24.Bennett

Demand for household water in Buon Ma Thuot, Viet Nam evidence from households’ revealed and stated preferences ESTIMATING HOUSEHOLD WATER DEMAND USING REVEALED AND CONTINGENT BEHAVIORS EVIDENCE FROM[.]

ESTIMATING HOUSEHOLD WATER DEMAND USING REVEALED AND CONTINGENT BEHAVIORS:

EVIDENCE FROM VIET NAM

Jeremy Cheesman 1 , Jeff Bennett 2 and Tran Vo Hung Son 3

Abstract: This article separately estimates water demand by

households utilizing (i) municipal water exclusively and (ii)municipal water and household well water in Buon Ma Thuot, VietNam Demand estimates are obtained from a panel datasetformed by pooling household-level data on observed municipalwater purchases and stated intended water usage contingent onhypothetical water prices Estimates show households usingmunicipal water exclusively have very price inelastic demand,whereas households using both municipal and household wellwater have more price elastic, but still inelastic, simultaneouswater demands and readily substitute between water sources inresponse to increasing prices Household water usage isconditioned by water storage and supply infrastructure, incomeand socio-economic attributes The demand estimates are usedfor forecasting municipal water usage as well as the municipalwater supply company’s likely revenue stream following anincrease to the municipal water tariff and also for modelingconsumer surplus losses from municipal water supplydisruptions

Keywords: urban water demand, household production

function, revealed preference, contingent behavior

1 Research Associate, Crawford School of Economics and Government, Australian National University, Canberra 0200 AUSTRALIA

2 Professor, Crawford School of Economics and Government, Australian National University, Canberra 0200 AUSTRALIA

3 Head, Environmental Economics Unit, University of Economics Ho Chi Minh City, Ho Chi Minh City, VIET NAM.

1 INTRODUCTIONHousehold water demand analyses are an economiccornerstone for demand side water management, developingefficient water tariff schedules and water infrastructure costbenefit analyses Meta-analyses profiling the household waterdemand literature concentrate on developed country applications(Espey et al 1997, Arbues et al 2003, Dalhuisen et al 2003) These applied studies from developed countries mainly estimatedemand from households’ observed water purchases from asingle municipal water supplier, municipal water’s multi-partblock tariff, household income, socio-economic attributes andsometimes climatic and structural factors, typically findinghousehold water demand is both price and income inelastic.Household water’s price and income inelasticity is normallylinked to water being a non-substitutable input in manyhousehold uses and also because household water expendituresonly account for a small percentage of most households’ budgets(Arbues et al 2003)

Less work has been directed towards estimating householdwater demand in less developed countries (LDC’s) Strand andWalker (2005) estimated a –0.32 household own price elasticityusing a survey dataset from 17 cities in Central America andVenezuela Their analysis shows households drawing water frommore than one source have source specific water demand andalso that in-household water infrastructure is a stronger demanddeterminant than water price Using data from seven Cambodiantowns, Basani, Isham et al (2008 forthcoming) estimatedhouseholds’ own price elasticity for municipal water suppliesbetween -0.40 and -0.50 Combining household data from ElSalvador and Honduras, Nauges and Strand (2007) estimatednon-tap water demand elasticities as a function of water cost,defined as the sum of water’s purchase price and hauling costs,

between -0.40 and -0.70 Rietveld, Rouwendal et al (2000)estimated an own price elasticity of -1.2 for a cross-section ofIndonesian households Acharya and Barbier (2002) estimatedlinear water demands for Nigerian households that (i) exclusivelycollected water, (ii) exclusively purchased water from vendors, or(iii) hauled and purchased water Households purchasing waterexclusively had an estimated own price elasticity of –0.067,whereas collecting and purchasing households’ own priceelasticity for purchased water was –0.073.

Estimating price elasticity requires that water’s price varies.However, water may be purchased at a constant price, as is thecase when a municipal water supplier charges the same tariff forevery cubic meter of water it delivers, or unpriced, in the sense

of not having a tariff, as occurs when a household draws itswater from a private well Both of these situations complicatehousehold water demand estimation, but both, and especially thelatter, are frequently features of household water use in LDC’s.Stated preference techniques can be applied for constructing theprice usage relationships needed for estimating household waterdemand functions in both these situations (Freeman 2003).Stated preference techniques construct hypothetical markets,using these for simulating respondents’ preferences for scarceresource allocation When available, households’ real waterpurchasing histories, such as their water bills, can be used as anempirical anchor point for investigating each household’s likelywater usage in novel water pricing situations Confirmingconvergent validity between a household’s observed waterpurchases and stated preferences shows the same underlyingpreference structure is being used for making actual andhypothetical water purchases Analyses pooling revealed andstated preference data (Adamowicz et al 1994, Ben-Akiva et al

1994, Englin and Cameron 1996, Adamowicz et al 1997, Huang

et al 1997, Acharya and Barbier 2002, Boxall et al 2002,Earnhart 2002, Hanley et al 2003) generally show poolingincreases estimated parameters’ efficiency and robustness,especially when estimates are based on small datasets (Englinand Cameron 1996, Haab and McConnell 2003, Hanley et al.

2003, Birol et al 2006)

This article estimates demand for delivered water byhouseholds using (i) municipal water exclusively and (ii)municipal water and household well water in Buon Ma Thuot(BMT), Viet Nam Buon Ma Thuot is located in Viet Nam’s CentralHighlands region and is Dak Lak Province’s largest town Themunicipal water supply system was upgraded and expanded in

2003, resulting in connected households increasing theirmunicipal water usage, and thereby diverting scarce water awayfrom the region’s irrigated agriculture sector The Buon Ma ThuotWater Supply Company (BMTWSC), the autonomous State agencyresponsible for operating the municipal water supply system, ismeant to operate at full cost recovery The fixed VND2,250 (USD1

 VND15,500) per cubic meter tariff it charges is less than theVND4,000 per cubic meter average cost it estimates it incurs fordelivering water to BMT’s households however All householdsreceiving municipal water supplies in BMT are metered and havetheir monthly household water bills calculated from theirmetered usage

Approximately 75 percent of all permanent households in BMTare now connected to the municipal water supply system Apercentage of households already connected to the municipalsystem combine municipal water and water from at least onealternative source, such as private wells or water vendors Little

is known about households’ usage patterns from non-municipalwater sources in BMT nor why households may prefer thesesources’ water to municipal water Madanat and Humplick (1993)

found that households had preferences for water by source inspecific uses and it is reasonable to expect the same thing here.For example, BMT’s households may prefer using municipal waterfor cooking and well water for drinking because they believemunicipal water tastes and smells of chemicals Nothing is knownabout how households using secondary water sources would alterusage between sources when responding to changes in theattributes of either the municipal or secondary source’s water.These substitution strategies carry important economic andwater planning implications in BMT however, meaning a system

of conditional water demands for households not using themunicipal water source exclusively must be estimated

This article’s main contributions lie first in developing thesparse literature on single and multiple source household waterdemand in Southeast Asia and second in the novel revealed andstated preference approach the article applies for estimating ownand cross price elasticities for water when faced with an invariantmunicipal water price and unpriced household well water.Household water demand estimates are constructed from asurvey dataset pooling households’ actual observed water usage

at the existing municipal water tariff and their stated waterusage preferences contingent on hypothetical water prices Thestated preference approach is based in the contingent behaviormethod, which works by eliciting individuals’ intended behavioralresponse to a hypothetical situation occurring, such as anincrease in water price (Hanley et al 2003) Acharya and Barbier(2002) have previously employed a contingent behaviorapproach in estimating Nigerian households’ water demand as afunction of real and hypothetical vendor water prices and waterhauling times

In the remainder of this article the conceptual householdwater demand model, estimation and survey approaches are first

described Following a brief descriptive analysis of the surveydata, household water demands are estimated from the paneldataset Policy implications are discussed in section five and thedemand estimates are used to forecast household municipalwater usage and the BMTWSC’s revenue following an increase tothe municipal water price The consumer surplus losses imposed

by binding water supply constraints are evaluated in section six

in light of dry season water shortages that have historicallyplagued BMT Section seven concludes

2 SPECIFICATION AND ESTIMATION TECHNIQUE

2.1 MODELING HOUSEHOLD WATER USAGE

Household water usage is a function of an underlying decisionmaking process that takes water usage preferences andconstraints on acquiring water into account (Larson et al 2006).When household labor is needed for collecting and preparingwater, a household water demand model accounting for having

to choose between allocating scarce household labor betweenwater collecting and preparing usages and income generatingwork is required Acharya and Barbier (2002) formally model thejoint consumer producer household’s decision making when twowater sources are available, with one source being free butrequiring labor input and the other priced and not requiring laborinput The household seeks to maximize utility from water giventhe water sources available and the household’s income andlabor constraints The end result is the household water demandfunction, conditional on water source:

 p, c,Ap,,Ac,,Z

j

where Q is the water quantity used from source j, j p is the p

purchased water’s price, s is the collected water’s shadow price, c

which is the marginal opportunity cost of foregone income fromwork, Ap,,Ac,are two vectors describing water quality attributessuch as turbidity, smell and taste of priced and collected water

respectively and Z is a vector of household specific

characteristics, including income and labor potential When water

is perfectly substitutable between sources, the utility maximizinghousehold consumes water from both sources until the marginalrate of substitution from purchasing water and collecting waterare equal, meaning the marginal opportunity cost of foregonework income equals the marginal water price This householddecision framework includes two corner solutions: firstly, whenthe opportunity cost of foregone work income due to watercollection and preparation always exceeds water’s marginal pricethe household consumes priced water only and secondly, whenthe marginal water price always exceeds labor’s marginalopportunity cost then the household always collects water

2.2 DEMAND ESTIMATION

Obtaining unbiased water demand estimates requires thathouseholds drawing water from wells in BMT do so as a result of

a random selection process It is possible however that latentvariables determine whether a household has a well or not Thispotential source of sample selection bias is controlled for usingHeckman’s (1979) two step estimation procedure In the first

step the discrete choice dependent variable (d i) equals one if thehousehold has a private well and zero if they do not Assuming a

normal probability distribution for the error term (u i), the decisionmodel in probit form is:

d i  1 Prxiβu i xiβ

)where x is a matrix vector of explanatory variables describingi

the household’s well status, β a vector of unknown coefficients

to be estimated and xiβ is the cumulative normal distribution.The inverse Mill’s ratio is calculated with the probit model’sestimated parameters and included in the second stagehousehold water demand estimates The inverse Mill’s ratio is:

 

 

β x

ˆ 1

ˆ

i

i i

For households using the municipal water supply only, theirconditional household demand function is assumed to be:

1 2 1

1 ln

lnQ m c a p maZ  (4

)Whereas the households using water from both municipal andprivate well sources have the conditional simultaneous demands:

1 3

2 3 2

1 2

ln ln

ln

ln ln

m w m m m m

b s b p b c Q

b s b p b c

)(6)Where the municipal water price is p , m s is well water’s w

shadow price, Z describes household socio-economiccharacteristics including water supply infrastructure such asstorage tanks and booster pumps and also the household’sinverse Mill’s ratio, i the normally distributed idiosyncratic errorterm and the remainder are coefficients for estimating Thesedemand specifications exclude costs from preparing water foruse, such as filtering or boiling water before drinking, because

descriptive analyses, to be discussed subsequently, suggestthese are likely immaterial The demand equations also excludewater quality attributes, again because descriptive analysesshowed BMT’s survey respondents viewed water quality as beingnear equal between municipal and household well sources andalso because water quality perceptions are likely correlated withincome and education (Whitehead 2005)

3 EMPIRICAL APPLICATIONSchedules of household water usage as a function of waterprices are constructed in this analysis by pooling observed andcontingent behavior data from Buon Ma Thuot’s urban and peri-urban households The observed behavior data is municipalwater usage by households at the existing municipal water tariff.The contingent behavior data is estimated by constructing howeach household changes its water usage following hypotheticalchanges in water pricing Because all households receiving theBMTWSC’s municipal water are metered, this data can be usedfor cross-validating households’ own water usage estimates andalso for anchoring the contingent behaviour scenarios

Survey development is discussed in detail in Cheesman, Son

et al (2007) The survey’s main objective was collectinghousehold background data, including details on in-householdwater supply infrastructure, and estimating actual andcontingent household water usage for BMT households’ sevenmain water usages, with these defined in pre-testing: (i) bathingand washing; (ii) preparing meals; (iii) drinking; (iv) cleaning; (v)laundry; (vi) outside (generally gardening); and (vii) homebusiness For estimating households’ revealed and statedpreferences for water by household usage, the surveyenumerator first assisted the respondents in estimating theiraverage daily household water usage by source for the seven

household usages To do this, the enumerator walked through therespondents’ household, identifying with the respondents whereactivities using water were occurring Following this initialidentification, the enumerator worked with the respondents toestimate the amount of water used in each activity during anormal day Because different household members are generallyresponsible for specific water usages, both the male and femalehousehold heads participated where possible Having bothhousehold heads responding may reduce the potential forstrategic behaviour because the respondents audited the other’sanswers and there was open discussion on points of difference(Thomas and Syme 1988) The household members estimatedtheir daily water usage via observation and demonstration Forwater usages that were not daily, weekly usage figures wereestimated

After household daily or weekly water usages in the sevenmain household usages were estimated, the enumeratorextrapolated monthly household water usage and waterexpenditure by water source As a first step, the household’sestimated municipal water usage was compared to their latestavailable municipal water bill to check whether the respondentsaccurately estimated their monthly municipal water usage Then,for estimating the monthly municipal water cost by usage, eachusage’s estimated monthly municipal water usage was multiplied

by the VND2,250 per cubic meter tariff charged by the BMTWSC.For calculating well water’s monthly cost by the seven householdusages, estimated well water usage was multiplied by avolumetric shadow price of VND450 per cubic meter, which wasthe representative household’s calculated well water extractioncost defined by pre-testing The shadow price was constructedusing labor and pumping fuel costs only, with these beingconstructed from the average daily wage and fuel price observed

from the pre-testing respondents Separately estimating wellwater shadow prices for each responding household would bepreferable to the averaging approach used, however in practice

we found this preferable approach was prohibitively timeconsuming, distracting and often lead to enumerators incorrectlycalculating shadow prices Because the survey focus groups, pre-tests and discussions with local authorities suggested householdswere relatively homogenous in their acquiring, storing and wellwater usage (a finding also supported by this article’s descriptivestatistics), we ultimately favored using a common shadow pricefor all households using well water This simplified approachobviously has its limitations

After the enumerator checked that the respondentsunderstood their monthly water cost by household usage andsource, this water usage expenditure schedule was used as ananchoring point for evaluating the household’s demandresponsiveness to hypothetical changes in water prices.Municipal water’s hypothetical price change was an increasing ordecreasing fixed municipal water tariff, whereas private wellwater’s price change was an increasing or decreasinggroundwater shadow price, defined without directly specifyingthe basis for passing on these cost changes to the household Foreach water source used, households were presented with twocontingent behaviour scenarios, resulting in three observationsper household per water source - one revealed preference based

on actual water usage at the existing price and two statedpreference contingent behaviour responses Municipal waterusers each received one hypothetical price lower than thecurrent VND2,250 tariff, either VND500, VND1,000, VND1,750,and one higher hypothetical price, either VND2,500, VND5,000,VND7,500, VND10,000, VND15,000 or VND25,000 The sameapproach was followed for eliciting household well water users’

stated contingent water usages, with the hypothetical pricesVND100, VND250, VND1,000, VND1,500, VND2,000, VND2,500,VND3,000, VND4,500 or VND7,500.

For each hypothetical water price, the enumerator firstcalculated and told the respondents their household’s newmonthly water expenditure assuming household usage by sourcedid not change This approach allowed households to see theirnew monthly water expense by household usage and also bywater source Respondents were then asked whether they wouldchange household water usage given their new waterexpenditure For respondents indicating they would changehousehold water usage, the enumerator worked with thehousehold in determining how the household would change theirwater usage in each of the seven household usages Behavioral,technical or structural modifications can be employed forchanging water usage, however most respondents focused onshort-term behavioral adjustments either changing the amount ofwater used, adopting water recycling or substituting usagebetween their available water sources After respondents hadrevised their household water usages, the enumerator calculatedthe household’s new water expenditure Respondents satisfiedwith their new water expenditure proceeded to the next scenario.The enumerator worked with unsatisfied households in revisingtheir water use, with this procedure being repeated until therespondents accepted their water expenditure The procedurallogic was the same for the well water scenarios

4 RESULTS, DISCUSSION AND POLICY IMPLICATIONS

4.1 DESCRIPTIVE STATISTICS

The household survey was completed in mid 2006 andobtained 291 usable responses Descriptive analyses revealedresponding households are characterized by a dependency on

municipal water; view both municipal and household well waterquality favorably but with some seasonal and income basedvariation; predominantly use municipal and well water forhousehold usages; have in-house water storage infrastructureprimarily to stock against municipal supply outages; have mainlyautomated household well water extraction; excepting drinkingwater are not devoting labor to preparing water; and do not knowthe municipal water tariff (Table 1)

With an average household size of 4.66 persons, the 55percent of households using municipal water exclusivelyconsume approximately 120 liters water per capita per day The

32 percent of households augmenting municipal water withhousehold well water only or with well water and water fromanother source have lower daily per capita usage from themunicipal source at 70 liters Almost nine out of ten respondenthouseholds reported having some form of in-house water storageinfrastructure In-household cement storage tanks proliferate,with these installed in almost seven out of every ten householdssurveyed These storage tanks have a 2.4 cubic meter averagestorage capacity, which is sufficient for supplying 4.5 days water

to the statistically average sized household consuming 120 litersper capita per day Households using water from wells havelargely automated this process with approximately 85 percentusing motorized pumps Even though households using both welland municipal water recorded similar perceived quality levels formunicipal and well water, less than 10 percent of householdswith water storage blend municipal and well water in the samestorage facility For more detailed descriptive analyses seeCheesman, Son et al (2007)

4.2 HOUSEHOLD WATER DEMAND ESTIMATES

Comparing the descriptive statistics and results from thecontingent behavior scenarios showed a percentage ofhouseholds who reported not having access to a private well inthe survey’s initial background section stated they would drawwater from a private household well in response to an increasingmunicipal water price For estimation purposes, respondents whowere using municipal water and stated they would use ahousehold well in at least one of the contingent behaviorscenarios where categorized as households having access tomunicipal and household well water Households indicatingthrough the scenarios that they would only use municipal waterwere categorized as municipal only households Categorizing onthis basis results in a 133 household sub-sample using municipalwater exclusively and 92 households drawing from both ahousehold well and the municipal source The remaining 66households draw from other several other secondary sources areexcluded, mainly due to small numbers in each sub-group.Eleven of the 133 municipal supply only households had missingincome replaced with their sub-sample’s average income.Similarly, 4/92 households drawing on both well and municipalwater had missing household income replaced by theirsubgroup’s average Three influential outlier observations weredropped from the municipal water sub-sample and two from thewell water group This procedure results in a final sample using

130 municipal water only households and 90 households usingmunicipal and household well water

The household water demand estimates’ veracity depend inpart on respondents being able to accurately estimate theirmonthly household water usage Pair-wise correlations betweenhouseholds’ own estimated monthly usage from the water usageanalysis and actual usage from the household’s most recent

municipal water bill on hand tested this assumption The wise correlation for households using municipal water only was0.86, significant at the one percent level, whereas householdsusing both municipal and well water had a pair-wise correlation

pair-of 0.93, also significant at the one percent level Thesecorrelations suggest responding households could estimate their

household water usage with an acceptable accuracy Assuming

that households using both municipal and well water canestimate their daily well water usage with equal accuracy astheir municipal water usage suggests these households use justunder 100 liters of household well water per capita per day in anormal sized household Aggregate well and municipal usage forthese households is then roughly 170 liters per capita per day.These results suggest that at current prices, households usingprivate wells in addition to municipal water get around 60percent of their daily water requirements from their well

In constructing the panel dataset, dummy variables were usedfor identifying the revealed preference scenarios and included inthe system of demand equations to test the null hypothesis thatthese variables’ coefficients equaled zero, thereby supporting aconclusion that households’ revealed and stated preferencesshare a common underlying preference structure In all estimatesthe null hypotheses that the revealed preferences coefficientswere not statistically different from zero could not be rejected

The best fitting probit estimate for the 220 municipal only andmunicipal and well households is significant at the one percentlevel (Table 2) Increasing household income decreases theprobability that a household has a well, which is consistent withobservations from the household water usage profile.Households’ listing farming as their main occupation are more

likely to have a well, which is unsurprising given farms arelocated primarily in BMT’s peri-urban areas and most farms areusing dug wells for irrigation Pair-wise correlations betweenfarming and income and self-employment and income show thesevariables are not significantly correlated The inverse Mill’s ratio

is calculated using the probit model’s estimated parameters forincluding in the water demand models to control for selectionbias

Water demand for households using municipal water only isestimated using random effects generalized least squares,because this allows for including time invariant householdspecific explanatory variables The balanced panel datasetincludes 390 observations, comprising the two contingentbehavior responses and one revealed preference response foreach of the 130 households using municipal water only Thedependent variable is monthly household water usage in cubicmeters’ natural log Several functional forms were evaluated andonly the best fitting model is reported here The model for at-site

household municipal water demand for a household (panel) (i) elicitation (‘time’) (t) using municipal water only is defined by:

t i i

i i

i i

store i

t knowm i

know t

m t

m

w mills a know a own a farm

a

store a

D a hhsize a

inc a p

a D a p a c

Q

, 11

10 9

8

7 , 6 5

4 , 3

, 2 , 1 1

ln denotes logarithms to base e; Q is the dependent variable m

describing the monthly amount of water in cubic meters that therespondent household consumes and the explanatory variablesare in order: municipal water price, a dummy variable describingwhether the respondent knew the municipal water tariff beforethe survey, an interaction variable testing whether own priceelasticity for households knowing the municipal water tariffdiffers from those who don’t, income, household size, here

measured as the number of people living in the household formore than five months a year, a dummy variable describingwhether the household has in-house water storage, thehousehold’s water storage capacity in cubic meters, a dummyvariable identifying farming households; a dummy variableidentifying households deriving their main income from homebusinesses, a dummy variable describing whether therespondent knew the water tariff when asked in section one andthe calculated inverse Mill’s ratio The additive composite error

term w comprises a term for individual specific unobserved

heterogeneity u i , and e,t, which is the usual idiosyncratic

have a zero mean and constant variances The explanatoryvariables D know, p knowm , D store and store are coded using Battese’s

(1997) coding approach which overcomes potential estimationbiases resulting from assigning small values to zero valuedobservations before transforming these data into naturallogarithms Roughly 75 percent of respondent householdsinstalled their water storage infrastructure before 2003 whenBMT’s municipal water supply system upgrade and expansionwas completed, and we therefore assume water storageinfrastructure is exogenous to current water usage

The estimated model is significant at the one percent leveland has an adjusted R-square of 0.43 (Table 3) The retainedmodel coefficients are generally significant and signed consistentwith expectations A Hausman test confirms the orthogonalityconditions imposed by the random effects estimator were notviolated The Breusch Pagan Lagrange multiplier test rejects thenull hypothesis that variance of u is equal to zero, showing that i

there are significant individual effects, meaning estimating with

pooled ordinary least squares would be inappropriate in this case(Baum 2006)

The –0.059 own price elasticity estimate is significant at theone percent level, showing households using municipal wateronly have highly inelastic water demands For example,increasing the municipal water tariff by 20 would result inhouseholds reducing monthly usage by approximately 1.2percent on average over the short run This household estimate

is lower than previous own price elasticity estimates forhouseholds using piped water exclusively in LDC’s Householdscorrectly stating the municipal water tariff during the survey aremore responsive to changing municipal water prices, with an ownprice elasticity of –0.081 Income elasticity is significant at theten percent level, indicating a ten percent increase in monthlyhousehold income increases monthly household usage by 1.4percent on average Household water usage is also increasing inthe number of permanent residents, such that doubling thepermanent residents increases monthly household usage byapproximately 50 percent The significant dummy variable for in-household storage shows households with storage consume morewater than households without storage irrespective of theirstorage capacity Moreover, the significant water storagecapacity elasticity shows increasing in-household water storagecapacity also increases these households’ total monthly waterusage Coefficients for operating a home based business, a farm,knowing the household water tariff and the Mills ratio areinsignificant The inverse Mill’s ratio estimate suggests there is

no selection bias in the model due to the household’s well status