Introduction
Policy decisions in any sector of the economy produce changes in society’s welfare; consequently, the potential benefits generated have to be quantified to justify the costs associated with their implementation (^{Uribe, Mendieta, Jaime, & Carriazo, 2003}). One of the most promising methodologies in this field is the choice experiments (CEs) that are part of multi-attribute valuation methods (^{Holmes & Adamowicz, 2003}).
Interest in multi-attribute valuation methods has increased, in part, as a response to the questions posed to the contingent valuation method (CVM). Multi-attribute valuation techniques are divided into two categories that differ according to the measurement scale used. In the first category are the preference-based approaches, where individuals rank alternative scenarios on a cardinal scale. The second category consists of choice-based approaches, where individuals are asked to choose (using an ordinal scale) the preferred scenario (^{Ben-Akiva & Lerman, 1985}; ^{Holmes & Adamowicz, 2003}).
Regarding the use of CEs, some studies have been conducted in the field of environmental economics. One of them is that of ^{Tudela (2010)}, who suggests technical criteria to prioritize management policies in protected natural areas of Mexico. Along the same lines, ^{Villota (2009)} determined the economic value of the Lenga wetland in Concepción, Chile. Another noteworthy study was conducted by ^{Birol, Karousakis, and Koundouri (2006)}, who estimated the monetary values of several ecological, social, and economic functions that the Cheimaditida wetland provides to the Greek population. Likewise, ^{Hanley, Wright, and Álvarez-Farizo (2006)} estimated the economic value of ecological improvements in the Wear River in the city of Durham, England, and in the Clyde River in central Scotland. ^{Carlsson, Frykblom, and Liljenstolpe (2003)} identified the attributes that increase and decrease the welfare of citizens considering their preferences regarding a water wetland in Staffanstorp, in southwestern Sweden.
On the other hand, several studies involving the use of CEs stand out in the area of economic valuation of improvements in sanitation services. ^{Lucich and Gonzales (2015)} disaggregated the value of the quality of the potable water supply service in the city of Tarapoto, Peru. As a result, they found that the attributes of the potable water supply service with the greatest economic value for the users are: quality of the drinking water with respect to its levels of turbidity, hours of water supply and availability of the water resource through the conservation of its source. ^{Justes, Barberán, and Farizo (2014)} valued water uses in the city of Zaragoza, Spain, concluding that the number of household members, employment status, age, income and water consumption level are important variables that must be taken into account in setting rates. ^{Tarfasa and Brouwer (2013)} estimated the willingness to pay for improvements in water supply services in an urban area of Ethiopia; the CE design allowed them to estimate the value of water availability for consumption and future supply. ^{Birol and Das (2010)} estimated the willingness to pay for improvements in the capacity and technology of a wastewater treatment plant in the municipality of Chandernagore, located on the banks of the Ganges River in India. As for the valuation of water supply preferences, ^{Blamey, Gordon, and Chapman (1999)} estimated the environmental value in the context of a consumer evaluating future water supply options in the Australian capital.
Based on a literature review of the subject, it is clearly possible to place a value on the economic benefits generated by public policies aimed at improving the population’s welfare. Therefore, the objective of this research was to estimate the potential economic benefits of an improvement in the provision of basic sanitation services (water, sewage and treatment) in the city of Puno, Republic of Peru. Specifically, two results on which we worked on in this research are of interest: the first is to prioritize the intervention alternative that generates the greatest welfare for users, and the second is to identify the socioeconomic variables that condition the choice of options. The results are expected to provide useful information for social actors and become a tool for decision-making in the allocation of resources and management of basic sanitation services in the city of Puno.
Materials and methods
The CE has a theoretical basis in Lancaster’s consumer choice model and an econometric basis in the random utility models. "Lancaster breaks with the traditional theory of consumer behavior by assuming that it demands goods by virtue of their characteristics or properties and that it is these characteristics, and not the goods themselves, that generate utility. On the other hand, the theory of random utility assumes a perfectly rational individual that always chooses the alternative that implies a greater expected utility" (^{Tudela, 2010}). Consequently, if different attributes are considered for the different choice alternatives, individuals will express their preferences for a selection of possible combinations. For example, if individuals express their preferences by making choices between alternatives j = 1, 2, …J of choice set C, then the utility for the choice of alternative j for each individual will be given by:
In each alternative of the choice set, the indirect utility function depends on the levels taken by the attributes Z
_{
ij
} , the socioeconomic characteristics of the users S
_{
i
} and the income M
_{
i
} . The user i will prefer alternative h over any of the options j in the choice set C, if the utility reported for alternative h is superior to that offered by each of the options; that is, if
The utility is composed of a deterministic component v_{ ih } and an unobserved component of random error e _{ ih } . The observed component of utility (indirect utility function) can be expressed as a linear function of the explanatory variables:
where,
α = |
specific constant for each alternative |
β = |
vector of utility coefficients associated with the Z vector of explanatory variables |
γ = |
coefficient associated with the price of alternative j, COST _{ j } |
δ = |
vector of coefficients associated with socioeconomic variables (^{Blamey et al., 1999}) |
Therefore, the probability that individual i prefers alternative h ( C equals the probability that the sum of the observed and random components of that option will be greater than the same sum for the rest of the presented alternatives, that is:
Welfare measures are obtained by estimating the parameters that define the indirect utility function, for which it is necessary to define a probability function. ^{McFadden (1974)} observed that if the error terms of the above equation are independent and identically distributed with a Gumbel distribution or extreme value type I, the probability of choosing alternative h has the following representation:
This expression, in which the attributes to be valued and the characteristics of the individuals are present, is known as multinomial logit or conditional logit; where, ( is a scale parameter inversely proportional to the standard deviation of the error term of the distribution and is typically normalized as one (^{Ben-Akiva & Lerman, 1985}). The main problem with the multinomial logit model is the implicit assumption of independence of the irrelevant alternatives; that is, the choice probability ratio of two alternatives is independent of any other real or potential alternative. Such an assumption gives rise to biased results when it is not met (^{Louviere, Hensher, & Swait, 2000}). The parameters of the indirect utility function (α, β y δ) are estimated using the maximum likelihood method (^{Greene, 2003}).
On the other hand, the mixed logit model is very flexible and can approximate any random utility model (^{McFadden & Train, 2000}). This model circumvents the limitations of the multinomial logit model, allowing random variation of preferences, unrestricted substitution patterns and correlation between unobserved factors over time. Mixed logit probabilities are the integrals of the multinomial logit probabilities over a probability density of parameters. In a mixed logit model, the choice probabilities are expressed in the form (^{Train, 2009}):
where,
L _{ ih } (β) = |
logit probability evaluated at parameters β |
f(β) = |
probability density function |
v _{ ih } (β) = |
observed part of utility, which depends on the parameters β. |
If utility is linear in β, then . In this case, the probability of the mixed logit model takes the following form:
The probability of the mixed logit model is a weighted average of the multinomial logit formula evaluated in different values of β, with the weights given by the density f(β). For the econometric estimation of the parameters in the mixed logit models, the simulated maximum likelihood method is used (^{Train, 2009}).
The parameters of the indirect utility function for both types of models were estimated using the procedure described and, subsequently, the monetary welfare measures were estimated. According to ^{Alpizar, Carlsson, and Martinsson (2001)}, for a linear utility function, the marginal rate of substitution between two attributes is simply the ratio of their coefficients, and the marginal willingness to pay (MWTP) for a change in attribute Z _{ a } is given by:
Choice card design
Choice cards in survey format were made based on an experimental design derived from the diagnosis of the problem in the provision of basic sanitation services in the city of Puno, taking into account the recommendations of ^{Hensher, Rose, and Greene (2005)}. Below, the aspects considered most important in the preparation of the choice cards are presented.
Identification of attributes and levels
Based on management documents obtained from the Empresa Municipal de Saneamiento Básico (Municipal Basic Sanitation Company, EMSAPUNO), responsible for the administration of drinking water and sewage services in the city of Puno, three aspects that should be prioritized in the design of public investment policies or projects were identified: (1) improvement in the continuity of water provision, (2) improvement in the sewerage network and (3) wastewater treatment. ^{Louviere et al. (2000)} recommend introducing one more attribute, which manages to restrict choices by demanding an economic compensation for the improvement actions. The levels of this monetary attribute were determined from an open-question pilot survey, which allowed obtaining the minimum and maximum value of the possible rate increase. These values are: 4 PEN, 6 PEN and 8 PEN. Table 1 summarizes the attributes and levels used in the choice experiment, in operational terms.
Attributes | Variables | Levels |
---|---|---|
Water supply | Increase in continuity to 24 hours (ACA24) | Excellent (24 hours) |
Increase in continuity to 12 hours per day (ACA12) | Good (12 hours) | |
Deficient (no change) | ||
Sewage | Upgrading of 100 % of sewer system (R100) | Excellent (100 %) |
Good (50 %) | ||
Upgrading of 50 % of sewer system (R50) | Deficient (no change) | |
Treatment | Optimal wastewater treatment (construction of a wastewater treatment plant) (TOAR) | Excellent (new plant) |
Partial wastewater treatment (periodic cleaning of sludge with machinery - dredges) (TPAR) | Good (periodic cleaning) | |
Deficient (no change) | ||
Rate | Additional COST | 4 PEN |
6 PEN | ||
8 PEN |
Source: Author made
Generation of experimental design
According to Table 1 there are 81 combinations of different scenarios (3 x 3 x 3 x 3); carrying out the survey with this number of cards would not be practical, so fractional factorial analysis, which minimizes the correlation between attributes (^{Bennett & Adamowicz, 2001}), was used. The combination of attributes, by means of orthogonal statistical design, was carried out with SPSS version 22 software (^{IBM SPSS Statistics, 2014}). According to Table 2, nine cards or alternatives were generated; these optimal scenarios are orthogonal (there is no correlation between levels and attributes) and balanced (each level appears in the attribute the same number of times).
Card number | Water | Sewage | Treatment | Rate (PEN) |
---|---|---|---|---|
1 | Excellent | Good | Excellent | 4 |
2 | Excellent | Excellent | Deficient | 6 |
3 | Good | Deficient | Excellent | 6 |
4 | Good | Excellent | Good | 4 |
5 | Good | Good | Deficient | 8 |
6 | Deficient | Excellent | Excellent | 8 |
7 | Deficient | Deficient | Deficient | 4 |
8 | Excellent | Deficient | Good | 8 |
9 | Deficient | Good | Good | 6 |
Source: Author made based on SPSS software results.
The orthogonal design illustrated in Table 2 contains a combination (card 7) identical to the status quo (which is characterized by having deficient levels in all attributes). In the face of "no improvement" scenarios and with an economic contribution, the choice of card 7 is meaningless, so it was discarded and at the end there were eight optimal combinations.
Coding of attributes to be valued
Effects codes and dummy codes were used in determining the effects of the attributes. The effects codes case is coded taking into account that each attribute has three levels of improvement (deficient, good and excellent); the variable that corresponds to "deficient" is the base level to compare, so finally, in the econometric analysis, two variables are worked with for each attribute (Table 3).
Quality level | Change attributes | |||||
---|---|---|---|---|---|---|
Water | Sewage | Treatment | ||||
IWC24 | IWC12 | U100 | U50 | OWT | PWT | |
Excellent | 1 | 0 | 1 | 0 | 1 | 0 |
Good | 0 | 1 | 0 | 1 | 0 | 1 |
Deficient | -1 | -1 | -1 | -1 | -1 | -1 |
Source: Author made. IWC24 = Increase in continuity to 24 hours, IWC12 = Increase in continuity to 12 hours per day, U100 = Upgrading of 100 % of the sewer system, U50 = Upgrading of 50 % of the sewer system, OWT = Optimal wastewater treatment, PWT = Partial wastewater treatment.
In the second case, dummy codes are used to code the variables associated with the attributes, for which another survey format is not necessary. The dummy variables (0, 1) replace the effects codes (1, 0, -1).
Implementation of choice card
Operationally, the eight choice sets considered optimal in the orthogonal design were divided into blocks of four different versions on which the users proceeded with their choice: card 1 (1 and 2), card 2 (3 and 6), card 3 (4 and 8) and card 4 (5 and 9). Table 4 illustrates a type of card shown to respondents.
Card 1 (1&2) | Alternative A | Alternative B | Alternative C |
---|---|---|---|
Continuity in the water supply | Increases water continuity in the home to 24 hours | Increases water continuity in the home to 24 hours | No change |
Improvement of sewer system | Upgrading of 50 % of the sewer system | Upgrading of 100 % of the sewer system | No change |
Wastewater treatment | Optimal treatment (construction of a new plant) | No change | No change |
Additional rate increase (PEN·month^{-1}) | 4 | 6 | 0 |
Please choose your preferred option: | |||
Alternative A ( ) | Alternative B ( ) | Alternative C ( ) |
In total, 392 surveys were applied to heads of households with water and sewage connections. Due to the technical characteristics of the survey format (presentation of cards to each head of household for the choice of the preferred alternative), it was necessary to train the applicators. All surveys were conducted in January 2017 in the city of Puno.
The experiment had four replications, thus obtaining a panel data-type data structure. For each respondent, 4 x 3 = 12 observations were obtained, distributed in 392 surveys, generating a database with 4 x 3 x 392 = 4 704 observations. The surveyed individuals made 1 568 choices (392 x 4).
Results and discussion
Table 5 presents a synthesis of the main results of the estimated econometric models. The mixed logit model with dummy codes was selected based on the econometric criteria; in general, in this model, the signs of the coefficients that accompany the explanatory variables are the expected ones. The highly significant variables (P ≤ 0.01) were IWC24, U100, OWT, PWT and COST, and the variables significant at 5 % (P ≤ 0.05) were IWC50 and U50. In addition, there was a good fit (16.62 %) in terms of the adjusted pseudo R^{2} (does not come too close to the unit) and the likelihood ratio statistic (Chi-square) rejects the hypothesis that all the model’s slopes are zero (P < 0.01). The parameters of the multinomial logit and mixed logit models were estimated using NLOGIT version 4 software (^{Econometric Software, Inc., 2007}).
Variables | Multinomial logit | Mixed logit | ||
---|---|---|---|---|
Effect codes | Dummy codes | Effect codes | Dummy codes | |
IWC24 | 0.437 | 0.927 | 0.754 | 0.965 |
(5.116)^{***} | (5.814)^{***} | (3.888)^{***} | (5.531)^{***} | |
IWC12 | 0.108 | 0.615 | 0.060 | 0.582 |
(1.406) | (2.883)^{***} | (0.620) | (2.454)^{**} | |
U100 | 0.137 | 0.394 | 0.213 | 0.422 |
(2.333)^{***} | (3.931)^{***} | (2.337)^{**} | (3.173)^{***} | |
U50 | 0.198 | 0.475 | 0.259 | 0.449 |
(2.761)^{***} | (3.384)^{***} | (2.310)^{**} | (2.299)^{**} | |
OWT | 0.501 | 1.070 | 0.907 | 1.155 |
(7.207) ^{***} | (8.389) ^{***} | (3.653) ^{***} | (6.779)^{***} | |
PWT | 0.154 | 0.655 | 0.004 | 0.724 |
(2.162)^{**} | (5.163)^{***} | (0.031) | (3.665)^{***} | |
COST | -0.373 | -0.417 | -0.467 | -0.432 |
(-17.189)^{***} | (-13.298)^{***} | (-7.627)^{***} | (-11.017)^{***} | |
1_EDUC | 0.090 | 0.097 | ||
(2.495)^{**} | (2.418)^{**} | |||
1_INC | 0.0002 | 0.0002 | 0.0003 | 0.0002 |
(4.758)^{***} | (3.028)^{***} | (4.148)^{***} | (2.907)^{***} | |
2_EDUC | 0.111 | 0.114 | ||
(2.932)^{***} | (2.790)^{***} | |||
2_INC | 0.0003 | 0.0002 | 0.0004 | 0.0002 |
(5.997)^{***} | (3.687)^{***} | (4.999)^{***} | (3.575)^{***} | |
Log-likelihood | -1 434.758 | -1 429.698 | -1 429.815 | -1 428.107 |
Chi-square | 414.360 | 424.480 | 585.618 | 589.034 |
Pseudo R^{2} | 0.1261 | 0.1292 | 0.1699 | 0.1709 |
Adjusted Pseudo R^{2} | 0.1236 | 0.1262 | 0.1657 | 0.1661 |
Number of observations | 4 704 | 4 704 | 4 704 | 4 704 |
Z statistic in parentheses: *** P ≤ 0.01 and ** P ≤ 0.05. Variables: IWC24 = increase in continuity to 24 hour, IWC12 = Increase in continuity to 12 hours per day, U100 = Upgrading of 100 % of sewer system, U50 = Upgrading of 50 % of sewer system, OWT = Optimal wastewater treatment, PWT = Partial wastewater treatment, COST = Rate increase, EDU = Educational level, INC = Income. Source: Author made based on NLOGIT software results.
The parameters of the attributes of the improvements in basic sanitation services have the expected signs; that is, the increase in continuity to 24 hours (IWC24), the increase in continuity to 12 hours per day (IWC12), the upgrading of 100 % of the sewage system (U100), the upgrading of 50 % of the sewage system (U50), optimal wastewater treatment (OWT) and partial wastewater treatment (PWT) are improvements that positively affect the user’s utility.
The coefficient of the cost (COST) variable, which reflects the rate increase for water and sewage services, is negative as expected; the higher the rate, the lower the disposable income and, therefore, the lower its indirect utility will be.
On the other hand, the socioeconomic characteristics of the users reflect the interaction effect with the specific constants for each alternative. Both the educational (EDU) level and the income (INC) level were highly significant; that is, the higher the educational level and the higher the income levels, the greater the indirect utility for the improvements in basic sanitation services.
Analysis of marginal willingness to pay
The CEs allow the estimation of changes in welfare due to a variation in any of the levels of the attributes. The MWTP or the implicit price of a non-monetary attribute of the good is the willingness to pay for a unit change in this attribute, keeping the rest constant. The results of the econometric estimations of the mixed logit model with interaction indicate that the estimated indirect utility function has the following form in its random and non-random part:
Table 6 shows the MWTP of the non-monetary attributes used in the choice experiment. When adding the MWTP, the treatment and water attributes are greater than the sewage attribute. The total WTP for the improvements in the four attributes is 9.95 PEN.
Basic sanitation services | MWTP by levels of improvement (PEN·month^{-1}·dwelling^{-1}) | Total WTP (PEN·month^{-1}·dwelling^{-1}) | WTP (%) | |
---|---|---|---|---|
Good | Excellent | |||
Water | 1.35 | 2.23 | 3.58 | 36 |
Sewage | 1.04 | 0.98 | 2.02 | 20 |
Treatment | 1.68 | 2.68 | 4.35 | 44 |
Total | 4.07 | 5.89 | 9.95 | 100 |
Source: Made by authors based on the econometric mixed logit-dummy codes model.
In another study on the same subject, ^{Tudela-Mamani (2017)} applies the double-bounded CVM in estimating the WTP for improving the wastewater treatment system in the city of Puno, Peru, finding an average WTP of approximately 4.38 PEN·month^{-1}·dwelling^{-1}. This figure is similar to the MWTP obtained in this work for the treatment attribute (4.35 PEN·month^{-1}·dwelling^{-1}).
On the other hand, ^{Lucich and Gonzales (2015)}, when applying the CE in the city of Tarapoto, Peru, conclude that water distribution service users would be willing to pay the sum of 7.00 PEN·month^{-1}, as an additional amount on their bill, for improving the quality of the potable water supply service and for conserving the current water source through reforestation. When breaking down their results, the attribute "water quality: turbidity" and the value of "increasing the water supply hours" add up to a total of 3.91 PEN. This result is close to that reported in the present research for the water attribute (3.58 PEN).
Conclusions
The mixed logit-dummy codes model showed greater theoretical consistency due to the greater individual and joint significance of the parameters. Through this model, an aggregate marginal willingness to pay of 9.95 PEN·month^{-1}·dwelling^{-1} (3.32 USD), which, considering the total number of favored households, represents a measure of economic benefit in the cost/benefit evaluation of the improvements proposed. This study shows that the wastewater treatment attribute is more valued than the water supply and sewage attributes; therefore, any policy aimed at improving basic sanitation services in the city of Puno should be focused on solving the issue of wastewater treatment. The choice of alternatives for improving basic sanitation services is strongly conditioned by the educational level and the monthly income of the users.