Rural water supply in Kerala, India: How to emerge from a low-level equilibrium trap

WATER RESOURCES RESEARCH, VOL. 29, NO. 7, PAGES 1931-1942, JULY I993

Rural Water Supply in Kerala, India: How to Emerge From a Low-Level Equilibrium Trap BHANWARSINGH,1 RADHIKA RAMASUBBAN, 1 RAMESHBHATIA,2 JOHN BRISCOE,2 CHARLESC. GRIFFIN,3,4.5AND CHONGCHUNKIM 4 Large quantities of financial and human resourceshave been devoted to improving rural water suppliesin developingcountriesover the pasttwo decades.Many projectshave been successful,but many have failed to meet the needs of the intendedbeneficiaries.Evidence of the failures lies in the unusedand poorly maintainedsystemsthat are scatteredthroughoutrural areasof the developing world. The current situation in water supplyin rural Kerala, India, reflectsthis general observation and can be describedas a "low-level equilibriumtrap." Water systemsprovide a low level of service with few yard taps. The monthly tariff for water from householdconnectionsis low. With few connectorsand low tariffs, little revenue is generatedbeyondsubsidiesprovidedby the government. The water authority can affordto maintainthe systemup to a level at which the reliability of service is low, forcingconsumersto supplementpipedwater from traditionalsources.This study analyzes

contingentvaluationdata collectedin threeareasof Kerala to evaluatethe possibilityof lifting the systemout of thistrap. The analysisshowsthatby makinga few criticalpolicychanges,encouraging privateconnections andfinancing thoseconnections throughhighertariffs,the systemcanratchetup to a "high-levelequilibrium"in which there are many connectors,monthlyrevenuesare greatly increased,and consumerwelfare improves.Such a systemwould be better financed, making it possibleto improve the reliability and qualityof the service.

households'tastes and constraints on policy options for improvingmodernrural water systems? New water supply systemsin rural towns of developing This study examineswillingnessto pay for yard taps, or countriestend to be designed as inexpensive communal house connections,in several towns in the Indian state of systems with a few publictapsor deepwells.Suchsystems Kerala. Rural systems,includingthose in Kerala, are parINTRODUCTION

areoftenpoorly maintainedand have muchlessimpacton tially or fully fundedfrom centralgovernmentsources.All usepatternsof the intendedbeneficiaries thanis expected. such projectsmust conform to inflexible designcriteria Evenpeoplewho use water from the publicsystemoften specified by thecentralgovernment, including a capacityof continueto use traditional sources, such as shallow wells, 40 L per-capitaper day, capital costsno higherthan 200 rivers,ponds,rainwater, and vendors.These other sources rupees(Rs;aboutU.S. $14.29in 1988)perbeneficiary, and There is typically maysupplement the low volumesavailablefrom the public (for the mostpart) no houseconnections. source or maybe chosenfor convenience,loweropportunity no chargefor water from public taps. By officialestimates, cost,better taste, and other characteristics.Typical water

between 50 and 70% of these rural water systems are in a

consumption patternsin ruraltownsof developing countries state of disrepair,and only 50% of the populationwith arethusquite differentfrom thosein industrialized coun- accessto an improvedwatersourceare in fact usingit. As for household connections, once the schemes are

tries,andthe designchosenfor moderncommunitysystems

applications for yard tapsare acceptedand hasnot been very successful in steeringusersto a safer, commissioned, connections are given.Paymentto the water authorityfor a reliablesourceof water [Briscoeand de Ferranti, 1988]. Donorsandgovernment officialsarebeginning to focuson connectionis generallymodest,but the connectinghousefinancialissues, especiallythe generationof revenues holdbearsthe full costof runningthe pipelinefrom the water throughdomesticconnections,as a way to escapefrom the mainto thehouse,plusa watermeter,plusin-houseplumbThereisa smallmonthlycharge forwater problems createdby thistraditionalapproach.Thisemerging ingif it isinstalled. consensus bringsconsumer behaviorto theforein thedesign used. The additional connections cause deterioration in

andreliabilitybecause the systems aredesigned to ofwatersystems, whichhaspr'•viously notbeenthecase. pressure provide water volumes adequate only for a limited number of Theprincipalissueis thefollowing:howwill ruralpeoplein

differentsettingsrespond,in terms of their willingness to public taps. In Kerala, pipedwater serviceis mainlythroughfree helpfinancewatersystems, to differentconfigurations (such Publictapsare locatedat specified disasa choiceof yard tapsor publictaps),levelsof service, publicstandposts. tances along the main pipe; every 200 M is a common tariffs,andconnectioncosts?in short,whatis the impactof

spacing. Somestandposts may serveas manyas 70-80

ICentre forSocialandTechnological Change, Bombay, India. 2TheWorldBank,Washington, D.C. 3TheUrbanInstitute, Washington, D.C. 4Department of Economics, University of Oregon, Eugene. 5NowatWorldBank,Washington, D.C. Copyright1993by the AmericanGeophysical Union. Papernumber92WR02996. 0043-! 397/93/92 WR-02996505.00

householdswhile othersare usedby only 5 or 6 households.

Commonserviceproblemsat theseoutletsincludeleaking

standposts, sealedstandposts awaitingrepairformonths ata time, low waterpressure,poor qualitywater, andfrequent

pumpfailures. Moststandposts maintain a flowof waterfor only4 or 5 hoursa day,oftenat unpredictable times.The inevitableresult of thesefactors, as is true in most develop-

ingcountries, islongqueues at publicstandposts, rationing, 1931

1932

SlNGHET AL.' RURALWATERSUPPLYIN KERALA,INDIA

and use of multiple sources of water [Ramasubbanand Singh, 1989].

The conventionalwisdom in Kerala is that this type of water systemdoes not meet the needsof the populationvery well and that a large proportionof the populationcontinues to rely on traditional water sources. The system imposes high time and coping costs on the intended beneficiaries. Many problemsobservedin Kerala couldbe solvedif private house connections were more widely available and if residentswere willing to pay enoughfor piped water to finance an improved level of service. Barriers to more widespread use of yard taps are thought to include (1) the cost of connecting to the water main, (2) the monthly tariff, and (3) unreliableand poor quality water supply. This situation can be described as a low-level equilibrium trap, and it characterizes piped water supply systems in much of the developing world [Briscoe and de Ferranti, 1988]. Systems provide a low level of service with few yard taps. Because there are few connectors and because tariffs are low, little revenue is generated. The water authority can afford to maintain the system up to a level at which the reliability of service is low, forcing consumersto supplement piped water from traditional sources.

willingto payfor the commodity underconsideration. Respondents' answers are likelyto be biasedin a number of ways:hypothetical biasdueto thehypothetical nature ofthe

question (e.g.,description of a hypothetical commodity, paymentmethod,and market conditions);strategic bias becausethe respondent may perceivean opportunity to manipulatethe outcome(e.g., highlyvaluinga goodin the

hopeit willbeprovided, butwithnoexpectation ofactually havingto pay the amountbid); compliancebiasbecause the respondenttries to anticipateresponsesthe interviewer

wants;andstarting pointbiaswithbidsbeinginfluenced by interviewers'suggestions. Many innovations in survey methods havebeendeveloped over the yearsto manage theseproblems, suchasadding structured willingness-to-pay questions to help respondents arrive at reasonableresponses,procedures that mimic market conditions,and approachesthat give subjects time to reconsidertheir answers. The usefulness of contingent valuation methodshas been firmly established, and there is agreement that the problems describedabove do not represent insurmountable obstacles [Cummings et al., 1986]. The contingentvaluationapproachhas now beenapplied

to the analysisof water policy in a numberof developing

Is there any chance for a such system to lift itself out of this trap? We use contingent valuation methods to analyze the potential for a yard tap-based strategy to work in Kerala by testing the sensitivity of householdsto the monthly tariff for water from a yard tap, the cost of connectingthe house

countries[Briscoeet al., 1990;Whittingtonet al., 1990a,b; 1991, 1992]. The resultshave been variable, showinghigh willingness to pay in some communities and little or no willingnessto pay for improved water suppliesin others.The major determinantsappear to be the condition of existing

to the water main, and paying more for improved quality of

water sources, income, and the cost to clients of the im-

service.

proved service[WorldBank Water Demand ResearchTeam,

Pairs of towns in Kerala, comprising one town where pipedwater is currentlyavailableand anotherwherea piped water system is planned but not yet operational, were chosenfor the study. Three pairs of towns were surveyed. One pair is located in a water-abundantarea; a secondis in an area with plenty of water, but where traditional water s0.urcessuffer from salt water intrusion; and a third is in a water-scarce area where traditional water sourcesdry up for

1993].

A Model of the Choice of Water Source

Typically, the analysisof contingentvaluation datauses the maximum stated willingness to pay as the dependent

variablein a regression dquation.However,that approach has a number of limitations for our purpose, the most

part of the year. The contingentvaluationmethodallowed important being that information from related contingent respondentsto considerhypotheticalchangesin water sup- valuation modules (such as monthly tariff and connection ply characteristicsand to respondto questionsabout the cost) cannotbe combined,and price elasticitiescannotbe effect of the cost of a connection, the monthly tariff, and derived. As a result, we model the respondent'schoices improvedquality of serviceon the decisionto purchasea within a discrete choice random utility framework.The yard tap. followingmodel is adaptedfrom Train [1990]and Varian The remainderof the paper is organizedinto five sections. [1984]. The next (second)sectiondiscussesthe contingentvaluation The analystimperfectlyobservesthe indirectutilitythat questions, the economicmodel,the estimationproblem,and respondent j receives fromchoosing to purchase a yardtap the variablesused in the analysis.The third sectionanalyzes or continuing to useexisting sources. Theobserved partof the qualitativeresultsof the estimation.In the fourth sec- indirect utility is modeled as a function of water source tion, we use simulationsto examinethe policy implications characteristics(x), householdcharacteristics(z), an unof the quantitativeeffects of monthly tariff, connection knownvectorof parameters (/5or y), andan errore. The costs,andimprovedservicefor boththewaterauthorityand indirectutility of choosing a yardtap is shownin (1): consumers. The fifth section reiterates the major findings.

Vyt,j•- V(xyt, j, zj, • ) + Eyt,j METHODS

(1)

Theindirectutilityassociated withnotchoosing a yardtap (or of usingthe currentsource)is shownin (2):

The ContingentValuationApproach

V77,•= V(x77,2, z•, 'v) + e

Contingentvaluationstudieshave been used over the

(2)

chooses ayardtapif (1)exceeds (2),and(3) yearsprimarilyto valuepublicgoodsandotheruntraded Therespondent commodities, suchas environmental amenities.In its crud-

estform,a contingent valuationsurveysimplyaskspeople an open-ended questionabouthow muchthey wouldbe

holds:

V(Xyt, j, Zj,•) + Eyt,j > V(X•,j,Zj,3/) + E•,j ('3!

SINGHET AL.: RURALWATERSUPPLYIN KERALA,INDIA TABLE !.

1933

Location and Types of Survey Sites, With Sample Size B Sites: No

A Sites:

Improved Water SourceAvailable Nonconnectors

Ezhuvathuruthy 66

Ezhuvathuruthy 819

Nannamukku 1497

66

I00

200

Connectors

Area

improved Water Source Available

Water Abundant

Location Households

Household sample

Water

Location

Scarce

Elapully

Elapully

Elapully

Households

86

723

876

Household sample

86

100

200

Water Abundant but Saline Intrusion

Ezhuvathuruthy

Location Households

Household sample Total household sample

98 98 250

Ezhuvathuruthy 768 100 300

Thus theprobability of choosing a yardtapis givenin (4):

=

Vallikkunnu 1313 200 600

monthlytariff for improvedservice.A modifieddoubleboundeddichotomous choiceapproachwasused[Haneman

et al., 1991].Eachsetof questions is discussed below,and

> *F,•- *y•,•]

(4)

the discussion is summarized in Table 2.

Connectioncost. NonconnectingA-site and all B-site householdswere administeredthe connectionchargemodjointnormal distribution, thismodelcanbeestimated statis- ule first. Interviewersfor B-site householdsdescribedthe

Assuming thattheerrorsin observed indirectutilityhavea ticallyasa probitmodel,following(5):

p(Yri)=xj($x;-Yx;) +zj(a,q-•lo)+(e•7,j-eyt,j) Simplifying the notation,we estimate(6):

P(YTj)= xia + zfi•+ •i

typeofwatersystem tobeinstalled in thevillage,whichis similarto the typicalsystempreviously described in this (5) paper.Respondents weretoldthattheirresponses would affectneitherthe designnor the price of water from the plannedsystem.BothB-siteandA-sitenonconnectors were (6) informed,"I wouldnow like to ask you somequestions

aboutyourreactions to changes in the costof takinga TheSamplingStrategy Six sitesin northernKerala are coveredby the survey.As

domesticconnection.As we discussedearlier, the present

cost of a domesticconnectionfor you is about Rs. X

repeats theamount calculated earlier).In addiis shownin Table 1, eacharea includesa site wherethe (interviewer

for reasonable useis aboutRs.5 improved watersupplyhasbeenin existence fora fewyears tion,themonthlycharge Thentherespondent wasasked if heor shewould andwherea number of houseconnectionshave been made ($0.36)." (theseare labeledA sites).In eachA sitethereare connec- connectif the connectioncostwere Rs. 700 ($50). All respondents werestartedouthigh,at Rs.700.If an tors and nonconnectors.Table 1 shows the number of households fallingintoeachcategory among households that affirmativeanswerwas made for Rs. 700, that was the maximum willingness to pay.A "no"to Rs. arecloseenoughto the watermainto be ableto connect. respondent's

of whetherhe or shewouldpayRs. Thesecond typeof site,in contrast, is currently without 700eliciteda question a Rs.500question, anda animproved watersystem (thesearelabeled B sites).They 100.A "yes"at Rs. !00elicited werechosento be similarto the A sitesin other ways, such

"no"putmaximum willingness topayatlessthan100.The

chargeof Rs. 200for associal, economic, andenvironmental factors.All of theB finaliterationcamewith a suggested to payRs. 500but willingto payRs. 100. sites havebeentargeted for improved watersupplysystems thoseunwilling withinthe near future. Each set of A and B sitesis catego- Thuseachrespondent canbe assigned a yesor no response rizedin thetableaccording to the watercharacteristics de- tofourprice levels, orthemaximum response canbeplaced scribed in theintroduction: waterabundant,waterscarce,and within a range: lessthan100,100-199,200-499,500-699, and saline.

700 or higher.

Thesampling design isalsoshowninTable1.All connect- Monthly tariff. Themonthly tariffmodule wasadminisinghouseholds in the A siteswere sampledbecause there teredto all households. It wasintroduced to theconnecting weresofewof them(250);100nonconnecting households in A-siterespondents asfollows: "! wouldnowliketoaskyou eachof theA sitesweresampled, and200households in some questions regarding yourresponse to changes in the eachof theB sitesweresampled. Thetotalsample sizeis charge madefor water.If the supply continued asit is 1150houseBolds. and..." It wasintroduced to nonconnecting A-siteand

B-siterespondents asfollows: "I wouldlikeyouto tellme whatyouwoulddoif thecostof takinga connection was fixed at only Rs. I00 but the monthly charge was more than Threesetsofcontingent valuation questions wereasked, werethenledthrough covering theconnection charge, themonthly tariff,andthe Rs.5." All threetypesof households TheContingentValuationModule

1934

SINOH ET AL.: RURAL WATER SUPPLYIN KERALA, INDIA

TABLE 2. Descriptionof Contingent ValuationQuestions

Water SystemCharacteristics Variedin the Contingent ValuationModules Module

Survey Site

Tariff A-Site

Currently connected to an existing scheme, with a yard tap

tariff

Service

ConnectionCost

Connectors

rangeup

currentservicelevel

NA

(Rs. 10-50)

improved

rangeup

betterservice described NA

(Rs. 10-50) A-Site

Access to the same scheme as the

Nonconnectors

current(Rs. 5)

currentservicelevel

rangedown (700-100rupees)

tariff

range up

current service level

held constantat 100rupees

improved service

(Rs. 10-50) range up (Rs. 10-50)

better service described

held constantat 100rupees

connection cost

current (Rs. 5)

current service level

range down (700-100rupees)

tariff

rangeup

currentservicelevel

held constant at 100rupees

connection

cost

A-site connectors, but not

currently connected

B Site

New scheme planned or under construction;will have accessin the future

(Rs. 10-50)

Range upmeans thattheexisting price(Rs.5)istheminimum, andbidsranged upfromthatlevel.Range 'down means thatconnection

costwasstartedat Rs. 700andreducedin increments to thefinaloptionof Rs. 100.NA, notapplicable.

to a setof contingentvaluationquestions in whichthemonthly The resultingstrategicbias would overstatewillingness tariff was moved from Rs. 50, down to Rs. 10, up to Rs. 30,

pay.

By startingeveryonehigh, we are unableto knowif and downto Rs. 20, just as in the connectioncostmodule. mighthavevariedif somehouseholds hadbeen All respondents werestartedhighandwerequotedthesame responses so prices.The maximum bidscanbe placedwithinoneof the startedlow. No follow up of respondentswas possible, we do not know if bids would have changed after giving them followingranges'lessthan10, 10-19,20-29,30-49,and50 or higher.

time to think further about their responses.

Improvedservice. The finalmoduleassessed theeffect The total effect of these biases is unknown, and the effectsmaybe randomenoughthattheycancel of improvedservice,but only for thoseresiding in the A offsetting out. Two approaches areusedto compensate forpossible sites.This modulewasnot appropriate in the B sites,where no service is currently available. The respondentswere bias,whichthe analystcannotobserve.First,we include variables to identifytheA-sitenonconnectors and giventhefollowinginformation' "Now I wouldlikeyouto dummy tell mewhatyouwoulddo if the servicethroughthepiped B-site householdsin the estimation.Thus we can detect from thoseof the water systemwas greatlyimproved.Supposethe water whethertheir bids vary nonrandomly A-site connectors. The A-site connectors know the water supply wasavailable everydayformostoftheday,thatthe

best(reducing hypothetical bias)andhavetheleast flow in the taps was alwaysgood,andthat the waterwas system for strategic biasbecause theirbids(except inthe cleanand tastedall right. If the monthlychargefor this incentive

watermodule) cannot affecttheoutcome. They improved servicewas..." Nonconnecting A-siterespon- improved awarethatmonthly tariffsarecollected bythe dents were also informed that the connectioncharge was arealready

fixed at Rs. 100. The range of possibleresponses was the sameas in the tariff module.Everyonewas startedhigh.

watersystem. Thusweassume thattheeffects ofbiaswillbe leastamong these households, andthedummy variables for

biasrelativeto thisgroup.Second, Biasproblems.Responses fromthe B sitesmightbe theothersitesmeasure resurveys of the households werecompleted inthefallof expected to sufferfromhypothetical biasin thesemodules

a yardtapwhen because of thehypothetical natureof thepipedwatersystem 1991to findoutwhodidor didnotchoose they became available in two of the B communities. Ournext for them. The A-site households shouldbe morefamiliar step in the research, yet to be executed, is to compare actual withthe currenttypeof system.However,bothA-nonconwiththehypothetical resultsfromthecontingent nectorandB-sitehouseholds mighthavehypothetical biasin behavior

theimproved service module because theydonotcurrentlyvaluation ownyardtaps.Thesebiasesare compensated, withunknowneffect,by the description of the systems thatprecededeachcontingent valuationmodule.We expecthypothetical bias to depresswillingnessto pay for these

survey.

The Estimation Problem

Ourgoalistoestimate (6)using information fromallthree

contingent valuation modules simultaneously. Each yes/no response is therefore considered as an observation, x¾ith All respondents maysense someadvantage toproviding price, connection cost, and improved reliability appearing as highbidsinthehopethatsuchbidswould elicittheresponse variables. Equation (7)shows thebasic setup ofa wellfunctioning watersystem withpoorlyenforced fees. independent

households.

SlNGH ETAL.:RURAL WATER SUPPLY INKERALA, INDIA TABLE 3. Reproduction of 12 Observations for a Single Household

1935

the estimated covariance matrix and a correction formula

shownin (9) [Gallantand White,1988;Guilkey,1992]:

HookUp/

Cov* (•)= Coy(•r)-•A(•r)Cov

Keep Connection YardTap? Cost* Tariffs Improved?$

Module

where

0 1 1 1 0 1 0 0

700 500 200 100 100 100 100 100

5 5 5 5 50 10 30 20

0 0 0 0 0 0 0 0

connection cost connection cost connection cost connection cost tariff tariff tariff tariff

0 1

100 100

50 10

1 1

improved improved

0 1

100 100

30 20

1 1

improved improved

A(rr)= X Ai(•r)Ai(•')'

(9)

i

Ai(tr) =E O•' O3•j/O•is the derivativeof the log likelihoodandrris the vector of estimatedcoefficients.We report the probit coef-

*Maximumwillingnessto pay in the connectioncost module: Rs.

ficientestimates usingall observations. The reportedstandard errorsand t statisticsare correctedusing(9).

500.

tMaximum willingnessto pay in the tariff module: Rs. 10. •;Maximumwillingnessto pay in the improvedmodule:Rs. 20.

Variables Used in the Analysis

The list of variables in Table 4 shows the categories, providesa definition,and indicatesthe expectedsignfor each variable. Referring to that table, the price variables withthewater sourcecharacteristicsspelledout. Tariff is the associatedwith the improved system (tariff and connection monthlycostfor reasonableuse quoted in the tariff module, charge)are expectedto reducethe probabilityof connecting. connectionis the connection cost quoted in the connection The quality variable(improvedservice)is expectedto raise costmodule,and improved is a dummy variable for whether the tariff quote comes from the improved service module. Distanceand queue measure, respectively, reported distanceand waiting time for the traditional source:

the probability of connecting.The time cost variables associated with the primary traditional source used by the household (distance and queuing time) are expected to

increasethe probabilityof choosinga yard tap. The household variables (income, electricity, number of rooms in the house, and adult education)measuringincome, wealth, and human capital are expectedto increase the probability of

P(YTj)= a 1q-Ot2(tariffj)+ a 3(connection/) + a 4(improved/) + a 5(distancej)

dummyvariablefor whetherthe serviceis improvedis setto

hookingup. The occupationvariables(governmentemployment by females and males) are intended to capture the effect of modernsectoremploymentin raisingthe opportunity cost of time; hence, they are expected to raise the probability of hookingup to the water system. The religion

0. In the tariff module, connection cost is set at 100.

variable (Hindu) is a control variable for which we have no

+ a6(queuej) + zfi3+ ej

(7)

In the connection cost or simple tariff modules, the

Connection cost is treated as a sunk cost for the A-site

expectationas to the sign. Sex of both the respondentand the household head is included because many observers speculatethat females benefit more from yard taps and thus possible responses,so eachhouseholdcouldappearup to 12 are more likely than men to provide positive hook-up retimesin the data usedfor theseestimates.As an example, sponsesat every price. Dummy community variables differconsiderthe single household (from an A-site connector) entiate community water characteristics (abundant, scarce, appearing in Table 3. The top four observationscorrespond saline), with the expectation that households in scarce or to the connection cost module. The middle four observations salinewater areaswill be more likely to choosea yard tap at correspondto the simple tariff module, and the last four each price, everythingelse equal. Finally, dummy variables correspondto the improved service module. Each of the distinguishingthe type of household (A nonconnectors,B) connectinghouseholds,so it is always 0 for them. Each of the three contingentvaluation moduleshas four

A-siteconnectorshas8 observations,A-site nonconnectors are included to control for bias for households have12 observations,and B-site households have 8 obser- currently connected, relative to connectors.

that are not

vations.

In effect,thisapproach addsa second,household-specific componentto the error term in (7). We now expandthe error

RESULTS

term to differentiatebetween the responses(j) and the

household (i), asis shownin (8). Thenewerrorcomponent, Maximum Willingnessto Pay Statistics /•ij, represents the within-household error,andej represents Table 5 shows average maximum willingnessto pay dethe across-household error:

rived from the variouscontingentvaluation modules.In the simple monthly tariff module, the mean bid for connectors P( YTij) = xijotq-Zij• q-Igij q-Ej (8) was Rs. 19.3. The current average tariff is about Rs. 5. Probitestimatesof the coefficientsin this model are unbi- Nonconnectorsbid Rs. 8.7 on average, and B-site responased,but the standard errors are underestimated. Consistent dents averaged only Rs. 5.5. The second row of numbers estimates of the correctstandarderrorscanbe formedusing shows that 56% of connectorsbid higher than the current

1936

$INGHETAL.:RURALWATER SUPPLY IN KERALA, INDIA TABLE

Category

Characteristics of the improvedwater

4.

List of Variables Used in the Analysis

Variable

Expected Sign

Description

tariff

tariff quoted in the contingent valuation module

connection charge

connectionchargequotedin the contingentvaluation

source, given in the contingent valuation module module

improved service

+

whetherthe contingentvaluationmoduleindicatedthat

distance

+

distance to thecurrent source of wateror,if hooked up,

+

average queuing timeoverseasons at thecurrent source;

the service would be improved (0/1) Characteristics of the current water

sourceor alternative to a yard tap

to current

distance to the primary alternative source

source

queue at current

if hookedup already,queuingtime at the primary

source

alternative

electricity number of rooms

number

females in government

whetheranyfemalesin the household areemployed by

+

characteristics

of rooms in the house

the government (0/1)

service

Household

source

estimated household income divided by household population whether the householdhas electricity (0/1)

per capita income

males in government

whetherany malesin the householdare employed bythe

service Hindu sex of HH

government (0/1) if the household's religion is Hindu (0/1) head

if the householdhead is female (0 = male/1 = female) if the respondent to the survey is female (0 = male/1-female)

sex of respondent some primary school

+

if the maximum education of adults in the household is

primary school complete middle school complete

+

if the maximum education of adults in the household is

+

if the maximum education of adults in the household is

+

if the maximum

some primary school (without finishing) (0/1) completion of primary school (0/1)

completionof middle school (0/1) secondary school complete more than secondary

Traditional water supply

education of adults in the household is

completionof secondary(0/1) +

if the maximum

area

+

at least some college (0/1) household is in a scarce water area (0/1)

saline water area

+

household is in an area where salt water has intruded

A-nonconnecting

-

scarce

water

education

of adults in the household is

characteristics

Dummy variables to account for contingent valuation module bias

into traditional sources (0/1) household is a nonconnector in villages with improved water already available (0/1)) household is in a village'without an improved water source (0/1)

household

B-village household

-

Expectedsign:the effect,positive(+), negative(-), or unknown(?), on the probabilityof choosing a yardtap.

tariff, Rs. 5. Only 43% of the nonconnectors,and34% of the B householdshad bidsgreaterthan their currenttariff, Rs. 0.

would make a contribution for the connection chargewith

the tariff kept at Rs. 5, suggestingthat respondents are

In the second module, when the connection charge was willingto maketradeoffsbetweenthe tariff andthe connecarewillingtopay varied from Rs. 100 to Rs. 700, the averagebid falls near the tion cost.Currently,connectedhouseholds middle of the range, at Rs. 355 for the A-site nonconnectors an average of' nearly 4 times the current tariff for an

and well below the midpoint, at Rs. 267 for householdsin site B. However, 78% of the A-site nonconnectors and62% of the B householdsare willing to pay at least Rs. 100for a connection(when the tariff is held at Rs. 5). The third groupof statisticsshowsthat about85% of the currentlyconnectedhouseholds wouldpayan averageof Rs. 25 for the improvedsystemdescribedby the interviewer. Both the bids and the percentage who would pay are substantiallyhigherthan for the simpletariff module.Nonconnectors, in contrast, would pay only Rs. 1 more on

unimproved systemandabout5 timesthe currenttarifffor an improved system. Probit

Estimates

Theabovedescriptive statistics do notreflectpureprice effectsbecausethis is not a controlledlaboratoryexperiment,nor dotheyallowusto examinetradeoffs people are

willingto makeamong themonthly tariff,connection charge,

andqualityof service. Table6 contains theresults ofthe

estimation forthefullsample, whichisolates theprice averagefor the improvedservice,andthe percentage who probit effects by controlling for other variables that affect the would pay anythingis the sameas in the simpletariff module.

Overall,thesedescriptivestatisticsshowlittle willingness

responses andallowssimultaneous consideration of the contingent valuation responses. Thefollowing information is

inthetable: theestimated coefficients, corrected to paymuchmorethanthecurrenttariffby households that reported are not connectedto a water system.Yet a largeproportion

standard errors,andasymptotic t statistics; an asterisk

!937

SINGH ET AL.: RURAL WATERSUPPLYIN KERALA,INDIA

TABLE 5.

Average Maximum Willingnessto Pay by Site B Sites: No

A Site:

Improved

Improved Water Source Available

Water Source

Contingent Valuation Module

Connectors

Available

Nonconnectors

Monthly Tariff !9.3 56

Average maximum bid (rupees) Percent of respondentswith a bid greaterthan current tariff

5.5 34

8.7 43

Connection Charge NA NA

Average maximum bid (rupees) Percent of respondentswith a bid greater than zero

267 62

355 78

Monthly Tarifffor Improved System 9.7 Average maximumbid (rupees) 25.0 43 Percent of respondentswith a bid 85

NA NA

greater than zero

NA, not applicable.

changein the connection indicating whethereachcoefficient is statistically significant only0.3%.Thusa smallpercentage fora two-tailedtest at the 10% level; the elasticityestimated costappearsto have lesseffect on the probabilityof conchangein the tariff. for continuousvariables at the means of all independent nectingthandoesan equalpercentage variables; andthe meanof eachvariablein the sampleused This issue will be discussedin more detail later. The other water service characteristic, whether the sys-

for the estimation.

Characteristics of the improvedwater source. The price tem's reliabilityis improved,is not statisticallysignificant variables,tariff and connectioncost, have the expected and has the wrongsign.We will also explorethis issuein

negative effectson theprobability of hookingup, andthey more detail later. Characteristics of the current water source. The variarestatistically significant at lessthanthe 1%level.Thetariff of the primarytraditional elasticity is large:a 1%increasein the monthlycostreduces ablesmeasuringcharacteristics source, distance and queuing time, are not statistically theprobability of choosing a yardtapby 1.5%.TheconnecWhilethisresultis contraryto ourexpectations, tioncostelasticityis substantially smaller:a 1% increasein significant. thecostof hookingup reducesthe probabilityof doingsoby

if household locationis partiallydeterminedby characteris-

TABLE 6. Probit Estimatesof Probabilityof Choosinga Yard Tap

Variable

Dependent variable: hookup?

Coefficient

Constant Tariff

-0.301 -0.060

Connection charge

-0.001

Improved service

Distance to currentsource

Queue atcurrent source Percapita income Electricity Numberof rooms

-0.058

0.00002

SE

t Statistic

0.472 0.003

0.637 20.312'

-1.465

1.000 17.633

0.00007

13.212'

-0.289

2!8.747

0.064

0.001

31.597

1.542 1.933' 3.832'

0.032 0.083

8.412 2613.400 0.461

0.053

0.0002

0.003 0.002 0.00002 0.00001 0.335 0.087 0.086

0.025

1.106

3.495*

Elasticity

Mean

0.302

0.377

0.114

3.188

Females ingovernment service -0.100 Males ingovernment service 0.!66

0.176 0.099

0.567 1.783'

0.054 0.262

Sexofrespondent Some primary school Primary school complete Middle school complete Secondary school complete Morethansecondary

0.074 0.150 0.!41 0.148 0.140 0.!6!

3.702* 3.391' 4.452* 6.510' 8.086* 7.993*

0.595 0.110 0.197 0.181 0.264 0.178

Hindu Sexof household head

-0.191 0.057

Scarce waterarea Saline waterarea

0.347 -0.232

-0.275 0.509 0.629 0.961 1.132 1.290

A-nonconnecting household -0.307 B-village household -0.492

0.092 0.086

0.098 0.106

0.125 0.127

2.070* 0.662

3.542* 2.192'

2.454* 3.873*

0.463 0.240

0.253 0.359

0.315 0.666

A totalof9720observations withnomissing values foranyvariables wereused intheestimation.

Estimates areweighted bythepopulation ofthesampling unit. Theprobit asawhole issignificant at

better than the0.00001 level foralikelihood ratio test(chisquare). Standard errors arecorrected using equation(9) in the text.

*Coefficient issignificant atthe0.10levelorbetter fora two-tailed test.

1938

SINOHETAL.:RURALWATERSUPPLY IN KERALA, INDIA TABLE 7.

IncrementalEffectsof Schoolingon the Probabilityof Choosinga Yard Tap Percent at Each Level

Percentage Increase in Probability Over

Schooling Level Primary Middle

Secondary At least some college

the Previous

Level

of

Schooling A-Nonconnectors

5.9

!8

21

13.1

9

21

4.7 3.5

33 28

22 !2

ticsof the traditionalwater sources,the behavioralimpactof those characteristicsmay be blunted by adjustmentsthat have already taken place by the household.

model,althoughlow levelsof schooling (seeTable7)provide

icant lower bids.

impedimentto more widespreaddemandfor houseconnections. Indeed, 58% of the nonconnectingA-site households cited connection cost as a reason they had not already connected to the existing system. The responsesto the contingent valuation modules, however, indicate that the price elasticity of demand with respect to the connection chargeis only about 20% of the price elasticity with respect to the monthly tariff. Are respondentsreally lesssensitiveto the connection charge than to the monthly tariff? The short answer is no. Because the connectionchargeis for a durable good, there is a combinationof an interestrate

an opportunityto use a slow4o-changehouseholdcharacteristicsto targetpossiblesubsidiesor to designotherintervenHousehold characteristics. The household income and tions aimed at reducing constraintsrelated to thosecharacasset measures, per capita income, whether the household teristics.We concentratein this sectionon the quantitative has electricity, and the number of rooms in the house, all effects of variables that policy makers can control in the have statisticallysignificantpositive effectson the probabil- short run: the connectioncharge, tariff, and qualityof ity of choosinga yard tap. The civil service occupational service. dummy is positive and significant. The gender variables do not behave as hypothesizedin the Effects of Connection Cost on the Probability literature. Female head of household is not statistically of Choosing a Yard Tap significant,and female respondentsgave statisticallysignifIn Kerala, connectioncost is thought to be the major All of the adult education dummies have statistically significant positive effects. Table 7 shows the incremental effects of these variables. The two largest increments in probability come at the lower levels of schooling:finishing primary school raisesthe probability of hookingup by 5.9% over having some primary schooling, and finishing middle schoolraises it by another 13.1%. Adding secondaryschool or college further increasesthe probability, but at a sharply declining rate. The effects in the table are approximately cumulative, so a household containing someonewho went to

college is (other things equal) nearly 30% more likely to choose a yard tap relative to a householdwith maximum schoolingof some primary. Table 7 also showsthe percentageof A-site nonconnectors and B-site householdsfalling into each education category. About 61% of the A-site

nonconnectors

and 34% of the

and amortization

schedule at which the amortized connec-

tion chargewouldhavethe sameimpacton the probability of choosinga yard tap as the monthlytariff. If the amortization scheduleis arbitrarily fixed at 30 years with an interestrate of 20%, for example,a 10% increasein the averageconnection charge would be equivalentto a 2% increasein the

B-site households are in the two top education groups.

averagetariff in the followingsense:they wouldhavethe sameimpacton the annualexpenditurefor waterandonthe probability of choosinga yard tap. Although we cannot observethe creditmarketconditionsthe respondents face,it is clearthat, on average,their responses to the connection costmoduletook into accountthe costof capital.Thusit is nottheconnection costper sethatis an obstacle to choosing primaryschool.In contrast,responses fromthe salinewater a yardtapbutthe costof credit.If the waterauthority's cost area are significantlylower than in the excludedabundant of capitalis lowerthanthe averagehousehold's, which

Traditionalwater supplycharacteristics. Householdsin the scarcewater area are substantiallymore likely to choose a yardtap (otherthingsequal)thanarethosein the excluded abundantwater site. The magnitudeof the effect, which is highly significant,is approximatelythe same as having electricityor completingmiddleschoolinsteadof stoppingat

water site, a result that is unexpected.

seemslikely,foldingthe connection costintothemonthly

Contingentvaluationbiases. As is shownin Table5, the tariff may increasesubstantially the numberof yardtaps averagebids for the nonconnecting households and B-site demanded. households are much lower than for the connectors, as

would be expectedif their bids are biaseddownward.In Table6, the coefficients on the dummyvariablesincludedto controlfor possiblebiasesare negativeandsizable.It seems reasonabletherefore to interpret the responsesfrom these

Usingthe coefficients in Table6, we cansimulate the effectsof differentconnectioncost and tariff combinations.

The methodis as follows.The probabilitythateachhouse-

holdwouldhookupis calculated astheconnection cost is firstsetto 100,thento700,whilevarying themonthly tariff sites as underestimates of the true probability of their fromRs.0 to Rs.50.OnlytheA-sitenonconnectors andthe choosinga yard tap. B-sitehouseholds areused,of course.Whether service is improved issetto zero.Theotherindependent variables are the actualvaluesfor eachhousehold. If the resulting estiof connecting exceeds 0.50,the Water policy decisionmakersobviouslycannotin the mateof theprobability SIMULATIONS

shortrun affect most of the control variablesincludedin the

household iscounted asconnecting. Foreach price thetotal

SINGHET AL.' RURALWATERSUPPLYIN KERALA,INDIA 400

1939

--

Connection Cost =100

350

300

250

B Site

Number of Connectors 200 B rite 150

100

Nonconnectors

50

A Nonconnectors

Connection Cost =700[ I

0

5

10

15

20

25

30

35

MonthlyTariff(Rupees) Fig. 1. Simulated demand for varying connectioncost and tariff combinations.Derived from Table 6.

numberof connectors is counted, which measures the de-

maximum at a Rs. 10 tariff for each site in the Figure 2. For

mandfor yard taps at each price/connectionchargecombi- the B sites, for example, monthly revenue would be about 4 nation.

Figure1 showsthe result. The monthlytariff appearson thehorizontal axis, and the number of connectionsappears on the left vertical axis. The thin lines are for the A-site

nonconnectors,and the thick lines represent the B house-

holds.The higher pair of lines correspondsto the 100 Rs connectioncharge; the lower set, to the 700 Rs charge.

times higher at a Rs. 10 tariff with a Rs. !00 connection cost instead of a Rs. 700 connection cost. Furthermore, reflecting

respondents'sensitivityto the monthlytariff, revenuewould be nearly as high with a Rs. 5 tariff as with a Rs. 10 tariff because there would be so many more connectors. An obviousquestionis therefore how would consumersand the water authority fare under a scenario in which connection

Comparing the effects of the differencein the connection cost were folded into the tariff? chargefor each site indicateshow significantis the connectionchargeas a deterrentto connecting.At a Rs. 5 tariff, for Higher Tariffs, Lower ConnectionCharges, example,3 times as many B-site respondentswould connect and Consumer Surplus with a connection cost of Rs. 100 relative to Rs. 700. Figure3 putsthe quantityof connectionson the horizontal The effect of the connection cost on demand translates axis and price on the vertical axis. The demandcurve is the into radically different revenue estimates for the water same as that for the Rs. 100 connection charge in Figure 1, authority,as is shown in Figure 2. Revenue is near the exceptthat the numberof connectionshasbeen scaledup to

2000

Connection Cost =100 I

1800

1600

BSite

1400

A Nonconnectors 1200

MonthlyRevenue1000 A Nonconnectors

800

600

B Site

400

200

IConnection Cost =760'1

0

0

5

10

15

20

25

30

35

MonthlyTariff

Fig.2. Simulated water authority revenue forvarying connection cost and tariff combinations. Derived from Table 6.

1940 5O

SINGH ETAL.:RURAL WATER SUPPLY IN KERALA, INDIA

Monthly Tariff

thatconsumer surplus in Figure3 is Rs.5,500compared to

Current Supply of Connections

45

Rs.25,000 inFigure 4,againof450%. Consumer surplus by

thosewhopreviously wereconnected fallsbyRs.1,250 even though overallconsumer surplus increases somuch.Such a largeincreasein welfarecouldbe usedasjustification for

4O

35

subsidization if thenewsystem couldnotbeselfsupporting;

3o • •Consumer Surplus

underany circumstances it suggests that expansion of the watersystemwill makepeoplemuchbetteroff evenif it costs them Rs. 10/month.

15

lo•

The increasein welfareis suchthat thereis alsoroomto

CurrentRevenue

Current

5{ • ""/ , , , , , , 0

,

0

500

looo

1500

2000

2500

3000

Price

Oneapproachwouldbe to pay themcashrebatesequalto

•

3500

compensateexistingconnectorsfor their loss of welfare.

4000

4500

Estimated Connections in Population

theaverage connection costforthenewconnectors, sothey would not feel unfairly treated by the "subsidization"of new

Anobvious question iswhether thisanalysis ismistakenly

Fig. 3. Currentavailabilityof yard tapsandconsumersurplus.

the whole eligiblepopulationusingthe data reportedin Table 1. The current supply of connectionsis shown as a vertical line at 250. The supply curve crossesthe demandcurve at slightly more than Rs. 25, which by our estimate is the monthly tariff the water authority could chargefor the few connectionscurrently provided. The current price appears as a horizontal line at Rs. 5. At that tariff, with a Rs. 100

connectors.

makinginferencesbasedon the upper bound case,wherethe demand curve translates into the maximum number who would connect under the best of circumstances.Doubtful.

First, the demandcurve is estimatedon the assumption that households pay the first Rs. 100 of the connection cost. However, given the example above, there may be no need for them to pay any up-front capital costs, which wouldshift the demand curve out. Second, suppose only half of the

projectednumberof householdsactually hook up at a Rs.10 tariff. The samerevenuewould be availablefor subsidizing connectionsand paying for recurrent costs on a per household basisfor the half who do connect, so the examplewould not change in that respect. The only question is to what that would be collectedif currentconnectorscouldbe charged degree the costs to the water authority of expandingthe the price they, or others, are willing to pay for available supply of connections are fixed or variable. If most are connections. Because connectors currently pay only Rs. 5 variable, the'totalnumberof connectors is lessimportant; if each per month, the water authority is essentiallyproviding most are fixed, a smaller number of connectorswouldpose them a gift (consumer surplus) equivalent to the shadedarea. a more serious problem. Third, if A-site nonconnectorand Supposethat the water authority decides to raise the tariff B-site responsesare biaseddownward, as we expect,the to Rs. 10/month, using about half the monthly tariff to pay numbers used in this example may actually representa for connections and half for recurrent costs. This situation is lower-bound situation. The basic result is apparentwide shown in Figure 4 as the "new price" line. Supposefurther scope for increasing household connections, and welfare, that the supply of connections is expanded to 2,500, which through financial means. clears the market at that price. The revenue effect of the tariff hike and expansion of connections is an estimated Rs. Improved Quality of Service 25,000/month, about 90% of the maximum feasible revenue, The secondproblem commonly cited as an impedimentto given this demand curve. purchasing a house connection is the poor quality of the With this revenue projection, we speculate that the water rupee connection charge and no restrictions on connections, about 3,500 householdsare predicted to connect. The shaded area in Figure 3, if added to the small area showing existing water system revenues, showsthe revenue

authority could borrow Rs. 1,746,390 for 30 years at a real interest rate of 5%, servicing the loan with monthly payments of about Rs. 10,000. This loan would allow it to pay for connections for the 2,250 new customers at an average of Rs. 776/connection, about 1.5 times the estimated average cost for the A nonconnectors

and B-site households in the

sample.Another Rs. 15,000would be availablemonthlyfor i-ecurrentcosts or to finance other capital costs (comparedto Rs. 1,250 today in total). The water company benefits through a twelvefold increase in revenuesto cover recurrent

costs,and more peopleare hookedup, but are consumers

50

Tariff

45 40 35

New Supply of Connections

Consumer

3O

Surplus

25 2O

better off?.After all, they are paying much more for water (at

15

least in terms of cash).

10

Those who previously were connected are worse off becausethe monthly tariff has risen from Rs. 5 to Rs. 10 for them. However, this small loss of consumersurplusis more than offset by the large increasein householdswho benefit

Monthly

New

Price

5

0

500

1000

1500

Estimated

2000

2500

Connections

3000

3500

4000

4500

in Population

from private connections.The new consumersurplus.is Fig.4. Simulated change inconsumer surplus witha higher price and unconstrained connections. shownin Figure 4 as the shadedarea. We estimateroughly

SINOHET AL.: RURALWATERSUPPLYIN KERALA,INDIA

3500-

1941

limproved Service]

3000

2500

-

2000

MonthlyRevenue 1500

1000

500

0

5

10

15

20

25

30

35

40

45

50

MonthlyTariff

Fig.5. Simulated waterauthority revenue fortheA-site connecting households withandwithout improved service.

nected,but high willingnessto pay for improvedquality connected, the currentsituationis perceivedas quitebad. amongthosewho are connected,introducesa temporal Only17%reported thattheyweresatisfied withservice from dimensionto the water authority's planning and to some thepolicyproblem.Earlyinvestments by theiryardtaps.In contrast,62%of the nonconnectors and degreesimplifies be devotedto providing 58%of the B-sitehouseholds reportedthat they were satis- thewaterauthoritymayreasonably fiedwiththeirprimarysourcesof water.Of A-sitenoncon- yard taps at low cost to a muchwider baseof users, in the scarcewaterareas,andlaterinvestments nectors usinga publictap as the primarysource,45%are especially satisfiedwith it. Moreover, A-site connectors report con- mightbe devotedto upgradingthe systemas new customers that sumption of anaverage of 117L/dayfromtheyardtapand becomewillingto payfor betterservice.The question longtermplansof thatnatureis thecapital an additional195 L/day from a secondsource,so they are arisesin making ableto drawonly38%of their dailyconsumption fromthe cost differentialbetween a minimal quality system and a andhowrecurrent costsvaryoverthe tap,onaverage.Thusdissatisfaction withthewatersystem goodqualitysystem, life of the alternative systems. It may be possiblethatmuch is concentrated amongthosewho currentlyown yardtaps.

water service. For the A-site households that are now

canbe purchased withinthe conTheimproved servicedummyvariable in Table6 hasthe higherqualitysystems wrong sign,although it isnotstatistically significant. In light straintsof consumers'existingwillingnessto pay for yard in theprevioussection. ofthedescriptive information in theprevious paragraph, and taps,asdiscussed The scarce-water connectorsvalue an improvedservice because theimproved servicemodulewasnotconducted in B sites,the probitmodelwas rerunonly for the A-site muchmorethan do householdsin the otherareas,andthey Differhouseholds, usinginformationonlyfrom the tariff andim- arewillingto paymuchmoreto getbetterservice. ences in willingness to pay across sites creates the opportuprovedservicemodules.In that estimation, the improved to chargedifferentially acrossthemto service dummyhas a positivesignbut is not statistically nity, if necessary, significant. All othervariables behaveasin theoriginal Table financedifferentlevelsof service. If it costsmore in the area to improveservice,which seemslikely, 6 estimates.A third run with only for the A-site connectors scarce-water

wasthen tried. In that estimation,the improvedservice the sameconditionsthat give rise to the highercostsalso

to pay. In contrast,cross variable is positiveandstrongly significant. To economizegiverise to higherwillingness onspace, theseotherrunsarenotreportedin thispaperbut subsidiesfrom abundantor salinewater areasto the scarcewaterareaare probablynot feasiblegiventhe relatively are available from the authors. to payfor an improvedsystemin those Figure5 showsestimatedrevenuefor the existingsystem lowerwillingness andfor an improvedqualitysystembasedon theseunreportedestimates for the A-siteconnectors only. Improving

areas.

CONCLUSION

theservice hasa clearpositiveeffecton revenue (and,of intheintroduction a typical setof"stylized course, ontheprobability of choosing a yardtap)at each Wepresented tariff.Fortheexisting system, revenues eventually decline facts" aboutrural water systemsin India, and in fact, in world:thatsystems suffer from asthepricerises,butfortheimproved system, revenue hits mostpartsofthedeveloping a plateaubecausemost scarce-waterhouseholdsare pre-

poorqualityservice, restrictions (dueto bothlegaland

dicted to continue to choosea yardtapevenasthetarifffor theimprovedsystemreachesthe Rs. 50 maximum. Thefindings of no significant willingness to payfor an improvedqualitysystemamongthosewho are not con-

nections, and low revenues.The problemsare intimately connected to oneanotherandcreatea low-levelequilibrium trapwiththefollowing characteristics.

financialbarriers)on availabilityof private householdcon-

1942

SINGH ETAL.:RURAL WATER SUPPLY IN KERALA, INDIA

1. Residentscopewith the poor qualitypublicwater cooperation of theGovernment of IndiaandtheKerala Water supplysystem by eithernotusingit or supplementing it with Authority. Theauthors appreciate theimprovements suggested by reviewers. Manythanksto DavidGuilkey for traditional sources. Thereis littledemand foryardtapsfrom threeanonymous assistance withtheestimation. Thedataareavailable fromThe the current system. World

Bank.

2. The water authoritylacks the financialresourcesto improvethe system. REFERENCES 3. There is little hopefor changebecausegovernment allocations to thewaterauthorityarelargelyindependent of Briscoe,J., andD. de Ferranti,Waterfor RuralCommunities: HelpingPeopleHelp Themselves, WorldBank,Washington. demand for better services. Little revenue can be derived

fromthe only othersourceof funds,yardtaps,because of low demand for them and low tariffs for water.

Our analysissuggests thatthereis indeedpotentialfor the systemto rise out of its currenttrap. The real constraintin preventinghook-upsby respondents whoare deterredby the

highcost of a connectionis probablycreditmarketconditions rather than the connection cost itself. The water

D. C., 1988.

Briscoe, J., P. F. deCastro, C. Griffin,J. North,andO.Olsen, Towardequitable andsustainable ruralwatersupplies: A contin-

gentvaluation study inbrazil,World BankEcon.Rev.,4, 115-134, 1990.

Cummings, R., D. Brookshire,andW. Schultze,ValuingEnviron.

mentalGoods: A Stateof theArt Assessment of theContingent

ValuationMethod,RowmanandAllanheld,Totowa,N.J., 1986. Gallant,R., and H. White, A Unified Theoryof Estimation and

for NonlinearDynamicModels,Blackwell, Oxford, authoritycan solve this problemby foldingthe connection Inference England, 1988. costinto the monthlytariff. The resultshouldbe manymore Guilkey,D., Correctingfor unobservedcommunityeffectsin demohousehold connections and a tremendous increase in graphicandhealthsurveydata,report,27 pp., Carolina Popul. Cent., Chapel Hill, N. C., 1992. monthly revenue to the water authority. We show that M., J. Loomis,andB. Kanninen,Statistical efficiency consumers'welfare, based on responsesto the contingent Hanemann, valuation questions, rises with the increase in connections and monthly tariff.

of double-bounded dichotomous choicecontingent valuation, Am.

J. Agric. Econ., 73, 1255-1263, 1991. Maddala, G., Limited-Dependent and Qualitative Variablesin Econometrics,CambridgeUniversity Press, New York, 1983.

This studythereforetracesout a "new" pathwhichwater R., and B. Singh,The efficacyof improvedsupply supplyplannersin Kerala (and their counterpartsin other Ramasubban, system in meeting rural water demand for domestic needs:Evistatesand countries)could follow. By making a few critical dence from northern Kerala, 139 pp., Cent. for Soc. and Tech. policychanges,encouragingconnectionsandfinancingthose Change, Bombay, India, 1989. connectionsthroughhigher tariffs, the systemscan ratchet Train, K., Qualitative Choice Analysis: Theory, Econometrics,and an Applicationto AutomobileDemand, MIT Press,Cambridge, up to a "high-level equilibrium" in which there are many connectors, and revenues are greatly increased. Such a systemwill be better financed,makingit possibleto improve the reliability and quality of the service. Over the longer term, as more households are connected, the analysis suggests that they will demand, and be willing to pay for, improved service. The critical ingredient in effecting such a revolution is a changein perceptionsabout the financingand purposeof a public water supply system. Rather than trying to provide a heavily subsidized, minimal-service-to-all system, planners have the tools they need to provide an improved service that may cost consumers more but better meets their needs. At least for this sample, economic and design variables that take advantage of consumers' willingness to pay for specific characteristics point to a clear path out of the low-level equilibrium trap. The remaining questionsare (1) whether the demand-side information generated in this study is compatiblewith supply-sideinstitutionaland economicconstraints, (2) if not, what actions can be taken to solve the asymmetries, and (3) how well the contingent valuation methodhas predictedactual behaviorin the surveyedcom-

Mass., 1990. Varian, H. R., Microeconomic Analysis, 2nd ed., Norton, N.Y.,

munities.

Washington, DC 20433.

1984.

Whittington, D., J. Briscoe, X. Mu, and W. Barron, Estimatingthe willingness to pay for water services in developing countries:A case study of the contingent valuation method in southernHaiti, Econ. Dev. Cult. Change, 38, 293-311, 1990a. Whittington, D., A. Okorafor, A. Okore, and A. McPhail, Strategy for cost recovery in the rural water sector: A case studyof Nsukka district, Anambta state, Nigeria, Water Resour. Res.,26, 1899-1913, 1990b. Whittington, D., S. T. Lauria, and X. Mu, A study of water vending and willingnessto pay for water in Onitsha, Nigeria, WorldDev., 19, 179-198, 1991.

Whittington,D., V. K. Smith, A. Okorafor, A. Okore,J. L. Liu,and A. McPhail, Giving respondentstime to think in contingent valuationstudies,J. Environ. Econ. Manag., 22,205-225, 1992. World Bank Water Demand ResearchTeam, Towards a newrural

watersupplyparadigm:Implications of a multi-country study of households'willingnessto pay for improved water services, World Bank Res. Observ., 8, 47-70, 1993.

R. Bhatia and J. Briscoe, The World Bank, INUWS, 1818H Street, NW, Washington, DC 20433.

C. C. Griffin, Populationand Human ResourcesSectorOpera-

tions,EasternAfricaDepartment, WorldBank,1818H Street, NW, C. Kim, Department of Economics, Universityof Oregon, Eugene, OR 97403.

Acknowledgments. Financial supportwasprovidedby the Danish International Development Agency (DANIDA), the United NationsDevelopmentProgramme,the SwissDevelopmentCorporation, the Norwegian Agency for InternationalDevelopment,and the World Bank. The study benefitted from the assistanceand

R. Ramasubban andB. Singh,Centrefor SocialandTechnological Change, Bombay, India.

(Received April 17, 1991; acceptedApril 27, 1992.)