Nature’s Call: Health and welfare impacts of sanitation choices in
Orissa, India1
Running title: Nature’s Call: Impacts of sanitation
Katherine L. Dickinsona, Subhrendu K. Pattanayakb*,
Jui-Chen Yangc, Sumeet R. Patild, Christine Poulosc
a
Advanced Studies Program, National Center for Atmospheric Research
b
Duke University, 126 Rubenstein Hall, Sanford School of Public Policy & Nicholas
School of the Environment, Durham, NC 27708, Phone: 919-613-9306; Fax: 919-6849940, [email protected]
c
RTI International
d
NEERMAN
* Corresponding Author
1
This study is funded by the World Bank, South Asia Environment and Sustainable Development division
(Task Managers, Kseniya Lvovsky and Priti Kumar). Dickinson acknowledges support from the Robert
Wood Johnson Health and Society Scholars program. We also thank Erin Sills, Doug Evans, Kanchan
Chopra, Prasanta Pattnaik, Anil Deolalikar, Ted Miguel, Kurt Schwabe, Peter Berman, John Mullahy,
Soma Ghosh-Moullick and Chris Timmins for their feedback on drafts of this paper. Seminar participants
at the Delhi School of Economics, University of California (Berkeley), Institute of Economic Growth
(Delhi), University of California (Riverside), University of Minnesota (International Economic
Development conference), Cornell University (Infectious Diseases in Poor Countries conference),
University of Wisconsin (Madison), Emory University and Duke University also provided helpful
discussion. The opinions reflected in this paper are the opinions of the authors, not those of their
institutions.
1
ABSTRACT
Worldwide, over 2.5 billion people lack access to basic sanitation, a situation that contributes to 2
million annual diarrhea-related child deaths and substantial morbidity. Yet rigorous evaluation of
sanitation behaviors and their health and welfare impacts are relatively rare. This paper uses a
randomized sanitation intervention to evaluate the effects of sanitation improvements on child
health and household welfare in Orissa, India. Following the sanitation campaign, households’
ownership of sanitary latrines increased by about 25%. Resulting reductions in acute illness
(diarrhea) are not statistically significant, but we observe significant improvements in children’s
nutritional status as measured by mid-upper arm circumference (MUAC). Switching from open
defecation to latrine use also resulted in substantial time cost savings and increased satisfaction in
sanitation conditions. We examine the implications of these results for costs and benefits
affecting this seemingly unglamorous and mundane household choice.
Keywords: sanitation, diarrhea, nutrition, child health, program evaluation, randomized
community trial
JEL Classification: H4 publicly provided goods; I1 health; R2 household analysis; O1
economic development; O2 development planning and policy; Q56 environment and development
DRAFT MAY 2011
Do not cite or quote without prior permission.
2
I. Introduction
About half of the world’s population lacks access to basic sanitation facilities (Watkins
2006). Lacking toilets, billions of people across the world respond to nature’s call in fields,
ditches, rivers, and roadsides – a practice called “open defecation” which contributes to the
global burden of infectious diseases (Pruss-Ustun et al. 2006; Pruss-Ustun et al. 2008). Although
there has been growing attention to sanitation within public health and development communities
(2008 was the International Year of Sanitation), the body of careful empirical research on this
subject is thin. Recent meta-analyses by Fewtrell et al. (2005), Waddington et al. (In press),
Norman et al. (In press), Cairncross et al. (2007), and Clasen et al. (2007) have been able to
identify only a few studies that examine the effect of sanitation interventions on health outcomes
in developing countries. Most of the studies that have been conducted do not provide rigorous
effect measures using control groups or robust experimental designs.
Our paper responds directly to the need for better estimates of the impact of latrines on
child health and household welfare. In particular, we draw on a randomized community-led total
sanitation campaign in Orissa, India, to study the impacts of households’ transition from open
defecation (still practiced by over 90% of the population in this state) to use of latrines. Using
data from this study, we contribute to a small but growing body of knowledge on the links
between sanitation, health, and welfare (see Altaf et al. 1994; Persson 2002; Woodruff et al.
2007; Whittington et al. 2008).
We find that diarrhea rates decreased following the sanitation intervention, but these
effects are not statistically significant at conventional levels for two main reasons. First, our
sample may be underpowered to detect a change in a binary outcome such as diarrhea. Second,
3
in the second year of our study there appears to be some exogenous factor that causes diarrhea
rates to be substantially lower than expected in both treatment and control villages, further
reducing our study’s ability to detect significantly different rates between groups. Despite these
challenges, the direction of effects is consistent with our hypothesis that improvements in
collective sanitation led to improvements in child health, and stronger evidence in support of this
hypothesis comes in the form of anthropometric measures. Using mid-upper arm circumference
as a more stable and continuous measure of nutritional status, we find statistically and
substantively significant improvements as a result of the sanitation campaign and associated
latrine use. Thus, our results indicate that increased latrine use improved health outcomes for
children in our study area.
We also observe sizeable non-health benefits for households that adopted latrines in
terms of substantial time cost savings and increased satisfaction with sanitation conditions.
Given that time cost savings alone appear to be greater than latrine construction and maintenance
costs, the benefits of latrine construction would seem to exceed the costs for households in this
area. However, the fact that latrine use remains low in the region suggests that we need further
research on the motivations for sanitation behaviors.
Section II provides a background on the sanitation context in Orissa, India, and describes
the randomized intervention that was implemented in Bhadrak district. Section III outlines our
empirical methods for estimating the effects of the sanitation campaign and subsequent sanitation
improvements on health and welfare outcomes. Section IV presents the study’s results, and
Section V synthesizes these results to compare realized costs and benefits for households that
were induced to adopt latrines following the campaign. We conclude by discussing the
4
implications of these results for households’ sanitation decisions and government sanitation
policies.
II. Sanitation Context and Study Design
In 2006, 2.5 billion people used “unimproved” sanitation facilities that did not
hygienically dispose of human waste; of these, 1.2 billion people (nearly one fifth of the world’s
population) continued to practice open defecation (WHO/UNICEF 2008). The prevalence of this
behavior is concentrated in Southern Asia. In India alone, over 650 million people (58% of the
country’s population) lack access to even the most basic sanitation facilities. Sanitation
conditions also vary substantially within India. The prevalence of open defecation in urban areas
was only 18% in 2006, compared to a startling 74% in rural areas (WHO/UNICEF 2008). In the
state of Orissa, where our empirical study took place, only about 10% of rural households had
access to a toilet facility in 2007 (Government of India 2007).
At a national scale, India established the Central Rural Sanitation Program in 1986
(DDWS 2007). Early attempts to address the sanitation challenge tackled the problem from an
engineering perspective. Supply-side policies focused on building latrines, only to find that
many households failed to use and maintain these facilities (Kar 2003; Figueroa et al. 2006). In
response to a growing recognition that effective policies would need to address demand for
sanitation as well as its supply, India restructured its efforts and launched the Total Sanitation
Campaign in 1999. This new approach was intended “to increase awareness among the rural
people and [generate] demand for sanitary facilities” (DDWS 2007, p.2).
In this spirit, one particularly promising model that emerged was the Community-Led
Total Sanitation (CLTS) approach. Originally developed and implemented in Bangladesh (Kar
5
2003), this method focused on “empowering local people to analyze the extent and risk of
environmental pollution caused by open defecation” (Kar 2003). More viscerally, one CLTS
advocate described the approach as “getting people to realize they are eating each other’s’ shit.”
Observational studies and informal reports suggested that this method could be particularly
effective in changing sanitation behaviors, and the approach spread beyond Bangladesh to Indian
states like Maharashtra (Sanan et al. 2007).
To provide a more rigorous evaluation of the effects of a CLTS-inspired intervention on
sanitation and related health and welfare outcomes, we conducted a randomized intervention in
Bhadrak district, Orissa, between 2005 and 2006. Within Orissa, we chose this district as our
study area for three reasons. First, Bhadrak still had a sufficiently large number of blocks and
villages where the Government of India’s existing Total Sanitation Campaign interventions had
not been implemented. Second, the use and maintenance of latrines in the area remained
unsatisfactory despite adequate water availability, and third, the Government of Orissa agreed
that no special water, sanitation or hygiene programs would be implemented in control villages
during the study period.
Out of about 1200 total villages in Bhadrak district, we selected 40 study villages using
the following criteria.1 First, we limited our study to two adjacent blocks, Tihidi and Chandbali,
that were accessible by road. Next, we excluded villages with less than 70 or more than 500
households to ensure that included villages would be similarly rural and would provide enough
households with at least one child under the age of five, since the main health outcomes we are
measuring include child diarrhea rates and anthropometrics. Finally, to minimize spillover
effects, we randomly selected one village per panchayat,2 mapped the remaining villages, and
removed villages that were adjacent to one another.
6
In order to assess the impact of the sanitation intervention on household sanitation
behaviors, child health outcomes, and welfare measures, we implemented a repeated measures
cohort design. We randomly selected 20 of the 40 sample villages and assigned them to the
“treatment” group, while the other 20 villages served as “controls.” In August of 2005,
enumerators listed and mapped eligible households in each village (i.e., households with at least
one child under five), and then randomly selected 28 households from each villages and
collected baseline data using a household survey. Between January and May of 2006, the
intervention took place in the 20 treatment villages, and post-intervention data were collected
from the same set of households in August and September of 2006. The 2005 baseline survey
covered 1086 households (treatment = 534, control = 552), and 1050 of these (treatment =529,
control =521) completed follow-up surveys in 2006.
Our data were collected by approximately 30 local enumerators employed by TNS, an
international survey organization. Surveys were conducted in the local language (Oriya). The
study protocol was approved by an ethics review board, an external technical oversight group
from leading public health agencies, and a local steering committee. Throughout the design stage
of this study, the evaluation team worked closely with the Government of Orissa as well as the
sanitation campaign implementation team to ensure consistency and coherence across all aspects
of the study, including integrity of the design and measurement.
Table 1 presents descriptive statistics for a number of household and village
characteristics in 2005 (prior to the intervention). In general, treatment and control villages are
fairly similar, with few significant differences in observable covariates prior to the sanitation
intervention. Qualitatively, treatment villages appear to be slightly worse off initially in terms of
a few indicators (e.g., distance from roads, ownership of TVs, and, more notably, ownership of
7
latrines – see Table 2). However, these differences are not statistically significant at
conventional levels, and the availability of pre-intervention data allows us to control for these
differences in subsequent estimation of the intervention’s effects.
The Intervention
Inspired by the Community-led Total Sanitation model, the intervention that was
implemented in Bhadrak focused squarely on the concept that targeting entire communities for
behavior change is more effective than trying to encourage latrine use on a house-by-house basis
(Pattanayak et al. 2009). This approach is motivated in part by an understanding of the
transmission pathways for diarrhea and the ways in which sanitation improvements affect these
pathways. In particular, sanitation practices in a community affect the overall environmental
burden of fecal matter and associated parasites. These parasites are then transmitted through
human consumption of both drinking water and food, affecting health outcomes. By
appropriately disposing of feces through latrine use, the level of fecal matter in a community is
reduced. Crucially, then, the collective level of sanitation in a community is a key determinant
of individual health outcomes (Alderman et al. 2003; Pattanayak et al. 2009).
The Bhadrak sanitation campaign promoted community-wide latrine adoption (i.e., an
end to open defecation) in the 20 randomly selected treatment villages through a number of
participatory activities. The purpose of these activities was to create a sense of disgust and
shame about open defecation and a desire for an immediate, village-wide end to open defecation.
One activity, called the “Walk of Shame,” consisted of a march through the village where
campaign motivators pointed out areas where people had openly defecated. In the “Fecal
Calculation” exercise, participants estimated the volume of feces generated by the village and
deposited into the community environment over a given period of time. These activities were
8
meant to call attention to the level of contamination in the village, and each activity emphasized
the collective nature of the problem: unless everyone stopped open defecation by using latrines,
everyone would continue to be exposed to fecal matter. The goal of these activities was to
induce entire villages to commit to becoming “open defecation free” by a collectively agreedupon date.
While the overarching philosophy of the campaign focused on community-level
outcomes, campaign organizers also recognized that inducing households to change their
behaviors required increasing demand for sanitation by both reducing costs and increasing
perceived benefits of latrine use. On the cost side, the campaign subsidized materials and labor
for latrine construction. The typical construction cost for the offset pit latrine promoted under
this campaign was Rs. 1500 (about US$30), of which households below the poverty line (BPL)
were only required to pay Rs. 300 (about US$6). These subsidies were intended to relax the
budget constraint faced by poorer households. Furthermore, the campaign reduced search costs
by providing technical know-how to guide household construction of latrines. The supply of
both materials and expertise was increased through production centers that were operated by
local NGOs in each village.
On the benefit side, campaign activities emphasized the links between sanitation and poor
health outcomes. In addition, the campaign placed much emphasis on the non-health benefits of
latrine use. In particular, messages about latrines’ convenience highlighted the potential time
savings households could enjoy from changing their sanitation behaviors: open defecation
generally requires walking to specific areas (fields, ditches, rivers, forests, etc.) and can thus be a
time consuming activity. Other less tangible benefits like privacy and dignity, particularly for
women, also played a prominent role in the campaign’s messages. By highlighting the
9
downsides of open defecation – including health and non-health factors – and promoting latrines
as a solution to these problems, the primary goal of CLTS to create an overwhelming sense of
dissatisfaction that would spur widespread behavior change.
III. Methods for Measuring Health and Welfare Impacts
We assess what impacts, if any, the randomized sanitation campaign and resulting
sanitation improvements had on households’ health and welfare in Bhadrak.
Sanitation Behaviors
We present results on two main outcomes of interest: whether or not each household had
a latrine in each year (binary outcome), and the proportion of household members who “usually”
or “always” use a latrine for defecation. We constructed the latter indicator using responses to
questions in the survey that asked how frequently men, women, and children used a private
latrine3 during the day and at night. We combined responses to these questions with information
on the composition of the household to construct an index of the proportion of household
members regularly using latrines.
Health and Welfare Measures
To measure the impacts of the Bhadrak campaign and resulting sanitation improvements,
we focus on two sets of indicators: 1) child health, and 2) sanitation-related welfare. Descriptive
statistics for these different indicators, by treatment and control groups, are presented in Table 2.
We measured child health impacts in terms of diarrhea prevalence and improvements in
an anthropometric indicator of nutritional status (mid-upper arm circumference). Diarrheal
10
diseases are a leading cause of death among children under five; Prüss-Üstün and Corvalán
(2006) estimate that about 1.5 million child deaths per year can be linked to water, sanitation,
and hygiene. To measure diarrhea prevalence in our sample, we collected data on whether or not
each child suffered from diarrhea in the two weeks prior to the survey. Primary caregivers were
interviewed in order to get the most accurate data on child health conditions.
Reductions in diarrheal disease burden are expected to decrease mortality as well as
morbidity in affected populations. While our study was underpowered to measure changes in
mortality, anthropometric measurements allowed us to gauge morbidity by assessing changes in
children’s nutritional status. The measure we focused on was mid-upper arm circumference
(MUAC). MUAC gauges both fat reserves and muscle mass, and (unlike height) can decrease as
well as increase in response to nutritional shocks (Alderman 2000). Low MUAC-for-age has
been shown to predict subsequent child mortality (Vella et al. 1993).
We computed z-scores for MUAC-for-age by comparing observed values for each child
with published WHO standard growth benchmarks for males and females of different ages
(WHO Multicenter Growth Reference Study Group 2007). MUAC distributions tend to be
skewed to the right, so that z-scores must be calculated using the LMS method (WHO
Multicenter Growth Reference Study Group 2007). In this case, we computed:
[
⁄ ]
where M, S, and L are the age- and gender-specific median, generalized coefficient of variation,
and the power in the Box-Cox transformation, respectively. (As recommended by the WHO
(WHO Multicenter Growth Reference Study Group 2007), there is a slight adjustment for
observed z-scores beyond ± 3 standard deviations.)
11
Finally, we also examined sanitation-related welfare indicators. While health impacts
may motivate efforts to improve sanitation from a public health perspective, latrines may also
have a range of more immediate, tangible costs and benefits for individual households
(Pattanayak et al. 2009). Measuring these impacts is important in order to understand why
household may (or may not) adopt latrines in a given setting. The two measures we analyzed
were changes in the time spent walking to defecation sites for adult household members, and
households’ subjective level of satisfaction with their sanitation situation – i.e., whether the
survey respondent was “completely dissatisfied” with her family’s sanitation facilities.
Data Analysis
To estimate the impact of the sanitation campaign in Bhadrak on each of the outcomes
outlined above, we used a difference-in-difference “intention to treat” (ITT) estimator:
1
where Yit is the household- or individual-level outcome of interest, Postt is a dummy variable that
takes a value of 1 in 2006, Treatmenti records whether the unit was in a village that received the
campaign. Because the sanitation campaign was randomly assigned, observed differences
between the two groups after the campaign can safely be interpreted as a causal treatment effect.
That is, β3 provides an estimate of the effect of the Bhadrak campaign on the outcomes of
interest. To increase the precision of our estimates, we included a set of individual, household,
and village covariates (Xit).
Most models, including those for binary outcomes (e.g., latrine ownership, diarrhea) were
run using ordinary least squares on the full panel of households (or individuals) in the sample.4
The exception is the time cost model, which was run using a Tobit specification to account for
12
the fact that this cost was censored at zero for those households that decided to build and use a
latrine. Because the treatment was applied at the village level, we clustered all standard errors at
this level as well.
ITT measures provide useful data on the effectiveness of this particular community-level
intervention in changing health and welfare outcomes. However, policymakers and researchers
may also be interested in the specific impacts of sanitation improvements on various outcomes.
In particular, establishing the relationship between community latrine use and individual health
may allow a comparison of the relative impacts of sanitation and other interventions (e.g., water
treatment) on child health. Assessing the impact of community latrine ownership on child health
is slightly more complicated than measuring the effect of the sanitation campaign because the
decisions to own latrines were made by individual households and were not randomly assigned.
It is quite possible that some of the factors that influence the extent of latrine adoption within a
village will also be correlated with health outcomes. Thus, simply regressing diarrhea or
anthropometrics on the percent of households owning latrines in each village may result in a
biased estimate of the treatment effect in this case.
Fortunately, the sanitation campaign provided an exogenous source of variation in
village-level latrine ownership across villages, thus providing us with an instrument that we
could use to measure the effect of sanitation improvements on child health outcomes (Duflo et al.
2006). We estimated these effects using two-stage least squares (2SLS). In the first stage, we
estimated the effect of the campaign on village-level latrine ownership, and then used results to
estimate a predicted level of latrine ownership for each village. In the second stage, we
estimated the effect of predicted village-level latrine ownership on child health outcomes.
13
First stage:
2
Second stage:
̂
3
All analyses were conducted using Stata 11 (StataCorp 2009).
IV. Results
Table 2 shows village-level means for each outcome in treatment and control villages
before and after the sanitation campaign. A simple difference-in-difference estimate based on
these village-level means is also presented for each outcome. We discuss each of these outcomes
in more detail below.
Sanitation Impacts
Following the sanitation campaign, Table 2 shows that levels of latrine ownership and
use increased substantially in treatment villages with very little change in the control villages.
Difference-in-difference estimates for latrine ownership are 28% and 26% in models without and
with controls for household characteristics, respectively (see Tables 2 and 3). Results suggest
that corresponding treatment effects for latrine use are 19% and 17%. Thus, the Bhadrak
sanitation campaign was indeed effective in inducing many households to build latrines. Use of
latrines also increased significantly, though results indicate that use does lag behind ownership to
some extent.
14
Child Health Impacts
Turning first to diarrhea, Table 2 shows that, overall, diarrhea rates in this sample were
quite high in 2005 with about one quarter of children experiencing a diarrhea episode in the two
weeks prior to the survey. Average diarrhea rates were also somewhat higher in treatment
villages (27%) compared to controls (23%), but the difference in these village averages is not
statistically significant. In 2006, two things happen: 1) diarrhea rates decrease substantially in
both treatment and control groups to an overall average of about 14%, and 2) the difference
between the two groups becomes much smaller, with rates in control villages becoming slightly
higher than in treatment villages.
The first three columns of Table 4 present ITT and 2SLS estimates of the impacts of
sanitation on diarrhea in Bhadrak. In particular, these models assess the impact of the Bhadrak
campaign and village-level latrine use on diarrhea among children under 5.5 In each of these
regressions, the dependent variable is a binary measure of individual diarrhea illness in the two
weeks prior to each survey. The first column presents a simple ITT effect, while the second
column adds controls for village-, household-, and individual characteristics. The third column
presents the 2SLS results for % latrine ownership at the village level. Across these different
models, we observe that the estimated effects of sanitation and latrine adoption on child diarrhea
outcomes are negative, indicating that latrine adoption is associated with decreased diarrhea
rates. However, none of these results is statistically significant at the 10% level. Thus, we are
not able to show a clear impact of the sanitation campaign or latrine use on diarrhea rates. As we
have noted elsewhere, this may be due in part to the binary nature of this outcome measure,
along with the unexplained drop in diarrhea rates across treatment and control groups in the
second year of our study.
15
Turning next to anthropometric measures, we examine mid-upper arm circumference
(MUAC). Looking at the simple difference-in-difference estimate using village-level means,
Table 2 shows that MUAC z-scores increased by about 0.17 standard deviations in treatment
villages relative to controls, but this difference is not statistically significant. We also note that
the children in our sample are quite malnourished by international standards. In fact, at every
age group, average MUAC-for-age in the sample falls below the WHO standard 15th percentile
growth curves.
Disaggregating and using individual-level data, columns 4-6 of Table 4 present ITT and
2SLS effects of the impacts of sanitation on MUAC-for-age. The ITT results indicate that
MUAC z-scores increased by 0.2 standard deviations in treatment villages relative to controls
following the sanitation campaign, and this difference is statistically significant at the 5% level.
Meanwhile, the 2SLS estimate indicates that going from 0 to 100% latrine ownership would
produce a full standard deviation increase in MUAC-for-age. At the average increase in latrine
ownership of about 30%, we would expect to see MUAC-for-age increase by about a third of a
standard deviation.
Impacts on Household Welfare
We turn finally to the impacts of the sanitation campaign on household welfare
indicators.6 Table 2 shows that the time households spent walking to defecation sites decreased
substantially following the campaign, as did levels of dissatisfaction with sanitation facilities.
These results are consistent across household-level regressions controlling for other covariates
(see Table 5). For time costs, the ITT point estimates of close to 30 minutes measure the total
16
time that households saved one way per defecation trip. Assuming each household member
makes one trip (both ways) per day, the total cost savings per treated household is 40-60 minutes.
Meanwhile, the 2SLS point estimate suggests that going from universal open defecation to 100%
latrine use would increase an average household’s one-way walking time by 99 minutes.
Respondents in treatment villages are also about 22% less likely to be “completely dissatisfied”
with their sanitation situation following the sanitation campaign.
V. Discussion
Do toilets improve health and welfare?
What do our results tell us about the impacts of the sanitation campaign and associated
changes in latrine ownership on various outcomes of interest? The basic story appears to be that
decreasing open defecation reduced fecal contamination of the community environment. These
improvements may have led to moderate reductions in diarrheal disease, and clearer
improvements in children’s nutritional status. Households also experienced time cost savings
and increased satisfaction.
Breaking this story down, a few points are worth noting. First, our results highlight the
importance of using multiple indicators of children’s health outcomes. As previously discussed,
our inability to detect substantial changes in diarrhea were due in part to the binary nature of the
diarrhea indicator, and in part to the across-the-board reductions in diarrhea that reduced the
statistical power of our study design. If diarrhea were our only measure of health outcomes, we
would thus conclude that the sanitation campaign had no discernible impact on children’s health
status. However, because we also collected data on anthropometrics, we are able to use more
stable, continuous indicators and detect improvements in z-scores for arm circumference.
17
Putting the diarrhea and anthropometric data together builds a more complete story. Our results
are consistent with the hypothesis that reductions in the overall parasite load in the population
decreased both acute disease and malnutrition among children exposed to the campaign.
Are toilets a cost-beneficial investment?
While our results indicate that toilets generate benefits, it is critical to evaluate if these
benefits exceed costs for both households who must make the choice to invest in toilets and
policy planners who are trying to influence households’ investment choices. In addition to
health effects of toilets, households also consider more immediate, private non-health costs and
benefits. Following Whittington et al. (2008), we attempt to quantify some of these payoffs to
understand households’ decisions in the Orissa setting.
Costs
We compute the monthly costs to the household of constructing and maintaining a latrine.
Under the Bhadrak campaign, the unsubsidized cost of latrine construction was 1500 rupees
(about $30). Households classified as below the poverty line (BPL) paid 300 rupees (~$6).
Following Whittington et al. (2008) we compute the monthly construction costs as the cost of
construction times the capital cost recovery factor, which is given by:
CCRF  (r (1  r ) d ) /((1  r ) d  1)
where r is the discount rate and d is the amount of time latrines will last before needing to be
replaced (6 years in our case). The choice of a discount rate is, as usual, somewhat tricky.
Whittington et al. use a 4.5% discount rate, but they analyze costs from the government or social
planner’s perspective. Because we are interested in perceived costs to the household, a
18
somewhat higher discount rate seems appropriate. Using 4.5% as a lower bound and a 20%
discount rate as an upper bound, we calculate that construction costs are between $0.48 and
$0.75 per month for an unsubsidized latrine, and between $0.13 and $0.15 for a subsidized
latrine.
Whittington et al. estimate that households spend about 10 hours per year on latrine
maintenance. For simplicity, we adjust this number slightly and assume households devote one
hour per month on latrine upkeep. The wage rate in Bhadrak is between $1 and $1.50 per day;
valuing household members’ time at 30% of the wage rate gives a monthly maintenance cost of
$0.04-$0.06 per household.7 Whittington et al.’s cost estimates also include the time household
members spend engaged in campaign activities. We omit these costs for the moment. Under
these assumptions, the total costs for latrine construction and maintenance are between $0.52 and
$0.81 per household per month for an unsubsidized latrine, and between $0.17 and $0.21 for a
subsidized latrine.
Benefits
We turn now to the benefits of latrine construction, including both health and non-health
benefits. Regarding health benefits, the typical approach is to express these in terms of cost of
illness (e.g., diarrhea) and value of statistical life (e.g., reductions in mortality related to
diarrhea). Although we do not find a statistically significant impact on diarrheal diseases, we
find improvements in terms anthropometrics (MUAC and height z-scores). There is less
precedent for monetizing these health indicators, so we set aside this issue for now. Instead, we
focus on the non-health benefits of latrine construction. By comparing these non-health effects
with the costs of building and maintaining a latrine, we can assess how large the expected health
benefits would need to be to justify latrine construction from the household’s perspective.
19
Non-health benefits of latrine construction include psychic benefits (e.g., satisfaction,
sense of privacy) as well as time savings relative to open defecation. As Pattanayak et al. (2006)
suggest, some of these benefits are harder to monetize than others (e.g., requiring contingent
valuation or similar stated preference methods) so we take the approach of monetizing what we
can and discussing implications for the magnitude of unquantified benefits.
In terms of time savings, we found that households saved approximately 60 minutes per
day in “travel cost” by switching from open defecation to latrine use. Using the same average
daily wage of $1-$1.50 and again valuing time at 30% of the wage rate, this amounts to $0.04$0.06 per household per day or $1.17-$1.75 per month. Given that such a household spends (and
earns) about $50 per month, these savings represent 2 to 3.5% of household income.
Calculus of coping
Prior to the sanitation campaign in Bhadrak, very few households had adopted latrines.
Following the campaign, we see a marked increase in latrine adoption among households in
treatment villages with no change in latrine ownership within control villages. A reasonable
conclusion would be that before the campaign, many households perceived the costs of latrine
adoption to exceed the benefits, and that the campaign achieved its success by either reducing
costs or increasing benefits (or both) enough to tip the balance in favor of adoption. Do our
results support this story?
The most striking result from our analysis is that time cost savings alone (ignoring health
and other psychic benefits) appear to exceed even the unsubsidized construction and maintenance
costs for households that adopted latrines in Bhadrak. Furthermore, these time savings should
not have been affected by the sanitation campaign – they simply represent the time households
were devoting to the practice of open defecation (valued at the pre-existing wage rate). Thus,
20
households in control villages and households who built latrines prior to the campaign could
have reaped these benefits, yet they did not. How can we explain this result?
Maintaining the standard economic assumption of rational choice based on perceived
costs and benefits, there are two main possibilities: either we have underestimated the costs of
latrine construction in the absence of the campaign, or we have overstated the perceived benefits.
We address each of these hypotheses in turn.
Turning first to costs, it is possible that the unsubsidized latrine construction cost under
the campaign (i.e., the cost paid by APL households in treatment villages) is a poor proxy for the
cost of building a latrine before the campaign or in the control villages. The sanitation campaign
increased the supply of latrine construction materials and decreased search costs for households
interested in building latrines, facilitating construction for APL and BPL households alike. In
other words, the campaign may have substantially reduced transaction costs for latrine
construction. Indeed, one way to interpret our results is to use the estimated time savings
(which, again, should not be affected by the campaign) as a lower bound on the perceived cost of
latrine construction for households in control villages. Since we estimate that the average
household could save a minimum of $1.17 per month by building a latrine, the perceived
monthly construction, maintenance, and transaction costs must have exceeded this amount for
households that did not build latrines. Keeping maintenance costs constant at $0.04/month and
assuming a 20% discount rate, this implies that the perceived construction costs (including initial
transaction costs) for the average household not exposed to the sanitation campaign exceeds $45
or 2250 rupees. This implies that even for subsidy-ineligible APL households, the Bhadrak
campaign reduced the cost of latrine construction by at least $15 or 33%, while BPL households
saw costs drop from at least $45 to $6 – an 87% reduction.
21
At the same time, we also acknowledge the possibility that we have overestimated the
benefits of latrine construction. For example, it is possible that households do not value their
time savings as highly as the Whittington et al. method assumes (30% of the wage rate).
Furthermore, while we have not valued the health and psychic benefits of latrine use, it is also
possible that there are important psychic costs associated with changing behavior. Indeed,
comments from people we interviewed in Bhadrak suggested that this may be the case for some
households. In informal interviews, the majority of people we spoke to emphasized the benefits
of latrines (such as privacy and convenience, especially for women), but there were a small
number of respondents that also emphasized the difficulties in changing behavior. One man said:
“If [open defecation] was good enough for the Maharajas, it’s good enough for me.” Some
women said that going out together in the evenings for open defecation gave them a chance to
spend time together and gossip. Another man spoke about his preference for using “Open Sky
Latrines,” rather than newer, man-made facilities. Interestingly, these same individuals often
acknowledged that using latrines could result in health benefits, but this knowledge alone was
not enough to motivate behavior change.
All of these considerations lead to the general point that while we report results in terms
of averages, it is likely that costs and benefits vary across the population distribution. That is,
while latrine adoption may look like a good decision for the marginal household, there may be a
significant proportion of the population for whom costs exceed benefits. If latrines were a purely
private good, our analysis might end here – allowing adoption up to the point where private
benefits equaled private costs would generate an efficient outcome. However, given our finding
that improvements in sanitation have health benefits that are distributed across the community,
the concern is that the public good aspect of latrines creates the potential for sub-optimal
22
sanitation levels.8 Indeed, this kind of logic likely guided planners in the United States who
delivered macro-public health interventions (e.g., sanitation, chlorination, and vaccination) that
some say help explain at least 50% of the improvements in longevity experienced in this country
over the past decade (Milgram 1967; Preston 1996). In any case, understanding the drivers of
households’ sanitation choices and applying this knowledge to the design of sanitation policies
may help close the “sanitation gap” between rich and poor countries, improving health and
welfare for some of the world’s most disadvantaged populations.
23
VI. References
Alderman, H. 2000. "Anthropometry". In Designing Household Survey Questionnaires for
Developing Countries, ed. M. Grosh and P. Glewwe. Washington, DC: The World Bank.
Alderman, H., J. Hentschel, et al. 2003. "With the help of one's neighbors: externalities in the
production of nutrition in Peru." Social Science & Medicine 56(10): 2019-2031.
Altaf, M. A. and J. A. Hughes. 1994. "Measuring the Demand for Improved Urban Sanitation
Services: Results of a Contingent Valuation Study in Ouagadougou, Burkina Faso."
Urban Studies 31(10): 1763-1776.
Barrett, C. B. and A. D. Foster. 2007. "The social dimensions of microeconomic behavior in
low-income communities: Introduction to a symposium." Journal of Development
Economics 83(2): 253-255.
Cornes, R. and T. Sandler. 1984. "Easy riders, joint production, and public goods." Economic
Journal 94(375): 580-598.
Cornes, R. and T. Sandler. 1994. "The comparative static properties of the impure public good
model." Journal of Public Economics 54(3): 403-421.
DDWS. 2007. "Guidelines on Central Rural Sanitation Programme Total Sanitation Campaign".
M. o. R. D. Department of Drinking Water Supply, Government of India.
Duflo, E., R. Glennerster, et al. 2006. "Using randomization in development economics research:
A toolkit". In Handbook of Development Economics.
Fewtrell, L., R. B. Kaufmann, et al. 2005. "Water, sanitation, and hygiene interventions to reduce
diarrhoea in less developed countries: a systematic review and meta-analysis." Lancet
Infectious Diseases 5(1): 42-52.
Figueroa, M. E. and D. L. Kincaid. 2006. "Social, cultural, and behavioral correlates of
household water treatment and storage", World Health Organization.
Government of India. 2007. "District Level Household and Facility Survey (DLHS-3): Orissa -Fact Sheet". Mumbai, India, International Institute for Population Sciences.
Kar, K. 2003. "Subsidy or self-respect? Participatory total community sanitation in Bangladesh".
IDS Working Paper Series. Sussex, England, Institute of Development Studies.
Kotchen, M. J. 2005. "Impure public goods and the comparative statics of environmentally
friendly consumption." Journal of Environmental Economics and Management 49(2):
281-300.
Milgram, S. 1967. "The small world problem." Psychology today 2(1): 60-67.
Pattanayak, S. K., C. Mansfield, et al. 2006. "Unpackaging Demand for Water Service Quality:
Evidence from Conjoint Surveys in Sri Lanka." World Bank Policy Research Working
Paper Series(3817).
24
Pattanayak, S. K. and A. Pfaff. 2009. "Behavior, environment, and health in developing
countries: Evaluation and valuation." Annual Review of Resource Economics 1: 27.2127.35.
Pattanayak, S. K., J.-C. Yang, et al. 2009. "Shame or subsidy revisted: Randomized evaluation of
social mobilization for sanitation in Orissa, India." Bulletin of the World Health
Organization 87(8): 580-587.
Pattanayak, S. K., J. C. Yang, et al. 2005. "Coping with unreliable public water supplies:
Averting expenditures by households in Kathmandu, Nepal." Water Resources Research
41(2): -.
Persson, T. H. 2002. "Welfare calculations in models of the demand for sanitation." Applied
Economics 34(12): 1509-1518.
Preston, S. H. 1996. "American longevity: past, present, and future". Syracuse University Policy
Brief No. 7.
Pruss-Ustun, A., R. Bos, et al. 2008. Safe water, better health. Geneva: World Health
Organization.
Pruss-Ustun, A. and C. F. Corvalan. 2006. "Preventing disease through healthy environments:
Toward an esitmate of the environmental burden of disease", World Health Organization.
Rohner, D. In press. "Reputation, group structure and social tensions." Journal of Development
Economics.
Sanan, D. and S. G. Moulik. 2007. "Community-Led Total Sanitation in rural areas: An approach
that works". A. S. Gupta, Water and Sanitation Program -South Asia, The World Bank.
Santos, P. and C. B. Barrett. In press. "Persistent poverty and informal credit." Journal of
Development Economics.
StataCorp. 2009. "Stata Statistical Software: Release 11. ". College Station, TX., StataCorp LP.
Vella, V., A. Tomkins, et al. 1993. "Anthropometry and childhood mortality in northwest and
southwest Uganda." American Journal of Public Health 83(11): 1616-1618.
Vicary, S. 1997. "Joint production and the private provision of public goods." Journal of Public
Economics 63(3): 429-445.
Watkins, K. 2006. "Beyond Scarcity: Power, Poverty and the Global Water Crisis. Human
Development Report 2006.". New York, NY, United Nations Development Programme.
Whittington, D., W. M. Hanemann, et al. 2008. "Copenhagen Consensus 2008: Sanitation and
Water Challenge Paper", Copenhagen Consensus Center.
WHO Multicenter Growth Reference Study Group. 2007. "WHO Child Growth Standards:
Head-circumference-for-age, arm-circumference-for-age, triceps-skinfold-for-age and
scapular-skinfold-for-age: Methods and development.". Geneva, World Health
Organization.
25
WHO/UNICEF. 2008. "Progress on Drinking Water and Sanitation: Special Focus on
Sanitation". New York and Geneva, World Health Organization and United Nations
Children's Fund Joint Monitoring Programme for Water Supply and Sanitation.
Woodruff, C. and R. Zenteno. 2007. "Migration networks and microenterprises in Mexico."
Journal of Development Economics 82(2): 509-528.
Yamauchi, F. 2007. "Social learning, neighborhood effects, and investment in human capital:
Evidence from Green-Revolution India." Journal of Development Economics 83(1): 3762.
26
1
Sample size calculations indicated that 40 villages with 25 eligible households per village would provide sufficient
statistical power (i.e. 80% or greater) to identify meaningful differences between treatment and control villages on
one of the primary outcomes, prevalence of diarrhea among children younger than five years. We assumed a base
rate of 25% and an anticipated program effect of 0.30. Moderate attrition was considered and a design effect
(DEFF) of 2.0 was introduced to account for variance inflation.
2
Gram Panchayat is the lowest level administrative unit in India consisting of a cluster of 3-6 villages.
3
The survey also asked about use of community toilets, but these were virtually non-existent in the study area.
4
We also estimated these models using other functional forms, such as probit and logistic regressions. Results were
qualitatively identical to those presented here, so we report results from the simpler linear models.
5
Given the variability in diarrhea rates by age, we also estimated models separately for different age cohorts.
Sample sizes become smaller as more narrow age ranges are examined, however, and results are qualitatively
similar to the overall under five results presented here.
6
In addition to the two welfare indicators discussed here, we initially examined two other indicators: days
unproductive due to diarrhea illness and medical costs incurred to treat diarrhea. Not surprisingly, given that we did
not find significant impacts of the campaign on diarrhea illness itself, we also do not find significant impacts in
terms of these cost measures.
7
Pattanayak et al. (2005; 2006) discuss how to value time inputs and review of the valuation literature related to
drinking water.
8
More accurately, the fact that latrines generate private benefits (times savings, privacy) along with their
contribution to the public good of improved environmental quality makes them an impure public good (Cornes et al.
1984; Cornes et al. 1994; Vicary 1997; Kotchen 2005). Cornes and Sandler (1984) show that under some conditions
(e.g., complementarity between private and public characteristics of the good), the existence of private benefits can
have “a privatizing effect, not unlike the establishment of property rights” (p. 404). That is, increasing the actual or
perceived private benefits of latrines may reduce free riding and increase public good provision.
27
Table 1: Comparison of Means for Selected Household & Village Characteristics
Variables
Village population
Village population density
(people/acre)
% BPL
% Hindu
% Scheduled caste/tribe
Distance from all-weather road
(minutes by foot)
% Female-headed households
% of HH heads with >primary
education
Average HH Size
Average number of children<5
Average expenditure in past 30 days
(Rs.)
% owning TV
Overall
1318 (224)
Treatment
1003 (68)
Control
1618 (426)
T-C
-615 (432)
21 (5.2)
12 (3.8)
30 (9.2)
-18* (10.0)
58% (4.3%)
97% (1.8%)
61% (5.7%)
96% (3.5%)
55% (6.5%)
98% (1.1%)
6.3% (8.6%)
-1.6% (3.7%)
28% (4.1%)
29% (6.7%)
27% (5.2%)
1.8% (8.4%)
46 (5.1)
51 (6.9)
41 (7.3)
10 (10)
10% (1.5%)
8.6% (2.0%)
11% (2.2%)
-2.9% (3.0%)
52% (1.7%)
51% (2.5%)
52% (2.4%)
-1.1% (3.4%)
7.0 (.13)
1.5 (.02)
7.0 (.20)
1.5 (.03)
7.1 (.16)
1.4 (.04)
-.09 (.26)
.07 (.05)
2491 (96)
2388 (137)
2589 (134)
-201 (192)
14% (2.2%)
9.2% (2.3%)
18% (3.3%)
-8.8%** (4.1%)
Standard errors are in parentheses. Sample sizes: Villages: 20 treatment, 20 control; Households: 534
treatment, 552 control
* = significant at 10% level, **= significant at 5% level, ***= significant at <1% level
28
Table 2: Comparison of Village Means for Child Health and Household Welfare Indicators
Variables
Overall
Treatment
Control
T-C
DID
10%
7.0%
13%
-6.3%
2005
(2.0%)
(1.6%)
(3.6%)
(3.9%)
28.2%***
% owning latrine
22%
31%
14%
18%**
(0.088)
2006
(4.0%)
(6.7%)
(3.7%)
(7.6%)
7.0%
4.4%
9.6%
-5.2%*
2005
% of household members who
(1.5%)
(1.3%)
(2.6%)
(2.9%)
18.5%***
always use latrines
13%
19%
7.7%
12%**
(0.060)
2006
(2.7%)
(4.7%)
(2.2%)
(5.3%)
25.2%
27.3%
23.0%
4.3%
Proportion of children under
2005
(1.8%)
(3.1%)
(1.8%)
(3.6%)
-4.9%
five having diarrhea in past
14.2%
14.0%
14.6%
-0.6%
(4.2%)
two weeks
2006
(1.1%)
(1.7%)
(1.6%)
(2.3%)
-1.39
-1.42
-1.37
-.054
Average mid-upper arm
2005
(.037)
(.053)
(.051)
(.074)
.172
circumference (MUAC)-for-age
-1.31
-1.25
-1.37
.118
(.122)
z-score for children under five
2006
(.049)
(.067)
(.070)
(.097)
Average time (minutes) spent
60.3
65.9
54.7
11.2
2005
walking to defecation site
(3.5)
(4.8)
(4.8)
(6.8)
-20.0**
(total for all adults in
37.5
33.1
41.9
8.8
(8.7)
2006
household)
(2.7)
(4.6)
(2.8)
(5.4)
Proportion of respondents
65.0%
69.2%
60.8%
8.3%
2005
that are completely
(2.9%)
(4.8%)
(3.1%)
(5.7%)
20.5%**
dissatisfied with sanitation
45.7%
39.6%
51.8%
-12.2%
(9.2%)
2006
facilities
(3.7%)
(5.4%)
(4.8%)
(7.2%)
Table compares village-level means for each indicator, so the effective sample size for these
comparisons is 40 (20 treatment, 20 control). Standard errors for these village-level means are in
parentheses. Underlying sample sizes are as follows:
Households: 1050 (534 treatment, 552 control) (both years)
Households with water quality data: 530 (264 treatment, 266 control) (both years)
Children under five: 2005: 1572 (797 treatment, 775 control); 2006: 1256 (641 treatment, 615 control)
* = significant at 10% level, **= significant at 5% level, ***= significant at <1% level
29
Table 3: Impacts of the sanitation campaign on latrine ownership and use
Y=Household owns latrine
Treatment village
2006 dummy
Treatment x 2006
Observations
R-squared
-0.010
(0.018)
-0.019
(0.025)
0.26***
(0.030)
1,947
0.194
Y= % of household members
usually/always using latrine
-0.017
(0.014)
-0.020
(0.020)
0.17***
(0.024)
1,947
0.134
Models include the following control variables: population density, road distance, open caste, land
owner, education of primary caregiver, expenditure on food and non-food items, TV ownership,
handwashing, improved water source, water treatment.
Results from linear regression models. Robust standard errors in parentheses.
* = significant at 10% level, **= significant at 5% level, ***= significant at <1% level
30
Table 4: Impacts of Sanitation on Child Health Indicators
Treatment village
2006 dummy
Treatment x 2006
% using latrines (IV)
Control variables
included?
Observations
R-squared
Y= diarrhea (past 2 weeks)
(1)
(2)
(3)
ITT
ITT +
2SLS
controls
0.044
0.046
(0.036)
(0.037)
-0.081*** -0.085*** -0.099***
(0.026)
(0.030)
(0.026)
-0.045
-0.055
(0.045)
(0.047)
-0.11
(0.11)
No
Yes
Yes
2,828
0.018
2,629
0.040
2,629
0.039
(1)
ITT
-0.061
(0.072)
-0.0095
(0.071)
0.20**
(0.093)
No
2,489
0.003
Y = MUAC z-score
(2)
(3)
ITT +
2SLS
controls
-0.035
(0.069)
0.097
0.083
(0.081)
(0.072)
0.22**
(0.10)
0.98***
(0.28)
Yes
Yes
2,355
0.054
2,355
0.059
Control variables: population density, road distance, open caste, land owner, education of primary
caregiver, expenditure on food and non-food items, TV ownership, handwashing, improved water
source, water treatment, age and age squared, current breastfeeding, female. MUAC regressions also
control for mother’s MUAC and whether the child had diarrhea in 2005.
Results from linear regression models. Village-clustered standard errors in parentheses.
* = significant at 10% level, **= significant at 5% level, ***= significant at <1% level
31
Table 5. Impacts of Sanitation on Time Costs and Satisfaction
Y = Time spent walking to
defecation site (minutes one way)
(1)
ITT
Treatment village
2006 dummy
Treatment x 2006
10.6
(7.42)
-13.6***
(3.77)
-27.4***
(8.66)
(2)
ITT +
controls
4.79
(6.44)
-4.38
(4.52)
-29.1***
(8.36)
% owning latrines (IV)
Pop density
Distance to road
Open caste
Land owner
ln(expenditure)
TV
Respondent’s ed:
primary
Respondent’s ed:
secondary +
Mother handwashing
Improved water
source
Treats water
Observations
R-squared
2,079
-0.052
(0.063)
0.12**
(0.048)
-0.17
(3.55)
-8.55**
(3.64)
-14.0***
(3.13)
15.7***
(3.69)
4.96**
(2.46)
-24.7***
(7.26)
-0.80
(0.68)
-9.85**
(4.55)
-12.5**
(5.34)
1,930
(3)
2SLS
-6.73
(4.36)
-99.2***
(27.4)
-0.0079
(0.048)
0.047
(0.047)
1.61
(3.75)
-8.72**
(3.69)
-14.5***
(3.07)
13.5***
(3.65)
5.51**
(2.44)
-22.8***
(7.12)
-0.64
(0.65)
-7.89*
(4.75)
-11.0**
(5.17)
1,930
Y = Respondent is completely
dissatisfied with sanitation
conditions
(1)
(2)
(3)
ITT
ITT +
2SLS
controls
0.11**
0.075
(0.045)
(0.055)
-0.087**
-0.020
-0.040
(0.041)
(0.059)
(0.051)
-0.22***
-0.22***
(0.069)
(0.079)
-0.75***
(0.18)
-0.00090
-0.00072
(0.00063) (0.00050)
0.00043
-0.00012
(0.00030) (0.00036)
-0.082*** -0.071**
(0.028)
(0.028)
-0.077**
-0.078**
(0.032)
(0.033)
-0.060*
-0.063**
(0.030)
(0.030)
0.079**
0.064**
(0.032)
(0.031)
-0.0088
-0.0055
(0.016)
(0.015)
-0.16***
-0.15***
(0.034)
(0.032)
-0.022*** -0.021***
(0.0042)
(0.0047)
-0.015
-0.00019
(0.038)
(0.036)
0.0036
0.014
(0.042)
(0.042)
2,100
1,947
1,947
0.052
0.103
0.115
For “time spent walking,” results are from a tobit model to account for censoring at 0 due to
latrine ownership. For “respondent is completely dissatisfied,” results reported are coefficients
from linear probability model (OLS). Village-clustered standard errors in parentheses.
* = significant at 10% level, **= significant at 5% level, ***= significant at <1% level
32