased percentage of
households with access to improved water by 5% (at the 0.01 significance level) in both the
stratification matching and kernel matching methods. In the regression-based model, the
27
increase in access to improved water was estimated to be 6 percent at the 0.05 significance
level.
An important and direct benefit of using piped water is time saving. The three PSM
techniques yielded similar estimated time savings. With access to piped water, a treated
household could save approximately 6 minutes (nearest-neighbor matching), 6.7 minutes
(stratification matching) or 7 minutes (kernel matching) daily when fetching water (Table 5);
the results were all statistically significant. Using the regression-based technique, a
household was estimated to save 9 minutes a day because of access to piped water.
Although the time saving was not substantial, access to piped water also had other
implications on the treated households, such as shorter distances to water collection point
and hygiene issues. The treated households needed to travel shorter distances than the
control households to their nearest drinking water source. A treated household had to
travel approximately 0.6 m less than a control household to collect water, depending on the
matching technique used for estimation; the results were statistically significant. The
regression-based technique estimated that a household had to travel 0.5 m less than a
control household to collect water.
Access to piped water also significantly increased water storage capacity. Households in the
treatment group generally used more containers for storing water than those in the control
group. It was found that access to piped water increased a household’s water storage
capacity by between 7.8 and 8.7 liters, depending on the matching technique used for
evaluation; the results were all statistically significant. The treated households’ tendency to
store more water could be a result of water rationing in the areas with access to pipedwater network. Similar results were obtained using the regression-based model; a treated
household’s water storage is about 8.6 liters larger than that of a control household (Table
5). Households with access to piped water also paid more for water than those without
access to piped water. Using the three matching techniques, it was estimated that a treated
household paid between BDT 615 and BDT 630 more for their water consumption (Table 5).
The regression-based model showed a similar result (BDT 633); the result is statistically
significant at the 0.01 level (Table 5). Sanitation was also an important aspect. It was
observed that access to piped water did not significantly increase the access to improved
sanitation among the treatment group. It was expected that piped water would positively
influence access to improved sanitation, especially in terms of access to flush toilets or
other improved toilets.
5.2 Impact on hygiene practices
Water and hygiene are very much related because improved water is essential to proper
hygiene practices. Therefore, the study also looked at how access to piped water changes
28
hygiene behavior. It was found that access to piped water did not significantly increased the
likelihood of a household practicing proper handwashing (i.e., with soap) after using toilet
and before child feeding. The different matching techniques gave different estimates of the
improvement in handwashing practices before child feeding where only kernel matching
estimated a 3.5% improvement at 10% significance level (Table 5). This result is not
consistent with the other estimated results and we can’t say that the handwashing with
soap before feeding child has been improved due to piped water use.
In terms of other hygiene indicators, such as cleaning water container with soap and
monthly soap consumption, the treatment group and the control group did not show any
statistically significant difference in all types of analyses. The only exception was the
difference between the monthly soap consumption of a treated household and that of a
control household as estimated using the kernel matching method.
5.3 Impact on health
In this study, impact on health was measured in terms of child diarrhea incidence, the cost
of treating children and the cost of treating adults. Unlike the study by Klasen et al. (2012),
none of the analyses found significant difference between the treatment group and the
control group in terms of child diarrhea incidence; that is, the t-statistics and the p-value
were not statistically significant (Table 5). The same was true for the cost of treating
children and adults. Access to piped water did not significantly reduce the amount of money
spent on treating illnesses in children and adults.
5.4 Impact on child growth
Child growth is a measure of the nutritional status of under-five children. Improving child
growth is an important goal of providing piped water to rural households. Access to piped
water did not have any significant impact on any of the child growth indicators measured.
The difference in height-for-age z-score, weight-for-age z-score and weight-for-height zscore between the treatment group and the control group were not statistically significant
in our analysis. Even following the WHO z-score classification system, it was found that no
statistically significant difference was observed between the treatment group and control
group in the categories of child growth indicators (such as stunting, severe stunting,
severely underweight, wasting, severe wasting).
29
5.5 Sensitivity analysis
A sensitivity analysis is performed to check how strongly unobserved variables affect the
selection process either causing an under- or overestimation of the matching results. In our
sensitivity analysis, the p-value was not exactly the same as the matching results and as the
different matching pairs because of outliers. The sensitivity analysis was conducted using
Rosenbaum bounds. The results for continuous and binary variables are reported
separately. For the continuous outcome variables, only the significant variables from
matching results are shown in Table 7. The analysis was based on the assumptions that (1)
there was no unobserved confounder due to selection bias and (2) all relevant
characteristics were matched so that the treatment group and the control group both had
the same basis for analysis. When gamma equals one and the p-value is significant, it implies
that there is no hidden bias due to unobserved confounder. For the variable water
collection time, the upper and lower bounds remained equal. But if the gamma is increased
to two (i.e., if the odds of a household being in the piped water program are doubled
because of different values of unobserved factors), despite being identical in the matched
covariates, the inference in the upper bound remains significant but the lower bound fails to
hold it significance level. So if the gamma is doubled, some unobserved factors may have
affected the impact.
The result obtained by calculating with different gamma values shows the level of sensitivity
of the produced results, but it does not imply that unobserved heterogeneity exists and
there is no effect of treatment on the outcome variables (Becker and Caliendo, 2007). The
result only show the confidence interval of the treatment effect would include zero if the
odds ratio of the treatment assignment differs between the treatment and control groups
by the gamma value. One should be cautious while interpreting the result of the matching
analysis and the sensitivity analysis. For example, water collection time shows if gamma is
equal to one then the result is significant but if the gamma is doubled then it losses its
significance level. The Hodges-Lehman Point estimate supports the result of significance
level which shows both the upper and lower bounds changes its sign if gamma is doubled,
which means that the result is sensitive when the odds are doubled. The sensitivity analysis
of the variables “distance of drinking water source” and “drinking water container capacity”
followed the same trend as the analysis of the variable “time to collect drinking water”. On
the other hand, the sensitivity analysis of the variable “water cost” indicated that the
matching results were not sensitive to unobserved factors or that the variable was not
affected by hidden bias. For the individual-level analysis of child anthropometrics (Table 9),
the result of its sensitivity analysis have to be interpreted in a similar way. It is notable that
the lower bounds of the two variables “weight-for-age z-scores” and “weight-for-height zscore” couldn’t hold significance level when gamma value was increased to two.
30
For the sensitivity analysis of the binary variables, Mantel-Haenszel statistics is shown with
its significance levels in Table 8 and Table 10. The variable “access to improved drinking
water” was explicitly sensitive when the gamma value was doubled but underestimated.
This shows that there might some unobserved heterogeneity or hidden bias for this
variable. The impact on the variable “handwashing with soap before feeding” became
insignificant when gamma was one; as gamma increased to 1.2, the impact became
insignificant when gamma is 1.7 and at 2 it is still significant but underestimated. This
variable was non-sensitive in the beginning but became sensitive at higher gamma (Table 8).
For the variable “percentage of underweight children” (Table 10), the result was significant
at all values of gamma. The result of sensitivity does not necessarily mean that there is no
treatment impact on this variable, but rather it shows that the result becomes sensitive at
different values of gamma. So one should be cautious when interpreting the results of the
sensitivity analysis and its relation to matching results.
31
6. Discussion of results and policy implications
Groundwater is the only source of potable water in Bangladesh. In north-western
Bangladesh, groundwater is becoming scarcer as it is depleting at a high rate (Figure 1).
Many households use piped water supplied by the BMDA, which charges the households a
nominal fee per month for the service. Other sources of public piped water obtained from
deep tube wells are community tube well and private tube wells. Many households that use
piped water from the BMDA complained that the BMDA rations water. Households get
discontinuous water supply and hence store water for later use. The water supplied by the
BMDA is generally clean and originates from deep tube wells. However, piped-water
rationing may encourage households to practice unhygienic water handling and storage. The
data in this study did not capture the frequency and amount of water rationing.
The study results suggest that supplying piped water to the marginalized communities as a
form of public intervention could improve access to improved water and reduce the time a
household spent collecting water, but it could not guarantee water quality at the point of
use. Similar results were also found by Devoto et al. (2012). The level of E. coli in drinking
water was not significantly improved by having access to piped water. Similarly, the
microbiological quality of kitchen utensils also did not improve with access to piped water;
there was no significant difference the level of E. coli on kitchen utensils between the
treatment group and the control group. Therefore, access to piped water by itself cannot
ensure good microbiological quality in water and on kitchen utensil. Piped water needs to
be treated before it is consumed or used for washing kitchen utensils. For example, boiling
or filtering piped water can reduce the level of E. coli in water. The knowledge of proper
hygiene practices needs to be improved in the rural households to ensure that their drinking
water and kitchen utensils are safe for use. This study also found that the treated
households tended to store more water than the control households. Because improper
water storage may offer a conducive environment for bacterial growth, proper handling of
water storage containers and regularly cleaning the container with soap may help reduce
the risk of water contamination. However, this is not the main focus of the paper.
The risk of child diarrhea and other waterborne diseases could be reduced by ensuring that
water and food are safe for consumption. But the study showed that under-five children in a
household with access to piped water generally didn’t have better weight-for-age and
weight-for-height z-scores than their counterpart in a household without access to piped
water. This finding contradicts the study by Briscoe et al. (1986). The percentage of
underweight children in the treated group was also lower than that in the control group.
Under-five children in the treated group were also less likely to be underweight than their
counterpart in the control group. However, this result is only significant at 10% level in
Kernel matching, other matchings do not show any significant results which implies the lack
32
of consistency. However, access to piped water did not offer any advantage in terms of
increasing access to improved sanitation, improving the microbiological quality of water and
kitchen utensils, improving handwashing practices after defecation, lowering child diarrhea
incidence, decrease the cost of treating illness and, more importantly, reducing the
prevalence of stunting and wasting in under-five children. This paper also investigated other
possible gains from having access to piped water, such as the quantity of water use, the
amount of leisure time, the number of working day lost, and school absenteeism. However,
no statistically significant difference was found in any of these variables. Also, the data for
these variables are not available for all observations, which restricted the analysis to only
some of the outcome variables.
33
7. Conclusion
Access to piped water generated a positive impact on access to improved water and
significantly reduced the amount of time a household spent collecting drinking water.
However, access to piped water by itself could not ensure adequate drinking water quality
at the point of use because the treatment households tended to store piped water in
reaction to discontinuous water supply. Using the level of E. coli and coliforms as measures
of microbiological quality, the study found that access to piped water did not have any
significant impact on the microbiological quality of drinking water and of kitchen utensils.
Therefore, proper household hygiene practices and good drinking water supply
management are vital for maintaining drinking water quality at the point of use. This raises
the question of how much piped water does a household need to be able to stop using
water from unsafe sources and therefore improve their food and water hygiene.
Unfortunately, the data collected for study is inadequate for addressing this particular issue.
Hygiene practices among household members did not get improved in the treatment group.
Washing hands before feeding child and after defecation are not significantly different in
the treated and control group. Monthly soap consumption among the treated households
remained low and was not significantly improved compared to the control households. This
hints that the root cause of contaminated household drinking water may be improper
hygiene practices. Further, dirty water storage containers may have also contributed to
unsafe point-of-use drinking water. A water storage container may be improperly cleaned
because of its design. For example, a container may have an open mouth, allowing water to
easily contaminated, or be too narrow to be properly cleaned. However, the study data
does not allow us to explore this issue further. The results also showed that a treated
household tended to have a larger water storage capacity than a control household, which
makes proper cleaning of water containers even more important. Although a piped-water
connection does not ensure good water quality, households still have to pay for the piped
water. As a result, a treated household spent significantly more on water services than a
non-treated household.
Access to piped water did not bring about any significant immediate impact on health.
Diarrhea incidence in under-five children was not significantly reduced by having access to
piped water. Also, the cost of treating illnesses in adults and children was not significantly
lower in the treatment group. The short-term health impact of piped water may also
manifest itself in fewer sick days and fewer days of school absence. However, no significant
difference was found in these two variables between the treatment and control groups. This
could have been caused by the following reasons: First, the data on the number of days
under-five children were absent from school because of water- and sanitation-related
diseases were limited as children start school at the age of five or six. Second, the data on
34
the number of sick days taken by adults due to waterborne diseases were limited because of
its low prevalence. This paper also found no significant changes in the number of working
hours and the amount of leisure time between the treatment and control groups. The time
saved by not having to travel to a distant water source was not reflected in an increase in
leisure time. Hence, this paper adhere strictly to analyzing daily time spent by a household
on collecting water.
We also observed the long-term health impact of piped water in child anthropometrics. It
was found that under-five children in the treatment group had similar anthropometric
measures, for example- weight-for-age, weight-for-height and height-for-age z-scores than
their counterpart in the control group. These observations indicated that access to piped
water couldn’t improve the long-time development of under-five children, which is the
expected outcome of a water-sanitation intervention. Similarly, the two groups did not
show any significant difference in terms of the prevalence of stunting, underweight and
wasting among under-five children. However, both type of measurements show the similar
results.
Overall, the BMDA piped water project has been a success because the state supplies water
to some marginalized households in rural areas, where water availability is low. Access to
piped water generated much benefit, such as improving access to improved water,
decreasing the amount of time spent on collecting water, decreasing the distance to a
drinking water source. Despite not having a significant impact on health outcomes, the
piped water network has brought about significant water infrastructure, and therefore we
recommend that the government should expand the piped water network to other
marginalized communities.
35
Table 5: Impact of access to BMDA piped water on different outcome variables based on Propensity Score Matching
Outcome variables
Water-Sanitation facilities
Access to improved sanitation
Access to improved drinking-water
Time to collect drinking water (min/day)
100ml drinking water E.Coli count (cfu)
100ml drinking water Coliform count (cfu)
E.Coli count in the food utensils (cfu)
Coliform count in the food utensils (cfu)
Distance of drinking water source (meter)
Drinking water container capacity (liter)
Water cost (BDT)
Hygiene situation
Hand wash with soap after toilet (%)
Hand wash with soap before feeding child
Clean water container with soap
Total soap consumption per month
Health outcomes
Child diarrhoea in last one month (age<59months)
Cost for illness for adults (Thousand BDT)
Cost for illness for children (Thousand BDT)
Nearest-Neighbour
Matchingb
(Treatment=186;
Control=116)
ATT
SE
Stratification
Matching
(Treatment =183;
Control =328)
ATT
SE
ATT
SE
Coefficient
SE
0.065
0.027
-5.89***
1.94
98.21
-43.55
-32.175
-0.645**
7.82*
630.6***
0.06
0.03
2.02
33.35
47.37
22.09
25.1
0.16
3.73
40.42
0.027
0.05***
-6.73***
2.18
30.73
-12.5
-17.44
-0.56***
8.7**
615.03***
0.04
0.02
1.56
17.6
41.83
13.11
17.97
0.12
3.91
41.71
0.03
0.048***
-6.931***
-0.251
24.53
-14.27
-20.18
-0.57***
8.36**
628.12***
0.04
0.01
1.76
18.14
43.92
13.31
16.43
0.11
3.52
43.11
-0.00
0.06**
-9.35***
-25.12
-23.64
2.61
-19.3
-0.45***
8.55*
632.60***
0.06
0.02
2.21
23.26
52.03
17.35
25.44
0.14
4.25
50.36
0.097
0.038
0.075
0.21
0.06
0.03
0.06
0.14
0.049
0.035
0.056
0.224
0.04
0.02
0.05
0.12
0.053
0.035*
0.058
0.253*
0.04
0.02
0.04
0.11
0.06
0.03
-0.03
0.24
0.05
0.02
0.06
0.15
-0.011
-0.109
0.041
0.03
1.35
0.12
-0.006
-0.786
0.035
0.03
0.79
0.14
-0.024
-0.104
0.072
0.03
1.22
0.12
-0.00
0.00
0.06
0.04
1.18
0.18
Kernel Matchingb
(Treatment =186;
Control =325)
Regression based
nearest-neighboring
matching (N=512)
b
Source: Authors’ calculation. represent Bootstrapping 50 times. Matching variables are: Household savings, per capita expenditure, number of livestock, number
of cow, number of goat, number of poultry, total land, wage earning households, agricultural household, non-agricultural household, age of household head,
household size, electricity, distance from road, distance from small market, distance from big market, distance from health center, distance from town.
Note: *** p<0.01, ** p<0.05, * p<0.1
36
Table 6: Impact of access to piped water on child growth based on Propensity Score Matching
Child health outcome
Height-for-age z-score
Weight-for-age z-score
Weight-for-height z-score
Stunted (dummy)
Severely Stunted (dummy)
Underweight (dummy)
Severely underweight (dummy)
Wasted (dummy)
Severely wasted (dummy)
Nearest-Neighbour
Matchingb
(Treatment=207;
Control=139)
ATT
SE
-0.010
0.164
0.138
0.117
0.184
0.170
0.006
0.068
0.010
0.034
-0.053
0.050
0.010
0.030
-0.012
0.037
0.014
0.014
Stratification
Matchingb
(Treatment=207;
Control=356)
ATT
SE
-0.077
0.102
0.083
0.103
0.173
0.113
0.015
0.039
0.013
0.025
-0.065
0.050
0.003
0.029
-0.018
0.030
0.007
0.010
Kernel Matchingb
(Treatment=207;
Control=356)
ATT
-0.076
0.102
0.195
0.003
0.007
-0.071*
0.001
-0.022
0.004
SE
0.100
0.100
0.109
0.043
0.028
0.037
0.021
0.030
0.012
Regression based
nearest-neighboring
matching (N=569)
Coefficient
-0.04
0.12
0.21
0.01
0.03
-0.07
0.01
-0.07
0.01
SE
0.15
0.13
0.13
0.06
0.04
0.05
0.03
0.04
0.01
b
Source: Authors’ calculation. represent Bootstrapping 50 times. Matching variables are: Household savings, per capita expenditure, number of
livestock, number of cow, number of goat, number of poultry, total land, wage earning households, agricultural household, non-agricultural
household, age of household head, household size, electricity, distance from road, distance from small market, distance from big market, distance
from health center, distance from town. Note: *** p<0.01, ** p<0.05, * p<0.1
37
Table 7: Sensitivity analysis- Rosenbaum bounds for continuous variables (only significant variables from the matchings are shown)
Outcome
Time to collect drinking
water (min/day)
Distance of drinking water
source (meter)
Drinking water container
capacity (liter)
Water cost (BDT)
Gamma
1
2
3
1
2
3
1
2
3
1
2
3
*
Matched
pairs
183
183
183
183
Significance level
Upper
bounds
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0007
0.7557
0.9985
0.0000
0.0000
0.0000
Lower
bounds
0.0000
0.4036
0.9689
0.0000
0.0337
0.5150
0.0007
0.0000
0.0000
0.0000
0.0000
0.0000
Hodges-Lehman Point
estimate
Upper
Lower
bounds
bounds
-5.0000
-5.0000
-10.0000
0.0000
-13.5000
2.5000
-0.5000
-0.5000
-1.0000
0.0000
-1.0000
0.0000
3.5000
3.5000
-1.0000
8.0000
-2.5000
11.0000
528.0000
528.0000
432.0000
648.0000
390.0000
726.0000
95% Confidence interval
Upper
bounds
-7.0000
-13.0000
-17.0000
-1.0000
-1.0000
-1.5000
1.5000
-2.5000
-5.0000
480.0000
396.0000
354.0000
Lower bounds
-2.0000
2.0000
6.0000
-0.5000
0.0000
0.5000
5.5000
11.0000
14.5000
582.0000
720.0000
810.0000
Source: Authors’ calculation
* gamma is the log odds of differential assignment due to unobserved factors
38
Table 8: Sensitivity analysis- Rosenbaum bounds for binary variables (only significant variables from the
matchings are shown)
Outcome variables
Access to improved
drinking-water
(dummy)
Hand wash with soap
before feeding child
(dummy)
Gamma*
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
Mantel-Haenszel statistic
Overestimation
Underestimation
1.525
1.525
1.404
1.654
1.294
1.772
1.195
1.884
1.105
1.990
1.022
2.091
0.946
2.188
0.876
2.281
0.810
2.370
0.748
2.456
0.690
2.540
2.053
2.053
1.896
2.221
1.753
2.375
1.625
2.520
1.507
2.658
1.400
2.789
1.301
2.914
1.209
3.035
1.123
3.151
1.043
3.262
0.968
3.370
significance level
Overestimation
Underestimation
0.064
0.064
0.080
0.049
0.098
0.038
0.116
0.030
0.135
0.023
0.153
0.018
0.172
0.014
0.191
0.011
0.209
0.009
0.227
0.007
0.245
0.006
0.020
0.020
0.029
0.013
0.040
0.009
0.052
0.006
0.066
0.004
0.081
0.003
0.097
0.002
0.113
0.001
0.131
0.001
0.148
0.001
0.167
0.000
Source: Authors’ calculation
* Odds of differential assignment due to unobserved factors
39
Table 9: Sensitivity analysis- Rosenbaum bounds for continuous variables (only significant
variables from the matchings are shown here)
Outcome Gamma*
Weightfor-age
z-score
Weightforheight zscore
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
Matched
pairs
205
205
Significance level
Upper
bounds
0.048
0.142
0.296
0.483
0.660
0.799
0.892
0.946
0.975
0.989
0.996
0.022
0.078
0.189
0.348
0.525
0.686
0.811
0.895
0.946
0.974
0.988
Lower
bounds
0.048
0.012
0.003
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.022
0.005
0.001
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
Hodges-Lehman Point
estimate
Upper
Lower
bounds
bounds
0.160
0.160
0.105
0.225
0.050
0.270
0.005
0.315
-0.035
0.365
-0.080
0.400
-0.115
0.440
-0.150
0.475
-0.180
0.505
-0.215
0.535
-0.245
0.570
0.200
0.200
0.140
0.260
0.090
0.320
0.040
0.375
-0.010
0.430
-0.055
0.475
-0.095
0.515
-0.135
0.555
-0.175
0.595
-0.205
0.635
-0.240
0.665
95% Confidence interval
Upper
bounds
-0.025
-0.085
-0.140
-0.180
-0.235
-0.275
-0.310
-0.350
-0.390
-0.420
-0.450
0.005
-0.060
-0.120
-0.175
-0.225
-0.275
-0.320
-0.365
-0.410
-0.445
-0.475
Lower
bounds
0.355
0.405
0.460
0.510
0.555
0.600
0.640
0.670
0.710
0.745
0.775
0.415
0.480
0.540
0.595
0.650
0.695
0.745
0.780
0.820
0.860
0.900
Source: Authors’ calculation
* gamma is the log odds of differential assignment due to unobserved factors
40
Table 10: Sensitivity analysis- Rosenbaum bounds for binary variables (only significant variables
from the matchings are shown here)
Outcome variables
Underweight children
(dummy)
Gamma*
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
Mantel-Haenszel statistic
significance level
Overestimation Underestimation Overestimation Underestimation
1.310
1.310
0.095
0.095
1.748
0.877
0.040
0.190
2.148
0.480
0.016
0.316
2.517
0.115
0.006
0.454
2.860
0.003
0.002
0.499
3.181
0.317
0.001
0.376
3.482
0.611
0.000
0.271
3.766
0.887
0.000
0.188
4.035
1.147
0.000
0.126
4.290
1.394
0.000
0.082
4.534
1.629
0.000
0.052
Source: Authors’ calculation
* Odds of differential assignment due to unobserved factors
41
References
Abadie, A., Drukker, D., Herr, J., & Imbens, G. (2004). Implementing matching estimators for
average treatment effects in Stata. Stata Journal, (1), 1–18. Retrieved from
http://scholar.harvard.edu/files/imbens/files/sjpdf-1.html_.pdf
Anderson, A., & Harris, E. (2005). An Introduction to Theory of Change. The Evaluation
Exchange (Produced by Harvard Family Research Project), XI(2), 12.
Becker, S., & Caliendo, M. (2007). Sensitivity analysis for average treatment effects. The
Stata Journal, 7(1), 71–83. doi:The Stata Journal
Becker, S., & Ichino, a. (2002). Estimation of average treatment effects based on propensity
scores.
Stata
Journal,
2(4),
358–377.
Retrieved
from
http://www.labor-
torino.it/pdf_doc/ichino.pdf
Briscoe, J., Feachem, R. G., & Rahaman, M. (1986). Evaluating Health Impact: water supply,
sanitation, and hygiene education. Ottawa, Ontario: IDRC, ICDDR,B and London School
of Hygiene and Tropical Medicine.
Caliendo, M., & Kopeinig, S. (2008). Some practical guidance for the implementation of
propensity
score
matching.
Journal
of
Economic
Surveys,
22(1),
31–72.
doi:10.1111/j.1467-6419.2007.00527.x
Curtis, V., Schmidt, W., Luby, S., Florez, R., Touré, O., & Biran, A. (2011). Hygiene: new hopes,
new horizons. The Lancet Infectious Diseases, 11(4), 312–21. doi:10.1016/S14733099(10)70224-3
Devoto, F., Duflo, E., Dupas, P., Parienté, W., & Pons, V. (2012). Happiness on tap: Piped
water adoption in urban Morocco. American Economic Journal: Economic Policy, 4(4),
68–99. doi:10.1257/pol.4.4.68
DiPrete, T. a, & Gangl, M. (2004). Assessing bias in the estimation of causal effects:
Rosenbaum bounds on matching estimators and instrumental variables estimation with
imperfect instruments. Sociological Methodology, 34, 271–310. doi:10.1111/j.00811750.2004.00154.x
Esrey, S. A. (1996). Water , Waste , and Well-Being : A Multicountry Study. American Journal
of Epidemiology, 143(6), 608–623.
Esrey, S. A., Potash, J. B., Roberts, L., & Shiff, C. (1991a). Effects of improved water supply
and sanitation on ascariasis, diarrhoea, dracunculiasis, hookworm infection,
42
schistosomiasis and trachoma. Bulletin of the World Health Organization, 69(5), 609–
621.
Esrey, S. A., Potash, J. B., Roberts, L., & Shiff, C. (1991b). Effects of improved water supply
and sanitation on ascariasis, diarrhoea, dracunculiasis, hookworm infection,
schistosomiasis, and trachoma. Bulletin of the World Health Organization, 69(5), 609–
621.
European Commission. (2013). Investing in health (No. SWD(2013) 43 Final). Brussels.
Fewtrell, L., Kaufmann, R. B., Kay, D., Enanoria, W., Haller, L., & Colford, J. M. (2005a). Water,
sanitation, and hygiene interventions to reduce diarrhoea in less developed countries: a
systematic review and meta-analysis. The Lancet Infectious Diseases, 5(1), 42–52.
doi:10.1016/S1473-3099(04)01253-8
Fewtrell, L., Kaufmann, R. B., Kay, D., Enanoria, W., Haller, L., & Colford, J. M. (2005b).
Water, sanitation, and hygiene interventions to reduce diarrhoea in less developed
countries: a systematic review and meta-analysis. The Lancet Infectious Diseases, 5(1),
42–52. doi:10.1016/S1473-3099(04)01253-8
Gamper-Rabindran, S., Khan, S., & Timmins, C. (2010). The impact of piped water provision
on infant mortality in Brazil: A quantile panel data approach. Journal of Development
Economics, 92(2), 188–200. doi:10.1016/j.jdeveco.2009.02.006
Gatzweiler, F. W., & Baumüller, H. (2014). Marginality—A Framework for Analyzing Causal
Complexities of Poverty. In J. Von Braun & F. W. Gatzweiler (Eds.), Marginality:
Addressing the Nexus of Poverty, Exclusion and Ecology (pp. 27–40). SpringerOpen.
GoB. (2012). The MDGs: Bangladesh progress report 2011. Dhaka.
GoB, & UNDP. (2009). Millennium development goals need assessment & costing 2009-2015
Bangladesh. Dhaka.
Guiteras, R., Levinsohn, J., & Mobarak, A. M. (2015). Encouraging sanitation investment in
the developing world: A cluster-randomized trial. Science, 348(6237), 903–906.
doi:10.1126/science.aaa0491
Heinrich, C., Maffioli, A., & Vazquez, G. (2010). A Primer for Applying Propensity-Score
Matching: Impact-Evaluation Guidelines. doi:NEP-ECM-2010-10-23
Hoque, B., Juncker, T., Sack, R., Ali, M., & Aziz, K. (1996). Sustainability of a water, sanitation
and hygiene education project in rural Bangladesh: a 5-year follow-up. Bulletin of the
World
Health
Organization,
74,
431–7.
Retrieved
from
43
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2486877&tool=pmcentrez
&rendertype=abstract
Jalan, J., & Ravallion, M. (2003). Does piped water reduce diarrhoea for children in rural
India. Journal of Econometrics, 112, 153–173.
Jalan, J., & Ravallion, M. (2003). Estimating the Benefit Incidence of an Antipoverty Program
by Propensity-Score Matching. Journal of Business & Economic Statistics, 21(March
2015), 19–30. doi:10.1198/073500102288618720
James, C. (2011). Theory of Change Review. A report commissioned by Comic Relief.
Retrieved from http://mande.co.uk/blog/wp-content/uploads/2012/03/2012-ComicRelief-Theory-of-Change-Review-FINAL.pdf
Javaid, N., Haschke, F., Pietschnig, B., Schuster, E., Huemer, C., Shebaz, A., … Secretin, M. C.
(1991). Interactions between infections, malnutrition and iron nutritional status in
Pakistani infants. A longitudinal study. Acta paediatrica Scandinavica. Supplement (Vol.
374).
Kaiser,
J.
(2015).
For
toilets,
money
matters.
Science,
348(6232),
272–272.
doi:10.1126/science.348.6232.272
Keusch, G. T., Fontaine, O., Bhargava, A., Cynthia BoschiPinto, Zulfiqar A. Bhutta, E. G.,
Rivera, J., Chow, J., … Laxminarayan, R. (2006). Diarrheal Diseases. In D. T. Jamison, J. G.
Breman, A. R. Measham, G. Alleyne, M. Claeson, D. B. Evans, … P. Musgrove (Eds.),
Disease control priorities in developing countries (2nd ed., pp. 371–387). Oxford
University Press and the World Bank.
Klasen, S., Lechtenfeld, T., Meier, K., & Rieckmann, J. (2012). Benefits trickling away: the
health impact of extending access to piped water and sanitation in urban Yemen.
Journal
of
Development
Effectiveness,
4(4),
537–565.
doi:10.1080/19439342.2012.720995
Konikow, L. F. (2011). Contribution of global groundwater depletion since 1900 to sea-level
rise. Geophysical Research Letters, 38(17). doi:10.1029/2011GL048604
Mantel, N., & Haenszel, W. (1959). Statistical Aspects of the Analysis of Data from
Retrospective Studies of Disease. Journal of the National Cancer Institute, 22(4), 719–
748.
44
Prüss, A., Kay, D., Fewtrell, L., & Bartram, J. (2002). Estimating the burden of disease from
water, sanitation, and hygiene at a global level. Environmental Health Perspectives, 110,
537–542. doi:10.1289/ehp.02110537
Prüss-Ustün, A., Bartram, J., Clasen, T., Colford, J. M., Cumming, O., Curtis, V., … Cairncross,
S. (2014). Burden of disease from inadequate water, sanitation and hygiene in low- and
middle-income settings: A retrospective analysis of data from 145 countries. Tropical
Medicine and International Health, 19(8), 894–905. doi:10.1111/tmi.12329
Rodell, M., Velicogna, I., & Famiglietti, J. S. (2009). Satellite-based estimates of groundwater
depletion in India. Nature, 460(7258), 999–1002. doi:10.1038/nature08238
Rosenbaum, P. R. (2002). Observational Studies. Observational Studies (Springer Series in
Statistics). doi:10.1016/B978-0-08-044894-7.01354-3
Rosenbaum, P., & Rubin, D. (1983). The central role of the propensity score in observational
studies for causal effects. Biometrika, 70(1), 41–55. doi:10.1093/biomet/70.1.41
Rubin, D., & Thomas, N. (2000). Combining Propensity Score Matching with Additional
Adjustments for Prognostic Covariates. Journal of the American Statistical Association,
95(450), 573–585. doi:10.2307/2669400
Shamsudduha, M., Taylor, R. G., Ahmed, K. M., & Zahid, A. (2011). The impact of intensive
groundwater abstraction on recharge to a shallow regional aquifer system: evidence
from Bangladesh. Hydrogeology Journal, 19(4), 901–916. doi:10.1007/s10040-011-07234
Shiklomanov, I. A., & Rodda, J. C. (2003). World water resources at the beginning of the
twenty-first
century.
Rivers.
Retrieved
from
http://catdir.loc.gov/catdir/samples/cam034/2002031201.pdf
Smith, J. a., & Todd, P. E. (2005). Does matching overcome LaLonde’s critique of
nonexperimental
estimators?
Journal
of
Econometrics
(Vol.
125).
doi:10.1016/j.jeconom.2004.04.011
Stern, E., Stame, N., Mayne, J., Forss, K., Davies, R., & Befani, B. (n.d.). Broadening the range
of designs and methods for impact evaluations. Report of a study commissioned by the
Department of International Development. London, UK.
Taylor, R. G. (2013). Ground water and climate change. Nature Climate Change, 3(4), 322–
329. doi:10.1038/NCLIMATE1744
45
Tsegai, D., Mcbain, F., & Tischbein, B. (2013). Water, sanitation and hygiene: The missing link
with agriculture (No. 107). Bonn. doi:ISSN 1864-6638
Unicef, & WHO. (2015). Progress on Sanitation and Drinking Water- 2015 Update and MDG
Assessment. Geneva, Switzerland: WHO Press.
United Nations. (2015). The Millennium Development Goals Report 2015. New York, NY.
Retrieved
from
https://visit.un.org/millenniumgoals/2008highlevel/pdf/MDG_Report_2008_Addendum
.pdf
van der Hoek, W., Feenstra, S. G., & Konradsen, F. (2002). Availability of irrigation water for
domestic use in Pakistan: its impact on prevalence of diarrhoea and nutritional status of
children. Journal of Health, Population, and Nutrition, 20(1), 77–84. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/12022163
Waddington, H., Snilstveit, B., White, H., & Fewtrell, L. (2009). Water , sanitation and
hygiene interventions to combat childhood diarrhoea in developing countries.
Wolf, J., Prüss-Ustün, A., Cumming, O., Bartram, J., Bonjour, S., Cairncross, S., … Higgins, J. P.
T. (2014). Systematic review: Assessing the impact of drinking water and sanitation on
diarrhoeal disease in low- and middle-income settings: systematic review and metaregression.
Tropical
Medicine
&
International
Health,
19(8),
928–942.
doi:10.1111/tmi.12331
46