1

Pricing Policy Effectiveness is Domestic Water Demand Management

Estimation of Domestic Water Demand Function in Lahore

Tamkinat Rauf1*

and

M.Wasif Siddiqi**

ABSTRACT. This study examines the management of household water demand through a pricing

policy for achieving the objectives of cost recovery, efficient water use, and equitable allocation

of water resources. To this end, a demand function is estimated using household level data about

water consumption and socio-economic characteristics of 156 households supplied by WASA,

Lahore, in the period 2004-2006. The results show that domestic water demand is highly elastic

to price at the existing tariff rates. An average increase of up to 30% in the consumption-based

part of existing tariffs as well as an increasing non-volumetric rate based on property value or

the size of dwelling is recommended.

1 The writers are, respectively, Economic Analylist,Research Dept.State Bank of Pakistsn,Karachi and Associate professor Govt. College University, Lahore. Tamkinat.rauf@gmail.com

2

1. INTRODUCTION The population of Lahore has roughly doubled over the past twenty years, and an increase of

two million is expected by the year 2020 (UN, 2005). This has important implications for city

planning as demand for housing, electricity, water, sanitation, public health, education, and

infrastructure grows accordingly.

WASA, the city’s official water supplier, has often responded to the growing demand by

offering the supply-side solution: augmenting supply capacity by exploiting new water resources.2

Such investments are costly, but in view of the public good nature of water, WASA has kept tariffs

well below the cost-recovery level, relying on heavy loans and subsidies. While this arrangement may

have worked in the past, it is now becoming increasingly unsustainable, because 1) WASA is facing

severe financial constraints and which has led to poor service and underinvestment, and 2) the

environmental cost of extracting water is increasing.

With its low tariff rates and continually increasing costs, the WASA Lahore is unable to meet

even its O&M costs (WASA, 2007). WASA has been receiving financial assistance from the

provincial and Lahore district governments as well as international donors in the form of grants and

loans with the grant element gradually diminishing over the passage of time. WASA currently owes

Rs. 5.6 billion to these agencies and is in no position to make the repayment (WASA, 2007).

Deteriorating financial situation has also led to short-term planning, reactive operational strategy, and

underinvestment in asset maintenance, future capacity, IT equipment, management and accounting

information system, and training (IFC, 2005). Consequently, WASA has shown suboptimal

performance: low pressure and irregular supply, leakages, poor customer service, etc.

Secondly, indiscriminate and unplanned exploitation of water resources may result in severe

water shortages in future. Water supply in Lahore depends essentially on groundwater pumped

through privately or publicly-owned tubewells and hand pumps. However, groundwater is a limited

resource, recharged only once a year during the monsoon. Reportedly, the groundwater table has been

falling.3 Falling water tables increase the cost of pumping, as more energy is required to pump deep

water. Furthermore, the International Water Management Institute has predicted severe water shortage

in the country by the year 2025, that will threaten even the sustainability of agriculture (IWMI, 2000).

2 In line with the demands of the growing population, WASA has continuously expanded its supply capacity in the past. Currently, WASA produces 350 million gallons of water per day with its 400 tubewells. Over 25 more tubewells have been approved under the 2007-08 budget of the agency (WASA, 2007). 3 Over the year 2002-03, the groundwater table in Punjab fell on average by 0.61 percent (Govt. of the Punjab, 2005). More recent estimates were not available.

3

Relentless extraction of water may also lead to an irreversible decline in the ability of the region to

store water in the ground (Gleick, 1998).

Water produced by WASA is not being efficiently utilised. The basic water requirement for

drinking, sanitation, bathing, and food preparation is 13.2 gallons pcd, while WASA produces 80

gallons pcd – an excess of over 66 gallons pcd (WASA, 2007).4 Evidently, there is an excess demand

– people demand more quantities of water than they would if they were made to pay the true

environmental and supply cost of water. Clearly, the supply solution discussed earlier is not the best

answer to the apparently growing demand.

Instead, WASA should be looking at demand management that involves pricing policies and

rationing – notionally allocating a fixed amount of water to each household, based upon lifeline and

household size considerations. However, though rationing brings a definite change in demand, it is

difficult to implement and may not be widely acceptable. Pricing policies, on the other hand, have

been successfully implemented in countries like Brazil, Canada, France, Spain, the United Kingdom,

and the United States.

Through pricing policies existing demand patterns are modified to achieve various objectives,

such as cost recovery, conservation, and equitable allocation of water among different income groups.

To implement such a policy successfully, the value that consumers place on water must be known.

This value is reflected by the price elasticity of water demand – the percentage change in demand that

will be caused by a percentage change in price. If the demand is inelastic, it shows that at the existing

prices, the consumer highly values water and will be willing to pay a higher price in order to consume

the same amount of water. On the other hand, if elasticity is high, the consumer indicates a

willingness to reduce/increase the use of water with changes in its price. And if price is increased, he

would shift a part of his expenditures elsewhere. Clearly, this information is fundamental in deciding

the manner in which tariff should be structured.

Another important aspect of a pricing policy is that households with different socio-economic

settings, especially different income levels, will be affected differently, often giving rise to issues of

equity – fairness in the distribution of cost and conservation burdens. How water demand varies

across different types of household is central in estimating the implications of a pricing policy on

equity.

4 The internationally recommended lifeline supply is 50 litres pcd. 1 litre = 0.2642 gallons

4

This study estimates household water demand in Lahore in order to explore the potential of a

pricing policy to increase revenues and discourage inefficient water use while ensuring a fair

distribution of water in the community.

The thesis is organised as follows: Section two reviews literature on water demand. Section

three gives a description of the study area. This is followed by the methodology: section four,

sampling framework; section five, theoretical framework; and section six, data description. Results

are presented in section seven, followed by policy recommendations in section eight, the last section.

2. LITERATURE REVIEW

The estimation of urban residential water demand has been an area of wide and growing

interest world-wide for the past three decades. However, most of the existing literature pertains to the

developed world: the United States and Europe (Agthe and Billings, 1987, Arbues and Villanua,

2006, Batchelor, 1975, Chicoine and Ramamurthy, 1986, Foster and Beattie, 1979, 1981, Hansen

1996, Headley, 1963, Hewitt and Hanemann, 1995, Nauges and Thomas, 2000, Neiswiadomy and

Molina, 1989, 1991, Renwick and Archibald, 1998, Wong, 1972). So far, no comprehensive study has

been conducted in Pakistan to estimate the urban residential water demand and neither has the

researcher come across any such study of similar-income countries. Therefore, the reviewed literature

has limited usefulness in many aspects. What follows is an analytical review of the methodologies and

data types used in previous studies and a brief assessment of the suitability of these methods in the

present context.

Water demand estimation studies have used various sorts of data: time-series (Agthe and

Billings, 1987, Hansen, 1996), panel (Arbues and Villanua, 2006, Nauges and Thomas, 2000,

Neiswiadomy and Molina, 1989, 1991, Renwick and Archibald, 1998) and cross-sectional (Chicoine

and Ramamurthy, 1986, Foster and Beattie, 1979, 1981, Headley, 1963, Wong, 1972). Time series

data is useful to study the effects of a policy change such as restructuring of block-rates, rationing of

water supply, or the introduction of a new water-related appliance. Time series data also captures the

effect of weather. A drawback of such data is that it is aggregate: summing or averaging quantities,

such as consumption, income, and prices, for the entire community. For this reason, results derived

from time series data have limited usefulness.

Cross-sectional data on the other hand can be collected for disaggregate units, such as

individuals, households, or localities. This data holds more information than time-series data, and is

appropriate for estimating demand across different groups. However, cross-sectional units may have

5

too much variability which can cause heteroscedasticity, in which case the OLS estimators have high

variances.5

The most useful approach is perhaps the panel data, because it combines elements of both

cross-sectional and time-series: more variables can be studied while time effect is also captured. Panel

data also increases the number of observations, and hence the accuracy of the model. For these reason,

this study has used panel data.

Urban water is usually priced under ‘block-rate’ schedules: a volume-based rate consisting of

a sequence of marginal prices for different consumption blocks. Water use in each billing period is

divided into successive blocks with use in each ascending block charged at a different price. The

block rate schedules can be progressive or regressive with increased consumption. The schedules are

established to ensure efficient use of resource, as well as to achieve equity, environmental

conservation, cost recovery, and public acceptability. An important point of contention in water

demand studies has been the specification of the price variable in the model (Charney and Woodard,

1984, Chicoine and Ramamurthy, 1986, Foster and Beattie, 1981, Opaluch 1982, 1984). Water

demand studies use two alternative types of price specifications:

1. Marginal price of the block under which the consumer falls plus a ‘difference variable’ (following

Taylor, 1975 and Nordin, 1976). The difference variable is calculated as the difference between what

the consumer actually pays and what he would have been charged if all consumption units were

charged at the marginal price of the last unit of consumption. The difference variable is used along

with the marginal price.

2. Average price – the total water expenditure by the consumer in a billing period divided by the total

water consumed in that period.

Proponents of the difference variable specification (Agthe and Billings, 1987, Renwick and

Archibald, 1998) argue that the consumer is well-informed and therefore responds to the marginal

price and difference variable. On the other hand, those who favour average pricing argue that the

consumer does not devote time to studying the tariff structure and only has a rough idea of what he

pays for his consumption (Foster and Beattie, 1981).

Nieswiadomy and Molina (1991) and Opaluch (1982) have suggested statistical tests to

determine the price to which the consumers actually respond. The advantage of one test over the other

5 Heteroscedasticity is defined as non-constant variances of residuals. In the presence of heteroscedasticity, OLS estimators remain unbiased and consistent but they no longer have minimum variance.

6

was not readily apparent. This study has used the more recent Nieswiadomy and Molina (1991) price

specification test to determine the correct price specification.

It has been further argued that ill-informed consumers react to past rather than current prices

(Charney and Woodard, 1984), and hence the appropriate specification of price would be the lagged

(average) price. The lagged-price specification has not been used in the present study because the

WASA tariff schedule is rather uncomplicated (only three blocks) and has been revised only once

since 1998. However, past bills may have some impact on the consumers’ decision-making; a lagged

consumption variable has therefore been added as a regressor.

With multi-part block rates, prices are endogenously determined by the quantity demanded,

and hence a the model is based on simultaneous equations. Under simultaneity, the OLS method

yields biased and inconsistent estimates. Most water demand studies have used either instrumental

variables (Nauges and Thomas, 2000, Neiswiadomy and Molina, 1989, 1991) or two-stage least

squares (Agthe and Billings, 1987, Renwick and Archibald, 1998) to remove the ‘simultaneity bias’.

Arbues and Villanua (2006) have used a ‘dynamic panel model’ which is applicable to cases where

the price is lagged to a degree such that it is no longer correlated with the error term in the current

period. Hewitt and Hanemann (1995) have used a complex discrete/continuous model that builds on

the discontinuous nature of the budget constraint faced by the consumer under block-pricing. The

OLS method has been used in studies where uniform rates were charged (Hansen, 1996) or the

demand function was formulated under restrictive assumptions (Chicoine and Ramamurthy, 1986,

Foster and Beattie, 1979, Headley, 1963, Wong, 1972). This study has used two-stage least squares

method of estimation.

Different functional forms have been used in domestic water demand studies, including linear

(Agthe and Billings, 1987, Batchelor, 1975, Renwick and Archibald, 1998, Nauges and Thomas,

2000, Neiswiadomy and Molina, 1989), double-log (Foster and Beattie, 1979, 1981, Hewitt and

Hanemann, 1995, Neiswiadomy and Molina, 1991, Wong, 1972) and semi-log models (Hansen,

1996). However, there is no evidence to indicate which is the most appropriate form. Linear models

are easy to estimate while double-log models are useful because the coefficients give estimates of

elasticities. Following Arbues and Villanua (2006), in order to establish the most adequate functional

form, this study has estimated all three types of specifications: linear, double-log, and semi-log. The

most statistically and theoretically sound model has been selected for drawing conclusions.

7

3. WATER SUPPLY IN LAHORE Lahore is one of the oldest cities of South Asia and is the provincial capital of Punjab. The

Lahore district spreads over an area of 1,772 square kilometres with a population density of over

3,566 persons per thousand square kilometres (Govt. of the Punjab, 2005). Population-wise, Lahore is

the second largest city in Pakistan, fifth-largest in South Asia, and 23rd in the world (World

Gazetteer, 2007). The current urban population is over 6.6 million and is expected to exceed eight

million by the year 2020 (UN, 2004).

The Lahore Development Authority (LDA) is the chief municipal body responsible for

preparation and implementation of schemes for environmental improvements, housing, slum

improvement, solid waste disposal, transportation and traffic, health and education facilities, and

water supply and sewerage in Lahore City area under the LDA Act, 1975. The chief water supplier in

urban Lahore is WASA, formed under the Act.

The WASA service area extends to 350 square kilometres, supplying water and sewerage

services to a population of over five million (IFC, 2005). Other private water suppliers also exist in

Lahore City, but there is no official record of their number and coverage. For administrative purposes,

the area covered by WASA is divided into six blocks called ‘towns’: Allama Iqbal Town, Aziz Bhatti

Town, Ravi Town, Shalimar Town, Ganj Baksh Town, and Nishtar Town. Each town is further

divided into O&M sub-divisions.

Though located along the bank of River Ravi, water supply in Lahore depends on

groundwater, the river being the most polluted in the entire country.6 For Lahore, groundwater is an

ideal source of water because it is relatively free of impurities and therefore little or no treatment of

the water is needed before it is put to household use.

Table 3.1: WASA Administrative Division

Town O&M Sub-divisions

Allama Iqbal Town Allama Iqbal Town, Samanabad, Johar Town, Ichhra

Aziz Bhatti Town Taj Pura, Mustafabad

Ganj Baksh Town Ravi Road, Krishan Nagar, Shimla Hill, Mozang, Gulberg

6 River Ravi is the most polluted river in Pakistan, receiving 47 percent of the total municipal and industrial pollution discharged into all rivers in the country. According to the World Wildlife Fund, extreme pollution has destroyed around 42 fish species that the river was home to. Even contact with the river water has been reported to cause severe skin diseases. The water is certainly unfit for drinking. (EDC News: http://www.edcnews.se/Cases/PakRaviriver.html)

8

Nishtar Town Green Town, Industrial Area, Township, Garden Town

Ravi Town Shahdara, Data Nagar, City, Farkhabad, Misri Shah,

Shadbagh

Shalimar Town Baghbanpura, Mughalpura

Water Pricing under WASA

WASA water tariffs are apparently based on cost-considerations, but are well below the cost

recovery level. The objective is seemingly to recover an acceptable portion of the cost rather than the

full costs, that being a compulsion due to political considerations.7

Table 3.2: ARV-based Billing Structure for Domestic Unmetered Connections

Rate (Rs. per month) ARV (Rs.)

January 1998 May 2004

Up to 400 70.55 98.77

401-500 108.80 152.32

501-720 185.30 259.42

721-1000 323.00 452.20

10001-1500 455.00 637.84

1501-2388 479.40 671.16

2389-4370 510.00 714.00

4371-4499 533.00 747.32

4500 and above 84% of ARV 84% of ARV

Table 3.3: Water Tariff Structure for Domestic Metered Connections

Rate (Rs. per 1,000 gallons per month) Consumption (gallons)

January 1998 May 2004

Up to 5,000 9.20 12.88

5,001-20,000 14.90 20.86

20,001 and above 19.50 27.30

Source: WASA, Lahore.

7 The City District Govt. has not allowed WASA to raise tariffs for the past three years in spite a 10 percent increase in the electricity rates (WASA, 2007).

9

Tariffs for both metered and unmetered connections have been revised thrice over the past

decade: in July 1997, in January 1998, and then in May 2004. No annual inflation adjustments are

made.

Water is charged volumetrically where the connection is metered, while unmetered

connections are charged on the basis of the annual rental value (ARV) of the house.8 The ARV is

divided into nine bands ranging from Rs.400 to Rs.4, 500 and above. However, since ARV-based

charges do not directly affect consumption, we have selected only metered households for this study.

Currently only 30 percent of WASA connections are metered but WASA is making

substantial efforts to meter all existing connections (WASA, 2007).9 No new unmetered connections

have been issued since January 1997. Metered connections are charged with a two-part tariff: a

variable volume-based part and a fixed part. The fixed part includes monthly connection fee of Rs.12

plus a flat charge of Rs.3 per month. The volume-based charges are divided into three ascending

volumetric blocks, that is, consumption in each succeeding block is charged higher than the previous

block.

4. A MODEL FOR WATER DEMAND

The domestic demand for water arises from its use for sanitation, bathing, washing clothes,

cleaning homes and cars, cooking and drinking, watering lawns, cooling, and recreational activities.

Like other commodities, the demand for water is expected to fall with price and increase with income.

Some other factors, such as the prices of water-related appliances, household size, house size, and

weather, etc. are also expected to have some influence on water demand. Based on these

considerations, a model for domestic water demand is presented below.

As discussed in chapter two, the price effect under block-rate pricing enters the demand

equation indirectly. If consumers are well-informed of the price structure, they respond to the

marginal price (MP), that is, the price of their final consumption block. But fully informed consumers

are also aware of the benefit that they gain by having paid less on the initial blocks. This benefit

enters the demand equation as the ‘difference variable’ – the difference between what the consumers

actually pay and what they would have paid had all units been priced at the marginal price. The

difference variable (DV) is computed as follows: 8 Annual Rental Value (ARV) is defined as the gross annual rent at which a land or a building might be expected to be let from year to year, less deductions for repair and maintenance. (World Bank and The Urban Unit, Lahore, 2006.) 9 WASA allocated Rs.29 million for the procurement of domestic water meters in 2007-08. (WASA budget document, 2007.)

10

DV = [P1Q1 + P2Q2 + P3 (Q-Q1-Q2)] – [P3Q] [1]

Where P1, P2, and P3 are the respective prices charged under the successive blocks. Q1 and Q2 are

the respective consumption limits for the first two blocks.

Alternatively, if the consumers are not fully aware of the water charges, they approximate the

average price of water by dividing the total water expenditure in a billing period by the quantity of

water consumed in that period. The average price (AP) is computed as:

AP = [P1Q1 + P2Q2 + P3 (Q-Q1-Q2)] / Q [2]

The correct specification of price is largely a circumstantial question. For example, in

Zaragoza, Spain, where the tariff rate consists of 205 consumption blocks, it is reasonable to assume

that consumers cannot be fully aware of such a complex tariff structure, and therefore AP

specification can be used.10 In the present case there are only three blocks and, analogously, it can be

argued that the marginal price specification is correct. However, a more rigorous approach can also be

adopted.

Under increasing block rates, the perceived price (P*) must lie somewhere between the

average and marginal prices. Let ‘k’ be a parameter that measures price perception. We may write,

P* = MP (AP/MP) k

The value of ‘k’ will lie somewhere between zero and one. If ‘k’ is found to be zero, then MP is the

perceived price. If it is equal to one, the perceived price is AP. If neither of these cases holds, then the

perceived price lies between AP and MP.

To test the value of ‘k’, the following equation is estimated:

lnQ = a0 + a1lnMP + a1kln(AP/MP)+ ΣaiXi + µ [3]

Where Xi denotes predetermined variables.

Standard t-test can be applied to test the value of ‘k’.

Household water use is expected to increase with increase in income. More affluent

households are likely to use water less vigilantly than low income households. They are also more

likely to use water for washing cars, watering lawns, and swimming pools. A strong relationship has

also been found between use of water-based appliances and household water demand (Batchelor,

1975). However, reliable data of these indicators can be difficult to obtain. Therefore, information

10 Arbues, Fernando and Villanua Inmaculada (2006). Potential for Pricing Policies in Water Resource Management: Estimation of Urban Residential Water Demand in Zaragoza, Spain.

11

about the house value was used as an indicator of wealth, a proxy for income, hoping that it would

also take into account some of the other water-related variables.11

The size of the dwelling and the number of residents is also expected to influence water use.

Larger houses use more water for cleaning and irrigation of lawns, thus water use should increase

with the house size. The relationship of demand with the number of residents is, however, not so

direct. Generally, high income families are smaller than low income families. Per capita water

consumption may, therefore, be lower in low income families, in which case there may be a negative

effect of household size on consumption.

Three water-related activities are predominantly influenced by weather: watering lawns, using

room-coolers, and bathing. As temperatures rise, these activities become more frequent. A weather

variable has been included in the demand model to capture this effect.

Past bills, and therefore consumption patterns, are likely to have some influence a consumer’s

decision-making in the future. Moreover, this effect is likely to decline with time: older bills wielding

lesser influence than the ones that follow. In such a situation, Koyck (1954) has proposed substituting

the lagged explanatory variables by one single lagged dependent variable that appears among the

regressors.12 Following this proposition, a lagged consumption term has been included. The

coefficient of this variable must be positive and must not exceed one.

Domestic water-meters are owned by either WASA or the consumers. Entitlement to the

meter does not directly influence demand but it may influence the quality and accuracy of the meter:

if the meter overestimates consumption, the consumer would make sure that the fault is fixed. If, on

the other hand, WASA meters under-read consumption, WASA will ensure accuracy. A dummy

variable has been used to capture this affect. The dummy takes a unit value if meter is owned by

consumer, and a zero value if owned by WASA. Possibly, this dummy will be negatively related with

the recorded consumption.

Fixed Effects

As discussed above, many factors cause variability in demand patterns across households:

household income, per capita income, ages of residents, lawn size, car ownership, and use of water-

related appliances. However, not all these variable could be included in the demand model. The effect

of missing variables is reflected in the intercept term.

11 Following Batchelor (1975), and Nieswiadomy and Molina (1989, 1991). 12 See: Koutsoyiannis, A. (1972). Theory of Econometrics (2nd ed.). pp.304-6.

12

When missing variables differ significantly across areas or communities, this may give rise to

a ‘differential intercept model’ – a model in which the regression line has varying intercepts for

different population groups. To estimate these differences across groups, a ‘fixed effects model’ is

used, based upon dummy variables assigned on the basis of geographical location.13 Five intercept

dummies have been introduced for Aziz Bhatti, Allama Iqbal, Ganj Baksh, Ravi, and Shalimar towns.

Nishtar town is the benchmark category.

Table 4.1: Notation and Variable Description

Variable Indicator Description

Qt Quantity of water consumed in billing period t

AP Average price

MP Marginal price

DV Difference variable

W Wealth

NR Number of residents

L Plot size

T Average temperature in period t

Dmo Meter ownership dummy

Dait Dummy for residence in Allama Iqbal Town

Dabt Dummy for residence in Aziz Bhatti Town

Dgbt Dummy for residence in Ganj Baksh Town

Drt Dummy for residence in Ravi Town

Dst Dummy for residence in Shalimar Town

Functional Form

As discussed in chapter two, there is no theoretical basis for adopting any specific functional

form. Past studies have typically used linear, semi-log, and double-log forms.

The MacKinnon, White, and Davidson test (MWD test) was used to choose between linear

and double-log forms. The results recommend a double-log functional form. However, there were no

a priori grounds for choosing between semi-log and double-log functions, and therefore, both

functions were estimated. Double-log models have the advantage that their estimators give the

13 Fixed Effect Model (FEM): A model in which the intercept term varies over sampling units but is constant over time.

13

elasticities of the variables, thus simplifying interpretation. Semi-log models are useful to estimate

elasticity across different population groups, as the coefficients give ‘semi-elasticity’: elasticity of the

dependent variable with respect to varying values of the regressors.

The Estimated Models

The following demand models were estimated:

1. MP Models

DV = [P1Q1 + P2Q2 + P3 (Q – Q1 – Q2)] – P3Q [1]

Semi-log:

lnQ = β0 + β1MP + β2T + β3W + β4NR + β5L + β6Dmo + β7Dabt + β8Dait + β9Dgbt + β10Drt + β11Dst

+ β12 DVµ3 [3]

Double-log:

lnQ = β0 + β1lnAP + β2lnT + β3lnW + β4lnNR + β5lnL + β6Dmo + β7Dabt + β8Dait + β9Dgbt + β10Drt

+ β11Dst + β11D β11DV + µ2 [4]

2. AP Models

AP = [P1Q1 + P2Q2 + P3 (Q – Q1 – Q2)] / Q [2]

Semi-log:

lnQ = β0 + β1AP + β2T + β3W + β4NR + β5L + β6Dmo + β7Dabt + β8Dait + β9Dgbt + β10Drt + β11Dst +

µ3 [5]

Double-log:

lnQ = β0 + β1lnAP + β2lnT + β3lnW + β4lnNR + β5lnL + β6Dmo + β7Dabt + β8Dait + β9Dgbt + β10Drt

+ β11Dst + µ2 [6]

Estimation Technique

Since price is endogenously determined by the model a ‘simultaneous equation bias’ is likely

to arise.14 Hausman test was used to check for if the simultaneity problem existed. The results showed

that both price variables, as well as the difference variable created a simultaneous equation bias under

OLS.

14 Simultaneous equations bias: Inconsistency of OLS estimators in a simultaneous equations system.

14

The 2SLS technique was used to eliminate this bias. In the first stage, instrumental variables

were created for all the three variables, by separately regressing them on all pre-determined variables

as well as the three block prices. In the second stage, OLS was used to estimate equations 3, 4, 5, and

6, replacing AP, MP, and DV by their respective instrumental variables.

5. DATA DESCRIPTION

Three sets of information were collected: household level data about consumption and

household characteristics, tariff structures, and weather.

Household-level information: Primary data for the study was obtained from WASA, Lahore. The data

set contained the following information about 156 households:

Location: Information about the town in which the connection is located.

Consumption: Water consumption in gallons from January 2004 to December 2006. Meter reading is

carried out by WASA with a back-lag of two months: that is, for the billing cycle March-April, the

meter is read in the last week of February. For the subsequent cycle, May-June, the meter is read in

the last week of April. In this way, the March-April bill is based on January-February consumption,

and May-June bill is based on March-April consumption.

The variable used in this study is the consumption during the billing period, the previous

month’s meter-reading, rather than the meter-reading recorded in the billing period. Eighteen meter

readings were recorded per household.

Plot size: The plot size in marlas.

Property Value: Property value in Rupees.

Number of Residents: The number of residents living in a house.

Meter-ownership: Information about whether the meter is owned by WASA or the consumer.

Tariff Structure: The tariff structures for domestic metered connections since January 2004 were

provided by WASA.

Table 6.1: Descriptive Statistics

Variable Mean Std. Deviation Minimum Maximum

Consumption (1,000 gallons per billing period)

17.9 14.9 2.00 155.12

Average Price (Rs.) 17.7 2.4 9.2 26.0

Marginal PriceP (Rs.) 21.8 3.5 9.2 27.3

15

Household Size 6.3 1.8 2.0 11.0

Plot Size (marlas) 8.7 13.9 1 160

Property Value (Rs. lacs) 29.4 35.5 2.0 250.0

Water Expenditure (Rs) 340.1 373.6 18.4 4033.9

Temperature (Celsius) 30.4 6.2 18.5 38.6

Weather: Information about the weather in Lahore was provided by the Pakistan Meteorological

Department. The data-set contained the maximum average monthly temperature recorded in degree

Celsius from January 2004 to December 2006. The temperature values were averaged over two-month

periods in line with WASA billing cycles.

6. RESULTS

The estimated demand model explains 60 percent of the variation in water consumption and is

statistically significant.15 Price, plot size, household size, meter ownership, past water consumption,

and the difference variable that captures the income effect have been found to significantly influence

the household demand for water. The effect of weather is moderate while that of wealth, as measured

by property value, is negligible.

Whether water demand is more responsive to marginal or average price could not be

established from the data. Therefore, both AP and MP models were estimated. The AP models

showed a significant positive price-demand relationship, which is contrary to economic theory. Such

contradiction is often attributed to statistical errors, such as multicollinearity or weak instrumental

variables. However, no such error could be identified. Regardless of the reason for the discrepancy,

the AP models were not suited for drawing inferences. All following discussions and conclusions are

based on MP models.

The results show that the marginal price of water has a significant impact on water use: a

percentage increase in the marginal price brings a 2.33 percent decline in average water use. A 10

percent increase in the marginal price will bring down average water consumption by 23 percent.

Table 7.1: Model Summary

Statistic Value

15 Though this R2 value is not very high, it is acceptable for panel data models with a large cross-sectional element. Cross-sectional data has high variance due to the diversity of the sampling units, and it is therefore difficult to fit a regression line that explains all the variation.

16

R2 0.600

F-Statistic 304.163*

Variable Coefficient t-ratio

Intercept 0.54 (1.725)**

Meter-ownership (Dummy) -0.06 (-2.180)*

Difference Variable 2.89 (8.292)*

Plot Size 0.04 (3.297)*

Marginal Price -2.33 (-5.473)*

Household Size -0.14 (-4.380)*

Lagged Consumption 0.38 (10.402)*

Temperature 0.05 (1.457)***

Property Value -0.01 (-1.066)

Location Dummies

Aziz Bhatti Town -0.064 (-2.510)*

Allama Iqbal Town -0.128 (-3.901)*

Ganj Baksh Town 0.023 (0.942)

Ravi Town -0.070 (-2.751)*

Shalimar Town 0.010 (0.442)

* = significant at 5% level. ** = significant at 10% level. *** = significant at 15% level.

The size of the plot, which presumably affects water use because of higher cleaning and lawn

watering requirements, has a small but significant impact on domestic water use. For every percentage

increase in the size of the plot, water consumption goes up by 0.041 percent. However, this

information does not tell us about the individual impact of lawn size and house size and water use

separately attributed to each variable. Possibly, water demand for one of the purposes may be more

elastic than the other.

The relationship of water demand and the number of residents living in a household was

found negative, contrary to what was expected. As household size increases by one percent,

consumption falls by 0.14 percent. Apparently small, this connection is statistically significant and

also has important implications for policymaking.

Commonly, low income families are larger as compared to high income families. If large

families consume less than small families, then there must be some underlying income connection.

17

We may infer that low-income households consume less water per capita. However, more information

is needed to substantiate this inference.

The difference variable has an effect of increasing demand by 2.89 percent with every

percentage increase. In essence, the difference variable reflects the apparent increase in income from

having paid less for some units of a commodity under increasing block rates – what is known in

microeconomic theory as the ‘income effect’. As expected, this income effect is positive.16 This

means that as more discriminatory tariffs are charged for consumption in successive blocks, the

income effect induces consumers to increase water expenditure. However, as a policy variable, DV

has limited value as it cannot be directly manipulated.

Weather has a modest influence on domestic water consumption. A one percent rise in

temperature increases water demand by 0.05 percent. This means that water demand is higher in

summer than in winter.17

The intercept term reflects the average water consumption for Nishtar Town when all other

variables are set to zero is 860.6 gallons per household per month if the meter is owned by WASA

and 813.7 gallons if the meter is owned by the consumer. That means on average meter-ownership

reduces water consumption by 46 gallons per month.

According to the model, average household consumption of Nishtar Towm is statistically

similar to that of Ganj Baksh and Shalimar towns. Consumption in Aziz Bhatti and Ravi towns differ

from Nishtar Town by about 50 gallons per month, while Allama Iqbal Town households consume

around 100 gallons less water than households in Nishtar Town.

7. POLICY RECOMMENDATIONS

The results of this study show that there is considerable potential to use demand management

pricing policies. Therefore, a calculated re-arrangement of the tariff structure can be instrumental in

encouraging efficient use of water, improving revenues, and ensuring universal lifeline supply.

Conservation: Tariffs can be effectively used to induce efficient water use. At the current rate,

household water use is extremely elastic to changes in prices. A small ten percent increase in the

overall tariffs will cause an average household to conserve 2,000 gallons of water per month. A 30

16 Note that the difference variable was originally negative. Natural log of the absolute values was included in the computations. However, the positive relationship of the variable consumption was also found in linear and log-linear models (See Appendix). We have therefore interpreted the coefficient as positive. 17 The weather coefficient was significant at 15 percent level. If one adheres to the conventional five percent level, then the coefficient is statistically zero.

18

percent increase will bring down water use to 2,700 gallons – close to the level of lifeline

consumption. (See table 8.1).18

A drawback of these computations is that they are based on average consumption patterns,

and do not give any indication of how the tariff increase will affect households with varying

socioeconomic settings, particularly income levels. Taking into account these considerations may

make an overall equal increase in tariffs undesirable.

If an overall increase is not desirable, effective conservation can still be achieved by

increasing the tariff of the final block – monthly consumption of over 20,000 gallons. Currently,

average consumption in this block is over 1,800 gallons. A 30 percent increase in the price of this

block will induce average households to cut down over 1,000 gallons of monthly consumption. A

larger increase can virtually eliminate per household water consumption beyond 20,000 gallons. (See

table 8.2)

Table 8.1: Impact of Price Increase on Monthly Water Use

Tariff Increase Current

10% 20% 30%

Avg. MP (Rs.) 21.8 23.98 26.16 28.34

Avg. Monthly Use (gallons) 8,900 6,900 4,800 2,700

Table 8.2: Impact of Price Increase on Water Use in Block 3

Tariff Increase Current

30% 40% 50%

Block 3 Price (Rs.) 27.30 35.49 38.22 40.95 Avg. Monthly Use in Block 3 (gallons) 1,820 545 125 -300

Cost Recovery: Typically, a tariff structure that is meant for budget balancing consists of a large fixed

part coupled with declining block rates (Montginoul, 2006). Declining block rates promote

consumption leading to higher revenues and faster cost recovery on investments, while the fixed part

increases the certainty of revenue projections.

Though leading to immediate cost recovery, decreasing block rates are not compatible with

promoting efficient water use, and assuming that conservation is more important of the two

objectives, declining blocks rate are not advised. Moreover, though increasing block rates do not

18 Daily per capita lifeline is 13.21 gallons. The average household size is taken to be 7 (see Chapter 6 for descriptive statistics on household size). Lifeline water supply is calculated as: 13.21*7*30= 2,800 gpm.

19

generate equally high revenue, they do lead to a cut-down in O&M costs while at the same time

easing the need for new investments to increase supply capacity. Therefore, the management will not

be ill-advised to continue following increasing block rates.

The non-volumetric part can also be effectively used to increase revenues. Currently the

monthly non-volumetric charge consists of a connection fee of Rs.3 and meter charge of Rs.12, that

is, all households indiscriminately pay fixed charges of Rs.15 per month. The connection fee can be

increased, uniformly or discriminately, varying increasingly with the household’s ability to pay. The

best measure of ability-to-pay is household income, but it is difficult to oblige people to provide

correct information about their incomes. Alternatively, non-volumetric charges can be based on

property value, ARV, or plot size.

Allocation: The effect of a pricing policy may not be uniformly felt across all income groups. Past

studies have found that indiscriminate tariff increase causes greater reduction in water use in lower

income households than in higher income households (Agthe and Billings, 1987, Renwick and

Archibald, 1998). If this disparity gets large enough, it can threaten the lifeline supply of vulnerable

households.

As a rule, water expenditure should not exceed five percent of the household income (WASA,

2005). The lifeline water requirement for a seven-member household (2, 800 gallons per month) costs

Rs.36.06. If the monthly income is Rs.5, 000, this expenditure amounts to a bare 0.72 percent. If the

tariff in the first block is increased by 50 percent, the lifeline consumption cost will amount to only

1.1 percent of the expenditures of a household with above-mentioned features.

Table#: Impact of Block 1 Price Increase on Lifeline Consumption

Tariff Increase Current

10% 20% 50%

Price in Block 1 (Rs.) 12.88 14.17 15.46 19.32 Cost of Avg. Household Lifeline Supply (Rs.) 36.06 39.68 43.29 54.10

Summing the discussion on pricing policies, one can see that there is significant room for

increasing tariffs to achieve conservation and cost recovery objectives without risking lifeline supply.

20

APPENDIX

I. Sampling Framework The sampling frame consists of all domestic households that were metered prior to the beginning of the

study period, i.e., January 2004. The correct number of such households was not available, but an estimate was

made using the available information. Currently, WASA provides 480,000 domestic water connections of which

only 30 percent are metered. Therefore, the sampling frame comprises of 144,000 sampling units.

Sample Size: WASA Lahore covers 90 percent of the households under its official jurisdiction while the

remaining 10 percent being provided by other (private) sources (WASA, 2007). In a randomly selected sample,

the probability of a household having a WASA connection is, therefore, 90 percent. The sample size (n) was

computed in the following way:

n = PQ/SE(P) = 150

where P is the percentage of households having a WASA connection and Q the percentage of households

provided by others. If a six percent standard error is acceptable, a sample of 150 households is likely to contain

90 percent households with a WASA connection. For this study, the sample size was increased to 156

households in order to achieve a better allocation of sampling units.

Allocation of Sampling Units: The sampling frame was stratified on the basis of WASA’s administrative

division. As discussed earlier, the WASA area is divided into six towns which are further divided into O&M

units. Information about the exact population of these divisions was not available; therefore sampling units were

allocated proportionately according to a rough estimate of the population density under each subdivision.

Finally, the units were selected randomly using a random number table.

Time Frame: It was intended that the time frame should include at least one tariff change, in order to better

capture the effect of price on consumption. However, as mentioned earlier, tariffs have remained unchanged

since May 2004. Before that, tariffs were last increased in January 1998. In order to include both (or more) tariff

changes, the initial timing would have to be set at (or before) 1998. However, it was not practical to stretch the

study period that long.

As we go back in time, the number of metered connections, and hence the sampling frame, becomes

smaller. Moreover, socio-economic and demographic characteristics, such as household size and income, would

have significantly changed for each sampling unit over such a long time frame. Variations in these variables are

difficult to measure and would have created considerable errors in the data.

Because of these considerations, the time frame of the study was limited to the last three years, 2004 to

2006, to include one tariff change while allowing for consistency of socioeconomic and demographic variables

of each unit over this period.

21

Table 5.1: Allocation of Sampling Units

Town Sampling Units O&M Sub-division Sampling Units

Allama Iqbal Town 5

Ichhra 5

Johar Town 5

Sabzazar 5

Allama Iqbal Town 25

Samanabad 5

Taj Pura 13 Aziz Bhatti Town 25

Mustafabad 12

Gulberg 5

Krishan Nagar 5

Mozang 5

Ravi Road 5

Ganj Baksh Town 25

Shimla Hill 5

Garden Town 6

Green Town 6

Industrial Area 7

Nishtar Town 26

Township 7

City 5

Data Nagar 5

Farkhabad 5

Misri Shah 5

Shadbagh 5

Ravi Town 44 30

Shahdara 5

Baghbanpura 13 Shalimar Town 25

Mughalpura 12

Total 156 156

II. Hausman Test Ho: Coefficient of residuals is statistically zero; no simultaneity problem

H1: Coefficient of residuals is not statistically zero; simultaneity problem exists Level of Significance: α = 0.05 Test Statistic: t-ratio

Computations and Conclusions:

22

Regression 1 Regression 2 Variable

Dependent Independent Dependent Independent

Residual

Coefficient

Conclusion

1. AP AP Q_lag, P3, Dst, T,

W, Dmo, Drt, L,

Dabt, NR, Dgbt,

Dait

Q RES, PRE, T, Dabt,

Dmo, Dgbt, NR, L,

Dst, Dait, W, Q_lag,

Drt

5.953

(76.009)

Reject Ho.

Conclude that

simultaneity

problem exists.

2. MP MP Q_lag, P3, Dst, T,

W, Dmo, Drt, L,

Dabt, NR, Dgbt,

Dait

Q RES, PRE, Dst, T,

Dmo, Dgbt, L, Dabt,

NR, Dait, W, Q_lag,

Drt

1.727

(28.804)

Reject Ho.

Conclude that

simultaneity

problem exists.

3. DV DV Q_lag, P3, Dst, T,

W, Dmo, Drt, L,

Dabt, NR, Dgbt,

Dait

Q RES, PRE, Dst, T,

Dabt, P2, Dmo, Dgbt,

L, NR, W, Drt, Dait

-0.077

(- 30.451)

Reject Ho.

Conclude that

simultaneity

problem exists.

Details 1. AP Regression 1 Residual Statistics

Min Max Mean SD N PRE 11.7150 29.1553 17.9508 1.80555 2652 RES -8.19803 8.07184 .00000 1.13266 2652

Regression 2 Coefficients

Predictors B SE t Sig. (Constant) -3.347 1.420 -2.357 .018 T .101 .015 6.752 .000 NR .205 .255 .803 .422 Dmo -1.605 .365 -4.393 .000 L .051 .007 7.005 .000 W 6.39E-009 .000 .191 .848 Dgbt 1.197 .378 3.166 .002 Drt .003 .742 .003 .997 Dst .617 .413 1.495 .135 Dabt .503 1.086 .464 .643 Dait .496 .414 1.199 .230 Q_lag .791 .009 90.491 .000 PRE .201 .071 2.841 .005 RES 5.953 .078 76.009 .000

2. MP Regression 1 Residuals Statistics

Min Max Mean SD N PRE 13.8864 33.3924 22.1529 2.22821 2652 RES -13.73344 6.95036 .00000 2.30573 2652

Regression 2 Coefficients

Predictors B SE t Sig. (Constant) -3.110 2.113 -1.472 .141

23

T .101 .023 4.324 .000 NR .177 .399 .444 .657 Dmo -1.628 .570 -2.856 .004 L .051 .011 4.509 .000 W 1.07E-009 .000 .021 .984 Dgbt 1.182 .589 2.007 .045 Drt -.059 1.157 -.051 .960 Dst .590 .644 .916 .360 Dabt .432 1.692 .255 .799 Dait .358 .656 .546 .585 Q_lag .793 .013 62.615 .000 PRE .160 .088 1.823 .068 RES 1.727 .060 28.804 .000

3. DV Regression 1 Residuals Statistics

Min Max Mean SD N PRE -351.0884 -2.2660 -78.4510 39.70316 2652 RES -175.04536 186.92253 .00000 53.85123 2652

Regression 2 Coefficients

Predictors B SE t Sig. (Constant) 53.504 2.567 20.845 .000 T -.148 .023 -6.287 .000 NR -2.745 .395 -6.956 .000 Dmo -4.854 .557 -8.720 .000 L -.028 .011 -2.457 .014 W -3.68E-007 .000 -7.136 .000 P2 -1.742 .101 -17.265 .000 Dgbt 2.117 .580 3.650 .000 Drt -5.569 1.143 -4.873 .000 Dst -3.566 .636 -5.606 .000 Dabt -5.237 1.670 -3.136 .002 Dait -13.675 .663 -20.616 .000 PRE -.383 .005 -79.960 .000 RES -.077 .003 -30.451 .000

III. Price Perception Test Ho: k=1; consumers respond only to average price. H1: k≠1; consumers do not respond only to average price. Level of Significance: α= 0.05 Test Statistic: t* = (β1 – β2)/ S.Dβ1-β2 Critical Region: t*> 2 19

19 The degrees of freedom of the β population is unknown, therefore the rule of thumb is used.

24

Computations:

Regression

Dependent Variable: lnQ

Independent Variables: lnQ_lag, Dgbt, lnT, ln_MP_IV, ln(AP/MP), Dmo, Dabt, lnNR, Dst, lnL, Dait, , Drt, lnW

Variable Tested Coefficient S.E. Variance

MP20 0.756

(3.514) 0.215 0.046225

AP/MP -0.576

(-6.677) 0.086 0.000740

β1 – β2 = 0.756 – (-0.576) = 1.332 S.Dβ1-β2 = (Varβ1 + Varβ2 – 2Covβ1, β2)½ = 0.230913 21 t* = 1.332/0.230913 t* = 5.768

Conclusion: t* > 2 ; reject Ho. Conclude that k≠1. Consumers do not respond only to average price. [Note that a similar test cannot be used for the hypothesis that consumers respond only to MP (k=0). If the alternate hypothesis (k≠0) is false, it would not imply that consumers do not respond to AP. Average price has been introduced in the regression equation as a ratio of AP and MP. If the alternate hypothesis is false, we can only conclude that consumers do not respond to this ratio: a conclusion with little practical significance.]

Details Coefficients

B SE t Sig.

(Constant) -.588 .287 -2.050 .040

Dmo -.031 .025 -1.235 .217

Dgbt .042 .025 1.719 .086

Drt .002 .024 .090 .929

Dst .047 .023 2.085 .037

Dabt .010 .023 .407 .684

Dait .036 .024 1.457 .145

lnT .156 .030 5.278 .000

lnNR .020 .024 .850 .396

lnW -.010 .011 -.907 .365

lnL .087 .012 7.446 .000

ln_MP_IV .756 .215 3.514 .000

lnAP_MP -.576 .086 -6.677 .000

20 The positive sign of MP coefficient is possibly due to multicollinearity between MP and AP/MP variables. 21 Covβ1, β2 = -rσ2/[(1-r2)√(Σx1

2Σx22)]

25

ln_Q_lag .649 .016 41.784 .000

IV. Summary of Estimated Models AP Models MP Models

Statistic Semi-log Double-log Semi-log Double-log

R2 0.569 0.590 0.569 0.600

F-Statistic 290.669 315.841 290.669 304.163

D-W Statistic 1.911 2.315 1.911 2.289

Variables

Constant 1.658

(17.415)*

-0.494

(-1.826)**

2.331

(9.550)*

.543

(1.725)**

AP 0.010

(1.901)**

.255

(3.431)* -- --

Dmo

0.017

(0.644)

-.026

(-1.006)

-.069

(-2.643)*

-.056

(-2.180)*

DV --

--

1.292

(40.825)*

2.886

(8.292)*

Dabt 0.082

(3.504)*

0.033

(1.411)

-0.128

(-5.379)*

-.064

(-2.510)*

Dait 0.170

(6.755)*

0.066

(2.721)*

-0.251

(-9.373)*

-0.128

(-3.901)*

Dgbt -0.036

(-1.504)***

0.049

(1.986)*

-0.023

(-0.932)

.023

(0.942)

Drt 0.069

(2.903)*

0.018

(0.760)

-0.126

(-5.247)*

-0.070

(-2.751)*

Dst 0.122

(5.263)*

0.054

(2.365)*

-0.010

(-0.431)

0.010

(0.442)

L 0.002

(4.218)*

0.079

(6.697)*

-3.69E-005

(-0.069)

0.041

(3.297)*

MP --

--

-1.428

(-13.914)*

-2.327

(-5.473)*

NR 0.034

(8.451)*

0.044

(1.876)**

-0.042

(-9.374)*

-.139

(-4.380)*

Qt-1 0.022

(35.285)*

0.648

(39.479)*

--

(excluded)

0.380

(10.402)*

T 0.005

(5.049)*

0.155

(5.206)*

-.001

(-1.037)

.047

(1.457)***

W 5.57E-009

(2.331)*

-0.003

(-0.270)

-5.33E-009

(-2.222)*

-0.012

(-1.066)

* = significant at 5% level. ** = significant at 10% level. *** = significant at 15% level.

26

References

Agthe, Donald E. and Billings, R. Bruce (1987). Equity, Price Elasticity, and Household Income under Increasing Block Rates for Water. American Journal of Economics and Sociology, 46(3), July 1987, pp. 273-286.

Arbues, Fernando and Villanua, Inmaculada (2006). Potential for Pricing Policies in Water Resource Management: Estimation of Urban Residential Water Demand in Zaragoza, Spain. Urban Studies, 43(13) December 2006, pp. 2421-2442.

Bailey, Stephen J. (2002). Public Sector Economics, 2nd ed. Palgrave, New York.

Batchelor, R.A. (1975). Household Technology and the Domestic Demand for Water. Land Economics, 51(3), 1975, pp. 208-223.

Bathla, Seema (1999). Water Resource Potential in Northern India: Constraints and Analysis of Price and Non-Price Solutions. Environment, Development and Sustainability 1, Kluwer Academic Publishers, Netherlands, pp. 105-121.

Becker, Nir (2003). Water Pricing – Its Effects and Proposed Reforms: The Israeli Experience. International Journal of Public Administration, 26(3), pp. 247-263.

Boiteux, M. (1971). On the Management of Public Monopolies Subject to Budgetary Constraints. Journal of Economic Theory, 3, pp. 219-240. Reprinted in: The Economics of Public Utilities, Vol. 1. Edited by: Ray Rees (2006), Edward Elgar Publishing Limited, UK.

Charney, Alberta H. and Woodard, Gary C. (1984). A Test of Consumer Demand Responses to Water Prices: Comment. Land Economics, 60(4), November 1984, pp. 414-416.

Chicoine, David L. and Ramamurthy, Ganpathi (1986). Evidence on the Specification of Price in the study of Domestic Water Demand. Land Economics, 62(1), February 1986, pp. 26-32.

De Azevedo, Luiz Gabriel T. and Baltar, Alexandre M. (2005). Water Pricing Reforms: Issues and Challenges of Implementation. Water Resources Development, 21(1), pp.19-29.

Foster, Henry S. and Beattie, Bruce R. (1979). Urban Residential Demand for Water in the United States. Land Economics, 55(1), February 1979, pp. 43-58.

Foster, Henry S. and Beattie, Bruce R. (1981). Urban Residential Demand for Water in the United States: Reply. Land Economics, 57(2), May 1981, pp. 257-265.

Foster, Henry S. and Beattie, Bruce R. (1981). On the Specification of Price in Studies of Consumer Demand under Block Price Scheduling. Land Economics, 57(4), November 1981, pp. 624-629.

Gleick, Peter H. (1998). Water in Crisis: Paths to Sustainable Water Use. Ecological Applications, 8(3), August 1998, pp. 571-579.

Gujarati, Damodar N. (2003). Basic Econometrics (4th ed.). McGraw-Hill Ltd., New York.

Hanemann, Michael W. (2003). Willingness to Pay and Willingness to Accept: How Much Can They Differ? Reply The American Economic Review, 93(1), March 2003, p. 464.

Hansen, Lars Garn (1996). Water and Energy Price Impacts on Residential Water Demand in Copenhagen. Land Economics, 72(1), February 1996, pp. 66-79.

Headley, Charles J. (1963). The Relation of Family Income and Use of Water for Residential and Commercial Purposes in the San Francisco-Oakland Metropolitan Area. Land Economics, 39(4), November 1963, pp. 441-449.

Hewitt, Julie A. and Hanemann, Michael A. (1995). A Discrete/Continuous Choice Approach to Residential Water Demand under Block Rate Pricing. Land Economics, 71(2), May 1995, pp. 173-192.

International Finance Corporation (2005). Lahore Water and Sewerage Study: Technical and Environmental Consultant – Due Diligence Report.

International Finance Corporation. Lahore Water and Sewerage Services: Improving Service Delivery – Transaction Options and Structuring Report 2005.

Koutsoyiannis, A. (1972). Theory of Econometrics (2nd ed.). The Macmillan Press Ltd., London.

27

Lahore Development Authority Act 1975. Source: http://www.lda.gop.pk/act2.html

Lundqvist, Jan, Appasamy, Paul, and Nelliyat, Prakash (2003). Dimensions and Approaches for Third World City Water Security. Philosophical Transactions: Biological Sciences, 358(1440), Freshwater and Welfare Fragility: Syndromes, Vulnerabilities, and Challenges, December 29, 2003, pp. 1985-1996.

Montginoul, Marielle (2006). Analysing the Diversity of Water Pricing Structures: The Case of France. Water Resour Manage, 21, pp. 861-871.

Nauges, Celine and Thomas, Alban (2000). Privately Operated Water Utilities, Municipal Price Negotiation, and Estimation of Residential Water Demand: The Case of France. Land Economics, 76(1), February 2000, pp. 68-85.

Nieswiadomy, Michael L. and Molina, David J. (1989). Comparing Residential Water Demand Estimates under Decreasing and Increasing Block Rates Using Household Data. Land Economics, 65(3), August 1989, pp. 280-289.

Nieswiadomy, Michael L. and Molina, David J. (1991). A Note on Price Perception in Water Demand Models. Land Economics, 67(3), August 1991, pp. 352-359.

Opaluch, James J. (1982). Urban Residential Demand for Water in the United States: Further Discussion. Land Economics, 58(2), May 1982, pp. 225-227.

Opaluch, James J. (1984). A Test of Consumer Demand Response to Water Prices: Reply. Land Economics, 60 (4) November 1984, pp. 417-421.

Punjab, Govt. of, Federal Bureau of Statistics, and United Nations Children’s Fund (2004). District-Based Multiple Indicators Cluster Survey 2003-04.

Punjab, Govt. of, Bureau of Statistics (2005). 2005 Statistical Pocket Book of the Punjab.

Punjab, Govt. of, Bureau of Statistics (2006). Punjab Development Statistics 2006.

Renwick, Mary E. and Archibald, Sandra O. (1998). Demand Side Management Policies for Residential Water Use: Who Bears the Conservation Burden? Land Economics, 74(3), August 1998, pp. 343-359.

Sukhatme, Pandurang V. (1958). Sampling Theory of Surveys with Applications. Indian Society of Agricultural Statistics, New Delhi and Iowa State College Press, Iowa.

Terza, Joseph, V. and Welch, W.P. (1982). Estimating Demand under Block Rates: Electricity and Water. Land Economics, 58(2), May 1982, pp. 181-188.

World Bank, Infrastructure and Energy Department, South Asia Region and The Urban Unit, Lahore, 2006. Property Taxes in the Punjab, Pakistan.

Water and Sewerage Agency, Lahore Development Authority (2007). Budget 2007-2008. City District Govt., Lahore.

Wong, S.T. (1972). A Model on Municipal Water Demand: A Case Study of Northern Illinois. Land Economics, 48(1), 1972, pp. 34-44.

Yates, Frank (1965). Sampling Methods for Censuses and Surveys. Charles Griffin and Company Ltd., London.

Online Sources:

EDC News, Pakistan: The Death of River Ravi: http://www.edcnews.se/Cases/PakRaviriver.html

International Water Management Institute: www.iwmi.cgiar.org

Lahore, City District Govt. http://www.lahore.gov.pk

Oczkowski, Eddie (2003). Two-Stage Least Squares (2SLS) and Structural Equation Models (SEM). Online: csusap.csu.edu.au/~eoczkows/2slsnotes.pdf

Water and Sanitation Agency, Lahore Development Authority: http://www.lda.gop.pk/lda_wasa

United Nations Secretariat, Population Division, Department of Economic and Social Affairs. World Population Prospects: The 2004 Revision and World Urbanization Prospects: The 2005 Revision: esa.un.org/unup

World Gazetteer (2007). http://www.world-gazetteer.com