WATER RESOURCES RESEARCH, VOL. 29, NO. 5, PAGES 1353-1358, MAY 1993 A Model of Household Choice of Water Supply Systems in Developing Countries SAMER MADANAT School ofCivil Engineering, Purdue University, West Lafayette, Indiana FRANNIE HUMPLICK Latin America I Infrastructure Department, World Bank, Washington, D.C. Studies of pipe water demand in developing countries have traditionally analyzed household connection decisions to thepipe water system. Ontheother hand, empirical observations have revealed that often, after connecting, households do not use their pipe water supply, or augment it with alternative sources. Due to deficiencies in pipe water quality, pressure, or availability, households invest in coping strategies in the form of alternative supplies and storage facilities. Because these strategies have important economic implications, there is a need to develop an understanding of households' water demand thatgoes beyond connection decisions. Thispaper presents a model system of household water supply choices. The system accounts forthe fact that households may use different supply systems for different uses of water. Moreover, the relation between households' choices of water supply andtheirconnection decisions is explicitly modeled. The approach is illustrated using data from Faisalabad, Pakistan. 1. INTRODUCTION Residential water demand studies in the United States have analyzed the quantity of pipe water usedby a house- h01d as a function of variables such as incomeand price iDanielson, 1979;Agathe and Billings, 1980;Howe, 1982]. This emphasis, however, is not appropriate to developing countries, where it has been observed that households do not necessarily connect to existing pipe water systems [Attaf et al., 1989]. Asa result,water demandstudies in developing countries have attempted to identify the determinants of households' decisions to connect to the pipewater system, using binary choice modeling [Altaf et al., 1989; Singh et al., 199!;World Bank Water Demand Research Team, 1993]. The objective of such studies is to forecast the fraction of a population willing to payfor a pipeconnection at different price levels and fordifferent attributes of the pipewater,which would provide planners with an essential input for system design, operation, andpricing [Whittington et at., 1990]. However, empirical observations of existing pipewater systems in developing countries reveal that many of these systems are not functioning properly. In response to defi- ciencies in thepipe water quality, pressure, or availability, households invest in alternative supply sources or storage tanks or do not connect tothe pipe system. Moreover, it was observed thathouseholds who are connected to thepipe water system do not necessarily use pipe waterfor all, or even any,of their water consuming uses. A household may choose different sources ofwater for different uses (drinking, bathing, laundry, house cleaning, etc.). These observations have motivated a new look at the field of waterdemand analysis in developing countries, for both practical and methodological reasons. Copyright 1993 by the American Geophysical Union. Paper number 93WR00126. 0043.1397/93/93 WR-00126505.00 From a practical standpoint, the copingstrategies adopted by households have important economic implications. Households incur high fixed costs, in the form of invest- ments in alternatesupply sourcesand storagefacilities, and recurringcosts, in the form of water purchases from ven- dors, when coping with water supply deficiencies.More- over, they may also incur indirect costs such as diversion of labor away from income-generating activities to coping activities. It is thus of interest to understand the determi- nants of households'decisions to adopt various coping strategies. Methodologically, it became clear that pipe water systems must be viewed as one option in a water demand model, in whichalternative sources of water are explicitly accounted for. Such an analysis,using multinomial choice modeling, was conducted by Mu et al. [1990] for the case of rural residentialwater demand in Ukunda, Kenya. The authors analyzed the effect of price, collection time, water taste, income, education, and number of women in the household on the choice of primary water source. The alternatives included in their model included water purchased from vendors, wells, and kiosks. In this paper, we extend the work of Mu et al. in two directions. First, we recognize that in developing countries a household may use several water supply systems,though usually only one is chosen for each water use. Our analysis of water source choiceis thus usage specific.This enables us to identifythe relative weightsthat households place on different attributes of water (quality,pressure, reliability) for eachusage of water. Second, we jointly analyze household water source choices and connection decisions in a multidi- mensional model system. By doing so, we explicitly account for the interdependence betweenthese decisions. Namely, the effects of the expectedhousehold utility, associated with source choice, on connection decisions are identified. Thus our model of connection decisions identifies the water us- !353