Professional Documents
Culture Documents
AND DEVELOPMENT
Copyright 2022
ENVIRONMENTAL GOOD VALUATION: THE CASE
OF DRINKING WATER IN TUNISIA
Introduction
*Ali Bouchrika, Assistant Professor at the Institut Superieure de Gestion de Gabes (University of
Gabes), received his Ph.D. in economics from the University of Tunis El Manar. His research interests
include the economics of natural resources, economic development, economic justice, macroeconomics,
and growth in transitioning economies. The author has published more than 20 papers in various
journals such as Empirical Economics, International Journal of Sustainable Economies Management,
Journal of Reviews on Global Economics, Journal of Management Research, and Journal of
Environmental Sciences & Natural Resources, as a sampling.
Fakhri Issaoui, Full Professor of Economics and Management in the College of Business- King
Khalid University Saudi Arabia Asir – Abha (P.O. Box: 960 - Postal Code: 61421). The author is an
affiliate in the research laboratory of Prospective, Strategies et Developpement Durable PS2D
(University of Tunis El Manar), he received his Ph.D. in economics from the University of Tunis El
Manar. His research interests include public and labor economics, economic development, economic
justice, macroeconomics, and growth in transitioning economies. The author has published more than
50 articles in journals such as Panoeconomicus, The Journal of Energy and Development, Journal
of Cyber Warfare and Terrorism, Journal of the Knowledge Economy, International Journal of
Computational Economics and Econometrics, African Sociological Review, International Journal of
Green Computing, Economic Research Guardian, and International Journal of Sustainable Economies
Management, as a sampling. (continued)
Slah Slimani is an Assistant Professor in the College of Business - King Khalid University Saudi
Arabia Asir – Abha (P.O. Box : 960 - Postal Code : 61421). He received his Ph.D. in economics from
the University of Tunis El Manar. His research topics include economic policy analysis and growth,
macroeconomics, behavioral finance, health economics, and environmental economics. The author has
published several articles in journals such as the International Journal of Sustainable Economies
Management, International Journal of Sustainable Economy, International Journal of Computational
Economics and Econometrics, International Journal of Cyber Behaviour, Psychology and Learning,
and International Journal of Social Ecology and Sustainable Development.
DRINKING WATER: ENVIRONMENTAL GOOD VALUATION 179
have a form of valuation that measures both the environment and environmental
goods. This seems strategic to avoid any type of possible natural resource overuse
and waste and to strive to be on the path of sustainable development.
From an economic context, the environment appears to be one of the common
interfaces between consumers and producers. For environmental economics it is nec-
essary to provide forms of valuation for environmental goods and natural resources
through a variety of methods. Hence, economic theory has developed a methods
panel for evaluating consumer preferences in the case of market absence. Among
these methods, we can cite the contingent valuation method (CVM), which measures
all environmental goods values. Thus, CVM is an ex-ante evaluation method based
on the realization of a survey by which we seek to assess the amount that the con-
sumer would be willing to pay.
Although CVM has been widely used in cost-benefit analysis and environmen-
tal externality assessments for decades, it has been subject to a number of
criticisms. The main concern is that the validity of the results can be biased due to
a number of errors including overestimation of WTP (willingness to pay) since it
is hypothetical, the survey can be structured incorrectly, respondents can provide
untrue answers, and non-response bias. Despite the challenges of this method in
obtaining unbiased information on an economic agent’s preferences through sur-
veys, it still is a vital estimation tool for a public good. This information can then
be used in a cost-benefit analysis for a natural resource or environmental good,
which assesses the impacts of key government projects, policies, and public-sector
priorities. As B. Desaigues and P. Point highlight “it is necessary to attempt a mea-
sure, even if it is not perfect, of all the advantages of natural goods protection.”1
Indeed, the authors propose the careful evaluation of key factors and conditions to
allow for a more accurate valuation by defining and understanding the key varia-
bles of the population, the natural resource under study, and additional information
on households’ socio-economic characteristics to provide more robust results.
To determine the optimal amount of a public good, economic theory offers a
number of models. Based on individual utilities, two methods are used to calculate
this amount, namely, the indirect method is based on actions and the direct method
based on individual statements. If the indirect method consists of recording the util-
ities from the observed behavior, the direct method consists of asking the agents
by a survey. This last method can be implemented if the actions are difficult to
observe. Therefore, the direct valuation of an environmental good, widely applied
in environmental and health fields, is made in this case of contingent valuation.
Thus, the major objective of this study is to present a methodology to value an
environmental good, which is drinking water, in Tunisia. To do this, our paper is
organized as follows. The next section will deal with the economic valuation of an
environmental good followed by an econometric formalization of WTP models.
We will then present the empirical methodology followed by an estimation of the
WTP for the environment good of drinking water in Tunisia using the Logit and
180 THE JOURNAL OF ENERGY AND DEVELOPMENT
Tobit qualitative models. Last, we will present our conclusions and their implica-
tions for environmental good valuation within the Tunisian context.
Let Sbi be the sum of these surpluses for consumer i who benefits from an envi-
ronmental good.
X
n
Sbi 5 Sih
h:1
Let us assign to each consumer i a weighting coefficient ai, which measures the
greater or lesser importance given by the consumer to an environmental good.4 Thus,
ai is called consumer’s preference value for an environmental good. The weighted
collective surplus is then:
X
m
b5
W /i b
S 1 p where p is the monopoly profit:
i51
X
m X
n X
n
b 5
W ai Shi 1 Ph ðYh Þ2CT ðY1 ::: Yn Þ
i51 h51 h51
After choosing the scenario and the valuation question, we proceed to a series of
pre-tests which will have to validate the methodological choice and the valuation
question. In addition to this valuation question, the survey must present questions
relating to individuals’ socio-economic characteristics and questions relating to the
uses of the environmental good. Thus, from these different questions, explanatory
variables relating to WTP were created.
The objective of econometric analysis is to estimate a WTP model taking into
account explanatory factors from the survey. In the case of an open question,
econometric specification is a model in which the explained variable, consent to
pay (WTP), is continuous. In addition, in the case of a referendum technique, the
obtained econometric specification is a binary choice model.5 Thus, the dependent
variable (WTP) is qualitative with two modalities (yes and no). Therefore, econo-
metric analysis in contingent valuation method (CVM) can deal with many
182 THE JOURNAL OF ENERGY AND DEVELOPMENT
problems, such as zero or extreme values and the choice of the model functional
form. The specific econometric model of WTP must be compatible with economic
theorem hypothesis. D. Fadden and G. Leonard proposed a model which pro-
ceeds to a Box-Cox type transformation.6 In addition, B. Desaigues and P. Point
proposed a systematic bias correction (lack of experience of respondents) by a
Box-Cox transformation.7
Thus, the evaluation model is written as the following form:
WTP 5 X b 1 «
WTPðlÞ 5 X ðlÞ b 1 «
This model is estimated using the two-step maximum likelihood method. The
first step consists of determining a l estimator from an iterative algorithm. The sec-
ond step consists of estimating the b vector of parameters from the l estimation.
On the other hand, according to D. Fadden and G. Leonard, for an individual i, the
WTP model is written as follows:8
GðCTPi , Yi , aÞ 5 ZðXi , «i , gÞ
with CTPi denoting the consent to pay of the individual i; Yi is the income of individual
i; (a, gÞ is the parameter vector; and «i is the error term associated with individual i.
Additionally, the G (.) function is defined by:
8
>
> CTPi if a 5 0
> 12a
< Yi 2ðYi 2CTPi Þ12a
GðCTPi , Yi , aÞ 5 if aÞ0
>
> 12a
>
: 2log 12CTP if a 5 1
Yi
fZ ðGðCTPi , Yi , aÞ, gÞ
fCTPi ðCTPi , Yi , a, gÞ 5
ðYi 2CTPi Þa
This density is used to estimate the WTP model by the maximum likelihood
method, making assumptions about the error term law based on this density.
In the case of a continuous discrete choice question, the WTP distribution is
processed using a skillful censored model. Thus, if individual i declares a null
WTP, then the distribution function will be Fz(G(0, Yi, a), g) and if individual i
accepts an offer M, the likelihood contribution is 12 Fz(G(M, Yi, a), g).
The Logit Model: A dichotomous model is a statistical model in which the
dependent variable can only take two modalities. It is a question of explaining the
occurrence or not of an event. Therefore, the objective of dichotomous models is
to explain the occurrence of the considered event based on a characteristics number
related to the individuals who form the sample. In this type of model, we try to
specify the probability of the event’s occurrence. Indeed, the Logit model is one of
these types of models, using a 0 and 1 coding for the explained variable. This cod-
ing associated with the event occurrence and its probability is conditioned by the
exogenous variables. Thus, the models are constructed as follows:
1
Pi 5 P Yi 5 5 Fðxi bÞ
xi
where F (.) is the distribution function of the logistic law. Yi is the household WTP
and represents the explanatory variables. The maximum likelihood estimator
parameter in a dichotomous model is obtained by maximizing the likelihood func-
tion LogL(Yi , b) with:
natural asset. Thus, we obtain a positive or zero WTP. Indeed, to model this evalu-
ation type, J. Tobin proposed a model composed of two parts;9 a continuous part
corresponding to a linear regression model and a discrete part related to the censor-
ship point assumed to be equal to zero.
In addition, the model structure is then:
CTPi 5 Xi b 1 «i .0
CTPi 5
0 if not
The Tobit model is also a censored regression model. It provides a better estima-
tor of the explained variable, but it does not solve the hypothetical bias problem.
Well-Being Measurement: L. J udez et al. suggested that the well-being mea-
surement can be obtained by the truncated mean, with the following expression:10
ðT
mT 5 ð12F ðbÞÞdb
0
with b being the value offered; F(b) is the WTP distribution function; and T is the
truncation point.
This well-being measurement has been used in contingent valuation work with
dichotomous choices.11 The research of J. Duffield and D. Patterson indicates that
the expression for measuring well-being corresponds to the mean of the truncated
distribution of WTP for the values 0 and T.12 Indeed, to provide a well-being mea-
surement, the truncated mean must have three characteristics: theoretical consis-
tency, statistical efficiency, and aggregation properties.
b
To estimate the WTP distribution function, i.e., (F(b)), the Logit models are evalu-
ated in both their standard and logarithmic forms. The explanatory variable was the
offered value bi, and the explained variable was dichotomous with a 0 or 1 value.
Empirical Methodology
pi
5 f ðxi bÞbj
xji
with f (.) being the residuals’ density function. Since, by definition, f (.). 0, the
sign of this derivative is therefore identical to that of bj.13
If the explanatory variables are qualitative, we first measure the probability P1
of the first modality that an individual agrees to pay. In second step, we calculate
the probability P2 for an individual selecting the other modality. In this case, the
marginal effect is measured by the difference P12 P2.14
In addition, elasticity is defined as the percentage change in the probability Pi
of event occurrence yi 5 1 following a change in the explanatory variable. This
elasticity is calculated according to the type of explanatory variable (continuous or
qualitative).
A Tobit model can satisfactorily handle responses equal to zero. In this paper,
all null responses were considered equivalent to zero consent to pay (WTP). In the
case where the null responses are numerous, the application of the OLS method is
no longer valid because we cannot have a linear form. Thus, the most widely used
estimation procedure is that of maximum likelihood. In addition, the marginal
effects, in a censored regression model, correspond to the forecast distortions on a
continuous variable generated by a unit variation of one of the explanatory varia-
bles. In Tobit models, there are several possible predictions depending on whether
we are interested in the censored variable yi or the latent variable yi*. In addition,
the elasticities are calculated depending on whether we are interested in the latent
variable or the censored variable. In this paper, we are interested in the calculation
of the marginal effects and elasticities on the latent variable forecast.
WTP Estimation (Case of Drinking Water in Tunisia): To enhance the
“drinking water” value, which is considered an environmental good, we are inter-
ested in discovering the maximum amount that consumers are willing to pay (WTP)
to ensure the sustainable management, usage, and provision of “drinking water” as
resource. The respondent’s consent to pay (WTP) is linked to several conditions
such as revenue, appreciation of the available water quality, and households’
186 THE JOURNAL OF ENERGY AND DEVELOPMENT
socio-economic characteristics. For our study, the Logit and Tobit regression estima-
tion procedures were performed on a sample size of 750 households, constructed
from a consumer survey that covers part of the Tunisian population.
Data processing from the statistical survey first provides us with a structural break-
out of the diversity of households that consume our studied environmental good—
drinking water. The existence of alternative water sources available to the household,
household size, water quality appreciation, education level, and income variables have
a significant effect on consumer behavior. In addition, the aim of econometric analysis
is to identify the significant variables, based upon appropriate econometric methods
that can estimate the evaluation functions, in order to have highly significant results,
which can be consistent with the economic hypotheses and would be useful in the
design and implementation of appropriate policies for the sustainable management of
a resource (drinking water) that is becoming increasingly scarce.
The major goal of econometric estimation is to determine the relevant variables
that influence consumer behavior. To properly value the environmental good
“drinking water,” the following six variables have been selected for econometric
regressions and analysis in our study: (1) SAE 5 alternative sources available for
the household’s water; (2) WTP 5 willingness to pay; (3) Size 5 household size;
(4) Nivist 5 interviewee education level; (5) Income 5 household income level;
and (6) Quality 5 appreciation of water quality.
On the basis of these estimates, the objective will then be to seek an appropriate
pricing policy, which would be a consensus price reconciling the consumers’ pur-
chasing power levels with the operating water service providers’ return rates. This
price must depend upon the level of service quality and the households’ socio-
economic characteristics. This specific price can cover the water service providers’
operating costs and limit waste to ensure the resource is sustainably managed and
guarantee environment stewardship of this critical good.
Logit Model Estimation: On the basis of 750 households in our survey, the
WTP estimation results from our logistics function to enhance the provision of
“drinking water” (the environmental good) in Tunisia, are presented in table 1.
In this estimated Logit model, only the coefficients’ signs are interpretable. The
estimation results show that all variable coefficients are significant. In addition,
the coefficient for the household size variable (Size) is negative, which means that
the consent to pay (WTP) probability for large families is lower. The income varia-
ble’s coefficient is positive, which implies that the WTP probability increases with
household income as one would anticipate.
The alternative source variable (SAE) has a negative coefficient, implying that
households with this type of water consumption pattern have a lower consent to
pay (WTP) probability. This result is expected given that the existence of this alter-
native source forms a substitute for drinking water. The educational level variable
(Nivist) of the respondent has a positive coefficient. This result is expected
because, presumably, agents who have a high educational level have a higher
DRINKING WATER: ENVIRONMENTAL GOOD VALUATION 187
Table 1
ESTIMATION RESULTS OF THE LOGIT MODELa
Variables Coefficients t-statistics
Constant 224.13 211.92
Size 20.41 25.32**
Income 5.34 12.31***
SAE 20.94 22.75**
Nivist 0.37 3.12**
Quality 20.75 22.93**
chi-square (5) 5 572
a
*** 5 significance at 1-percent level; ** 5 significance at 5-percent level; * 5 significance at
10-percent level; Size 5 household size; Income 5 household income level; SAE 5 alternative sources
available for the household’s water; Nivist 5 interviewee education level; and Quality 5 appreciation
of water quality.
paying probability than the other agents. This variable can explain the conscious-
ness level of the interviewed agent. In addition, the negative coefficient of the qual-
ity variable shows that consent to pay (WTP) probability of the respondent is
higher if they consider the water quality to be poor.
The calculated chi-square statistic is much higher than the critical value given
by the chi-square table at 5 degrees of freedom with a 1-percent error risk, which
can reject, with a 99-percent confidence level, the null hypothesis that all the coef-
ficients are zero.
Marginal Effects and Elasticities: Thus far, only the signs of the variables can
be interpreted. Therefore, to study the sensitivities of an explained variable probability
according to variations in the explanatory variables, we proceed with calculating the
marginal effects and elasticities. For the purposes of results interpretation, we calcu-
lated the marginal effects on the explanatory variables’ mean values; the results are
given in table 2.
Table 2
MARGINAL EFFECTS OF THE LOGIT MODELa
Variables dy/dx t-statistics
Size 20.06 25.37
Income 1.03 12.21
SAE 20.25 22.56
Nivist 0.07 3.08
Quality 20.14 22.94
a
Size 5 household size; Income 5 household income level; SAE 5 alternative sources available
for the household’s water; Nivist 5 interviewee education level; and Quality 5 appreciation of water
quality.
188 THE JOURNAL OF ENERGY AND DEVELOPMENT
Table 3
LOGIT MODEL ELASTICITIESa
Variables ey/ex t-statistics
Size 20.51 25.12
Income 9.12 9.71
SAE 20.07 22.71
Nivist 0.25 2.95
Quality 20.09 22.94
a
Size 5 household size; Income 5 household income level; SAE 5 alternative sources available
for the household’s water; Nivist 5 interviewee education level; and Quality 5 appreciation of water
quality.
DRINKING WATER: ENVIRONMENTAL GOOD VALUATION 189
Table 4
RESULTS OF THE TOBIT MODEL ESTIMATIONa
Variables Coefficients t-statistics
Constant 24.43 29.84
Size 20.06 23.15***
Income 1.15 13.85***
SAE 20.45 24.43***
Quality 20.61 29.05***
Nivist 0.09 2.54**
Khi – square (5) 5 418 R 5 0.19
2
a
*** 5 significance at 1-percent level; ** 5 significance at 5-percent level; * 5 significance at
10-percent level; Size 5 household size; Income 5 household income level; SAE 5 alternative sources
available for the household’s water; Nivist 5 interviewee education level; and Quality 5 appreciation
of water quality.
190 THE JOURNAL OF ENERGY AND DEVELOPMENT
alternative drinking water source, are able to substitute the offered drinking water
consumption and would not agree to participate in a program to enhance an environ-
mental good. The appreciation of water quality (Quality) variable coefficient is nega-
tive, which means that as much as the drinking water quality is deemed as poor by
the consumer, the more the consumer is willing to pay in order to have access to a
better quality good.
What also emerges from this table and the above-mentioned interpretations is
that consent to pay (WTP) is mainly determined by two major variables: the house-
hold income (Income) variable (with a coefficient equal to 1.5) and the apprecia-
tion of water quality (Quality) variable (with a negative coefficient equal to
20.61). This clearly shows that Tunisian households WTP, even in the case of an
environmental good such as drinking water, remains dependent on classic demand
variables, which are income and quality. The highest coefficient attributed to
income describes the rational logic underlying households where the decision to
pay more depends mainly on their income (thus, ability to pay) and, to a lesser
degree, on water quality. A priori, a quality variable is considered as a conditioned
variable and a satellite of the income variable. In other words, even if the quality is
poor, the household cannot hope to pay money unless a precondition of ensured
income is met. It turns out that the Income variable is the necessary condition,
while Quality is intended as the sufficiency condition.
The fit quality of this model remains good for this type of contingent valuation
models. Indeed, the R2 value is equal to 0.19 and allows us to observe a certain
explanatory power. The chi-square test concluded that the hypothesis of simulta-
neous nullity of all coefficients was rejected, such a result being justified by the
calculated value of this test which is much higher than the theoretical value with
5 degrees of freedom.
Marginal Effects and Elasticities: In the case of a simple Tobit model, the
prediction of the latent variable is obtained by the marginal effect measured by
the partial derivative of the conditional expectation according to any component of
the explanatory variables vector.
Indeed, the marginal effect of a unit variation in the kth explanatory variable on
the latent variable forecast (WTP*) is measured by the quantity:
CTP
E xi i
ðkÞ
5 bk
xi
The results show that variations in household income (Income) and education
level of the interviewee (Nivist) have a significant effect on the latent variable pre-
diction (WTP*). Thus, these different variables can be considered as determining
factors in the realization of a valuation program of the “drinking water” good. On
the other hand, the variables for the household size (Size) and alternative sources
DRINKING WATER: ENVIRONMENTAL GOOD VALUATION 191
available for the household’s water (SAE) have a negative effect on the latent vari-
able forecast. In addition, the coefficient of the appreciation of water quality (Qual-
ity) variable is negative, but it has a positive effect on the latent variable
prediction. Therefore, this quality variable can be considered as a fundamental fac-
tor for a certain program to ameliorate environmental quality.
Otherwise, a 1-percent variation of the kth explanatory variable for the ith individ-
ual changes the latent variable prediction (WTP*), for these same individuals, by
«CTPi =Xik percent:
Therefore, the average elasticity over the entire sample is given by the follow-
ing equation with the Tobit model elasticities provided in table 5:
1X N
1X N
Xik bk
« CTPi =Xik 5 «CTPi =Xik 5 :
N i51 N i 5 1 Xi b
Thus, a 1-percent variation of the household income (Income) increases the latent
variable forecast (CAP*) by 6 percent. Also, the household size (Size) variable
causes a 3-percent decrease in the latent variable forecast following a 10-percent
increase of this variable. The education level of the interviewee (Nivist) variable
exhibits a positive effect on the latent variable prediction. Indeed, any variation of
this variable by 1 percent modifies the latent variable forecast by 0.19 percent in the
desired direction. The variation in the availability of alternative water sources for
households (SAE) modifies the latent variable forecast in the opposite direction.
That means that the existence of available alternative drinking water sources for
households presents a constraint on the household’s consent to pay (WTP).
Finally, appreciation for the drinking water quality (Quality) variable constitutes
a good indicator that acts on the latent variable forecast. Indeed, a 1-percent varia-
tion of this variable reduces the latent variable forecast by 0.23 percent. House-
holds that consider the “drinking water” quality as important are more likely to
Table 5
TOBIT MODEL ELASTICITIESa
Variables ey/ex t-statistics
Size 20.32 23.15
Income 6.03 13.17
SAE 20.09 24.63
Quality 20.23 28.38
Nivist 0.19 2.64
a
Size 5 household size; Income 5 household income level; SAE 5 alternative sources available
for the household’s water; Nivist 5 interviewee education level; and Quality 5 appreciation of water
quality.
192 THE JOURNAL OF ENERGY AND DEVELOPMENT
Conclusion
NOTES
1
B. Desaigues and P. Point, Economie de patrimoine naturel. La valorisation des actifs du pat-
rimoine naturel (Paris: Economica, 1993).
Measures Biased?” American Journal of Agricultural Economics, no. 61 (1979), pp. 926–30.
12
J. Duffield and D. Patterson, “Inference and Optimal Design for a Welfare Measure in
Dichotomous Choice Contingent Valuation,” Land Economics, vol. 67, no. 2 (1991), pp. 225–39.
13
C. Hurlin, Econom
etrie des variables qualitatives (Orleans: Universite d’Orleans, 2003).
14
I. Cadoret and C. Benjamin, Econometrie appliquee (Paris: Dunode, 2004).