"Estimation of non-tap water demand for connected and non-connected households in urban districts of Rwanda"

Claudine Uwera, Department of Economics, University of Gothenburg

Overview Few households connected to tap water in developing countries( Baisa,

Davis et al.2010) Complexity in sources choice and Specific modeling specification

(Whittington et al., 2008) Separate single water demand equations : (Aburizaiza, 1991); (Crane,

1994); (David and Inocencio, 1998); (Rietveld et al,2000); (Basani. et al.,2008)

Single equation not helpful System of simultaneous demand equations works better : (Cheesman. et al.,

2008); ( Nauges and Whittington 2010);

IDEA Motivation Water an heterogeneous good in DC :Different sources Main points Household’s decision on using a specific source among other alternatives.

Relationship between water consumption , price and other socio-economics

characteristics. Form of new improved service and policy implication. Contribution Short existing literature particularly on non-tap water demand in developing countries We assume access to non-piped sources not exogenous in the water demand model for non-

piped households.

Background

Water supply sector divided into 2 subsectors: Urban and rural water supply system.

We distinguish households connected to piped water into their houses; and

those who lack piped connections in Rwanda

Multitude of coping sources.

• Only 3.4% connected to piped water within house or plot. • Connected and non-connected deflect demand to the available coping

sources. • Daily per capita consumption very low (6 to 8 liters),

• Poor households are more the most affected

Survey design & Data Household survey conducted in 5 urban cities of Rwanda from January-April 2011 Sample:700 households in total from 3 districts that compose the capital city; and 2

other selected cities. Data set covers 2 groups of households: currently connected to the tap water ; those

unconnected and use different coping water sources. 205 connected households of which 83% rely on coping sources and 495 non tap

households.

Connected households : 30% in the capital city and 19% and 33% respectively in the two other districts.

91% of households who use tap water rely on water in yard

Descriptive statistics Non connected Connected

Variable Mean S.D. Mean S.D.

Monthly income(US$) 267.97 370.77 385.03 511.73

Years of schooling 7.78 4.947 9.36 5.01

Household size 5.49 2.38 5.59 2.68

Children less than five 1.24 1.67 1.33 1.69

Access to electricity(0/1) 0.60 0.49 0.80 0.40

Number of bedroom 3.20 1.16 3.24 1.27

Hauling time(minutes/cubic meter/month)

346.15 349.03 220.43 410.05

Source: Household’s survey in Rwanda

Average water consumption (m3/capita/month) & Average cost

Unconnect. Connect.

AWC AC AWC AC

Variable dwelling 0.0 0.0 1.54 0.10

yard 0.0 0.0 3.02 0.45 SE private tap 0.67 .09 0.0 0.0

public tap 0.60 1.27 0.02 0.54 Tubewell 0.02 0.07 0.01 0.07

Protected dug 0.01 0.02 0.01 0.02 Protected spring 0.08 0.45 0.04 0.21

Unprotected spring 0.17 0.12 .004 0.02 Cart with small tank 0.003 0.02 0.01 0.04

Surface 0.03 0.10 0.02 0.08 other 0.01 0.05 0.02 0.42

Total non-tap water 0.18 0.40 0.04 0.22 Overall 0.22 0.22 0.44 0.20

Source: Authors’ survey NB: Average tap water price is 0.25USD/m3

1. Econometric specification: Non-connected households

Assumptions: Household’s choice as a complex decision. Hh combines different types of coping sources but rely more on one source. Hh makes a choice of his preferred coping source 𝑗 among 𝐽 available water

sources. Set of explanatory variables: full cost of water as the sum of price of water (𝑃) and the pecuniary time cost 𝑇 .

income(𝐼) and a vector (𝑍) of socioeconomic characteristics

variable (𝑆) as money saving from using free water.

quantity of water used 𝑸 as the dependent variable

Multinomial logit-OLS regression : non-connected households Two-step estimators Lee method used to correct selection biases in the choice of

4-alternatives of coping sources Selectivity is modeled as a multinomial logit

Estimation run by step (multi logit, then linear regression

with selectivity. Selmlog adds to the explanatory variables a series of

variables labeled 𝑚1,𝑚2,𝑚3,𝑚4.

Multinomial logit model

Non connected

households

Marginal effects a Robust standards errors obability to use water from a public tap

Households income(US$) 0.0002* 0.000 Years of schooling -0.002 0.051

hhsize 0.028** 0.014 Number of bedroom -0.017 0.023

Access to electricity(0/1) (0/1) 0.060 0.049 Hauling time -0.500*** 0.002

Children less than five -0.007 0.018 obability to use water from a protected

spring

Households income(US$) 0.0001* 0.000 Years of schooling -0.001 0.034

hhsize -0.019** 0.008 Number of bedroom 0.005 0.014

Access to electricity(0/1) (0/1) -0.044 0.034 Hauling time 0.164*** 0.032

Children less than five -0.009 0.011 Probability to use water from surface

Households income(US$) -0.0001* 0.000 Years of schooling -0.007* 0.004

hhsize -0.012 0.011 Number of bedroom -0.003 0.015

Access to electricity(0/1) (0/1) 0.088 0.038 Hauling time(hours) 0.266** 0.042

Children less than five 0.012 0.013 obability to use water from a private tap

Households income(US$) 0.0001* 0.000 Years of schooling 0.005* 0.003

hhsize -0.002 0.006 Number of bedroom 0.015 0.014

Access to electricity(0/1) 0.103*** 0.034 Hauling time(hours) 0.068** 0.037

Children less than five 0.003 0.012 Marginal effects of each characteristic on the probability of using each of the four non-tap sources. a ***,** and * significance at 1,5 and 10% level, respectively

Source: Authors’ survey

Second step: Estimation of water demand function : non-connected households

Estimated

coefficientsa

Boostrapped standard errorsb

Student’s t-test

Constant -0.01 0.511 -0.01 Log(total cost(public tap)) -0.142** 0.063 -2.26

og(total cost (protected spring)) -0.014 0.052 -0.27 Log(total cost private tap)) -0.738*** 0.283 -2.61

Log(income) 0.033* 0.021 1.61 Log(savings) 0.199*** 0.071 2.82

g(lot size(number of bedroom)) -0.752** 0.359 -2.09 Log(kids under5 ) 0.254*** 0.091 2.81

Kicukiro dummy distr -0.011 0.110 -0.11 Gasabo dummy distr 0.064 0.086 0.74

Lee correction parameter 1c -0.863** 0.463 -1.87 Lee correction parameter 2 0.702 1.534 0.46 Lee correction parameter 3 -0.789 2.044 -0.39 Lee correction parameter 4 -0.011 0.731 -0.02

observations 495 Wald test of parameter equality(three sources)

14.66

p-value 0.002 Unconnected sub-sample in all districts

a ***,** and * significance at 1,5 and 10% level, respectively. b replications. c Water sources: Public tap, protected spring, private tap Source: Authors’ survey

2. Econometric specification : connected households System of simultaneous demand equations to estimate overall water

demand for connected-households. Assumptions : demand for water from the piped network 𝑞1 and a demand for water

from non-piped network 𝑞2 . 𝑞2 can be zero for connected households who don’t rely on coping

sources Ordinary Least Squares might be biased Equation for 𝑞2 as a tobit model for variable censored at zero The general system of water demand can be specified as follow:

�𝑞1 = ∑ 𝛾𝑗1

𝐽𝑗=1 𝑝𝑝 + 𝑥1𝛽1 + 𝑢1

⋮𝑞𝐽 = ∑ 𝛾𝑗

𝐽𝐽𝑗=1 𝑝𝑝 + 𝑥𝐽𝛽𝐽 + 𝑢𝐽

First step: Probability of having a piped in house for connected households

Two steps

1. The decision to have or not a piped connection.

The probit model : the probability of having a connection.

To control for selection bias, the estimated parameters from the first stage are used to compute the so-called inverse Mill’s ratio that will be added into the water demand model.

2. Tobit estimation of water demand of piped households

Probabilty of having a piped in house Coef. Std. Err. z P>z

income 0.001 0.000 4.56 0.000 If the piped water available (0/1) 1.172 0.211 5.56 0.000 Years of schooling 0.053 0.013 4.01 0.000 Kicukiro district (0/1) 0.860 0.206 4.16 0.000 Gasabo district (0/1) -0.191 0.156 -1.22 0.222 Nyarugenge district (0/1) 0.408 0.188 2.17 0.030 Huye district (0/1) -0.389 0.194 -2.00 0.046 _cons -1.422 0.252 -5.63 0.000 Number of observations 209 Likelihood-ratio test:test statistic (p-value)

a ***,** and * significance at 1,5 and 10% level, respectively Source: Authors’ survey

Second step: Tobit estimation of a system of water demand for connected households

Coef. Std. Err. z P>z

Dependent variable: Piped water consumption per capita per month Instrumented average price for households combining piped and non-piped water (log) -0.367 0.201 -1.82 0.068 Instrumented average price for households using piped water only (log) -0.365 0.176 -2.07 0.038

Income (log) 0.155 0.039 3.89 0.000

Full cost (log) 0.198 0.201 0.98 0.325

Household size (log) -0.965 0.121 -7.94 0.000

Mill’s ratio 0.033 0.037 0.88 0.377

Kicukiro district -0.077 0.159 -0.48 0.628

Gasabo district -0.096 0.160 -0.60 0.549

Constant 0.999 0.371 2.69 0.007 Dependent variable: Non- Piped water consumption per capita per month

Instrumented average price for piped water (log) 0.181 0.130 1.38 0.166

Income (log) -0.032 0.034 -0.95 0.340

Numebr of kids under five (log) 0.316 0.129 2.44 0.015

Number of bedroom (log) 1.150 0.234 4.91 0.000

Full cost (log) -0.545 0.215 -2.53 0.012

Mill’s ratio -0.237 0.046 -5.09 0.000

Kicukiro district -0.326 0.197 -1.65 0.099

Gasabo district 0.601 0.189 3.18 0.001

constant -1.665 0.361 -4.60 0.000

Number of observation 205

Source: Authors’ survey

Conclusion Cross sectional data collected in 5 urban areas of Rwanda ….

Substitutability between tap water and Public tap water….

Welfare effect of extending public tap connections might be very large…..

Connected households less sensitive to price change than non- connected.

Improving current price schemes as good instrument for extension….

However different reactions……………

Further applications:

Cost-benefit analyses for either extending current tap water system or improving current non-tap distribution system……………