Do Non-monetary Prices Target the Poor? Evidence
from a Field Experiment in India
Bridget Hoffmann∗
Northwestern University
November 24, 2014
Abstract
This paper uses a field experiment in India to study the targeting of health products
to the poor, using monetary prices and non-monetary prices (time costs) to allocate
the good. I model household demand for a health product at monetary prices and at
non-monetary prices. The model illustrates that monetary prices screen out the poor
and that it is theoretically ambiguous whether non-monetary prices screen out the rich
or the poor due to opposing income and substitution effects. Therefore, whether nonmonetary prices target the poor better than monetary prices is an empirical question.
In order to generate estimates for demand at monetary prices and at non-monetary
prices, I conducted a field experiment in India in which I elicited respondents’ willingness to pay (WTP) for a water purifier in Rupees and in working hours. I find that
a 10% increase in daily household income implies a 2.0% increase in monetary WTP.
In contrast, a 10% increase in daily household income implies a 0.5% decrease in nonmonetary WTP. Comparing across price types, non-monetary WTP falls 2.3% relative
to monetary WTP for a 10% increase in daily household income. Using demand curves
estimated from the WTP data, I evaluate the problem of a principal with a fixed budget whose objective is to increase take-up of the health product, in particular by those
with greatest need. In this case, the principal targets the poor because the poor are
less likely to own a water purifier at baseline and are more likely to report episodes
of water borne illness. Despite better targeting achieved by non-monetary prices, the
principal optimally chooses a monetary price for a large range of parameters.
∗
BridgetHoffmann2014@u.northwestern.edu, Northwestern University, Department of Economics, 2001
Sheridan Road, Evanston, IL 60208. I would like to thank Treb Allen, Lori Beaman, Matthias Doepke,
Seema Jayachandran, Cynthia Kinnan, Lee Lockwood, Ofer Cohen, Joel Gamoran, Juan D. Prada, and
seminar participants at Northwestern University for their helpful comments and their contributions to the
paper. I would like to thank the Russell Sage Foundation, the Buffett Center at Northwestern University,
and Northwestern University TGS for generous funding. All errors are my own.
1
1
Introduction
In developing countries, many health products have large private and social benefits but
very low take-up at the market price, particularly among the poor.1 Governments and Non
Governmental Organizations (NGOs) use price subsidies to increase take-up,2 but monetary
price subsidies do not select those with the highest returns (Cohen and Dupas, 2010; Kremer
and Miguel, 2007; and Ashraf et al. 2010). The poor often have the greatest benefit from
health products, but are screened out even at subsidized prices, because they lack the cash,
savings, or credit to purchase (Tarozzi et al., 2014; Dupas and Robinson, 2013; Devoto et
al., 2012). Therefore, organizations face a trade-off between higher monetary price subsidies
(at higher cost) and excluding the poor.
An alternative method of allocating health products is non-monetary prices (time costs),
often called “self-selection” mechanisms. Non-monetary prices may target the poor better
than monetary prices because selection occurs on willingness to pay in time rather than
money. If the time spent to purchase the product is productive, non-monetary prices may
allow organizations to recover partial product costs without screening out the poor.
This paper studies two methods of targeting health products to the poor, monetary prices
and non-monetary prices, using data from a field experiment in India. I develop a model of
household demand for a health product at monetary prices and at non-monetary prices that
illustrates three features of demand. First, I show that for any monetary price rich demand
exceeds poor demand, implying that the poor are underrepresented among demanders. Second, the model demonstrates that non-monetary prices may not result in greater demand
among the poor than the rich due to opposing income and substitution effects. When the
utility function is concave, a higher wage implies a greater loss of consumption for each hour
of work forgone to pay the non-monetary price, but also implies a lower marginal utility of
consumption. These opposing effects imply that whether the utility cost of the non-monetary
price is increasing in the wage is theoretically ambiguous. Third, regardless of whether the
utility cost of the non-monetary price is increasing in the wage, the model illustrates that,
relative to the rich, the utility cost of a monetary price is greater than the utility cost of
a non-monetary price for the poor. This implies that non-monetary prices target the poor
1
For high price sensitivity for health products, see J-PAL (2011), Berry, Fischer, and Guiteras (2012),
Dupas (2014a), Dupas (2014b), and Miguel and Kremer (2004). For large private and social benefits of health
products, see Miguel and Kremer (2004), Arnold and Colford (2007), Baird et al. (2011), and Thomas et al
(2006).
2
Information campaigns are an alternative to subsidies but have little effect on price sensitivity (Kremer
and Miguel, 2007; Dupas, 2009; Meredith et al., 2013).
2
better than monetary prices.
To empirically test whether non-monetary prices target health products to the poor better
than monetary prices, I conducted a field experiment in Hyderabad, India using the TATA
Swach Smart water purifier.3 The experiment included nearly 800 households in seventeen
slum neighborhoods of Hyderabad. At baseline, 70% of households did not own a water
purifier and 41% of households reported that at least one household member suffered from
an illness perceived to be caused by the drinking water during the last rainy season.
I used the Becker-DeGroot-Marshak (BDM) mechanism (Becker, DeGroot, Marshak,
1964) to elicit respondents’ willingness to pay (WTP) for a TATA Swach Smart water purifier
at monetary prices (rupees) and at non-monetary prices (hours sorting seeds by type). The
BDM mechanism is similar to a second-price auction in which the second-price is an unknown,
randomly drawn price. The respondent states her maximum willingness to pay for a good
and then a price is randomly drawn from a distribution of possible prices. If the price is
less than or equal to the respondent’s WTP, she purchases the good at the randomly drawn
price. If the price is above the respondent’s WTP, she cannot purchase the good.
I analyze the distribution of WTP for each price type across income levels. I find that
monetary WTP is increasing in income. A 10% increase in household daily labor income
implies a 2.0% increase in monetary WTP. For all monetary prices, rich demand exceeds poor
demand and the fraction of total demand that is poor falls as the monetary price rises. I can
reject equality of the rich and poor monetary demand curves. Household income is negatively
correlated with non-monetary WTP. A 10% increase in daily household income implies a 0.5%
decrease in non-monetary WTP, but I cannot reject that the rich and poor demand curves
are the same for non-monetary prices. In a pooled sample of households’ monetary WTP and
non-monetary WTP, I use household fixed effects to control for households’ fixed valuation
of the water purifier. Focusing on the differential effect of income on WTP by price type,
I find that non-monetary WTP falls relative to monetary WTP as income rises. Therefore,
non-monetary prices target the poor better than monetary prices (a larger fraction of total
demand is poor) even though non-monetary demand is similar across income levels.
Next, I evaluate non-monetary prices as a policy to target health products to the poor.4
3
The TATA Swach Smart water purifier follows US EPA guidelines and is economical in terms of both
the initial cost (retail price of 999 Rs) and the cost per liter relative to similar water purifiers.
4
Self-selection mechanisms are widely useful as a method of targeting the poor. For example, the National
Rural Employment Guarantee Scheme (NREGS/NREGA) is a targeted poverty alleviation program in India
that covers 11% of the world’s population (Niehaus and Sukhtankar, 2013). Although the purpose of NREGA
is to transfer resources to the rural poor, all households are eligible for NREGA benefits. The program relies
on poorer households self-selecting into the program while richer households self-select out of the program
3
I model the problem of a principal with a fixed budget who can choose either a monetary
price or a non-monetary price to allocate the health product. The principal’s objective is to
increase take-up of the health product, in particular by those with the greatest need. In this
case, the principal targets the poor because the poor are more likely to report episodes of
water borne illness and are less likely to own a water purifier at baseline.
I evaluate the principal’s problem in three steps. First, I use the experimental WTP
data to estimate the total demand and the poor demand curve for each price type. Second,
I find the optimal monetary price from the set of monetary prices, and I find the optimal
non-monetary price from the set of non-monetary prices. Third, I compare the value of the
objective using the optimal monetary price and optimal non-monetary price to determine
the optimal price.
The optimal price depends on the need of the poor relative to the rich. In the case that
the principal does not target the poor because there is no difference in need across income
levels, the principal simply chooses the price type with a lower net cost per unit at the
optimal prices. As the relative weight on the poor increases, the principal places increasing
weight on the difference in targeting achieved by the price types at the optimal prices. In
the case that the rich have no need for the health product, the principal only values health
goods distributed to the poor, and the principal compares the difference in the net cost per
unit to the difference in targeting achieved by the price types at the optimal monetary price
and the optimal non-monetary price.
I show that a non-monetary price is optimal only when the principal places much greater
weight on the poor than on total take-up. When the productivity of the non-monetary
price is 50 Rs per hour, the principal prefers a monetary price to a non-monetary price if
the rich have a relative weight of at least 0.136 (relative to the poor’s weight of 1). As
the productivity of the non-monetary price (hours worked in exchange for the health good)
increases the subsidy is effectively less costly and the non-monetary price is preferred to the
monetary price for a larger range of parameters. When the principal weighs the allocation
to the rich and poor according to the fraction of diarrhea episodes, the principal prefers the
monetary price to the non-monetary price until the non-monetary price has a productivity
of 85 Rs per hour.
Three recent experiments study the effectiveness of self-selection mechanisms in targeting
resources in developing counties. Alatas et al. (2013) provides evidence that an ordeal
combined with an asset test leads to a much poorer group of beneficiaries for Indonesia’s
because of the manual labor requirements.
4
Conditional Cash Transfer Program than a universal asset-based enrollment system. Ma,
Sylvia, Boswell, and Rozelle (2013) and Dupas, Hoffmann, Kremer, and Zwane (2013) study
the effectiveness of self-selection mechanisms in selecting those most likely to use the goods
for health purposes. Ma, Sylvia, Boswell, and Rozelle (2013) use distance from a student’s
home to the county seat, where a voucher can be redeemed, as an ordeal mechanism in the
distribution of eyeglasses to students in rural China. In western Kenya, Dupas, Hoffmann,
Kremer, and Zwane (2013) use a voucher for water treatment solution that can be redeemed
at a market center as a “micro-ordeal”. The micro-ordeal varies by whether households can
redeem the voucher in the market center nearest to their home. Both experiments find that
the ordeal reduces program costs without screening out a substantial number of people who
would have used the health product if given for free.
This paper builds on the previous results that non-monetary prices can improve targeting
and lower program costs. The WTP data used in this paper provide a much finer measure
of non-monetary WTP, which allows me to determine the optimal non-monetary price as
a function of the principal’s budget, the need of the poor relative to the rich, and the
productivity of the non-monetary price.
The remainder of the paper is organized as follows. In section 2, I model household
demand at monetary prices and non-monetary prices. The experimental design is described
in section 3. In section 4, I investigate the distribution of willingness to pay. In section 5, I
model the problem of a principal with a fixed budget dedicated to distributing preventative
health products. I evaluate the principal’s problem using estimated demand functions for
monetary and non-monetary prices and discuss the optimal pricing policy. In section 6,
I perform robustness checks, specifically concerning the distribution of willingness to pay.
Section 7 concludes.
2
Household Model
This section describes a simple model of a household’s decision whether or not to purchase
a water purifier through the subsidy program either for a monetary price, PM , or for a
non-monetary (time) price, PN M . PN M is denominated in hours.
Households have quasi-linear utility U = u(ci ) + αi 1[di = 1] with a common, increasing,
and strictly concave utility function over consumption goods. Households receive additional
utility αi from ownership of a water purifier which captures the household specific utility
received from a water purifier in the form of better health and increased earnings due to
5
reduced incidence of illness. d is an indicator variable for purchase of the water purifier.
Households earn income from labor at an exogenously determined wage, wi ∈ {wp , wr }
with wp < wr . There is no disutility of labor so all households choose to work the maximum
number of hours, h̄.
Household i maximizes its utility subject to its budget constraint, ci + 1[di = 1]·PM ≤ wi h̄
for a monetary price and ci ≤ wi ∗ (h̄ − 1[di = 1] · PN M ) for a non-monetary price. Household
i purchases a water purifier if the additional utility received from a water purifier exceeds
the loss in utility from reduced consumption. This creates a cut-off level of α for each wage
level, j, and price type, k, denoted αkj such that a household will purchase a water purifier if
its value of α exceeds the cut-off value and will not purchase otherwise. The cut-off level of
α for households facing a monetary price is given by u(wi h̄) − u wi h̄ − PM and the cut-off
level of α for households facing a non-monetary price is given by u(wi h̄) − u wi ∗ (h̄ − PN M ) .
Suppose that αi is distributed uniformly among households with each wage. Without loss
of generality, assume that the support of α is [0,1].
The following proposition shows that demand among the rich exceeds demand among
the poor for any monetary price.
Proposition 1. For the set of monetary prices, PM , such that Dr (PM ) > 0 and Dp (PM ) < 1
the proportion of households that purchase a water purifier among those with the high wage
is greater than the proportion of households that purchase a water purifier among those with
the low wage: Dr (PM ) > Dp (PM ).
p
p
r
and Dp (PM ) = 1 − αM
Proof. Since Dr (PM ) = 1 − αM
, the result holds since αM
=
r
by concavity of the utility function
u(wp h̄) − u(wp h̄ − PM ) > u(wr h̄) − u(wr h̄ − PM ) = αM
for consumption. Strict concavity of u implies that for every λ ∈ (0, 1) and for every pair
0 ≤ x < x0, u(x0) + u(x) < u(λx + (1 − λ)x0) + u(λx0 + (1 − λ)x). The result follows by
PM
, x = wp h̄ − PM , and x0 = wr h̄.
choosing λ = wr h̄−w
p h̄+PM
The conditions Dr (PM ) > 0 and Dp (PM ) < 1 are needed to rule out boundary cases in
which no households or all households demand a water purifier at price PM .
Intuitively, households with the higher wage have a lower marginal utility of consumption
and suffer a smaller disutility of forgone consumption when they purchase a water purifier
at a monetary price.
The following corollary extends the result above to consider changes in demand in response to a monetary price change. The difference in demand at the lower monetary price
and at the higher monetary price is decreasing in the wage.
6
Proposition 1 implies that for any monetary price, demand is lower among the poor
p (P )
M
is always less than one-half. The above
than among the rich so the targeting ratio, DD(P
M)
proposition holds for all monetary price levels, showing the potential for non-monetary prices
to target the poor more effectively than monetary prices.
When considering demand at a non-monetary price, it is not necessarily true that for
a given value of α, households with a higher wage, and higher opportunity cost of time,
have lower demand. The classical literature on ordeal mechanisms has assumed that time
costs are more costly in utility terms for the rich (Barzel, 1974; Nichols, Smolensky, and
Tideman, 1971; Nichols and Zeckhauser, 1982; and Besley and Coate, 1992). For example,
when the utility function for consumption is linear, the utility cost of the non-monetary
price is proportional to the wage and demand is decreasing in the wage. However, when the
utility function is concave, the utility cost of a non-monetary price is not proportional to the
wage. There are two opposing effects of a higher wage: (1) the substitution effect: a higher
wage results in greater loss of consumption utility from forgoing an hour of work to pay
the non-monetary price since each hour of work can buy relatively more consumption; (2)
the income effect: a higher wage implies higher consumption, and therefore, lower marginal
utility of consumption, leading to a smaller utility loss from forgoing an hour of work to pay
the non-monetary price. Non-monetary prices are most effective in targeting the poor when
the substitution effect outweighs the income effect, and the marginal utility of consumption
does not decrease too rapidly as consumption rises in the relevant range.
The following proposition shows the conditions under which non-monetary prices will
result in greater demand among the poor than among the rich.
Proposition 2. Define v(w, PN M ) = u(w(h̄ − PN M )). If v satisfies increasing differences in
w and −PN M , then for the set of non-monetary prices, PN M , such that Dr (PN M ) < 1 and
Dp (PN M ) > 0 demand among the poor is greater than demand among the rich: Dp (PN M ) >
Dr (PN M ).
p
Proof. Demand for the poor is given by Dp (PN M ) = 1 − αN
M and demand for the rich is
r
r
given by D (PN M ) = 1 − αN M . The result holds if v satisfies increasing differences in w and
−PN M because this implies u(wr h̄) − u(wr (h̄ − PN M )) > u(wp h̄) − u(wp (h̄ − PN M )).
The conditions Dr (PN M ) < 1 and Dp (PN M ) > 0 are needed to rule out boundary cases
in which all households or no households demand the water purifier at price PN M .
For utility functions, such as CRRA with relative risk parameter γ ∈ (0, 1), which satisfy
the requirements above, the utility cost of the non-monetary price is increasing in the wage
7
and demand at the non-monetary price is decreasing in the wage. In this case the targetp (P
NM )
is always greater than one-half and increases as the non-monetary price
ing ratio, DD(P
NM )
increases.
Even if non-monetary prices do not result in greater demand among the poor than among
the rich, there is still scope for non-monetary prices to better target the poor than monetary
prices. The following proposition shows that the targeting ratio is always greater for a nonmonetary price than for a monetary price. This follows directly from the propositions above
but also holds under weaker conditions.
Proposition 3. For PM and PN M such that Dp (PM ) > 0 and Dr (PN M ) > 0, the targeting ratio for a non-monetary price is greater than the targeting ratio for the corresponding
p (P
p (P )
NM )
M
monetary price: DD(P
> DD(P
.
NM )
M)
p (P
Dp (PM )
p
p
p
NM )
r
]
> [1 − αM
] ∗ [1 −
>
implies
that
[1
−
α
]
∗
[1
−
α
]
+
[1
−
α
Proof. DD(P
M
N
M
M
)
D(P
)
NM
M
p
r
αN M ]+[[1−αN
M ] . Normalize wp = 1 so the poor are indifferent between paying a monetary
r
r
price or a non-monetary price of equal magnitude. Then [1 − αM
] > [1 − αN
M ] since wr > 1
and u0(·) > 0.
The boundary conditions Dp (PM ) > 0 and Dr (PN M ) > 0 ensure that demand is positive
for both the poor and the rich at both the monetary price and non-monetary price. The
proposition above shows that non-monetary prices will result in better targeting of the poor
than monetary prices.
This simple model illustrates that monetary prices screen out the poor and that the
targeting ratio for monetary prices is always less than the fraction of the population that
is poor. While the targeting ratio for non-monetary prices may be greater or less than the
fraction of the population that is poor, it always exceeds the targeting ratio of monetary
prices.
3
Experimental Design
The experiment was conducted in Hyderabad, Telangana from June 26, 2013 to July 18,
2013. Seventeen slums were randomly selected from a list of nearly 1,400 slums provided by
the Greater Hyderabad Municipality. The list was restricted to slums with a population of at
least 100 households.5 Door-to-door surveying was conducted from Monday to Friday, with
5
This excluded 5 slums with missing population data and 323 slums with fewer than 100 households.
Thirteen slums selected from the list were skipped because they could not be located (6 slums) or because
the neighborhood was located but did not contain a low-income, residential area (7 slums).
8
the survey team visiting a new slum neighborhood each day. Households were surveyed if an
adult (age 20 years or above) household member was available and the household contained
more than one person.6 On average, 47 households completed the main survey in each slum
with the number depending on the weather and surveyor absences.
Since the primary objective of the experiment was to study targeting of health investment
goods to the very poor, households in the top and bottom quartiles of the income distribution
were overrepresented in the sample. Eligibility for the main survey was determined by a presurvey that consisted of 6 questions about household asset ownership and one categorical
income question.7 1,233 potential respondents completed the pre-survey. Of these, 804
respondents were determined to be eligible for the main survey, and 798 completed the full
survey.
The main survey included sections covering income and assets, savings and credit, household expenditure, and education. An additional section asked questions regarding water
source, water treatment, and water borne illnesses. A detailed section collected information about all household members’ labor market activities.8 Table 1 presents household and
respondent summary statistics for the full sample, poor (below median income), and rich
(above median income).
After these sections, surveyors introduced the water purifier by reading a passage with
information on the TATA Swach Smart. The TATA Swach Smart follows US EPA guidelines and is economical in terms of both the initial cost of the water purifier (market price of
999 Rs) and the cost per liter compared to similar water purifiers (for example, the PureIt
brand).9 Surveyors showed respondents an assembled TATA Swach Smart and gave respondents time to examine the water purifier and its box which displays the Maximum Retail
Price. Figure 1 contains an annotated diagram of the TATA Swach Smart, and Appendix A
contains the passage that surveyors read to respondents.
Two sections of the main survey elicited respondent’s monetary willingness to pay and
non-monetary willingness to pay for the TATA Swach Smart using the Becker-DeGrootMarshak (BDM) mechanism. The BDM mechanism is similar to a second-price auction
6
Single occupant households were excluded because they were primarily bachelors who traveled to Hyderabad for school and would frequently return to their family outside of Hyderabad.
7
The asset section of the pre-survey asked questions regarding tablets/laptops/computers, washing machines, four-wheeled vehicles, motorbikes/motorcycles/scooters, coolers, and cots/beds.
8
In addition, the survey included two sections regarding mental accounting statements developed in
Soman (2011), with one section for time and one section for money. The order of these two sections was
randomized by household.
9
The market price of 999 Rs is approximately 16.50 USD.
9
in which the second-price is an unknown, randomly drawn price. The respondent states
her maximum willingness to pay for a good and then a price is randomly drawn from a
distribution of possible prices. If the price is less than or equal to the respondent’s WTP,
she purchases the good at the randomly drawn price. If the price is above the respondent’s
WTP, she cannot purchase the good. For expected utility maximizers, the BDM mechanism
should elicit respondents’ true maximum willingness to pay. The theory behind the BDM
mechanism and experimental evidence of its ability to elicit truthful WTP are described in
Appendix B.
Each willingness to pay section walked respondents through the BDM mechanism. First,
the respondent was asked to state her maximum willingness to pay (WTP) for the water
purifier in rupees or in time spent sorting seeds by type. Next, the respondent was asked
if she would purchase the water purifier at each price on a price list, beginning at 50 Rs (1
hour) lower than her initial bid. The prices on the price list increased by an increment of 50
rupees (1 hour). Once the respondent stated that she would not purchase at a given price,
no questions were asked regarding higher prices. The highest price at which the respondent
stated that she would purchase was confirmed as her maximum willingness to pay. If a
respondent indicated that she would not purchase the water purifier at a price 50 Rs (1
hour) lower than her initial bid or she refused to answer a question in the price list, the
process was started again by asking the respondent to revise her initial willingness to pay.10
The two willingness to pay sections were designed to be identical except for the references
to time and money. The order of these two sections was randomized by household. The script
for both sections clearly state that the respondent cannot change her bid after seeing the
price and emphasize that respondents must be willing and able to pay the amount that they
bid on the following Saturday when all transactions take place. These scripts are shown in
Appendix A.
After eliciting each respondent’s maximum willingness to pay in money and in time,
respondents were randomly assigned to treatment groups with 418 assigned to the monetary price group and 380 assigned to the non-monetary price group.11 Respondents in the
monetary group randomly drew a price in the range 50 Rs to 999 Rs and respondents in
10
In pre-trials, this method resulted in the fewest respondents refusing to purchase or reporting regret when
the price chosen was less than or equal to their stated maximum willingness to pay and the most respondents
stating that they preferred not to purchase when the price chosen was above their stated maximum willingness
to pay.
11
To assign treatments, without looking, respondents chose a tile from a box that contained two tiles
with the word money and two tiles with the word time. The tile that the respondent chose determined the
respondent’s treatment group.
10
the non-monetary group randomly drew a price in the range 1 hour to 18 hours. The nonmonetary price range was chosen to be approximately equal to the monetary price range at
the prevailing male casual labor wage (about 450-500 Rs per 8-9 hour day).12
If the random price was below the respondent’s maximum willingness to pay (of the
same price type), she was given a voucher for the water purifier. Respondents who received
a voucher for the water purifier (those whose willingness to pay was above the price) were
asked if they wished that they had not bid so high and respondents who did not receive
a voucher for the water purifier (those whose willingness to pay was below the price) were
asked if they wished that they had bid more. At the conclusion of the survey, all respondents
were compensated for participation with a steel dining plate.13
For logistical reasons, both monetary and non-monetary price vouchers were redeemed
the following Saturday in a temple, school, or community center in the neighborhood.14 This
also kept the time and conditions of transaction consistent across money and time treatment
groups. Respondents who redeemed a voucher with a monetary price simply paid a surveyor
the price on her voucher and received a TATA Swach Smart. Respondents who redeemed a
voucher with a non-monetary price sorted seeds into piles by type for the number of hours
on their voucher. Although vouchers could be redeemed by either the respondent or his/her
spouse, 86% of redeemed vouchers were redeemed by the respondent.15
4
Willingness to Pay
I begin by looking at the distributions of monetary WTP and non-monetary WTP. Figures 2
and 3 show the percentage of respondents who stated a willingness to pay at least as high as
the price for each price level. Willingness to pay was elicited for one water purifier through
the BDM mechanism. Separate distributions are shown for target, poor households and nontarget, richer households. The poor subsample consists of households with per capita total
income less than or equal to the median in the sample.16 The objective of the pre-survey
was to oversample households in the top and bottom quartiles of the population income
12
450 Rs is approximately 7.50 USD.
The plates were purchased at a bulk price of approximately 30 Rs. per plate (approximately 0.50 USD).
14
Redemption centers were open Saturdays from 10am to 6pm but stayed open past the closing time or
opened on Sunday if a respondent needed additional time to complete her non-monetary payment.
15
Figures 11 and 12 show the number of vouchers issued and redeemed by income level for monetary
prices and for non-monetary prices.
16
The poor comprise 51% of the sample because many households report the median per capita household
income.
13
11
distribution. Splitting the sample into rich and poor subsamples at the median ensures that
all households in the bottom quartile of the population income distribution are included in
the poor subsample. For simplicity, non-target households with per capita household income
above the median are referred to as “rich”.
Figure 2 shows the percentage of households in each income group with BDM bid greater
than or equal to the price for each monetary price. As predicted by Proposition 1, for every
monetary price except 50 Rs, demand is greater among the rich than among the poor. The
Kolmogorov-Smirnov test rejects equality of the rich and poor distributions of monetary
WTP (p-value < 0.001). Appendix Figure 1 shows the proportion of respondents with
monetary willingness to pay above the price by quartile of the income distribution. The
poorest quartile generally has the lowest demand and the richest quartile generally has the
highest demand. However, the richest quartile does not have the highest demand at very
low prices and the poorest quartile does not have the lowest demand at very high prices.
Figure 3 shows the percentage of each income group with BDM bid greater than or equal
to the price for each non-monetary price. Proposition 2 described the conditions under which
poor demand will exceed rich demand at non-monetary prices. The curves suggest that the
sufficient condition for non-monetary prices to result in greater demand among the poor than
the rich does not hold. Although the poor have greater demand than the rich for all nonmonetary price levels, the difference between the curves is small at all price levels, and the
Kolmogorov Smirnov test does not reject equality of the rich and poor distributions of nonmonetary WTP (p-value of 0.326). Appendix Figure 2 shows the proportion of respondents
with non-monetary willingness to pay above the price by quartile of the income distribution.
The richest quartile generally has the lowest demand while the demand of the first and third
quartiles are very similar.
Figure 2 suggests that monetary prices will select richer households while Figure 3 suggests that non-monetary prices will not provide strong selection on income. In order to
quantify the correlation between income and willingness to pay for each price type, I directly
analyze the relationship between income and willingness to pay.
Tables 2 and 3 present the results of OLS regressions of the natural logarithm of WTP
on the natural logarithm of income with robust standard errors. Table 2 presents the results
using monetary WTP (Rs) as the dependent variable, and Table 3 presents the results using
non-monetary WTP (minutes) as the dependent variable.17 In each table, Panel A uses
17
In order to facilitate comparison of coefficients across price types, non-monetary willingness to pay is
denominated in minutes. The market wage for male casual labor is approximately 450-500 Rs per 8-9 hour
day or a wage rate of approximately 1 Rs per minute.
12
variables measuring daily labor income, Panel B uses daily total income variables, and Panel
C uses hourly wage variables.18 Columns (1)-(2) look at the effect of household income on
WTP and columns (3)-(4) look at the effect of respondent income on WTP. Columns (1) and
(3) present baseline regressions. Columns (2) and (4) include neighborhood fixed effects.
Columns (1)-(2) of Table 2 show that household income is positively correlated with monetary WTP for the water purifier; when including neighborhood fixed effects, a 10% increase
in household daily labor income implies a 2.0% increase in monetary WTP. At the mean,
this implies that an additional 52 Rs of household daily labor income raises WTP by 5.8 Rs.
As shown in columns (3)-(4) of Table 2, respondent income is also positively correlated with
monetary WTP. When neighborhood fixed effects are included, a 10% increase in respondent
daily labor income implies a 1.7% increase in monetary WTP. At the mean, this implies that
an additional 22 Rs of respondent daily labor income raises WTP by 5.2 Rs.19
Columns (1)-(2) of Table 3 show that household income is negatively correlated with nonmonetary WTP for the water purifier, although the effect is small relative to the effect for
monetary willingness to pay. When neighborhood fixed effects are included, a 10% increase
in household daily labor income implies a 0.5% decrease in non-monetary WTP. At the mean,
this implies that an additional 52 Rs of daily household labor income lowers non-monetary
WTP by 8.6 minutes. Columns (3)-(4) show that respondent income does not appear to be
correlated with non-monetary WTP.
Tables 2 and 3 show there is stronger selection on household income than on respondent
income, particularly for non-monetary prices. The presence of a respondent who does not
work or who works for a low wage does not necessarily identify a poor household. Therefore,
the results in Tables 2 and 3 support the use of non-monetary prices to select poor households.
Corroborating this, Appendix Table 1 shows that the asset index is positively correlated with
monetary WTP and negatively correlated with non-monetary WTP.20 Further, selection
occurs on household income rather than on household characteristics that may affect a
household’s valuation of the water filter. Appendix Tables 2 and 3 shows that, with the
exception of owning a water purifier at baseline, household characteristics (the number of
children in the household under age 14, episodes of illness perceived to be caused by the
18
The hourly wage is calculated for salaried workers by assuming 26 working days per month. The number
of hours worked per day is taken from survey data.
19
I cannot reject that the coefficients for household income (column (2)) and respondent income (column
(4)) are equal in any of the three panels.
20
The asset index is the first principal component of variables representing household ownership of five
durable goods: automobiles, motorbikes/scooters, tablets/computers, washing machines, and evaporative
coolers. The asset index is normalized to have mean 0 and standard deviation 1.
13
drinking water, treating drinking water at baseline, and credit constrained) are not correlated
with WTP.21
Using each households’ monetary WTP and non-monetary WTP, I can focus on the differential effect of income on monetary WTP and non-monetary WTP. I use a pooled sample
in which the dependent variable is either the natural logarithm of monetary WTP or the
natural logarithm of non-monetary WTP scaled by the median respondent hourly wage (23
Rs per hour). The regression equation contains household fixed effects, an indicator variable for non-monetary WTP, and the interaction between the non-monetary WTP indicator
and the natural logarithm of an income variable with robust standard errors clustered by
household. Because the household fixed effect controls for household’s fixed valuation of the
water purifier, the general effect of income on WTP for the water purifier is not identified in
this specification. However, the coefficient on the interaction term identifies the differential
effect of income on monetary WTP and non-monetary WTP.
Table 4 presents the results of the pooled regressions using various measures of household
and respondent income. Columns (1)-(2) use daily labor income variables, columns (3)(4) use daily total income variables, and columns (5)-(6) use hourly wage variables. As
household income and respondent income rise, non-monetary WTP falls relative to monetary
WTP. Column (1) shows that increasing household daily labor income by 10% decreases
non-monetary WTP by 2.3% relative to monetary WTP. Similarly, column (2) shows that
increasing respondent daily labor income by 10% decreases non-monetary WTP by 2.0%
relative to monetary WTP. The difference in monetary WTP and non-monetary WTP is
increasing in income, implying that non-monetary prices may better target the poor than
monetary prices.
Because I observe each respondents’ monetary WTP and non-monetary WTP, I can
compare respondents’ value of time (monetary WTP divided by non-monetary WTP) to
their wage. Respondents who make the decision of whether to pay a non-monetary price or
work for a wage at the margin have a value of time close to their hourly wage. On average,
respondents who work for a wage and have the flexibility to work part of a day value their
time at 1.07 times their wage.22 Respondents who work but do not make the decision of
whether to pay a non-monetary price or work for a wage at the margin have a value of
time that is much greater than their hourly wage. On average, respondents who work for a
21
Households that own a water purifier at baseline have lower monetary WTP and non-monetary WTP
than households that do not own a water purifier at baseline.
22
There are 88 respondents who report working for a wage and report the flexibility to work part of a
day.
14
wage but cannot work part of a day value their time at 3.5 times their hourly wage.23 These
respondents have a value of time greater than the hourly wage because they must forgo more
than an hour of labor income to pay an hour of the non-monetary price.
The flexibility of hours worked is an important determinant of non-monetary WTP relative to monetary WTP, and therefore, differences in the flexibility of hours worked across
income levels is an important determinant of the targeting achieved by non-monetary prices
relative to monetary prices. The percentage of respondents who work for a wage is very similar across income levels, but the percentage of respondents who report the ability to adjust
hours worked downwards differs across income levels. 45% of rich respondents who work for
a wage report downward flexibility in hours worked and 30% of poor respondents who work
for a wage report downward flexibility in hours worked.24 This implies that non-monetary
prices will better target the poor in contexts in which the rich do not have greater flexibility
in hours worked than the poor.
Finally, I consider the targeting achieved by each of the price types. Consistent with
Proposition 3, Table 5 shows that for any level of take-up among the poor, non-monetary
prices result in fewer windfall beneficiaries (rich households) than monetary prices.25 For
example, poor demand at the monetary price 200 Rs is 277 households. This is nearly
identical to poor demand at the non-monetary price 2 hours, 273 households. However,
the monetary price results in demand of 297 rich households while the non-monetary price
results in demand of 257 rich households. Similarly, poor demand at the monetary price
350 Rs is 103 households which is nearly identical to poor demand of 99 households at the
non-monetary price 4 hours. The monetary price results in rich demand of 158 households
and the non-monetary price results in rich demand of 89 households. In both cases, the
non-monetary price results in fewer windfall beneficiaries.
Table 5 also compares prices on the basis of the targeting ratio (proportion of purchasers
who are poor). Table 5 shows that non-monetary prices have a greater targeting ratio than
monetary prices. Further, as prices rise, non-monetary prices better target the poor, while
the opposite occurs for monetary prices. Figure 4 graphs the targeting ratio versus poor
demand for each price type. Moving right along the x-axis, prices rise and poor demand
falls. The figure illustrates that as prices rise the targeting ratio for monetary prices falls
and the targeting ratio for non-monetary prices rises. Despite the better targeting achieved
by non-monetary prices, non-monetary prices may not be the most cost effective method of
23
There are 151 respondents who work for a wage and report that they cannot work part of a day.
36% of rich respondents work for a wage and 33% of poor households work for a wage.
25
Table 5 includes prices for which total take-up is at least 20 households.
24
15
targeting the poor.
5
Principal’s Problem
I describe the problem of a principal, such as an NGO or government health ministry, whose
objective is to increase take-up of a health good within its borders. The principal most
values reaching those with greatest need, in this case, the poor. The principal can choose
either a monetary price or a non-monetary price to allocate the good.
The principal has fixed budget B dedicated to purchasing the health good at the market
price, M . The total number of units that the principal can distribute at a price P is the
minimum of total demand at price P , D(P ), and the number of units the principal can
afford to purchase. For each price type, there is a unique market clearing price, because the
number of units the principal can afford to purchase is weakly increasing in the price and I
assume demand is decreasing in both price types.
Although non-monetary prices achieve better targeting than monetary prices for a given
level of poor demand, a non-monetary price is not necessarily optimal. Unless the nonmonetary price is highly productive, for a given level of demand, monetary prices recover
greater products costs than non-monetary prices. Therefore, the principal effectively has
a larger budget when using a monetary price. Because the optimal price is decreasing in
the budget, there may be little difference in targeting between the optimal monetary price
and the optimal non-monetary price, but the monetary price can reach substantially more
households.
The principal’s objective places weight δ ∈ [0, 1] on total allocation, A(P ), and places
weight 1 − δ on allocation to the poor, Ap (P ):
δ ∗ A(P ) + (1 − δ) ∗ Ap (P ).
Since the distribution to the poor is included in the total allocation, this implies that a unit
allocated to the rich receives weight δ while a unit allocated to the poor receive a weight of 1.
I examine the choice between using a monetary price or a non-monetary price to allocate the
health good as a function of the productivity of the non-monetary price, ω, and the relative
weight on a unit allocated to the rich, δ. This is distinct from the problem of a social planner
who maximizes a social welfare function.
If the principal uses a monetary price, the total number of units that the principal
B
]. Similarly, if the principal uses a
can distribute is given by A(PM ) = min[D(PM ), M −P
M
16
non-monetary price, the total number of units that the principal can distribute is given by
B
A(PN M ) = min[D(PN M ), M −ωP
], where ω represents the benefit to the principal of each
NM
hour worked in exchange for the the health good. If PN M is pure deadweight loss then ω = 0.
I assume the only tools available to the principal for targeting are prices. This implies
that units are randomly allocated among demanders when there is excess demand.26 When
there is excess demand, the number of water purifiers distributed to the poor with a monetary
p (P )
B
M
price is given by Ap (PM ) = DD(P
∗ M −P
, where Dp (PM ) is poor demand at price PM ,
M)
M
and the number of units distributed to the poor with a non-monetary price is given by
p (P
B
NM )
∗ M −ωP
, where Dp (PN M ) is poor demand at price PN M .
Ap (PN M ) = DD(P
NM )
NM
When δ = 1, the principal values only the total number of units distributed. In this case,
B
∗
∗
and similarly
the principal maximizes A(PM ) by choosing PM
such that D(PM
) = M −P
∗
M
B
maximizes A(PN M ) by choosing PN∗ M such that D(PN∗ M ) = M −ωP
.
∗
NM
When δ < 1, the principal values targeting in addition to maximizing total demand. In
this case, the principal will not necessarily choose the monetary price or the non-monetary
price at which demand equals the number of units that the principal can supply. In particular,
when the targeting ratio is decreasing in the monetary price, the principal may choose a low
price such that there is excess demand but a higher targeting ratio. However, a price for which
there is excess supply will never be optimal. If there is excess supply, lowering the price will
improve coverage and targeting by monotonicity of the demand function. For a monetary
B
] =
price, the total number of units distributed is given by A(PM ) = min[D(PM ), M −P
M
B
27
.
M −PM
Similarly, a non-monetary price for which there is excess supply is never optimal so the
B
total number of units distributed can be written as A(PN M ) = min[D(PN M ), M −ωP
] =
NM
B
28
. In addition, when the targeting ratio is increasing in the non-monetary price,
M −ωPN M
a non-monetary price for which there is excess demand is never optimal and the principal
B
. When the targeting
chooses the non-monetary price PN∗ M such that D(PN∗ M ) = M −ωP
∗
NM
ratio is increasing in the non-monetary price, increasing the non-monetary price allows the
B
B
When there is excess demand at a monetary price A(PM ) = M −P
, and A(PN M ) = M −ωP
when
M
NM
there is excess demand at a non-monetary price
27
Define P̄M such that D(P̄M ) = M −BP̄M . Suppose that there is excess supply at a price P̃M , then
26
A(P̃M ) = min[D(P̃M ), M −BP̄M ] = D(P̃M ). Since demand is decreasing in the price and
B
M −PM
is increasing
in the price, P̃M > P̄M . Then A(P̃M ) = D(P̃M ) < D(P̄M ) = A(P̄M ). Therefore, there is greater coverage
with price P̄M than with price P̃M . At both price P̃M and price P̄M , all demand is satisfied. Then Ap (P̃M ) =
Dp (P̃M ) < Dp (P̄M ) = Ap (P̄M ). Therefore, the principal will never choose a monetary price such that there
is excess supply.
28
This follows the same argument as above for monetary prices.
17
principal to allocate weakly more units and increase the targeting ratio resulting in more
units allocated to the poor.29
∗
The principal compares the value of the objective function with price PM
and with price
PN∗ M . The principal will choose the non-monetary price, PN∗ M , if:
∗
M − PM
>
M − wPN∗ M
δ + (1 − δ) ∗
δ + (1 − δ) ∗
∗ )
Dp (PM
∗ )
D(PM
∗
Dp (PN
M)
∗
D(PN
M)
The left hand side of the inequality represents the ratio of the per unit cost paid by the
principal with the monetary price and with the non-monetary price. When δ = 1, the
principal does not place any additional weight on the poor and the right hand side simplifies
to 1. Then the principal simply choses the price type with a lower net cost per unit. When
δ = 0, the principal values only units distributed to the poor and the right hand side simplifies
Dp (P ∗ ) Dp (P ∗ )
to D(P ∗M) / D(P ∗N M) . In this case, the principal compares the difference in the cost per unit to
M
NM
the difference in the targeting achieved by the price types. As δ decreases from 1 to 0, the
principal places increasing weight on the difference in targeting achieved by the price types.
Next, I use the data from the field experiment to evaluate the principal’s problem. The
survey data indicate that targeting the poor is desirable. The poor are significantly less
likely to report that they treat their drinking water, and among households that do report
treating their drinking water, the poor are more likely to report using inferior methods of
water treatment. The poor are also more likely to report that a household member suffered
from an illness perceived to be caused by the drinking water, such as diarrhea. Importantly,
these results appear to be driven by differential water purifier ownership across income
levels as opposed to differential water purifier use (conditional on ownership) across income
levels. The relationship between income and water treatment is discussed in greater detail
in Appendix C and the results are shown in Appendix Tables 3, 4, and 5.
Using the experimental data, I evaluate the principal’s problem in three steps. First, I
estimate the total demand curve and the poor demand curve for each price type. Households
with per capita total daily income less than or equal to the median are used to estimate
Consider a price PN0 M such that there is excess demand. Then PN0 M < PN∗ M where PN∗ M is defined
B
by D(PN∗ M ) = M −ωP
. By increasing the price from PN0 M to PN∗ M the principal can afford to purchase
∗
29
NM
B
B
a weakly greater number of units M −ωP
≤ M −ωP
. This holds with equality when ω = 0 because the
0
∗
NM
NM
number of units that can be purchased does not depend on the non-monetary price. The number of units
allocated to the poor is strictly greater at the higher price since the targeting ratio is increasing in the price:
0
D p (PN
D p (P ∗ )
B
B
M)
< D(P ∗N M) M −ωP
.
∗
D(P 0 ) M −ωP 0
NM
NM
NM
NM
18
the poor demand curve in my sample. The objective of the pre-survey was to oversample
households in the top and bottom quartiles of the population income distribution. Splitting
the sample into rich and poor subsamples at the median ensures that all households in the
bottom quartile of the population income distribution are included in the poor subsample.
Second, using the estimated demand curves, I determine the optimal monetary price and
the optimal non-monetary price for a budget of 120,000 Rs or approximately 150 Rs per
household within the borders.30 I assume that the relevant population has a total demand
curve equivalent to the total demand curve in my sample and that 1/4 of the population
has a demand curve equivalent to the poor demand curve in my sample. Third, I compare
the value of the principal’s objective using the optimal monetary and non-monetary prices
to determine the optimal price.
First, I estimate demand curves for each price type using the WTP data. I create
synthetic observations such that the dataset includes an observation for each respondent at
every possible price. I code take-up as an indicator variable equal to 1 if WTP is greater
than or equal to the price to represent the purchase decision at that price. I estimate smooth
demand curves by regressing take-up on a fourth order polynomial of price.
Table 6 presents the results of estimating demand curves in the full sample and the rich
and poor subsamples. Monetary price regressions include prices 50 Rs-700 Rs and nonmonetary price regressions include prices 1 hr-9 hrs.31 For monetary prices, the demand
curves differ significantly across income levels. I can reject the joint test that the coefficients
for rich and poor are equal (p-value < 0.0001). In contrast, the rich and the poor have
similar demand at non-monetary prices. I cannot reject the joint test that the coefficients
are equal across income levels (p-value of 0.623).
Appendix Figures 3-6 show the estimated demand curves and the underlying demand
data used to estimate the curves. Appendix Figure 3 plots total demand at monetary prices
and Appendix Figure 4 plots poor demand at monetary prices. Appendix Figure 5 plots total
demand at non-monetary prices and Appendix Figure 6 plots poor demand at non-monetary
prices. In each case, the estimated demand curves are a close fit of the data.
It is an assumption that the relevant population has a total demand curve equivalent to
the total demand curve in my sample because the sampling strategy excluded households
with middle incomes.32 However, 11 households reported incomes in the excluded range
30
This is approximately the budget that I had for the health product when implementing the field experiment.
31
Prices are included in the regression if at least 1% of the sample has WTP greater than or equal to the
price.
32
The pre-survey included a categorical income question. Households who reported total monthly income
19
but were included in the sample, and an additional 19 households reported incomes outside
the excluded range in the pre-survey but have income in the excluded range according to
the income section of the full survey.33 I estimate monetary price and non-monetary price
demand curves for these 30 households. Figures 5 and 6 show the total demand curves for
the sample and the demand curves for the 30 households with middle incomes. As shown
in these figures, although the estimated demand curve for the middle income households
is slightly below the confidence interval for low and high prices, the estimate is imprecise
due to the small sample size. For both price types, the confidence interval for the total
demand curve lies in the center of the confidence interval for the middle income households.
This supports the use of the total demand curve for the sample as an estimate for the total
demand curve for the population despite the fact that households from the top and bottom
quartiles were overrepresented in the sample.
Second, I find the optimal monetary price as a function of δ, and the optimal nonmonetary price as a function of δ and ω. Figure 7 plots the optimal monetary price as a
function of δ and shows that when nearly complete weight is placed on the poor (very low
δ), the principal’s monetary objective is maximized at a monetary price for which there
is excess demand and the pieces are randomly allocated among demanders, i.e. according
to the targeting ratio. This price is lower than the market clearing price, and therefore,
has a higher targeting ratio. As the weight on overall take-up increases relative to the
weight on targeting the poor, the principal moves toward the market clearing price which
maximizes overall take-up by equating the number of pieces the principal can purchase and
total demand. The principal slowly increases the monetary price from 278.75 Rs to the
market clearing price of 402.58 Rs as δ increases from 0 to 0.079.
Figure 8 plots the optimal non-monetary price as a function of δ for ω = 50. In this
case, the principal receives 50 Rs of labor for each hour of the non-monetary price. This
is approximately the hourly wage for females in formal employment. Because the targeting
ratio, using the estimated demand curves, is increasing in the non-monetary price, the principal’s non-monetary objective is maximized at the price that equates total demand and the
number of pieces that the principal can purchase with budget B. As shown in Figure 8, the
principal optimally chooses the market clearing non-monetary price, 4.382 hours regardless
between 7,001 Rs and 15,000 Rs were not eligible for the full survey.
33
Most likely the respondent made an error when answering the categorical income question in the presurvey. The full survey collected income data by household member. Of the households in which no household
members report working for a daily income, there are 19 households who did not reported income in the
excluded range but actually have monthly income in the excluded range. I do not consider households in
which one or more households members reports daily income because their monthly income may fluctuate.
20
of δ.
Third, I compare the value of the principal’s objective when using the optimal monetary price and the optimal non-monetary price to determine the price that maximizes the
principal’s objective. Holding ω fixed at 50 Rs, Figure 9 graphs the value of the principal’s
objective with the optimal monetary price and with the optimal non-monetary price for
δ ∈ [0, 1]. The principal’s objective is maximized using a monetary price for a large range
of values for δ. The previous section illustrated that the targeting ratio is higher for nonmonetary prices than for monetary prices. However, as shown in Figure 9, the principal’s
objective is maximized using a non-monetary price only when the principal places nearly
complete weight on the poor. When ω = 50, the monetary price is optimal for δ > 0.136.
When the principal places a relative weight of 0.136 or greater on the rich, the higher targeting ratio achieved by the non-monetary price is outweighed by the greater product cost
recoupment of the monetary price.
Holding δ fixed at 0.527, Figure 10 graphs the value of the principal’s objective with
the optimal monetary price and the optimal non-monetary price for ω ∈ [0, 100]. I focus
on δ = 0.527 because the rich report 52.7% of the number of instances of diarrhea cases
that the poor report. The principal prefers the monetary price until the productivity of the
non-monetary price is at least 85 Rs. For lower levels of productivity, the improved targeting
of the non-monetary price is outweighed by the greater recoupment of product costs using a
monetary price.
6
Robustness Checks
A concern with using the BDM method in practice is that it is difficult to enforce purchase
decisions, eroding the incentive compatibility underlying the method.34 If purchase decisions
are not enforced, respondents could purposely overstate their willingness to pay and then
decide whether or not to purchase the good after the randomly chosen price is revealed.
Surveyors emphasized that purchase decisions would be enforced, but due to logistical constraints during surveying, households had the built-in option of choosing whether or not to
redeem their voucher. 99 of 121 (82%) respondents who received a voucher in the monetary price treatment group redeemed their vouchers and 39 of 50 (78%) respondents who
received a voucher in the non-monetary price treatment group redeemed their vouchers. I
34
The theoretical limitations of using the BDM method to elicit willingness to pay are described in
Appendix B.
21
present evidence using the distribution of WTP, non-incentivized survey data, and voucher
redemption data to argue that respondents are not systematically overstating their WTP.
First, as shown in the pseudo-demand curves, no respondent in the sample states a
monetary WTP above 850 Rs or a non-monetary WTP above 16 hours. This is evidence
that respondents are not systematically reporting their WTP as the price that they guess
to be the maximum possible price since the script that surveyors read to respondents states
the market price (999 Rs) and explicitly addresses non-monetary bids above 16 hours.
Second, using non-incentivized survey data, a small fraction of respondents reported
overestimating their WTP.35 12% of respondents who received a voucher for the water purifier stated that they overestimated their WTP.36 Further, whether or not a respondent
redeems her voucher is not correlated with reported overestimation of WTP. Panel A of Table 7 presents the results in the sample of monetary vouchers, the sample of non-monetary
vouchers, and subsamples by price level.
Third, I test whether respondents systematically overstate their WTP using voucher
redemption data. If respondents were systematically overstating their WTP then, all else
equal, the probability that a voucher is redeemed should be increasing in the difference
between the respondent’s WTP and the price. I regress an indicator for voucher redemption
on the difference between respondent’s WTP and the price in the sample of monetary price
vouchers and the sample of non-monetary price vouchers. For a given WTP, a lower price
indicates a higher difference between WTP and the price. In general, low price vouchers are
more likely to be redeemed than high price vouchers which positively biases the coefficient
on the difference between WTP and the price, increasing the chance that I find evidence of
overstatement of WTP. Despite this bias, as shown in Panel B of Table 7, the coefficient on
the difference is small and insignificant in both samples.37 Across all three types of data,
there is no evidence that respondents systematically overstated their willingness to pay.
Even if the overall level of WTP is not overstated, the cost effectiveness of the nonmonetary program could be overestimated. The cost-effectiveness of a non-monetary price
35
A respondent is coded as overstating her WTP if she states that she wished that she had not stated a
WTP as high as the revealed price.
36
The percentage is similar across price types: 12% in the monetary price treatment group and 10% in
the non-monetary price treatment group.
37
Separating the effect of the difference between WTP and the price from the effect of price, I regress
an indicator for voucher redemption on an indicator which equals 1 if price equals the respondent’s WTP
and 0 if the price is strictly less than the respondent’s WTP controlling for price in the sample of monetary
price vouchers and the sample of non-monetary price vouchers. While the indicator for price equal to WTP
has less power than the variable representing the difference between WTP and price, I find no evidence of
systematic overstatement of WTP.
22
program will be overestimated if the targeting achieved by the non-monetary price is overestimated relative to the targeting achieved by the monetary price or if the partial recoupment
of product costs is underestimated.
The targeting achieved by the non-monetary price program relative to that of the monetary price program will be overestimated if the rich overstate their monetary WTP relative
to the poor or if the poor overstate their non-monetary WTP relative to the rich. I test for
relative overstatement of WTP between the rich and poor using three different methods.
First, I use voucher redemption data to test whether income or wealth has predictive
power for a household’s decision of whether to redeem its voucher. Figures 11 and 12 show
the number of vouchers issued and redeemed by price level. In particular, if the rich were
overstating their monetary WTP relative to the poor, then the rich should be less likely
to redeem monetary price vouchers than the poor. Similarly, if the poor were overstating
their non-monetary WTP relative to the rich, then the poor should be less likely to redeem a
non-monetary price voucher than the rich. Panel C of Table 7 shows the results of individual
regressions of an indicator for voucher redemption on per capita household income and the
asset index controlling for price. Neither per capita household income nor the asset index
significantly predict voucher redemption.
Additionally, I use demographic variables proxying income or wealth to test for differential
voucher redemption by income level. Controlling for price, the respondent’s education level,
the asset index, indicator for credit constrained, and an indicator for whether the respondent
works are not significant predictors of whether a household redeems its voucher in either the
monetary price treatment group or the non-monetary price treatment group. There is no
evidence of differential voucher redemption by income or wealth level. However, because
there are other reasons that a household may not redeem its voucher besides overstatement
of WTP, such as forgetfulness or budget shocks, that are potentially correlated with income,
I directly test for overstatement of WTP using the voucher data.
Second, I follow the same strategy used above to test for universal overstatement of
WTP using voucher redemption data. I regress an indicator for voucher redemption on the
difference between the respondent’s WTP and the price in the four subsamples created by
splitting the sample by income level and price type. Table 8 shows that in all four subsamples
the coefficient on the difference is small. The coefficient is positive and significant (at the
10% level) in the sample of rich households with non-monetary price vouchers. This indicates
that the rich may be more likely to overstate their non-monetary WTP, which would imply
that the targeting results are underestimated. However, overall, there is strong evidence that
23
neither the poor nor the rich are more likely to overstate their WTP.38
Third, I use non-incentivized survey data to determine whether income is correlated with
reported overstatement or understatement of WTP.39 I regress an indicator for reported
overstatement of WTP on an indicator for poor in the sample of households who received
a voucher. The indicator for income level is not significant in either the monetary or the
non-monetary price groups. Similarly, in the sample of households that did not receive a
voucher, the poor were no more likely to report that they underestimated their WTP for
either price type.
Across all three methods, there is no evidence that the rich are overstating their monetary
WTP relative to the poor or that the poor are overstating their non-monetary WTP relative
to the rich. Thus, the targeting results are not driven by systematic differences in reporting
of WTP across income levels.
Finally, I address whether the recoupment of product costs is underestimated. The
amount of product costs recovered is determined by the monetary price required to reach
a desired level of demand among the poor. Therefore, recoupment of product costs will be
underestimated if the poor’s monetary WTP is understated.
Testing for understatement of WTP relies on non-incentivized survey data. While there
may be an incentive for respondents to overstate their WTP because of the built-in redemption option, there is no incentive for respondents to understate their WTP. Using this survey
data, 14% of the poor respondents in the monetary price group who did not receive a voucher
reported that they understated their WTP at the conclusion of the main survey. However,
I expect that this overestimates true understatement of WTP because the question likely
captures other phenomena, such as regret, as well as true understatement of WTP.
38
In order to control for the price, I also use a regression specification in which the dependent variable
is an indicator which equals 1 if price equals the respondent’s WTP and 0 if the price is strictly less
than the respondent’s WTP. This indicator has lower power than the variable representing the difference
between the respondent’s WTP and the price. The indicator is positive and significant in the sample of poor
households with non-monetary price vouchers indicating that the poor may be more likely to overstate their
non-monetary WTP. The coefficient on the indicator is small and insignificant in each of the other three
samples.
39
A respondent is coded as overstating her WTP if she states that she wished that she had not stated a
WTP as high as the revealed price and is coded as understating her WTP if she states that she wished that
she had stated a willingness to pay at least as high as the revealed price.
24
7
Conclusion
Although health products have high private and social returns, especially for the poor, takeup rates are low even at highly subsidized monetary prices. Many governments and NGOs
distribute subsidized health products in an effort to increase take-up. Non-monetary prices
may better target the poor because selection occurs on willingness to pay in time rather than
money.
I develop a model of household demand based on households’ decision whether or not to
purchase the health product when the product is offered at a monetary price and at a nonmonetary price that illustrates three features of demand. First, I show that for any monetary
price rich demand exceeds poor demand. Second, the model shows that non-monetary prices
may not result in greater demand among the poor than the rich due to opposing income
and substitution effects. For a concave utility function, the utility cost of the non-monetary
price may not be increasing in the wage. Third, regardless of whether the utility cost of
a non-monetary price is increasing in the wage, relative to the rich, the utility cost of a
monetary price is greater than the utility cost of a non-monetary price for the poor. This
implies that non-monetary prices result in better targeting of the poor than monetary prices.
In order to empirically test whether non-monetary prices target health products to the
poor better than monetary prices, I conducted a field experiment in Hyderabad, India.
To obtain estimates of demand at monetary and non-monetary prices, I used the BeckerDeGroot-Marshak (BDM) mechanism (Becker, DeGroot, Marshak, 1964) to elicit respondents’ willingness to pay (WTP) at monetary prices and at non-monetary prices. I find that
monetary WTP is increasing in income and I can reject that the rich and poor have the same
demand at monetary prices. In contrast, non-monetary WTP in decreasing in income, but
I cannot reject that rich and poor demand are the same for non-monetary prices. Using a
pooled sample of monetary WTP and non-monetary WTP, non-monetary WTP falls relative
to monetary WTP as income rises, and non-monetary prices result in better targeting than
monetary prices.
In order to evaluate non-monetary prices as a policy to allocate health products, I model
the problem of a principal who has a fixed budget dedicated to distributing water purifiers.
The principal can choose to use either a monetary price or a non-monetary price to allocate
the health product. The principal’s objective is to increase take-up of the health product,
in particular by the poor because the poor have greater need for the health product.
Using demand functions estimated from the experimental willingness to pay data, I determine the optimal price given the additional weight that the principal places on the poor
25
and the productivity of the non-monetary price. I show that the monetary price is preferred
for a larger range of parameters. Increasing the productivity of the non-monetary price
increases the range of parameters for which the non-monetary price is optimal.
26
References
[1] Abdul Latif Jameel Poverty Action Lab (J-PAL). “The Price is Wrong: Charging Small
Fees Dramatically Reduces Access to Important Products for the Poor.” J-PAL Bulletin
April 2011.
[2] Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, Ririn Purnamasari,
and Matthew-Wai-Poi. 2013. “Self-Targeting: Evidence from a Field Experiment in
Indonesia.” Working Paper.
[3] Arnold, Benjamin F., and John Colford, Jr. 2007. “Treating Water with Chlorine at
Point-of-Use to Improve Water Quality and Reduce Child Diarrhea in Developing Countries: a Systematic Review and Meta-Analysis.” American Journal of Tropical Medicine
and Hygiene, 76(2), 354-364.
[4] Ashraf, Nava, James Berry, and Jesse Shapiro. 2010. “Can Higher Prices Stimulate
Product Use? Evidence from a Field Experiment in Zambia.” American Economic
Review, 100 (5): 2383-2413.
[5] Baird, Sarah, Joan Hamory Hicks, Michael Kremer, and Edward Miguel. 2011. “Worms
at Work: Long-Run Impacts of Child Health Gains.” Unpublished.
[6] Barzel, Yoram. 1974. “A Theory of Rationing by Waiting.” Journal of Law and Economics, XVII (April 1974): 73-96.
[7] Becker, Gordon, Morris DeGroot, and Jacob Marschak. 1964. “Measuring Utility by a
Single-Response Sequential Method.” Behavioral Science, 9, 226-236.
[8] Berry, James, Greg Fischer, and Raymond Guiteras. 2012. “Eliciting and Utilizing Willingness to Pay: Evidence from Field Trials in Northern Ghana.” Working Paper.
[9] Besley, Timothy and Stephen Coate. 1992. “Workfare versus Welfare: Incentive Arguments for Work Requirements in Poverty-Alleviation Programs.” American Economic
Review, 82 (1): 249-261.
[10] Cohen, Jessica, and Pascaline Dupas. 2010. “Free Distribution or Cost-Sharing? Evidence from a Randomized Malaria Prevention Experiment.” Quarterly Journal of Economics, 25 (1): 1-45.
[11] Devoto, Florencio, Esther Duflo, Pascaline Dupas, William Parienté, Vincent Pons.
2012. “Happiness on Tap: Piped Water Adoption in Urban Morocco. American Economic Journal: Economic Policy, 4(4): 68-99.
[12] Dupas, Pascaline. 2009. “What Matters (and What Does Not) in Households’ Decision
to Invest in Malaria Prevention?” American Economic Review, 99 (2): 224-30.
27
[13] Dupas, Pascaline. 2014a. “Short-Run Subsidies and Long-Run Adoption of New Health
Products: Evidence from a Field Experiment.” Econometrica 82 (1): 197-228.
[14] Dupas, Pascaline. 2014b. “Getting Essential Health Products to Their End Users: Subsidize but How Much?” Science, 345(6202): 1279-1281, 12 September 2014.
[15] Dupas, Pascaline, Vivian Hoffmann, Michael Kremer, and Alix Peterson Zwane. 2013.
“Micro-Ordeals, Targeting, and Habit Formation.” Working Paper.
[16] Dupas, Pascaline and Jonathan Robinson. 2013. “Why Don’t the Poor Save More?
Evidence from Health Savings Experiments.” American Economic Review, 103(4): 11381171.
[17] Horowitz, John. 2006. “The Becker-DeGroot-Marschak Mechanism is Not Necessarily
Incentive Compatible, Even for Non-random Goods.” Economic Letters, 93(1): 6-11.
[18] Kremer, Michael, and Edward Miguel. 2007. “The Illusion of Sustainability.” Quarterly
Journal of Economics, 122 (3): 1007-1065.
[19] Ma, Xiaochen, Sean Sylvia, Matthew Boswell, and Scott Rozelle. 2013. “Ordeal Mechanisms and Information in the Promotion of Health Goods in Developing Countries:
Evidence From Rural China.” Working Paper.
[20] Mazar, Nina, Botond Koszegi, and Dan Ariely. 2010. “Price-Sensitive Preferences.”
SSRN eLibrary.
[21] Meredith, Jennifer, Jonathan Robinson, Sarah Walker, and Bruce Wydick. 2013. “Keeping the Doctor Away: Experimental Evidence on Investment in Preventative Health
Products.” Journal of Development Economics, 105: 196-210.
[22] Miguel, Edward, and Michael Kremer. 2004. “Worms: Identifying Impacts on Education
and Health in the Presence of Treatment Externalities.” Econometrica, 72(1), 152-217.
[23] Miller, Klaus M., Reto Hoffstetter, Harley Krohmer, and Z. John Zhang. 2011. “How
should Consumer’s Willingness to Pay be Measured? An Empirical Comparison of
State-of-the-art approaches.” Journal of Marketing Resarch, 48(1): 172-184.
[24] Nichols, D., E. Smolensky, and T.N. Tideman. 1971. “Discrimination by Waiting Time
in Merit Goods.” American Economics Review, LXI (June 1971): 312-323.
[25] Nichols, Albert, and Richard Zeckhauser. 1982. “Targeting Transfers through Restrictions on Recipients.” American Economic Review, 72 (2): 372-377.
[26] Niehaus, Paul, and Sandip Sukhtankar. 2013. “Corruption Dynamics: The Golden
Goose Effect.” American Economic Journal: Economic Policy, 5(4): 230-269.
[27] Soman, Dilip. 2001. “The Mental Accounting of Sunk Time Costs: Why Time is not
Like Money.” Journal of Behavioral Decision Making, 14, 169-185.
28
[28] Tarozzi, Alessandro, Aprajit Mahajan, Brian Blackburn, Dan Kopf, Lakshmi Krishnan, and Joanne Yoong. 2014. “Micro-loans, Bednets and Malaria: Evidence from a
Randomized Controlled Trial in Orissa (India).” American Economic Review, 104(7):
1909-1941.
[29] Thomas, Duncan, Elizabeth Frankenberg, Jed Friedman, et al. 2006. “Causal Effect of
Health on Labor Market Outcomes: Experimental Evidence.” Online Working Paper
Series. California Center for Population Research, UCLA.
29
Tables
Table 1: Household and Respondent Summary Statistics
Full
Rich
Poor
(1)
(2)
(3)
Panel A: Household Statistics
Household contains children under 14
Household owns water filter at baseline
Household reports treating children’s water at baseline
Household reports treating adults’ water at baseline
Household treats children’s water with cloth or net
Household treats adults’ water with cloth or net
Household episode of diarrhea
Average household size
Average daily household labor earnings
Average daily household non-labor earnings
72%
60%
84%
30%
37%
24%
74%
78%
71%
72%
75%
69%
46%
37%
53%
56%
49%
64%
15%
9%
19%
4.7
4.6
4.8
515 Rs 780 Rs 261 Rs
39 Rs 64 Rs 15 Rs
Panel B: Respondent Statistics
Female respondent
Respondent has no formal schooling
Respondent earns labor income
Average daily respondent labor earnings (given work)
Average hourly respondent wage (given work)
86%
26%
34%
217 Rs
32 Rs
84%
89%
23%
28%
36%
33%
313 Rs 118 Rs
43 Rs 19 Rs
Note: Methods of water treatment include boiling, chemical treatment, water purifiers or filters, and
straining water through a plastic net or cotton cloth. The percentage of households that treat children’s drinking water is the percentage of households with children under the age of 14 who report
treating their drinking water. Households are coded as treating their drinking water if they report
that they treat their drinking water all year or in the rainy season. Households that treat their water
with cloth or net is reported as a percentage of households that treat their drinking water Households
are coded as treating their drinking water with a cloth or net if these are the only methods used.
30
Table 2: Income and Willingness to Pay Monetary Prices
(1)
(2)
Panel A: Daily Labor Income
Log household income
0.193***
(0.032)
(4)
0.164***
(0.043)
No
260
0.168***
(0.044)
Yes
260
0.174***
(0.043)
No
260
0.177***
(0.045)
Yes
260
0.179**
(0.077)
No
260
0.177**
(0.082)
Yes
260
0.196***
(0.033)
Log respondent income
Area fixed effects
N
(3)
No
730
Yes
730
Panel B: Daily Total Income
Log household income
0.190***
(0.028)
0.192***
(0.029)
Log respondent income
Area fixed effects
N
No
739
Yes
739
Panel C: Hourly Wage
Log household income
0.176***
(0.038)
0.174***
(0.038)
Log respondent income
Area fixed effects
N
No
725
Yes
725
Note: The dependent variable is log of monetary WTP measured in rupees. The mean of the dependent
variable in Panel A, column (2) is 290 Rs. The mean of household daily labor income in Panel A, column (2)
is 517 Rs. This implies that an additional 52 Rs of daily household labor income increases monetary WTP
by 5.8 Rs at the mean. The mean of the dependent variable in Panel A, column (4) is 307 Rs. The mean
of respondent daily labor income is 216 Rs in Panel A, column (2). This implies that an additional 21 Rs
of respondent daily labor income increases monetary WTP by 5.2 Rs at the mean. Robust standard errors
in parentheses. In Panel C, the independent variable is the mean household hourly wage of all household
members who work for a wage.
31
Table 3: Income and Willingness to Pay Non-monetary Prices
(1)
Panel A: Daily Labor Income
Log household income
-0.048**
(0.021)
(2)
No
683
(4)
-0.006
(0.046)
No
243
-0.006
(0.053)
Yes
243
-0.012
(0.045)
No
243
-0.014
(0.053)
Yes
243
0.001
(0.074)
No
243
-0.019
(0.070)
Yes
243
-0.046**
(0.020)
Log respondent income
Area fixed effects
N
(3)
Yes
683
Panel B: Daily Total Income
Log household income
-0.049**
(0.020)
-0.047**
(0.020)
Log respondent income
Area fixed effects
N
No
693
Yes
693
Panel C: Hourly Wage
Log household income
-0.094***
(0.028)
-0.086**
(0.031)
Log respondent income
Area fixed effects
N
No
678
Yes
678
Note: The dependent variable is the log of non-monetary WTP measured in minutes. The mean of the dependent variable is 171 minutes in Panel A, column (2). The mean of household daily labor income is 517
Rs in Panel A, column (2). This implies that an additional 52 Rs of daily household labor income decreases
non-monetary WTP by 8.6 minutes at the mean. Robust standard errors in parentheses. In Panel C, the
independent variable is the mean household hourly wage of all household members who work for a wage. In
each panel, I cannot reject that the coefficient on household income is equal to the coefficient on respondent
income.
32
33
Yes
1332
-0.170
(0.382)
-0.229***
(0.064)
(1)
-0.201***
(0.070)
Yes
478
-0.591*
(0.355)
(2)
Daily Labor Income
Yes
1348
-0.189
(0.365)
-0.223***
(0.061)
(3)
-0.218***
(0.072)
Yes
478
-0.501
(0.368)
(4)
Daily Total Income
(6)
-0.627** -0.888**
(0.267)
(0.357)
-0.263***
(0.144)
-0.221*
(0.112)
Yes
Yes
1322
478
(5)
Hourly Wage
Note: The dependent variable is log monetary WTP (Rs) or log non-monetary WTP measured in hours scaled by the median respondent
hourly wage, 23 Rs per hour. Column headings are the specific income variable used in the regression. Robust standard errors are clustered
by household.
Household fixed effect
N
Non-monetary*Respondent income
Non-monetary*Household income
Non-monetary
Measure of Income:
Table 4: Pooled: Income and Willingness to Pay
Table 5: Targeting Poor Households
Price
(1)
50
100
150
200
250
300
350
400
450
500
550
600
1
2
3
4
5
6
7
Poor Take-Up
Total Take-Up
Targeting Ratio
(2)
(3)
Panel A: Monetary Price (Rs)
(4)
380
742
0.51
360
712
0.51
317
637
0.50
277
574
0.48
197
443
0.44
145
356
0.41
103
261
0.39
74
201
0.37
44
144
0.31
36
122
0.30
12
54
0.22
6
28
0.33
Panel B: Non-Monetary Price (Hrs)
358
273
181
99
60
31
18
687
530
329
188
106
57
28
0.52
0.52
0.55
0.53
0.57
0.54
0.64
Note: Prices are included in the table if total demand is at least 20
households. Households with per capita income less than or equal to
the median in the sample are considered poor. The poor comprise
51% of the sample because many households report the median per
capita household income.
34
Table 6: Demand for Water Purifiers at Monetary and Non-monetary Prices
Monetary Price
Constant
Price
Price squared
Price cubed
Price fourth
N
Non-monetary Price
Full Sample
Poor
Rich
Full Sample
Poor
Rich
(1)
(2)
(3)
(4)
(5)
(6)
1.0000
1.0000
1.0000
1.0000
1.0000
1.0000
-0.0291
(0.0211)
-0.0988***
(0.0143)
0.0190***
(0.0032)
-0.0010***
(0.0002)
0.0007
(0.0301)
-0.1449***
(0.0201)
0.0306***
(0.0044)
-0.0018***
(0.0003)
-0.0601***
(0.0295)
-0.0501***
(0.0202)
0.0070
(0.0045)
-0.0002
(0.0003)
-0.0860***
(0.0188)
-0.0705***
(0.0092)
0.0143***
(0.0015)
-0.0007***
(0.0001)
11,172
5,698
5,474
7,182
-0.0661** -0.1067***
(0.0255)
(0.0278)
-0.0773*** -0.0635***
(0.0125)
(0.0136)
0.0151*** 0.1341***
(0.0021)
(0.0022)
-0.0008*** -0.0007***
(0.0001)
(0.0001)
3,663
3,519
Note: Each regression includes one observation for each household at each price level. The dependent variable
represents the household’s purchase decision and is equal to 1 if WTP≥ price. Rich households are households
with per capita daily household income above the median of the sample. Monetary price regressions include prices
50 Rs-700 Rs and non-monetary price regressions include prices 1 hrs-9 hrs. Monetary prices are in 100 Rs units
and non-monetary prices are in hours. Robust standard errors are clustered by household.
35
36
Price
< 150 Rs
Full
Non-monetary
86
-0.117
(0.146)
35
0.063
(0.044)
86
0.453 × 10−4
(0.374 × 10−3 )
35
0.351 × 10−3
(0.251 × 10−3 )
121
0.198
(0.175)
0.732 × 10−4
(0.244 × 10−3 )
86
0.256
(0.222)
35
0.032
(0.180)
0.123 × 10−3 −0.617 × 10−4
(0.311 × 10−3 ) (0.328 × 10−3 )
Panel C: Income and Voucher Redemption
121
0.758 × 10−4
(0.232 × 10−3 )
Panel B: WTP-Price and Voucher Redemption
121
-0.097
(0.121)
50
0.112
(0.262)
-0.001
(0.001)
50
0.021
(0.017)
50
-0.200
(0.232)
29
-0.278
(0.376)
(6)
Price
<3 Hr
29
21
0.093
(0.123)
29
-0.093
(0.530)
0.001
-0.001
(0.001) (0.001)
21
-0.117
0.031
(0.044) (0.022)
21
-0.167
(0.301)
(5)
Price
≥3 Hr
Non-Monetary
In Panel C, the household income variable is per capita daily total household income and the asset index is the first principal component of
variables representing ownership of five durable goods: automobiles, motorbikes/scooters, tablets/computers, washing machines, and evaporative coolers. The index is normalized to have mean 0 and standard deviation 1. Robust standard errors in parentheses.
N
Asset index
Household income
N
WTP-price
N
WTP overstatement
Price
≥ 150 Rs
(1)
(2)
(3)
(4)
Panel A: Reported Overstatement of WTP and Voucher Redemption
Full
Monetary
Monetary
Table 7: Voucher Redemption
Table 8: Relative Overstatement by Income Level
Monetary Price
WTP-price
N
Non-monetary Price
Rich
Poor
Rich
Poor
(1)
(2)
(3)
(4)
0.027*
(0.016)
22
0.006
(0.043)
28
0.294 × 10−5 −0.203 × 10−3
(0.348 × 10−3 ) (0.295 × 10−3 )
69
52
Poor households are households with per capita household income below the median of the sample. Robust standard errors in parentheses.
37
Know Tata Swach
|
Awards & Reviews
|
Tata Sw
Figures
How to assemble your Tata Swach Smart
TATA Swach Smart Design
Know your Tata Swach Smart
In t
Th
Pr
Te
Pu
Figure 1
The TATA Swach Smart follows US EPA guidelines for water quality. The purifier can hold 7.5
liters of purified water in the lower chamber and 7.5 liters of water waiting to be purified in the
upper purifier. The Swach Smart is economical in terms of both the initial cost of the water
purifier and the cost per liter compared to similar water purifiers (for example, the PureIt brand).
The market price of the TATA Swach Smart is 999 Rs (approximately 16.50 USD). Each bulb
can purifier approximately 3,000 liters of water before requiring a replacement which costs 450 Rs
(approximately 7.50 USD). This implies a cost per liter of 0.2 cents after purchasing the purifier.
Copyright © 2009-­12 TATA Chemicals Ltd. | Terms & conditions | Privacy Policy | Sitemap | Website maintained b
38
Pseudo-demand at Monetary Prices (by income level)
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850 900 950 999
Price (Rs)
Poor Pseudo-demand
Rich Pseudo-demand
Figure 2
This figure shows the percentage of respondents who stated a monetary willingness to pay at least
as high as the price for each price level. Willingness to pay was elicited for one water purifier
through the Becker-DeGroot-Marshak (1964) mechanism. The Kolmogorov Smirnov test rejects
equality of the rich and poor distributions of monetary willingness to pay (p-value < 0.001).
39
Pseudo-demand at Non-monetary Prices (by income level)
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1
2
3
4
5
6
7
8
9 10 11
Price (Hrs)
Poor Pseudo-demand
12
13
14
15
16
17
Rich Pseudo-demand
Figure 3
This figure shows the percentage of respondents who stated a non-monetary willingness to pay at
least as high as the price for each price level. Willingness to pay was elicited for one water purifier
through the Becker-DeGroot-Marshak (1964) mechanism. The Kolmogorov Smirnov test does not
reject equality of the rich and poor distributions of non-monetary willingness to pay (p-value of
0.326).
40
18
Targeting Ratios for Monetary and Non-monetary Prices
0.7
0.6
Targeting Ratio
0.5
0.4
0.3
0.2
0.1
0
400
350
300
250
200
150
Poor Demand
Monetary Prices
100
50
Non-monetary Prices
Figure 4
This figure graphs the targeting ratios for monetary prices and non-monetary prices by the level of
poor demand. The level of poor demand is decreasing on the x-axis because demand decreases as
prices rise. Therefore, moving right on the x-axis represents increasing prices. The targeting ratio
for monetary prices is always below the targeting ratio for non-monetary prices and the difference
in targeting ratios increases as prices rise.
41
0
Total Demand and Middle Income Demand at Monetary Prices
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
0
100
200
300
400
Monetary Price (Rs)
500
600
Middle Income Demand
Total Demand
Middle Income Confidence Interval
Total Demand Confidence Interval
Figure 5
This figure compares demand at monetary prices for the middle income households to the total
demand. The dashed lines show the demand curve and 95% confidence interval for the 30 middle
income households. The solid lines show the total demand curve and 95% confidence interval for
the sample. Although the estimated demand curve for the middle income households is slightly
below the confidence interval for low and high monetary prices, the estimate is imprecise due to
the small sample size. The confidence interval for the total demand curve lies in the center of the
confidence interval for the middle income households.
42
700
Total Demand and Middle Income Demand at Non-monetary Prices
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
0
1
2
3
4
5
6
Non-monetary Price (Hrs)
7
8
Middle Income Demand
Total Demand
Middle Income Confidence Interval
Total Demand Confidence Interval
Figure 6
This figure compares demand at non-monetary prices for the middle income households to the total
demand. The dashed lines show the demand curve and 95% confidence interval for the 30 middle
income households. The solid lines show the total demand curve and 95% confidence interval for
the sample. Although the estimated demand curve for the middle income households is slightly
below the confidence interval for low and high non-monetary prices, the estimate is imprecise due
to the small sample size. The confidence interval for the total demand curve lies in the center of
the confidence interval for the middle income households.
43
9
Optimal Monetary Price
450
400
350
Price (Rs)
300
250
200
150
100
50
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Relative Weight on Rich Demand (δ)
0.8
0.9
Figure 7
This figure shows optimal monetary price as a function of the relative weight on rich demand.
When the principal only values goods distributed to the poor, δ = 0, the optimal monetary price is
278.75 Rs, below the market clearing price of 402.58 Rs. The optimal price is the market clearing
price for δ >= 0.079.
44
1
Optimal Non-monetary Price (ω=50)
5.0
4.5
4.0
Price (Hrs)
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Relative Weight on Rich Demand (δ)
0.8
0.9
Figure 8
This figure shows optimal non-monetary price as a function of the relative weight on rich
demand for ω = 50. This assumes that the productivity of an hour of the non-monetary
price is 50 Rs. Regardless of δ, the principal optimally chooses the market clearing price of
4.382 hours.
45
1
Value of Principal's Objective (ω=50)
250
Value of Objective
200
150
100
50
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Relative Weight on Rich Demand (δ)
Monetary
0.8
0.9
Non-monetary
Figure 9
This figure shows value of the principal’s objective using the optimal monetary price and the
optimal non-monetary price assuming that the productivity of the non-monetary price is 50 Rs.
For δ <= 0.136, the principal prefers the non-monetary price, and for δ > 0.136, the principal
prefers the monetary price.
46
1
Value of Principal's Objective (δ=0.527)
140
120
Value of Objective
100
80
60
40
20
0
0
10
20
30
40
50
60
70
Productivity of Non-monetary Price (ω)
Monetary
80
90
Non-monetary
Figure 10
This figure shows the value of the principal’s objective using the optimal monetary price and the
optimal non-monetary price for δ = 0.527. For this level of δ, the principal prefers the monetary
price unless the productivity of the non-monetary price is at least 85 Rs per hour.
47
100
Monetary Price Vouchers Issued and Redeemed (by income level)
30
Number of Vouchers
25
20
15
10
5
0
50-100
150-200
250-300
350-400
Price Range (Rs)
450-500
Poor Redeemed
Rich Redeemed
Poor Not Redeemed
Rich Not Redeemed
550-600
Figure 11
This figure shows the number of monetary price vouchers for the water purifier that were issued
(willingness to pay≥ price) and redeemed in the monetary price group. The size of the solid plus
shaded bars represent the number of vouchers issued in each price bin. The solid bars represent the
number of vouchers redeemed in each price bin. Vouchers are redeemed on the Saturday following
the household survey at a neighborhood community center, temple, or school.
48
Non-Monetary Price Vouchers Issued and Redeemed (by income level)
16
Number of Vouchers
14
12
10
8
6
4
2
0
1-2
3-4
5-6
7-8
Price Range (Hrs)
Poor Redeemed
Rich Redeemed
Poor Not Redeemed
Rich Not Redeemed
9-10
11-12
Figure 12
This figure shows the number of non-monetary price vouchers for the water purifier that were
issued (willingness to pay≥ price) and redeemed in the non-monetary price group. The size of
the solid plus shaded bars represent the number of vouchers issued in each price bin. The solid
bars represent the number of vouchers redeemed in each price bin. Vouchers are redeemed on the
Saturday following the household survey at a neighborhood community center, temple, or school.
49
Appendix A
Eliciting Willingness to Pay for the TATA Swach Smart
TATA SWACH SMART
READ: The TATA swach smart water purifier uses new technology to effectively destroy bacteria and viruses. The
purifier follows United States Environmental Protection Agency guidelines. The TATA swach technology was
successfully tested at institutions in India, the UK, and the Netherlands.
After 3,000 litres (about 3 months), the bulb will need to be replaced. A new bulb costs 450 Rs. The purifier has a
storage capacity of 7.5 litres. The filter does not require electricity or your own water tap. The TATA swach smart has
a MRP of 999 Rs.
[Show respondent water purifier and let them examine it.]
Do you have any questions about the TATA Swach Smart?
[Answer respondent’s questions.]
We are thinking about selling this water purifier at a subsidized price. Now we are going to give you a chance to
purchase this water purifier. We are not here to make a profit. We are here for project work. There are two sections
to this part of the survey. In one section we will ask you how much money you are willing to spend for the water
purifier. In the other section we will ask you how much time you are willing to spend separating seeds for the water
purifier. One of these two sections will be chosen by lottery and your answers in that section will determine whether or
not you actually purchase the water purifier. The discount you receive will also be decided by lottery. Some people
will receive a larger discount than other people.
If you choose to purchase a water purifier, you will receive a voucher today. You will redeem the voucher, pay for the
water purifier and receive the water purifier on _____________________ at
_________________________________________________________________________________________.
Only you or your spouse can redeem the voucher. Other members of your household cannot redeem the
voucher.
50
18
Section J: Willingness to Pay in MONEY
Interviewer Instructions: Read the following Introduction to the respondent and answer all of their questions.
Introduction:
•
•
•
•
•
•
•
•
•
READ
[Give respondent water purifier.] Please examine this water purifier. The manufacturer’s retail price of 999
Rs is listed on the box.
First, I am going to ask you how much money you will pay to purchase this water purifier. Please think
carefully about how much money you will pay, even if it is a small amount.
Second, if this section is chosen, you will pick a price out of this box. [Give respondent box of prices to hold.]
There are many possible subsidized prices in this box. The prices in the box will not change.
If the price that you picked is less than or equal to the amount that you said you will pay then you will buy the
water purifier. You will pay the price that you have picked out of the box.
If the price is more than you said that you will pay then you cannot buy the water purifier. You will not pay
anything.
You cannot change your mind after you have picked a price from the box.
You or your spouse will pay for and receive the water filter on _______________________ between
10am and 6pm at _____________________________________________________________.
You must be able to pay the amount that you say that you will pay.
Do you have any questions?
THE FOLLOWING QUESTIONS AND RECORD ANSWERS.
A
J.01
What is the most that you will
pay for this water purifier on
_________________?
B
0 Rupees
RUPEES
0
J.04A
0 Rupees
RUPEES
Sect K
Refuse to answer.
RUPEES
-997
Sect K
Yes
.. 1
No
..
.....2
Not Applicable
..-777
Refuse to answer....-997
RUPEES
J.03A
J.01B
J.03A
J.01B
Yes
.. 1
No
..
.....2
Not Applicable
..-777
Refuse to answer....-997
RUPEES
J.03B
Sect K
J.03B
Sect K
Yes
... ...1
No
..
......2
Refuse to answer .-997
RUPEES
J.04A
J.01B
J.01B
Yes
.. .1
No
..
...2
Refuse to answer ..-997
RUPEES
J.04B
J.08
Sect K
Yes
.1
No
...2
Refuse to answer.....-997
RUPEES
J.05A
J.08
J.01B
Yes
.
No
.
Refuse to answer.
RUPEES
.1
...2
-997
J.05B
J.08
Sect K
Yes
.1
No
...2
Refuse to answer.....-997
J.06A
J.08
J.01B
Yes
.1
No
...2
Refuse to answer....-997
J.06B
J.08
Sect K
Refuse to answer.
RUPEES
J.02
Will you buy the water purifier
for [J.01-50]?
J.03
Will you buy the water purifier
for [J.01]?
J.04
J.05
Will you buy the water purifier
for [J.01 +50]?
Will you buy the water purifier
for [J.01 +100]?
-997
0
J.04B
19
51
RUPEES
J.06
Will you buy the water purifier
for [J.01 +150]?
J.07
Will you buy the water purifier
for [J.01 +200]?
J.08
Are any questions in column B
completed?
i) What is the most that the
respondent will pay?
Record the amount of
Rupees
from
the
last
question
with
response
yes=1 completed in Column
A. If none, then record 0.
J.09
RUPEES
Yes
.1
No
...2
Refuse to answer.....-997
RUPEES
J.07A
J.08
J.01B
Yes
.1
No
...2
Refuse to answer....-997
RUPEES
J.07B
J.08
Sect K
Yes
.1
No
...2
Refuse to answer.....-997
J.07a
J.08
J.01B
Yes
.1
No
...2
Refuse to answer....-997
J.07a
J.08
Sect K
Yes
No
..
..
..1
2
J.09ii
J.09i
RUPEES
ii) What is the most that the
respondent will pay?
Record the amount of
Rupees
from
the
last
question
with
response
yes=1 completed in Column
B. If none, then record 0.
20
52
Section K: Willingness to Pay in TIME
Interviewer Instructions: Read the following Introduction to the respondent and answer all of their questions.
Introduction:
•
•
•
•
•
•
•
•
•
•
READ
[Give respondent water purifier.] Please examine this water purifier. The manufacturer’s retail price of 999
Rs is listed on the box.
First, I am going to ask you how much time you spend separating seeds by type for this water purifier. Please
think carefully about how much time you will spend separating seeds, even if it is a small amount.
Second, if this section is chosen, you will pick an amount of time out of this box. [Give respondent box of
times to hold.] There are many possible amounts of time in this box. The amounts of time in the box will not
change.
If the amount of time that you picked is less than or equal to the amount of time that you said you will spend
then you will get the water purifier. You will spend the amount of time that you have picked out of the box in
exchange for the water purifier.
If the amount of time is more than you said that you will spend then you cannot get the water purifier. You will
not spend any time separating seeds.
You cannot change your mind after you have picked an amount of time from the box.
You or your spouse will separate seeds and receive the water purifier on __________________
between 10am and 6pm at _______________________________________________________________.
If you say that you will spend more than 8 hours separating seeds then you will continue on the next day. If
you say that you will spend more than 16 hours separating seeds then you will continue on the following day.
You must be able to spend the amount of time separating seeds that you say that you will spend.
Do you have any questions?
THE FOLLOWING QUESTIONS AND RECORD ANSWERS.
A
K.01
What is the most time that
you will spend separating
seeds for this water purifier
on
____________________?
B
0 Hours
HOURS
Refuse to answer.
0
K.04A
0 Hours
HOURS
-997
Sect L
Refuse to answer.
HOURS
K.02
Will you spend [K.01-1 hour]
hours separating seeds for
the water purifier?
K.04
Will you spend [K.01] hours
separating seeds for the
water purifier?
Will you spend [K.01 +1
hour] hours
separating
seeds for the water purifier?
0
K.04B
-997
Sect L
Yes
..1
No
..
.....2
Not Applicable
..-777
Refuse to answer .-997
K.03B
Sect L
K.03B
Sect L
HOURS
Yes
.. .1
No
..
...2
Not Applicable
-777
Refuse to answer ..-997
K.03A
K.01B
K.03A
K.01B
HOURS
K.03
.
HOURS
Yes
.. .1
No
..
...2
Refuse to answer ..-997
K.04A
K.01B
K.01B
HOURS
Yes
.. .1
No
..
...2
Refuse to answer..-997
K.04B
K.08
Sect L
HOURS
Yes
.1
No
...2
Refuse to answer.....-997
K.05A
K.08
K.01B
Yes
.
No
.
Refuse to answer.
.1
...2
-997
K.05B
K.08
Sect L
21
53
HOURS
K.05
Will you spend [K.01 + 2
hours] hours separating
seeds for the water purifier?
K.06
Will you spend [K.01 + 3
hours] hours separating
seeds for the water purifier?
K.07
Will you spend [K.01 + 4
hours] hours separating
seeds for the water purifier?
K.08
Are any questions in column
B completed?
HOURS
Yes
.1
No
...2
Refuse to answer.....-997
HOURS
K.06A
K.08
K.01B
Yes
.1
No
...2
Refuse to answer....-997
HOURS
K.06B
K.08
Sect L
Yes
.1
No
...2
Refuse to answer.....-997
HOURS
K.07A
K.08
K.01B
Yes
.1
No
...2
Refuse to answer....-997
HOURS
K.07B
K.08
Sect L
Yes
.1
No
...2
Refuse to answer.....-997
K.07a
K.08
K.01B
Yes
.1
No
...2
Refuse to answer....-997
K.07a
K.08
Sect L
Yes
No
..
..
..1
2
K.09ii
K.09i
i) What is the most that the
respondent will pay?
Record the number of
hours from the last
question with response
yes=1
completed
in
Column A. If none, then
record 0.
K.09
HOURS
ii) What is the most that the
respondent will pay?
Record the number of
hours from the last
question with response
yes=1
completed
in
Column B. If none, then
record 0.
22
54
Appendix B
The BDM Mechanism
For expected utility maximizers, the utility of a good can be summarized by its value under
certainty. In this case, the BDM mechanism will elicit “true” willingness to pay as the
dominant strategy is to truthfully reveal your maximum willingness to pay.40
However, the BDM mechanism will not necessarily elicit “true” willingness to pay for
individuals who are not expected utility maximizers. In this case, the individual’s bid may
depend on the distribution of prices (Horowitz, 2006). Intuitively, after the bid is reported,
there is still uncertainty regarding the price and whether the good will be received. Because
this uncertainty is determined in part by the bid, the bid may depend on the distribution of
possible prices. Mazar et al. (2010) studies the effect of information regarding the distribution of possible prices on BDM bids in a laboratory experiment. They find that titrationbased BDM in which respondents make single buy/not buy decisions for a sequence of prices
nearly eliminated the sensitivity of BDM bids to the price distribution. The procedure used
in this study to elicit willingness to pay is similar to their titration-based procedure.
Two recent experiments have addressed the practical question of whether demand as
determined by BDM bids is closely related to market demand. Berry, Fischer, and Guiteras
(2012) estimate demand for a water filter in Ghana using BDM bids and take-it-or-leave-it
offers. For each price considered, they find that a smaller share of households purchase under
BDM than under take-it-or-leave-it. However, the authors note that in this environment
most goods are purchased through bargaining so normal market demand could more closely
match demand in the BDM treatment or demand in the take-it-or-leave-it treatment. In
contrast, in a sample of Swiss consumers, Miller, Hofstetter, Krohmer, and Zhang (2011)
find that demand determined by BDM bids closely matches demand determined by real
purchase decisions with a single randomly-drawn price in an online shop.
40
“True” willingness to pay is the highest price at which the individual would purchase the good at a
known price.
55
Appendix C
Income and Water Treatment
As shown in Appendix Table 4, households with per capita income above the median are
significantly less likely to report that a household member suffered from diarrhea or from
another illness whose cause the household attributed to the drinking water. In addition, rich
households are more likely to report treating both their children’s drinking water and adults’
drinking water. Splitting the sample at the median value of the asset index, instead of per
capita household income, gives stronger results for water treatment but yields no significant
difference in reported illnesses between rich and poor households.
Appendix Table 5 shows the type of water treatment used for child and adult drinking
water among households who report treating their drinking water. Among households that
report treating their children’s drinking water, households with per capita income above the
median are significantly more likely to report using a water purifier to treat their children’s
drinking water and are significantly less likely to report using a plastic net as a method to
treat their children’s drinking water. Using the asset index instead of household income to
determine wealth level gives the same results.
Among households that report treating their adults’ drinking water, households with per
capita income above the median are significantly more likely to report using a water purifier
to treat their adults’ drinking water. Splitting the sample based on the asset index, instead
of per capita household income, strengthens the results. Households with values of the asset
index above the median are significantly more likely to report using a water purifier to treat
their adults’ drinking water, significantly less likely to report using a plastic net to treat
their adults’ drinking water, and significantly less likely to report using a cloth to treat their
adults’ drinking water.
Importantly, these results appear to be driven by differential water purifier ownership
across income levels as opposed to differential use (conditional on ownership) of water purifiers across income levels. Among the sample of households who own a water purifier at
baseline, per capita household income, an indicator for above the median per capita household income and an indicator for above the median asset index are not significant predictors
of whether a household reports using a water purifier to treat either children’s or adults’
drinking water.41 These results are shown in Appendix Table 6.
41
Using household per capita daily income or the asset index instead of indicators for rich also results in
small and insignificant coefficients.
56
Appendix Tables
Appendix Table 1: Asset Index and Willingness to Pay
(1)
(2)
Panel A: Monetary Willingness to Pay
Normalized asset index
0.395***
(0.102)
No
742
Area fixed effects
N
0.413***
(0.107)
Yes
742
Panel B: Non-monetary Willingness to Pay
Normalized asset index
Area fixed effects
N
-0.342***
(0.122)
No
-0.342***
(0.120)
Yes
694
694
Note: In Panel A, the dependent variable is the log of monetary WTP
measured in rupees. In Panel B, the dependent variable is the log of
non-monetary WTP measured in minutes. The asset index is the first
principal component of variables representing ownership of five durable
goods: automobiles, motorbikes/scooters, tablets/computers, washing
machines, and evaporative coolers. The asset index is normalized to
have mean 0 and standard deviation 1. Households may use more than
one method of water treatment. Robust standard errors in parentheses.
57
Appendix Table 2: Willingness to Pay Monetary Prices
Household Income
(1)
Panel A: Log Income
Daily labor income
0.193***
(0.031)
No
730
Area fixed effects
N
Respondent Income
(2)
(3)
(4)
0.196***
(0.031)
Yes
730
0.164***
(0.040)
No
260
0.168***
(0.040)
Yes
260
Panel B: Household Characteristics
Children under 14
0.006
(0.018)
-0.029
(0.048)
0.064
(0.054)
0.011
(0.052)
-0.005
(0.059)
No
739
Water illness (indicator)
Treats drinking water (indicator)
Household owns purifier (indicator)
Credit constrained (indicator)
Area fixed effects
N
0.010
(0.018)
-0.039
(0.048)
0.051
(0.055)
0.016
(0.054)
0.006
(0.059)
Yes
739
Panel C: Log Income and Household Characteristics
Daily labor income
0.188***
(0.032)
0.004
(0.018)
-0.023
(0.047)
0.043
(0.052)
-0.022
(0.053)
0.036
(0.057)
No
727
Children under 14
Water illness (indicator)
Treats drinking water (indicator)
Owns purifier (indicator)
Credit constrained (indicator)
Area fixed effects
N
0.192***
(0.032)
0.008
(0.018)
-0.034
(0.047)
0.034
(0.053)
-0.026
(0.055)
0.043
(0.058)
Yes
727
0.166***
(0.040)
0.007
(0.038)
-0.026
(0.079)
-0.004
(0.078)
-0.156*
(0.085)
-0.033
(0.088)
No
260
0.172***
(0.041)
0.011
(0.038)
-0.045
(0.082)
0.012
(0.081)
-0.173*
(0.090)
-0.074
(0.090)
Yes
260
Note: The dependent variable is the log of monetary WTP measured in rupees. Robust standard errors in
parentheses.
58
Appendix Table 3: Willingness to Pay Non-monetary Prices
Household Income
(1)
Panel A: Log Income
Daily labor income
-0.048
(0.034)
No
683
Area fixed effects
N
Respondent Income
(2)
(3)
(4)
-0.046
(0.034)
Yes
683
-0.006
(0.043)
No
243
-0.006
(0.046)
Yes
243
Panel B: Household Characteristics
Children under 14
-0.020
(0.020)
-0.013
(0.048)
-0.054
(0.059)
-0.149***
(0.054)
0.030
(0.067)
No
691
Water illness (indicator)
Treats drinking water (indicator)
Household owns purifier (indicator)
Credit constrained (indicator)
Area fixed effects
N
-0.024
(0.020)
-0.015
(0.048) )
-0.025
(0.060)
-0.180***
(0.055)
0.043
(0.069)
Yes
691
Panel C: Log Income and Household Characteristics
Daily labor income
-0.034
(0.058)
-0.019
(0.021)
-0.018
(0.049)
-0.053
(0.060)
-0.132**
(0.055)
0.027
(0.067)
No
680
Children under 14
Water illness (indicator)
Treats drinking water (indicator)
Owns purifier (indicator)
Credit constrained (indicator)
Area fixed effects
N
-0.030
(0.058)
-0.021
(0.020)
-0.019
(0.048)
-0.028
(0.060)
-0.162***
(0.056)
0.043
(0.069)
Yes
680
0.005
(0.092)
0.019
(0.044)
-0.035
(0.087)
-0.068
(0.088)
-0.096
(0.097)
0.109
(0.098)
No
243
0.000
(0.098)
0.028
(0.044)
-0.062
(0.087)
-0.035
(0.094)
-0.116
(0.098)
0.086
(0.099)
Yes
243
Note: The dependent variable is the log of non-monetary WTP measured in minutes. Robust standard errors in parentheses.
59
Appendix Table 4: Income and Water Treatment
Treat Drinking Water
Rich (income)
Baseline mean (poor income)
Illness
Diarrhea
Children’s
Adult’s
(1)
(2)
(3)
(4)
-0.069**
(0.035)
0.449
-0.090***
(0.025)
0.189
0.071*
(0.036)
0.716
0.064**
(0.032)
0.688
798
798
575
798
-0.027
(0.036)
0.416
-0.041
(0.025)
0.160
0.122***
(0.036)
0.702
0.136***
(0.031)
0.670
798
798
575
798
N
Rich (asset index)
Baseline mean (poor asset index)
N
Households are included in column (3) if the household contains children under age 14. Households classified
as rich based on income have per capita daily income above the median of the sample. Households classified
as rich based on asset ownership have values of the asset index above the median of the sample. The asset
index is the first principal component of variables representing ownership of five durable goods: automobiles, motorbikes/scooters, tablets/computers, washing machines, and evaporative coolers. Households may
use more than one method of water treatment. Robust standard errors in parentheses.
60
Appendix Table 5: Income and Type of Drinking Water Treatment
Water Filter
Plastic Net
Cloth
(1)
(2)
Panel A: Children’s Drinking Water
(3)
(4)
0.184***
(0.044)
0.273
0.020
(0.041)
0.254
-0.103**
(0.045)
0.670
0.004
(0.031)
0.123
472
472
472
472
0.252***
(0.045)
0.259
0.046
(0.042)
0.245
-0.111**
(0.046)
0.666
-0.051*
(0.030)
0.145
472
472
472
472
Rich (income)
Baseline mean (poor income)
N
Rich (asset index)
Baseline mean (poor asset index)
N
Boiling
Panel B: Adults’ Drinking Water
Rich (income)
Baseline mean (poor income)
N
Rich (asset index)
Baseline mean (poor asset index)
N
0.143***
(0.040)
0.281
0.026
(0.027)
0.101
-0.065
(0.040)
0.676
0.123 × 10−4
(0.029)
0.133
571
571
571
571
0.262***
(0.040)
0.246
0.010
(0.027)
0.110
-0.104**
(0.041)
0.685
-0.066**
(0.028)
0.160
571
571
571
571
Households classified as rich based on income have per capita daily income above the median of the sample.
Households classified as rich based on asset ownership have values of the asset index above the median of the
sample. The asset index is the first principal component of variables representing ownership of five durable
goods: automobiles, motorbikes/scooters, tablets/computers, washing machines, and evaporative coolers.
Households may use more than one method of water treatment. Robust standard errors in parentheses.
61
Appendix Table 6: Income and Water Purifier Use
Children’s Water
Adult’s Water
(1)
0.136
(0.049)
0.880
(2)
0.031
(0.044)
0.867
169
227
0.038
(0.050)
0.867
0.020
(0.044)
0.874
169
227
Rich (income)
Baseline mean (poor income)
N
Rich (asset index)
Baseline mean (poor asset index)
N
Households are included in Panel A if the households own a water purifier and have
children under age 14 in the household. Panel B includes households that own a water
purifier. Households classified as rich based on income have per capita daily income
above the median of the sample. Households classified as rich based on asset ownership have values of the asset index above the median of the sample. The asset index
is the first principal component of variables representing ownership of five durable
goods: automobiles, motorbikes/scooters, tablets/computers, washing machines, and
evaporative coolers. Using per capita daily household income instead of the indicator for above the median household income as the independent variable also results
in a small and insignificant coefficient. Similarly, using the value of the asset index
instead of the indicator for a value of the asset index above the median also results
in a small and insignificant coefficient. Robust standard errors in parentheses.
62
Appendix Figures
Pseudo-demand at Monetary Prices (by income level)
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
50
100 150 200 250 300 350 400 450 500 550 600 650 700 750 800
Price (Rs)
Poor (Q1)
Poor (Q2)
Rich (Q3)
Rich (Q4)
Appendix Figure 1
This Figure shows the percentage of respondents who stated a monetary willingness to pay at least
as high as the price for each price level. Willingness to pay was elicited for one water purifier
through the Becker-DeGroot-Marshak (1964) mechanism. Households are divided into quartiles of
the per capita daily household income distribution.
63
Pseudo-demand at Non-monetary Prices (by income level)
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1
2
3
Poor (Q1)
4
5
6
Price (Hrs)
Poor (Q2)
Rich (Q3)
7
8
9
10
Rich (Q4)
Appendix Figure 2
This figure shows the percentage of respondents who stated a monetary willingness to pay at least
as high as the price for each price level. Willingness to pay was elicited for one water purifier
through the Becker-DeGroot-Marshak (1964) mechanism. Households are divided into quartiles of
the per capita daily household income distribution.
64
Total Demand at Monetary Prices
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
95% Confidence Interval
Estimated Demand
Demand (WTP)
Appendix Figure 3
This figure shows the estimated total demand curve at monetary prices for the full sample. The
diamonds represent the percentage of households with monetary willingness to pay above the price.
The solid line is the demand curve estimated as a fourth order polynomial of price with intercept
constrained to equal 1. The grey shaded area is the 95% confidence interval.
65
700
650
600
550
500
450
400
350
300
250
200
150
100
50
Price (Rs)
Poor Demand at Monetary Prices
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
95% Confidence Interval
Estimated Demand
Demand (WTP)
Appendix Figure 4
This figure shows the poors’ estimated demand curve at monetary prices for the full sample. The
diamonds represent the percentage of households with monetary willingness to pay above the price.
The poor are households with per capita daily household income less than or equal to the median
of the sample. The solid line is the demand curve estimated as a fourth order polynomial of price
with intercept constrained to equal 1. The grey shaded area is the 95% confidence interval.
66
700
650
600
550
500
450
400
350
300
250
200
150
100
50
Price (Rs)
Total Demand at Non-monetary Prices
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1
2
3
4
5
6
7
8
Price (Hrs)
95% Confidence Interval
Estimated Demand
Demand (WTP)
Appendix Figure 5
This figure shows the estimated total demand curve at non-monetary prices for the full sample.
The diamonds represent the percentage of households with non-monetary willingness to pay above
the price. The solid line is the demand curve estimated as a fourth order polynomial of price with
intercept constrained to equal 1. The grey shaded area is the 95% confidence interval.
67
9
Poor Demand at Non-monetary Prices
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1
2
3
4
5
6
7
8
Price (Hrs)
95% Confidence Interval
Estimated Demand
Demand (WTP)
Appendix Figure 6
This figure shows the poors’ estimated demand curve at non-monetary prices for the full sample.
The diamonds represent the percentage of households with non-monetary willingness to pay above
the price. The poor are households with per capita daily household income less than or equal to the
median of the sample. The solid line is the demand curve estimated as a fourth order polynomial
of price with intercept constrained to equal 1. The grey shaded area is the 95% confidence interval.
68
9